Datasets:

bzantium

/

LongPerplexity

like 0

statistical mechanics is a bridge between mechanics and thermodynamics . indeed , it may be assumed as the microscopic mechanical basis for the macroscopic thermodynamical properties of a physical system . the key concept which links these two arenas is the notion of the phase space , that is the space of all distinct possible microstates . the dynamical evolution is nothing but a smooth transition from one microstate to another and statistical mechanics is about how to count the number of these microstates . in a more technical language , the phase space of a dynamical system is a symplectic manifold which naturally is equipped with a symplectic structure @xcite . the symplectic structure covariantly determines the poisson brackets and liouville volume and the kinematics is then defined in a coordinate - independent manner on the phase space . the dynamics will be also specified upon taking the hamiltonian function as the generator of time evolution in this setup . therefore , the hamiltonian determines how the system evolves through the microstates and the liouville volume specifies how one can count and measure them . consider , for instance , a mechanical system consisting of @xmath0 particles which are subjected to some forces . each particle moves in euclidean three - dimensional space @xmath1 and the corresponding configuration space is then @xmath2 . therefore , the phase space will be the cotangent bundle of this space , that is , @xmath3 , which naturally admits a symplectic structure . in terms of the positions and momenta of the particles , the associated symplectic structure takes the standard canonical form . such a canonical symplectic structure , which is also dynamically invariant , may be used to determine the fundamental canonical poisson brackets and the kinematics for the system under consideration . subsequently , based on the canonical symplectic structure one leads to the canonical measure which can be viewed as a mathematical expression for the principle of equal probabilities in standard statistical mechanics @xcite . although it might be thought that the symplectic formulation is nothing but a covariant reformulation of the standard statistical mechanics , its advantages may become more clear when one is dealing with the hamiltonian systems with nontrivial structures ( for instance when the standard @xmath4 topology of the phase space is replaced with a more complicated topology or the standard canonical poisson algebra is replaced with a deformed noncanonical one ) . such systems , are investigated in the context of doubly special relativity theories , as a flat limit of quantum gravity , where the four - momentum space has a curved ( de sitter or anti - de sitter ) geometry with nontrivial topology @xcite . the space - time structure then turns out to be noncommutative and the associated poisson algebra between the positions and momenta become noncanonical @xcite . the space - time with noncommutative coordinates was first proposed by snyder in ref . @xcite in order to regularize quantum field theory in the ultraviolet regime @xcite . in more recent times , it was also used by the stability theory of the lie algebras to be a minimal candidate for the relativistic quantum mechanics @xcite and by string theory at a fundamental level @xcite . the duality between curved momentum space and noncommutative structure was first pointed out by majid in ref . the direct consequence of space - time with noncommutative coordinates is the modification to the dispersion relation and the associated density of states @xcite . in symplectic formulation , the density of states is properly determined by the liouville measure which will be constructed from the symplectic structure @xcite . in noncommutative space - time , however , the corresponding symplectic structure seems to be noncanonical which induces a nonuniform probability distribution over the set of microstates through a deformed liouville volume element . this dynamically invariant liouville volume enters in the definition of the gibbs entropy and partition function from which all the thermodynamical properties of a system can be extracted . therefore , the topology of the phase space will play an important roll since all integrals are evaluated over the entire phase space . for instance , let us consider a system which has an ultraviolet cutoff due to the existence of a maximal momentum . it is easy to see that the topology of the spatial sector of the momentum space may be compact . moreover , systems with a totally compact phase space , which can be considered as the classical limit of quantum systems with finite dimensional hilbert space ( e.g , angular momentum for systems with fixed total angular momentum ) were recently studied in the context of loop quantum gravity @xcite . for such a compact phase space , the liouville volume is finite and consequently a finite number of microstates naturally arises @xcite . this shows the correspondence between the topology of the phase space and the number of microstates which can be realized when the symplectic geometry is implemented . the black hole physics is a good example in which the system obeys the thermodynamical laws and its statistical origin is not thoroughly formulated yet @xcite in the framework of hamiltonian formalism ( see also ref . @xcite ) . in this regard , formulating the statistical mechanics in its most fundamental form is important , at least , for two reasons : ( i ) finding a more precise and fundamental interpretation for the basis of the statistical mechanics , and ( ii ) studying the statistical mechanics for the hamiltonian systems with nontrivial structure , such as those we have mentioned above . motivated by the stated issues , in this paper we are going to formulate statistical mechanics for a general hamiltonian system in a covariant ( coordinate - independent ) manner . the paper is organized as follows . in section slowromancap2@ , we deal with the kinematics and dynamics of a many - particle system with general hamiltonian structure in the context of symplectic geometry . the statistical mechanics for such systems is formulated in section slowromancap3@. in section slowromancap4@ , by means of the constructed setup we study the statistical mechanics in the snyder noncommutative space - time with curved energy - momentum space as an explicit example of a physical system with nontrivial ( noncommutative ) hamiltonian structure . as a case study , the thermodynamical properties of a harmonic oscillator system in snyder space is also presented at the end of this section . section slowromancap5@ is devoted to the summary and conclusions . in this section , in order to study the statistical mechanics in the context of symplectic geometry , we consider the kinematics and dynamics of an @xmath0-particle system . let us consider a system of @xmath0 particles similar to conventional systems in statistical mechanics . the corresponding @xmath5-dimensional kinematical phase space may be obtained by coupling @xmath0 single - particle phase spaces @xmath6 with @xmath7 as @xmath8 the phase space @xmath9 naturally admits a symplectic structure @xmath10 which is a nondegenerate closed 2-form @xcite . the classical state of this @xmath0-particle system is determined by a point in @xmath9 ( phase point ) . the evolution of phase points is then determined by the hamiltonian vector field @xmath11 , whose integral curves are trajectories of the phase points ( phase trajectories ) . from the fact that @xmath10 is nondegenerate , one can assign a vector field to a function @xmath12 as @xmath13 and the poisson bracket between two observables ( real - valued functions on the phase space ) can be defined as @xmath14 these functions with the above structure constitute a lie algebra under poisson brackets on @xmath9 . we also have the so - called liouville volume on @xmath9 with its standard definition as @xmath15 in terms of positions and momenta @xmath16 of the particles , the liouville volume takes the local form @xmath17 where @xmath18 and @xmath19 are the @xmath20th component of the positions and momenta of the @xmath21th particle , respectively , and @xmath22 is the matrix representation of the symplectic structure @xmath10 with components @xmath23 . now , one may be able to decompose the symplectic structure as @xmath24 where @xmath25 is the symplectic structure on @xmath26 . although this assumption has no mathematical basis , from a physical point of view it is acceptable since the particles are assumed to be kinematically separated . in other words , the lie algebras of the functions defined by @xmath27 over @xmath28 are separately closed . therefore , the liouville volume over @xmath9 takes the form @xmath29 where the wedge products take place @xmath30 times . the volume element then works out to be @xmath31 where @xmath32 is the volume element of the corresponding @xmath33 , and we have also used the fact that @xmath34 , for @xmath35 . clearly , @xmath36 , that is , a function on @xmath9 , factorizes into products of @xmath37 . this can not be realized in the most general situation , where the symplectic structure can not be written as ( [ decomposition ] ) . as a special case , consider a system consisting of @xmath0 particles subject to some forces moving in euclidean three - dimensional space with standard @xmath38 topology . therefore , the phase space will be @xmath39 which is indeed the cotangent bundle of the configuration space , that is , @xmath40 . geometrically , the cotangent bundle is a symplectic manifold endowed with a canonical symplectic structure @xmath41 in which @xmath42 and @xmath43 are interpreted as the positions and momenta of particles , respectively . the summation is over both @xmath44 and @xmath45 . this is the canonical representation of symplectic structure and coordinates @xmath46 are known as the canonical coordinates . substituting the canonical structure ( [ canonical structure ] ) into the definition ( [ pbd1 ] ) , the poisson bracket will be @xmath47 which leads to the standard canonical poisson algebra @xmath48 also , with the help of ( [ canonical structure ] ) we get from ( [ vol - d ] ) the liouville volume as @xmath49 which is nothing but the standard volume element on @xmath50 . thus , the kinematics of the standard statistical mechanics can be recovered as a special case of this setup . taking into account the fact that @xmath51 , the measure associated to the standard volume element ( [ canonical measure ] ) assigns a uniform probability distribution to the set of microstates . this is justifiable in the light of laplace s principle of indifference , which states that in the absence of any further information , all outcomes are equally likely @xcite . this is the fundamental basis of the statistical mechanics . in the case where the space - time has a noncommutative structure , apart from the details of the different models , there is always a deformation parameter which can be linked to a minimal length associated with the system under consideration . the direct consequence of this setup is the modification of the dispersion relation and the density of states ( see ref . @xcite for the special case of the stable noncommutative algebra for the relativistic statistical mechanics ) . the density of states is determined by the symplectic structure and the associated liouville volume . since the symplectic structure takes a noncanonical form in terms of the physical positions and momenta @xmath52 in noncommutative spaces @xcite , we have @xmath53 . this result immediately shows that laplace s indifference principle is no longer applicable for a noncommutative space - time . this is because of the fact that there is extra information in these setups ( minimal length or maximal momentum ) and the liouville measure then assigns a nonuniform probability distribution over the set of microstates by means of the relation ( [ vol - d2 ] ) . this simple consideration of noncommutative space - time as an example of a kinematically deformed system shows the advantage of the coordinate - independent symplectic formulation of the statistical mechanics ( see refs . @xcite for explicit examples ) . we will explicitly consider such an example in section iv . it also should be noted that , according to the darboux theorem , there is always a local chart on @xmath9 in which any symplectic structure takes the canonical form with @xmath54 . in the case of kinematically deformed systems such as the noncommutative space - time , however , the new canonical coordinates can not be interpreted as the positions and momenta of the particles . moreover , in the case of standard statistical mechanics , one can also find a local chart in which the standard canonical structure @xmath55 takes a noncanonical form with @xmath56 . but again , these new noncanonical coordinates can not be interpreted as positions and momenta of particles . in all of the above cases , as we will see in the next section , the partition function and consequently resultant thermodynamical properties are independent of the chart in which the system is considered . as we have mentioned above , the time evolution of the system is determined by the hamiltonian vector field @xmath11 . it can be obtained by solving the dynamical equation @xmath57 where @xmath58 is the hamiltonian function of the system that is a real - valued function over @xmath9 and @xmath59 is an interior product of @xmath10 and @xmath11 . additionally , @xmath11 can be expanded as @xmath60 where @xmath61 is the projection of @xmath62 on @xmath26 . note that , in general , the domain of components of @xmath61 is the whole phase space @xmath9 and consequently the lie brackets between @xmath63 will not vanish : @xmath64\neq0\,.\ ] ] however , as a special case , one can consider a system consisting of noninteracting particles . therefore , the hamiltonian function can be written as the sum of individual hamiltonians , that is , @xmath65 so , from decomposition ( [ decomposition ] ) for the symplectic structure , the dynamical equation ( [ can - dy - eq ] ) becomes @xmath66 using the fact that @xmath67 , for @xmath68 , one is led to a set of @xmath0 independent differential equations @xmath69 now , it is clear that the domain of the components of @xmath70 will be @xmath26 and the lie brackets between @xmath71 and @xmath72 vanish @xmath73=0\,.\ ] ] this can be seen as a criterion for the _ kinematical and dynamical separability _ of the particles . in the appendix , we will see that this benchmark shows itself as a criterion for the statistical independence of the macroscopic subsystems which in turn is a definition of equilibrium . the other important property of the hamiltonian systems is the existence of an invariant measure that enables one to construct the equilibrium statistical mechanics through the well - known liouville theorem as @xmath74 where @xmath75 is the total time derivative and @xmath76 denotes the lie derivative with respect to @xmath62 . the conservation of @xmath77 can be traced back to the fact that @xmath78 and @xmath79 , where we have used @xmath80 and the closure of the symplectic structure @xmath81 . it is obvious that @xmath77 is not explicitly time dependent . note also that the liouville theorem arises from the particular dynamics of a hamiltonian system . consider the standard statistical mechanics as a special case , where the kinematics is given by canonical symplectic structure ( [ canonical structure ] ) . the dynamical equation ( [ can - dy - eq ] ) then leads to the familiar form of the hamilton s equations in terms of the physical positions and momenta as @xmath82 note that when the symplectic structure has a noncanonical form , the form of hamilton s equations would deviate from the relation ( [ h equation ] ) . the triple @xmath83 constitutes a general hamiltonian system and we explore the statistical physics of this general system in the next section by means of an invariant volume element @xmath77 which is covariantly defined in the relation ( [ vol - d ] ) . statistical mechanics links macroscopic properties of a system to microscopic laws that are quantum mechanical at the fundamental level . the basic concept is the von neumann entropy that is defined as @xmath84 where @xmath85 is the density operator @xcite . using the principle of maximum entropy and considering specific relevant statistical constraints , we can find the density operator which contains all the statistical information about the system at equilibrium . for example , for a system in contact with a thermal bath at temperature @xmath86 and therefore with a fixed mean energy , the resultant normalized density operator works out to be @xmath87\,,\end{aligned}\ ] ] where @xmath88 is the hamiltonian operator of the system and @xmath89 is called the canonical partition function which is defined as @xmath90=\sum_{i}\exp\big[-\varepsilon_i / t \big]\,,\end{aligned}\ ] ] where @xmath91 are the energy eigenvalues of @xmath88 and the summation is taken over all the accessible microstates for the system . at the fundamental level , the microscopic laws are quantum mechanical and the microstates are determined by quantum mechanics . on the other hand , in the classical limit , the microstate of a system represents itself as a point on the corresponding phase space and due to the infinite resolution of phase space points , the summation over microstates is really not well defined . consequently , one can not construct a full classical statistical mechanics in essence . however , quantum mechanics provides the semiclassical approximation in the spirit of the uncertainty principle . indeed , a phase space with finite resolution and well - defined number of classical microstates can be achieved through the approximation @xcite @xmath92 where @xmath93 is due to the fact that the particles may be indistinguishable and @xmath77 is the dynamically invariant liouville volume element of the @xmath5-dimensional symplectic manifold @xmath9 which is defined in the relation ( [ vol - d ] ) . it is important to note that the semiclassical approximation ( [ approx ] ) coincides with the full quantum consideration at high temperature regime . in this limit , the von neumann entropy ( [ von entropy ] ) is replaced with the gibbs entropy @xmath94 where @xmath95 is now interpreted as a probability density over the space of microstates @xmath9 . the maximum entropy principle for a system in heat bath at temperature @xmath86 leads to the gibbs state @xmath96\,,\end{aligned}\ ] ] where @xmath58 is the hamiltonian function of the system and @xmath89 is the semiclassical canonical partition function @xmath97\omega^{3n}\,.\end{aligned}\ ] ] the result of the integral over the phase space for the total partition function ( [ partition f ] ) is independent of a chart in which the system is considered when the dynamical equation ( [ can - dy - eq ] ) is satisfied in a chart - independent manner . while the local form of the symplectic structure determines the probability distribution over the set of microstates , from the statistical point of view , we expect that the resultant thermodynamical properties extracted from a partition function are independent of a chart in which the physical system is considered ( see , for instance , ref . @xcite in which the partition function in two different charts is calculated ) . by considering the decompositions ( [ decomposition ] ) and ( [ decomposition h ] ) for the symplectic structure @xmath10 and hamiltonian function @xmath58 , respectively , the probability density will also be decomposed as @xmath98 where @xmath99 is the gibbs state defined on @xmath6 . hence the integral over @xmath9 can be factorized into the products of the integrals over the phase spaces of the particles as @xmath100\,\omega^3_{\alpha}\bigg)\,.\end{aligned}\ ] ] now , if the particles are influenced by the same kinematics and dynamics , the partition function can be written in the well - known form @xmath101 where we have defined the single - particle state partition function @xmath102\,\omega_1 ^ 3\,.\end{aligned}\ ] ] it is usually claimed that the relation ( [ partition fghi ] ) is applicable when one ignores the quantum correlations between particles . this means that decompositions such as ( [ decomposition ] ) and ( [ decomposition h ] ) may be considered as conditions for the absence of quantum correlations . moreover , the general formalism which is formulated in this section , can also support a hamiltonian system with very nontrivial structure , _ e.g. _ a kinematically classical entangled system in which the decomposition ( [ decomposition ] ) is no longer applicable . existence of a minimal length is suggested by any quantum gravity candidate such as string theory and loop quantum gravity @xcite . it is then widely believed that a nongravitational theory which supports the existence of a minimal length would arise at the flat limit of quantum gravity . such a theory , would be reduced to the standard relativistic quantum mechanics at low energy regime where the effects of a minimal length are negligible . evidently , taking a minimal length into account naturally leads to the space - time with noncommutative coordinates . the first attempt in this direction was done by snyder in 1947 who formulated a lorentz - invariant discrete space - time with noncommutative coordinates @xcite . the space - time with noncommutative structure is also suggested by the theory of stability of the lie algebras @xcite . on the other hand , existence of a universal minimal length can not be supported by the special relativity since any length scale in one inertial frame may be different in another observer s frame through the well - known lorentz - fitzgerald contraction . thus , the doubly special relativity theories are formulated in order to take into account an observer - independent minimum length as well as the velocity of light @xcite . the curved four - momentum space then naturally appears to be a suitable framework to formulate the doubly special relativity theories @xcite . furthermore , the different doubly special relativity theories can be obtained from the different basis of the @xmath103-poincar algebra on the associated @xmath103-minkowski noncommutative space - time @xcite . in this respect , it is also shown that the snyder noncommutative algebra which is proposed in ref . @xcite and the stable algebra of relativistic quantum mechanics that is obtained in ref . @xcite can be obtained from a ten - dimensional phase space with curved geometry for the four - momentum space through the symplectic reduction process @xcite . thus , it seems that the space - time with curved four - momentum space and noncommutative coordinates is a fundamental framework for taking into account a minimal length scale in the flat limit of quantum gravity @xcite . it would also be claimed that the lorentz invariance is an approximate symmetry and will be broken at high energy regime . therefore , the lorentz - violating noncommutative algebra was proposed by camelia in ref . @xcite which supports the existence of a minimal observer - independent length scale . magueijo and smolin showed that the lorentz invariance can also be preserved by a nonlinear action of the lorentz group on the momentum space @xcite . in order to preserve the lorentz invariance , the following commutation relations for the generators will be held @xmath104 where @xmath105 is the minkowski metric and @xmath106 . the commutation relations between @xmath107 then classify the different lorentz - invariant algebras such as the snyder noncommutative algebra @xcite , the stable algebra @xcite and also the standard lorentz algebra . while the relativistic algebras are written in the eight - dimensional relativistic phase space , to formulate the statistical mechanics one needs to work in a six - dimensional nonrelativistic subalgebra of the deformed relativistic algebras in which the time parameter is fixed and the corresponding poisson brackets take a noncanonical form . by means of the resulting deformed hamiltonian system , one can study the statistical mechanics with the help of the constructed setup in the pervious sections . in recent years , many works have gone in this direction in order to study the effects of an ultraviolet cutoff ( minimal length and maximal momentum ) on the thermodynamical properties of the physical systems . for instance , thermodynamics of some physical systems in noncommutative spaces is studied in ref . for the special case of the snyder noncommutative space , see ref . @xcite , in which the thermodynamics of the early universe is explored . it is shown that the liouville theorem for the deformed phase space with general noncanonical poisson algebra is satisfied . however , as we have shown in section slowromancap3@ , the liouville theorem can be justified in a very simple manner in the covariant formalism , which also shows the advantage of the setup . furthermore , thermodynamical ideal gases are studied in the context of doubly special relativity theories @xcite . also , motivated by the string theory , the generalized uncertainty principle is suggested which supports the existence of a minimal length as a nonzero uncertainty in position measurement @xcite . for the statistical mechanics in this setup , inspired by loop quantum gravity , the polymer quantum mechanics is formulated which supports the existence of a minimal length known as the polymer length scale @xcite . thermodynamical properties of the ideal gases and harmonic oscillator are also studied in refs . @xcite . to see the effects of minimal length on the thermodynamics of the black holes in noncommutative space , the generalized uncertainty principle framework , and the polymer quantization scheme , @xcite . apart from the details of the above - mentioned effective approaches to quantum gravity phenomenology , all of them try to consider the effects of a minimal length in a relevant manner , and implementing the covariant formalism can clarify the consequences and applications in its most fundamental form in the language of symplectic geometry . as we have mentioned above , the advantages of the covariant formulation may be revealed when a hamiltonian system with nontrivial structure is considered . therefore , in this subsection we study the statistical mechanics in the snyder noncommutative space as a well - known example of a kinematically deformed hamiltonian system . the snyder relativistic algebra preserves the lorentz invariance and therefore the commutation relations ( [ algebra ] ) are satisfied by the corresponding generators . the commutation relations between @xmath107 which define the snyder algebra are given by @xcite @xmath108 where @xmath103 is the deformation parameter with the dimension of the inverse of length which plays the role of the universal quantum gravity scale in this setup . the four - momentum space of the eight - dimensional relativistic phase space of a test particle which moves on the space - time with noncommutative algebra ( [ snyder - algebra ] ) , is a de sitter space with topology @xmath109 @xcite . identifying the energy space with @xmath110 , the topology for the space of spatial momenta will be @xmath111 . in order to study the statistical mechanics in this setup , one should consider the nonrelativistic subalgebra of ( [ snyder - algebra ] ) and replace the standard canonical poisson algebra with them . the nonrelativistic snyder algebra for a particle moves in three - dimensional euclidean space @xmath38 and is then given by @xcite @xmath112 it is straightforward to show that the above commutation relations can be realized from the symplectic structure @xcite @xmath113 through the covariant definition ( [ pbd1 ] ) for the poisson brackets . the associated phase space of the particle is then simply the cotangent bundle of the configuration space which now has the nontrivial @xmath114 topology . indeed , the space of the spatial momenta has compact @xmath115 topology rather than the standard @xmath38 one ( see refs . @xcite for more details ) . since the quantum gravity cutoff ( which is defined by the deformation parameter @xmath103 ) is universal , it will be the same for all of the particles . thus , for the kinematically deformed system consisting of @xmath0 particles which obeys the snyder noncommutative algebra ( [ snyder - algebra - nr ] ) , the total symplectic structure can be obtained by the relation ( [ decomposition ] ) . the phase space @xmath6 for all of the particles is the same and has the nontrivial @xmath116 topology . the associated total phase space @xmath117 can be easily obtained through the coupling ( [ gamma split n ] ) , which has the following nontrivial topology : @xmath118 substituting the symplectic structure ( [ snyder - symplectic ] ) into the definition ( [ vol - d ] ) , the corresponding liouville volume takes the form @xmath119 where @xmath18 and @xmath19 are the @xmath20th component of the positions and momenta of the @xmath21th particle respectively . from this relation it is clear that the probability distribution is nonuniform in snyder noncommutative space and the standard uniform one can be recovered only at low energy ( low temperature ) regime @xmath120 . indeed , the existence of a minimal length , as an extra information , changes the probability distribution at the high energy regime . thus , in all of the kinematically deformed noncommutative phase spaces , the local form of the symplectic structure ( in terms of the positions and momenta of the particles ) is noncanonical and therefore the probability distribution will be nonuniform such that the microstates with higher momenta are less probable . this is the advantage of the symplectic covariant formulation of the statistical mechanics which shows that the probability distribution will be nonuniform on any phase space with noncanonical poisson algebra like the snyder algebra ( [ snyder - algebra - nr ] ) . however , in light of the darboux theorem , one can always find a local chart through the darboux transformation @xmath121 , in which the symplectic structure ( [ snyder - symplectic ] ) takes the canonical form @xmath122 for all of the particles . thus , the associated liouville volume becomes @xmath123 , which induces a uniform probability distribution over the set of microstates . but , it is important to note that @xmath124 are different from the standard canonical ones @xmath125 which are defined in relations ( [ canonical structure ] ) and ( [ canonical measure ] ) . indeed , while @xmath125 are the positions and momenta of the particles by definition , @xmath126 can not be interpreted as the positions and momenta of the particles . also , the noncanonical snyder liouville volume ( [ snyder - vol ] ) reduces to the standard canonical one ( [ canonical measure ] ) in the limit of @xmath127 and the noncanonical variables @xmath128 then coincide with standard canonical variables @xmath125 in this limit . however , the canonical variables @xmath126 are obtained through a darboux transformation and the functional form of the hamiltonian function will be modified ( compared with the standard functional form ) in this chart as @xmath129 since the dynamical equation ( [ can - dy - eq ] ) will be satisfied in a chart - independent manner . substituting the liouville volume ( [ snyder - vol ] ) in the relation ( [ partition f ] ) , the total partition function will be @xmath130}{\prod_{\alpha=1}^{n}\big(1+(p^\alpha/\kappa)^2 \big)}d^{3n}qd^{3n}p\,.\end{aligned}\ ] ] since the snyder liouville volume ( [ snyder - vol ] ) is decomposed similar to the relation ( [ decomposition 4 ] ) , for the noninteracting systems in which the total hamiltonian function also decomposes as the relation ( [ decomposition h ] ) , one can rewrite the total partition function ( [ snyder - tpf ] ) in terms of the single - particle partition function . taking the decomposition ( [ decomposition h ] ) into account and substituting the volume ( [ snyder - vol ] ) into relation ( [ snyder - tpf ] ) , the total partition function can be rewritten in the form of the relation ( [ partition fgh ] ) as @xmath131}{\big[1+(p^\alpha /\kappa)^2\big]}\,d^3q_{\alpha}d^3p^\alpha\bigg)\nonumber \\=\frac{1}{n ! } { \mathcal z_1}^n,\hspace{6cm}\end{aligned}\ ] ] where we have defined the single - particle partition function as @xmath132}{\big(1+(p^1/\kappa)^2\big)}\,,\ ] ] and we have also assumed that the hamiltonian functions are the same for all of the particles . then , all the thermodynamical properties of a system can be deduced from the total partition function ( [ snyder - tpf ] ) or ( [ snyder - d - tpf ] ) in which , as we have mentioned in the pervious section , the gibbs factor @xmath133 is considered for the nonlocalized systems such as the ideal gas and it should then be removed for the localized systems such as the harmonic oscillator which is the subject of the next subsection . now let us consider a system of @xmath0 independent and similar three - dimensional harmonic oscillators . since the oscillators are assumed to be independent , the relations ( [ decomposition ] ) and ( [ decomposition h ] ) will be satisfied and one then can implement the relation ( [ snyder - d - tpf ] ) in order to obtain the partition function of the system . the hamiltonian of the three - dimensional simple harmonic oscillator is given by @xmath134 , where @xmath135 is the mass of the oscillator , @xmath136 is the frequency , @xmath137 and @xmath138 . substituting this hamiltonian into the relation ( [ snyder - pf1 ] ) , gives the following single - particle partition function @xmath139\exp\bigg[\frac{\kappa^2}{2mt}\bigg]}\right)\,.\end{aligned}\ ] ] at the high temperature regime , the second term in the right - hand side of the above relation is negligible . under this condition the dominant contribution is due to the first term and the partition function is approximated by @xmath140 . this shows that in the domain of the high temperature , the number of degrees of freedom will be reduced since the partition function is proportional to @xmath141 rather than the standard one which is proportional to @xmath142 . we will show this feature in a more precise manner in this subsection . the total partition function can be obtained from the relation ( [ snyder - d - tpf ] ) as @xmath143 where the gibbs factor @xmath133 is dropped since the oscillators are localized . substituting the single - particle partition function ( [ ho - pf ] ) into this relation , the total partition function for @xmath0 three - dimensional harmonic oscillators can be obtained from which one is able to study the thermodynamical behavior of the system in snyder space . however , before doing this task , it is useful to consider the limit @xmath144 which is related to the first order corrections that would arise from the minimal length effects . this limit also allows us to compare our result with the full quantum partition function . in this limit , we may use the relation @xmath145 , to arrive at the following asymptotic expression for the partition function @xmath146 in which we have defined the dimensionless variable @xmath147 . in order to obtain the quantum partition function , one should solve the hamiltonian eigenvalue problem in snyder space which is not an easy task at all since the schrdinger equation takes a complicated form in this setup . however , in the context of the generalized uncertainty principle the problem of a @xmath148-dimensional harmonic is exactly solved @xcite and the results are applicable also for the snyder algebra ( [ snyder - algebra - nr ] ) as a particular case . the energy eigenvalues for a three - dimensional harmonic oscillator in snyder space are given by @xmath149,\hspace{0.5cm}\end{aligned}\ ] ] where @xmath150 , and @xmath151 and also we have considered relevant parameter identification to rewrite the result of the ref . @xcite in our notation and units . the associated quantum partition function then will be obtained from the definition ( [ qpf ] ) as @xmath152\,.\ ] ] the above summation can not be evaluated analytically . however , in ref . @xcite , it is shown that up to the first order of approximation , in the limit of @xmath153 , it leads to @xmath154 $ ] , which is nothing other than the first term in the summation ( [ ho - pf - asym ] ) . this coincidence shows that the quantum partition function ( [ ho - qpf ] ) correctly leads to the semiclassical partition function ( [ ho - pf ] ) up to the first order of approximation at the high temperature regime . it also reveals the advantages of semiclassical approximation ( [ approx ] ) in the sense that although there is not an analytical solution for the case of full quantum partition function ( [ ho - qpf ] ) , the relation ( [ ho - pf ] ) provides an analytical expression for the high temperatures . therefore , in dealing with the statistical considerations of the physical systems in the presence of a minimal length , since one is usually interested in the high temperature regime ( such as early universe thermodynamics ) , the usage of the semiclassical approximation seems to be quite reasonable . substituting the single - particle partition function ( [ ho - pf ] ) into the relation ( [ ho - tpf ] ) gives the total partition function . the internal energy then can be obtained from the standard definition @xmath155 as @xmath156\,{e}^{\theta^2}\right)^ { -1}-\theta^2\right].\hspace{.5cm}\end{aligned}\ ] ] in figure [ fig:1 ] we have plotted the internal energy versus temperature in comparison with its nondeformed counterpart . the specific heat can also be deduced from the definition @xmath157 which leads to a somehow cumbersome expression . + 3em internal energy versus temperature is plotted . the solid and dashed lines represent the internal energy for the snyder - deformed and nondeformed cases , respectively . the deviation from the standard case arises at the high temperature regime when the minimal length effects dominate . the figure is plotted for @xmath158.,title="fig:",width=288 ] an interesting feature here is the reduction of the number of degrees of freedom in snyder space . according to the well - known equipartition theorem of energy , for the hamiltonians of the form @xmath159 , each of the 6 degrees of freedom makes a contribution of @xmath160 towards the internal energy . thus one may consider @xmath161 as the number of degrees of freedom for each three - dimensional harmonic oscillator . in figure [ fig:2 ] the number of degrees of freedom for each three - dimensional harmonic oscillator is plotted . as this figure shows the number of degrees of freedom will be reduced from 6 to 4 in snyder space . this may be compared with the statistical mechanics of the ideal gases considered in ref . @xcite in which the same result is obtained . however , we would like to emphasize that the reduction of the number of degrees of freedom for the case of ideal gas can be interpreted as an effective dimensional reduction of the space at the high temperature regime which is a common feature of quantum gravity proposals @xcite and also phenomenological approaches to the minimal length conjecture @xcite . + 3em the number of degrees of freedom versus temperature . as it is clear from the figure , in snyder space the number of degrees of freedom for a three - dimensional harmonic oscillator is reduced from 6 to 4 at the high temperature regime.,title="fig:",width=288 ] considering the limit @xmath162 is useful at least for two reasons . first , we see that the modified thermodynamical relations recover the standard results at the sufficiently low temperature regime when the minimal length effect are negligible ( correspondence principle ) . second , one can estimate the magnitude of the order of the quantum gravity corrections to the thermodynamical quantities which may potentially lead to observable effects . in this limit , the series in the relation ( [ ho - pf - asym ] ) converges and one can use it to obtain all the minimal length ( quantum gravity ) corrections to the internal energy and specific heat . substituting the relation ( [ ho - pf - asym ] ) into ( [ ho - tpf ] ) and then using the definition @xmath163 , it is easy to show that the internal energy in the limit of @xmath164 will be @xmath165 where @xmath166 is the internal energy for the standard nondeformed @xmath0 independent three - dimensional harmonic oscillators . in this limit , the specific heat can be also obtained by substituting the above relation into the definition @xmath167 which gives @xmath168 where @xmath169 is the specific heat of the corresponding nondeformed case . as is clear from the relations ( [ ho - ie - asym ] ) and ( [ ho - sh - asym ] ) , the first minimal length ( quantum gravity ) corrections to the internal energy and specific heat of the harmonic oscillator are of the order of @xmath170 . to estimate the magnitude of the order of these corrections , consider the vibrational oscillations of a carbon monoxide molecule which may be modeled with the oscillators of mass @xmath171 and frequency @xmath172 . with this numerical value for the mass and also considering @xmath173 and @xmath174 , we get @xmath175 . therefore , the first minimal length corrections to the internal energy and specific heat of the harmonic oscillator in snyder space are of the order of @xmath176 which is too small to be experimentally detected for the accessible energy scales . appearance of the hamiltonian systems with nontrivial structure in the context of phenomenological quantum gravity candidates , such as the doubly special relativity theories with deformed noncommutative phase spaces , naively suggests the revision of the statistical mechanics formalism . in the first step , we have formulated the statistical mechanics of a general hamiltonian system in a covariant ( chart - independent ) manner by means of the symplectic geometry . the results show that the two properties of the phase space , as a symplectic manifold , play distinguished roles in statistical consideration of a system : the topology of the phase space as a global property and the local form of the symplectic structure that determines the geometry of the phase space . for a topologically trivial phase space with the canonical symplectic structure , the standard statistical mechanics emerges as it is expected . the subtleties arise when the phase space has nontrivial topology or geometry such as the phase spaces with curved momentum space and noncommutative structure . the topology of the phase space turns out to be related to the total number of microstates , for instance , the number of accessible microstates will be finite for a system with compact phase space . the symplectic structure also affects the probability distribution of the microstates . the canonical form of the symplectic structure in terms of the positions and momenta of the particles plays an important role in the standard statistical mechanics . it induces a uniform probability distribution over the microstates by defining a uniform measure on the corresponding phase space . however , the noncommutative phase space always provides a noncanonical symplectic structure in terms of the positions and momenta of the particles and the probability distribution then will be nonuniform . this is a general result for the phase spaces with noncanonical ( noncommutative ) structure which can be immediately realized from the covariant formulation of the phase space . we implemented the covariant formalism in order to study the statistical mechanics in snyder noncommutative space as an explicit example of a phase space with nontrivial topology and geometry . we obtained the associated partition function from which all the thermodynamical properties of the system can be obtained . as a particular example , we obtained the partition function for the three - dimensional harmonic oscillator and we have shown that our result is in good agreement with that which arises from the full quantum consideration at the high temperature regime . while there is no analytical expression for the full quantum partition function of the harmonic oscillator in snyder space , the semiclassical approximation provides an analytical partition function which is applicable at the high temperature regime . using the obtained partition function , we have studied the thermodynamical properties of the system of harmonic oscillators in snyder space and our analysis shows that the number of microstates will be drastically reduced in this setup due to the existence of a minimal length . this result justifies our general claim for the deformed spaces which take into account a minimal length scale . apart from the details of the models which deal with a universal minimal length such as the noncommutative spaces , doubly special relativity theories , the generalized uncertainty principle , and polymer quantization , the covariant formalism reveals the main role of the minimal length in statistical mechanics : the microstates with higher energy or momenta are less probable when there is a minimal length scale for the system . also , we calculated the quantum gravity corrections to the internal energy and specific heat of a system of harmonic oscillators . we estimated the order of magnitude of the first minimal length corrections which are of the order of @xmath177 for both of the internal energy and specific heat . the equilibrium as an essential criterion for the statistical system is also considered in the appendix by means of the dynamically invariant liouville volume which enters in the definition of the gibbs entropy and partition function of the statistical system . relation ( [ decomposition 2 ] ) can be seen as a particular case in which the system is divided into statically independent subsystems and subsystems were taken to be the particles themselves . it is clear that the statistical independence between particles will be spoiled for a system in which the particles interact with each other . however , when the system is in equilibrium , one can still divide it into @xmath178 macroscopic subsystems ( @xmath179 ) which are statistically independent . in this respect , the total phase space @xmath9 and the associated symplectic structure @xmath10 will be @xmath180 the corresponding liouville volume is then given by @xmath181 where @xmath182 is the number of particles of the subsystems and is assumed to be sufficiently large @xmath183 to guarantee that the subsystems are macroscopic . the statistical independence of the subsystems is then defined as @xmath184 this criterion is usually employed as an intrinsic characterization of an equilibrium probability distribution @xcite . indeed , the system is in equilibrium if components of the system are in relative equilibrium with respect to each other . based on this internal definition of the equilibrium , the gibbs state can be understood as an equilibrium state as follows . consider the total hamiltonian function of the system as @xmath185 where @xmath186 are the hamiltonian functions of the @xmath178 macroscopic subsystems and we also disregard the interaction between subsystems based on the fact that the relative number of particles which take part in the interactions is negligible compared to @xmath182 . so , using the relations ( [ can rho class ] ) , ( [ gamma split m ] ) , and ( [ decomposition 8 ] ) , the gibbs state turns out to be an equilibrium state based on ( [ decomposition 5 ] ) . additionally , it is straightforward to show that @xmath187=0\,,\ ] ] which is the criterion for the kinematical and dynamical separability of the subsystems . the relation ( [ equilibrium ] ) then can be viewed as the equilibrium criterion when the system is in a gibbs state . note that the decomposition ( [ decomposition 8 ] ) holds only over not - too - long intervals of time since the effects of interaction of the subsystems will eventually be dominated even if such interactions are too weak such as the universal gravitational effects @xcite . another more common definition for the equilibrium statistical state of the system is @xmath188 in which @xmath189 denotes the density that corresponds to the equilibrium state . the fact that the gibbs state is an equilibrium state then will be justified through the liouville equation @xmath190 in which the liouville theorem ( [ liouvillee ] ) is used . the liouville equation should be satisfied by any statistical state as well as gibbs one and it is , indeed , the necessary requirement for the consistent probabilistic interpretation of @xmath95 . using the fact that @xmath191 , the liouville equation ( [ liouville e ] ) can be rewritten in the more well - known form @xmath192 from the relation ( [ can rho class ] ) , it is clear that @xmath193 , and the gibbs state then satisfies the relation @xmath194 , which shows that it is indeed an equilibrium state through the definition ( [ equili ] ) . if we use the liouville equation ( [ liouville e ] ) and consider the fact that the liouville volume @xmath77 does not change explicitly with time , it turns out that @xmath195 which shows that the fine - grained entropy ( [ gibbs entropy ] ) does not change with time as expected for closed systems . the above relation and the liouville equation ( [ liouville e ] ) guarantee that the number of microstates does not change through the dynamical evolution of the system , which makes the statistical formulation a consistent setup to give the thermodynamical properties in any ensemble . j. kowalski - glikman , phys . b * 547 * ( 2002 ) 291 + j. kowalski - glikman and s. nowak , class . quantum grav . * 20 * ( 2003 ) 4799 + f. girelli , e. r. livine and d. oriti , nucl . phys . b * 708 * ( 2005 ) 411 + j. kowalski - glikman , int . phys . a * 28 * ( 2013 ) 1330014 + s. meljanac , a. pachol , a. samsarov and k. s. gupta , phys . d * 87 * ( 2013 ) 125009 . j. kowalski - glikman , phys . lett . a * 299 * ( 2002 ) 454 . g. amelino - camelia , l. freidel , j. kowalski - glikman and l. smolin , phys . d * 84 * ( 2011 ) 084010 + g. amelino - camelia , l. freidel , j. kowalski - glikman and l. smolin , gen . . grav . * 43 * ( 2011 ) 2547 . j. kowalski - glikman , phys . lett . a * 299 * ( 2002 ) 454 + x. zhang , l. shao and b. -q . ma , astropart . * 34 * ( 2011 ) 840 + s. das , s. ghosh and d. roychowdhury , phys . rev . d * 80 * ( 2009 ) 125036 + s. das and d. roychowdhury , phys . d * 81 * ( 2010 ) 085039 + n. chandra and s. chatterjee , phys . d * 85 * ( 2012 ) 045012 + g. amelino - camelia , n. loret , g. mandanici and f. mercati , int . j. mod . d * 21 * ( 2012 ) 1250052 . g. t hooft , nucl . b * 335 * ( 1990 ) 138 + e. witten , phys . d * 44 * ( 1991 ) 314 + l. susskind and j. uglum , phys . d * 50 * ( 1994 ) 2700 + a. strominger and c. vafa , phys . b * 379 * ( 1996 ) 99 + g. t. horowitz and a. strominger , phys . * 77 * ( 1996 ) 2368 + j. m. maldacena , [ arxiv : hep - th/9607235 ] + j. m. maldacena , a. strominger and e. witten , jhep * 9712 * ( 1997 ) 002 . c. rovelli , phys . * 77 * ( 1996 ) 3288 + a. ashtekar , j. baez , a. corichi and k. krasnov , phys . * 80 * ( 1998 ) 904 + m. domagala and j. lewandowski , class . quantum grav . * 21 * ( 2004 ) 5233 + k. a. meissner , class . quantum grav . * 21 * ( 2004 ) 5245 . d. j. gross and p. f. mende , nucl . b * 303 * ( 1988 ) 407 + d. amati , m. ciafaloni and g. veneziano , phys . b * 216 * ( 1989 ) 41 + m. kato , phys . b * 245 * ( 1990 ) 43 + l. garay , int . a * 10 * ( 1995 ) 145 . c. rovelli and l. smolin , nucl . b * 331 * ( 1990 ) 80 + c. rovelli and l. smolin , nucl . b * 442 * ( 1995 ) 593 + a. ashtekar and j. lewandowski , class . quantum grav . * 14 * ( 1997 ) a55 + a. ashtekar and j. lewandowski , class . quantum grav . * 21 * ( 2004 ) r53 . o. f. dayi and a. jellal , phys . lett . a * 287 * ( 2001 ) 349 + f. g. scholtz and j. govaerts , j. phys . a * 41 * ( 2008 ) 505003 + w. -h . huang and k. -w . huang phys . b * 670 * ( 2009 ) 416 + a. shariati , m. khorrami and a. h. fatollahi , j. phys . a * 43 * ( 2010 ) 285001 . m. maggiore , phys . b * 319 * ( 1993 ) 83 + m. maggiore , phys . d * 49 * ( 1994 ) 5182 + a. kempf , g. mangano and r. b. mann , phys . d * 52 * ( 1995 ) 1108 + a. kempf and g. mangano , phys . d * 55 * ( 1997 ) 7909 + k. nozari and a. etemadi , phys . d * 85 * ( 2012 ) 104029 . s. k. rama , phys . b * 519 * ( 2001 ) 103 + l. n. chang , d. minic , n. okamura and t. takeuchi , phys . d * 65 * ( 2002 ) 125028 + m. lubo , phys . d * 68 * ( 2003 ) 125004 + k. nozari and b. fazlpour , gen . reltiv . grav . * 38 * ( 2006 ) 1661 + p. wang , h. yang and x. zhang , jhep * 08 * ( 2010 ) 043 + d. mania and m. maziashvili , phys . lett . b * 705 * ( 2011 ) 521 + p. wang , h. yang and x. zhang , phys . b * 718 * ( 2012 ) 265 + b. vakili and m. a. gorji , j. stat . ( 2012 ) p10013 . a. ashtekar , s. fairhurst and j. willis , class . quantum grav . * 20 * ( 2003 ) 1031 + k. fredenhagen and f. reszewski , class . quantum grav . * 23 * ( 2006 ) 6577 + a. corichi , t. vukainac and j. a. zapata , phys . d * 76 * ( 2007 ) 044016 . g. m. hossain , v. husain and s. s. seahra , class . quantum grav . * 27 * ( 2010 ) 165013 + v. husain , s. s. seahra and e. j. webster , phys . d * 88 * ( 2013 ) 024014 + g. chacn - acosta , e. manrique , l. dagdug and h. a. morales - tcotl , _ sigma _ * 7 * ( 2011 ) 110 + e. castellanos and g. chacn - acosta , phys . b * 722 * ( 2013 ) 119 + g. chacn - acosta and h. hernandez - hernandez , int . d * 24 * ( 2015 ) 1550033 . p. nicolini , a. smailagic and e. spallucci , phys . b * 632 * ( 2006 ) 547 + s. ansoldi , p. nicolini , a. smailagic and e. spallucci , phys . lett . b * 645 * ( 2007 ) 261 + y. s. myung , y .- w . kim and y. -j . park , jhep * 0702 * ( 2007 ) 012 + k. nozari and s. h. mehdipour , class . quantum grav . * 25 * ( 2008 ) 175015 + r. banerjee , b. r. majhi and s. samanta , phys . d * 77 * ( 2008 ) 124035 + r. banerjee , b. r. majhi and s. k. modak , class . quantum grav . * 26 * ( 2009 ) 085010 + c. bastos , o. bertolami , n. c. dias and j. n. prata , phys . d * 80 * ( 2009 ) 124038 . f. scardigli , phys . b * 452 * ( 1999 ) 39 + a. j. m. medved and e. c. vagenas , phys . d * 70 * ( 2004 ) 124021 + g. amelino - camelia , m. arzano , y. ling and g. mandanici , class . quantum grav . * 23 * ( 2006 ) 2585 + y. s. myung , y. -w . kim and y. -j . park , phys . b * 645 * ( 2007 ) 393 + k. nozari and a. s. sefiedgar , phys . b * 635 * ( 2006 ) 156 + a. bina , s. jalalzadeh and a. moslehi , phys . rev . d * 81 * ( 2010 ) 023528 + l. xiang and x. q. wen , jhep * 0910 * ( 2009 ) 046 .

we will consider the ginzburg landau model of superconductivity . if a @xmath1-dimensional superconducting sample with ginzburg landau parameter @xmath0 is submitted to a uniform magnetic field of strength @xmath2 , then ( by a theorem of giorgi and phillips @xcite ) there exists a field strength @xmath3 such that if @xmath4 , then the sample will be in its normal state , i.e. superconductivity is lost altogether . it is at first sight natural to expect this phenomenon to mark a monotone transition , i.e. to expect that the material is in its superconducting ( possibly mixed ) state for all @xmath5 . indeed , such a monotonicity result has been proved recently in a number of geometric situations and in both @xmath1 and @xmath6 dimensional settings @xcite in the case where the ginzburg landau parameter @xmath0 is large ( it also follows from asymptotic expansions obtained in other works such as @xcite ) . however , nature does not support this monotonicity in general . the famous little parks effect @xcite shows that for narrow cylinders ( or @xmath1d annuli ) one has an oscillatory behavior instead of monotonicity . in this paper we will establish such ` oscillatory ' effects rigorously in different geometric settings . the lack of monotonicity comes from the topology / geometry of the annulus . it is natural to ask whether one can get such an oscillatory effect for ( non - vanishing ) magnetic fields defined on domains without topology . from the previous investigations @xcite we know this to be impossible for a uniform magnetic field , but how about more general fields ? the analysis of constant magnetic fields tells us that this question is linked to a purely spectral problem , namely whether the first eigenvalue of the schrdinger operator @xmath7 is monotone increasing in the parameter ( strength of the magnetic field ) @xmath8 for sufficiently large values of @xmath8 . this property has been called ` strong diamagnetism ' and has been proved for large classes of magnetic fields it is even ` generically ' satisfied @xcite . however , we produce counterexamples in the general case . the ginzburg landau theory of superconductivity is based on the energy functional @xmath9 here @xmath10 is a material parameter ( the ginzburg landau parameter ) , @xmath11 is a parameter measuring the intensity of the external magnetic field . the domain @xmath12 is the part of space occupied by the superconducting material . for @xmath13 there are two natural choices . one can take @xmath14 . that will not be our choice here because for reasons of simplicity we want to avoid an unnecessary technical complication connected with unbounded domains in @xmath15 ( for details on how to handle this issue see @xcite ) . one can also and that will be our convention here take @xmath13 to be the smallest simply connected domain containing @xmath16 , i.e. the union of @xmath16 and all the ` holes ' in @xmath16 . the function @xmath17 is the profile of the external magnetic field . in the setting of bounded @xmath18 the functional @xmath19 is naturally defined on @xmath20 . the functional is immediately seen to be gauge invariant , @xmath21 . the vector field @xmath22 models the induced magnetic vector potential . the function @xmath23 measures the superconducting properties of the material , with @xmath24 being a measure of the local density of cooper pairs . we say that a minimizer @xmath25 of the ginzburg landau functional is trivial if @xmath26 and @xmath27 . in each of the situations we will encounter , the notation @xmath28 will be reserved for a fixed choice of vector potential with @xmath29 . for trivial minimizers we clearly have @xmath30 . for a nontrivial minimizer the functional must be negative , since one gets from the euler - lagrange equations of a minimizer that @xmath31 if @xmath25 is a minimizer . we define the set @xmath32 following @xcite one typically defines the third critical field to be given by @xmath33 , which is finite by @xcite . however , unless @xmath34 is an interval , this definition is not the only natural one to take see @xcite for a discussion . we will see below that this is not always the case . let @xmath35 denote the annulus with inner radius @xmath36 and outer radius @xmath37 , let @xmath38 . in this case we will write @xmath39 i.e. the disc of radius @xmath37 . [ thm : main3 ] there exists an annulus @xmath16 and a @xmath40 such the set @xmath41 is not an interval . the mechanism behind this result is a convergence of the magnetic quadratic form on the annulus to the corresponding form on the circle . this convergence was already noticed in the works @xcite , where also ` annuli ' of non - uniform width were considered . it is likely that one could deduce theorem [ thm : main3 ] from these works , however , we prefer to give a simple independent proof which also emphasizes the connection to the bohm aharonov - effect . [ rem : largerkappa ] by shrinking the inner radius @xmath36 of the annulus , we can get @xmath42 as large as we want , since the eigenvalues of the limiting problem will then cross at a level @xmath43 . in particular it is possible to have @xmath44 , which means that theorem [ thm : main3 ] also applies to superconductors of type ii . one may criticize the result of theorem [ thm : main3 ] on two accounts . one could desire not to have the topology fixed a priori , but rather have it generated by localization properties of the minimizer . also most previous mathematical analysis has considered the limit of large values of @xmath0 . one can show that for sufficiently large values of @xmath0 the set @xmath34 of a superconducting sample in the shape of an annulus will behave as the one of the disc with the same outer radius , and it is known that for the disc and with constant magnetic field for sufficiently large values of @xmath0@xmath34 is indeed an interval @xcite . our next theorem remedies these defects . [ thm : hc3-oscillation ] let @xmath16 be the unit disc in @xmath45 . there exists an everywhere positive magnetic field @xmath46 such that for all @xmath40 there exists @xmath47 satisfying that @xmath34 is not an interval . in fact , the magnetic field can be chosen as @xmath48 , where @xmath49 is some sufficiently small constant . theorem [ thm : hc3-oscillation ] follows directly from theorem [ thm : main ] ( or theorem [ thm : intasymptot ] ) below using ( * ? ? ? actually , it easily follows from theorem [ thm : intasymptot ] below , that for all integers @xmath50 we can choose @xmath51 so small that @xmath52 will consist of at least @xmath53 intervals for all @xmath0 sufficiently large . for easy reference we collect the notation and assumptions concerning the magnetic fields that we will treat . we will work on an open set @xmath16 being one the following three cases @xmath54 . [ ass : magnetic ] suppose that @xmath55 , is a non - negative , radial magnetic field , possessing five continuous derivatives in an open neighborhood @xmath56 of the unit circle @xmath57 . define @xmath58 and assume that @xmath59 and write @xmath60 when @xmath61 , we assume that @xmath62 where @xmath63 is the spectral constant recalled in appendix [ sec : dg ] . when @xmath64 , we impose the stronger assumption that @xmath65 has a unique , non - degenerate minimum at @xmath66 and that @xmath67 the assumptions assure that ground state eigenfunctions will be localized near @xmath66 . for @xmath64 , we have @xmath68 by assumption , but that is not necessarily true in the cases with boundary . we define @xmath69 i.e. @xmath70 denotes the magnetic flux through the unit disc . for a magnetic field satisfying assumption [ ass : magnetic ] and @xmath71 , we study the lowest eigenvalue @xmath72 of the self - adjoint magnetic schrdinger operator @xmath73 in @xmath74 . here @xmath28 is a magnetic vector potential associated with the magnetic field @xmath75 . we refer the reader to section [ sec : prel ] for a more complete definition of this operator and the eigenvalue . we will study this eigenvalue problem in three cases , namely for @xmath16 the unit disc , the complement of the unit disc and the whole plane @xmath45 . if @xmath16 has a non - empty boundary we impose a magnetic neumann boundary condition . the next theorem states that if @xmath16 is the unit disc or its complement , then special choices of magnetic fields satisfying assumption [ ass : magnetic ] will give that the function @xmath76 is _ not _ monotonically increasing for large @xmath8 . before stating the theorems , we remind the reader that @xmath77 are universal ( spectral ) constants coming from the de gennes model operator this is recalled in appendix [ sec : dg ] . [ thm : main ] let @xmath16 be the unit disc or its complement . suppose that @xmath75 satisfies assumption [ ass : magnetic ] . assume that @xmath49 and @xmath78 then for all @xmath79 there exist @xmath80 and @xmath81 , with @xmath82 , such that @xmath83 on the other hand , if @xmath84 then there exists @xmath79 such that @xmath85 is monotone increasing on @xmath86 . in particular , holds for the magnetic field @xmath87 for all @xmath49 sufficiently small the flux in this case is @xmath88 . therefore , this magnetic field will not display monotonicity for large field strength . theorem [ thm : main ] is a consequence of the following precise asymptotic formulas for the ground state eigenvalue given as theorem [ thm : extasymptot ] and theorem [ thm : intasymptot ] . [ thm : extasymptot ] suppose that @xmath16 is the complement of the unit disc , that @xmath75 satisfies assumption [ ass : magnetic ] and that @xmath49 . then there are constants @xmath89 and @xmath90 such that if @xmath91 then , as @xmath92 , @xmath93 by a careful reading of the proof , one will realize that the constant @xmath89 is independent of @xmath51 but that @xmath90 depends on @xmath51 . however , for our purposes this extra information is irrelevant . a similar expansion holds in the interior of the unit disc . [ thm : intasymptot ] suppose that @xmath16 is the unit disc , that @xmath75 satisfies assumption [ ass : magnetic ] and that @xmath49 . then there exist constants @xmath94 and @xmath95 such that if @xmath96 then , as @xmath92 , @xmath97 notice that for the disc or its complement , the constant magnetic field @xmath98 satisfies assumption [ ass : magnetic ] , so theorems [ thm : extasymptot ] and [ thm : intasymptot ] imply this special case . this agrees with the calculations in @xcite ( see also @xcite ) . in the case of constant field is not satisfied , and one does get monotonicity of the ground state energy for large magnetic field ( this is discussed in detail in @xcite ) . we continue with @xmath99 . here , we are only able to destroy monotonicity in the case @xmath100 . [ thm : wholeplane ] let @xmath99 . then , for all @xmath101 and all magnetic fields satisfying assumption [ ass : magnetic ] there exists a @xmath79 such that @xmath72 is monotonically increasing for @xmath102 . however , if @xmath103 , then @xmath85 is not monotone increasing on any unbounded half - interval . as for the disc and the exterior of the disc , the proof of this result goes via asymptotic expansions . [ thm : r2deltapos ] suppose that @xmath99 , and that @xmath75 satisfies assumption [ ass : magnetic ] with @xmath49 . then , as @xmath92 , @xmath104 [ thm : r2deltazero ] let @xmath105 and @xmath106 be the spectral constants from and respectively . suppose that @xmath99 , and that and that @xmath75 satisfies assumption [ ass : magnetic ] with @xmath100 . there exist constants @xmath107 and @xmath108 such that if @xmath109 then , as @xmath92 , @xmath110 in all of the results above the ground state has angular momentum @xmath111 ( to leading order in @xmath8 ) . we recall that @xmath112 is the total flux through the unit disc the bounded domain enclosed by the curve where we have localization . the possibility to obtain non - monotonicity comes from the condition that @xmath113 must be an integer , which leads to frustration . this is similar to examples in @xcite . theorem [ thm : wholeplane ] raises the question whether one can break strong diamagnetism with a strictly positive magnetic field on the whole plane . in the next section we define the operators involved and perform the fourier decomposition reducing the study to a family of ordinary differential operators . in section [ sec : annulus ] we prove a non - monotonicity result for an annulus and use that to prove theorem [ thm : main3 ] . in section [ sec : section4 ] we work in the exterior of the unit disc and prove theorem [ thm : extasymptot ] . we indicate in section [ sec : disc ] how the proof of theorem [ thm : extasymptot ] can be modified to give the proof of theorem [ thm : intasymptot ] . in section [ sec : nonmondiscanddiscext ] we see how theorem [ thm : extasymptot ] and theorem [ thm : intasymptot ] imply theorem [ thm : main ] . in section [ sec : planeg0 ] we prove theorem [ thm : r2deltapos ] and in section [ sec : plane0 ] we prove theorem [ thm : r2deltazero ] . these two results are used to prove theorem [ thm : wholeplane ] . we consider the self - adjoint magnetic neumann schrdinger operator @xmath114 with domain @xmath115 here @xmath116 is the interior unit normal to @xmath117 , @xmath118 and @xmath119 is the strength of the magnetic field . in general , for a self - adjoint operator @xmath120 that is semi - bounded from below we will write @xmath121 for the lowest point of the spectrum of @xmath120 . in the case of the disc or if @xmath122 as @xmath123 the operator has compact resolvent ( see @xcite ) . if @xmath16 is unbounded and if @xmath124 , then the essential spectrum will be bounded below by @xmath125 ( see @xcite for the case of @xmath45 and @xcite for the case of the exterior of the disc ) . in any case , as it will follow by the results below , @xmath72 will be an eigenvalue . we will work in domains @xmath16 that are rotationally symmetric . for that reason , we will often work in polar coordinates @xmath126 here @xmath127 will be an interval . moreover , we will work with magnetic fields that depends only on @xmath128 . for a radial magnetic field @xmath129 we will work with the gauge @xmath130 where , so @xmath131 has an immediate interpretation in terms of the flux through the disc of radius @xmath132 . ] @xmath133 in calculations , we will often meet the expression @xmath134 . this we can write as @xmath135 where @xmath136 thus , under assumption [ ass : magnetic ] , as @xmath137 , @xmath138 with @xmath139 . the expression for the operator @xmath140 in polar coordinates becomes @xmath141 we decompose the hilbert space as ( here @xmath142 denotes any of the intervals @xmath143 , @xmath144 , @xmath145 or @xmath146 ) @xmath147 that is , for a function @xmath148 , we write @xmath149 where @xmath150 . next , we write the operator @xmath140 corresponding to this decomposition as @xmath151 where @xmath152 is the self - adjoint operator acting in @xmath153 , given by @xmath154 with neumann boundary conditions at the endpoints of @xmath142 . the quadratic form corresponding to @xmath152 is given by @xmath155=\int_i \bigl[|\psi'(r)|^2 + \bigl(\frac{m}{r}-ba(r)\bigr)^2|\psi(r)|^2 \bigr]r\ , dr.\ ] ] it holds that @xmath156 in this section we will let @xmath157 we aim to prove theorem [ thm : main3 ] . we first notice the non - monotonicity of the function @xmath76 . [ thm : mainlin ] let @xmath158 and @xmath159 . then the operator @xmath140 in the annulus @xmath16 satisfies @xmath160 in particular , the function @xmath76 is monotonically decreasing around @xmath161 . one might suspect that some properties of @xmath140 are carried over to some model problem on the circle , as @xmath162 . let @xmath163 be the self - adjoint operator @xmath164 in @xmath165 with periodic boundary conditions . its spectrum is easily seen to consist of eigenvalues @xmath166 . in particular @xmath167 our next theorem states that @xmath72 will tend to @xmath168 as @xmath162 . [ thm : main2 ] let @xmath71 . then @xmath169 as a direct consequence of theorem [ thm : main2 ] it is possible to find an annulus such that the function @xmath76 is monotonically increasing and decreasing alternatively as many times as desired . another direct consequence of theorem [ thm : main2 ] is that , although the diamagnetic inequality tells us that @xmath170 for all @xmath71 we can actually get @xmath72 to be arbitrary close to zero if @xmath171 , @xmath172 by choosing @xmath37 close enough to @xmath36 . theorem [ thm : main2 ] can easily be extended to thin cylinders in three dimensions , since the third variable then separates . in this section we will prove the spectral results theorem [ thm : mainlin ] and theorem [ thm : main2 ] . we will work in polar coordinates . we recall that here @xmath158 . let @xmath173 denote the potential in the quadratic form @xmath174 in . we start by showing that if @xmath175 and @xmath176 then @xmath177 the function @xmath178 is positive for @xmath179 . indeed , @xmath180 . if @xmath181 then @xmath182 is decreasing , and @xmath183 . if @xmath184 then @xmath185 which is clearly positive for all @xmath179 . the inequality follows by a comparison of quadratic forms . next , we show that if @xmath186 then @xmath187 by perturbation theory it holds that @xmath188 where @xmath189 denotes the eigenfunction corresponding to @xmath190 . moreover the factor @xmath191 is negative for all @xmath192 if @xmath193 . inserting this into gives it is now easy to finish the proof of theorem [ thm : mainlin ] . let @xmath159 . inequality and analytic perturbation theory imply that @xmath194 for @xmath8 in a neighborhood of @xmath195 . since , by , it holds that the derivative of @xmath196 is negative at @xmath197 the same is true for the derivative of @xmath72 . by continuity of the derivative this holds in a neighborhood of @xmath197 . in particular we conclude that the function @xmath198 is strictly decreasing for these values of @xmath8 . since @xmath199 theorem [ thm : main2 ] is a direct consequence of the fact that , for @xmath200 and @xmath119 , @xmath201 to get an upper bound we use a trial state . in fact , we use the simplest possible one . let @xmath202 . then @xmath203 . a simple calculation shows that @xmath204 & = \lim_{\ro\searrow\ri } \biggl ( \frac{2m^2}{\ro+\ri}\frac{\log\ro-\log\ri}{\ro-\ri } -bm + \frac{b^2}{8}\bigl(\ri^2+\ro^2\bigr ) \biggr)\\ & = \bigl(\frac{m}{\ri}-\frac{b\ri}{2}\bigr)^2 . \end{aligned}\ ] ] hence @xmath205 . the lower bound is obtained by using the potential @xmath206 . let @xmath189 be a normalized eigenfunction corresponding to @xmath190 . then @xmath207 \geq \int_{\ri}^{\ro } \bigl(\frac{br}{2}-\frac{m}{r}\bigr)^2|u|^2 r\,dr \geq \min_{\ri\leq r\leq \ro}\bigl(\frac{br}{2}-\frac{m}{r}\bigr)^2.\ ] ] since @xmath208 as @xmath162 we conclude that @xmath209 this completes the proof of , and thus finishes the proof of theorem [ thm : main2 ] . in this section we prove theorem [ thm : main3 ] . we recall the reader that @xmath210 below denotes the disc with radius @xmath37 , centered at the origin . we need the following lemma , and refer to @xcite for its proof . [ lem : curl ] let @xmath36 be fixed and let @xmath211 . there exists a constant @xmath212 ( independent of @xmath37 ) such that for all @xmath213 we have @xmath214 combining this with the sobolev embedding we get the existence of a constant @xmath215 ( independent of @xmath216 $ ] ) such that for all @xmath213 @xmath217 given @xmath218 , the cauchy inequality implies that @xmath219 and so @xmath220 here we used and @xmath221 to get the last inequality . if we choose @xmath222 , then we see that if @xmath223 , then for all @xmath37 sufficiently close to @xmath36 and all @xmath25 , @xmath224 on the other hand , if @xmath225 , then we have ( with @xmath226 and @xmath189 the normalized eigenfunction corresponding to @xmath227 ) @xmath228 for sufficiently small values of @xmath229 . therefore , by the explicit spectrum of @xmath163 we can choose @xmath40 and @xmath82 such that @xmath230 define @xmath231 . by the convergence of the spectrum given in theorem [ thm : main2 ] and we find the existence of @xmath232 such that @xmath233 has a non - trivial minimizer for all @xmath234 and @xmath235 . on the other hand , it follows from that the minimizer of @xmath236 is trivial for all @xmath237 sufficiently close to @xmath36 . we conclude the existence of @xmath237 such that there exist non - trivial minimizers when @xmath238 and @xmath239 but not when @xmath240 . since @xmath241 it is clear that @xmath41 is not an interval . in this section we consider the case @xmath242 and assume that the magnetic field satisfies assumption [ ass : magnetic ] with @xmath49 . our aim is to prove theorem [ thm : extasymptot ] . before continuing we give an agmon estimate for the lowest eigenfunction . [ prop : agmon_outside ] assume that @xmath75 satisfies assumption [ ass : magnetic ] with @xmath49 . let @xmath243 . then there exist positive constants @xmath244 , @xmath245 and @xmath246 such that if @xmath102 , and if @xmath23 is an eigenfunction of @xmath140 corresponding to an eigenvalue @xmath247 . then @xmath248 theorem 8.2.4 in @xcite gives the same estimate with the restriction that the domain should be bounded . however , since we give a similar agmon estimate in section [ sec : planeg0 ] with proof we omit the proof here . we recall that the quadratic form after decomposition is given by ( with @xmath249 from ) @xmath250=\int_{1}^{+\infty}\bigl(|u'(r)|^2 + \bigl(\frac{m}{r}-b a(r)\bigr)^2|u(r)|^2\bigr)r\,dr.\ ] ] notice that at @xmath66 the potential takes the value @xmath251 this suggests that we will find the lowest energy for @xmath252 . that this is the case is the content of the following lemma . [ lem : restrictiononm ] let @xmath243 . suppose @xmath253 is an eigenfunction of @xmath140 with eigenvalue @xmath247 . then @xmath254 we neglect the kinetic energy in the expression for @xmath174 . recall the calculation . for @xmath255 , we get @xmath256 so , estimating the quadratic form with the potential , combining and , and using proposition [ prop : agmon_outside ] , we get @xmath257 & \geq \int_1 ^ 2 \frac{1}{r^2}\bigl(m - b r a(r)\bigr)^2 |u_m(r)|^2 r\ , dr\\ & \geq \int_1 ^ 2 \frac{1}{r^2}\bigl[\frac{1}{2}(m-\phi b)^2-(cb)^2(r-1)^2\bigr]|u_m(r)|^2 r\ , dr\\ & \geq \frac{1}{8}(m-\phi b)^2\bigl(1+\mathcal{o}(b^{-\infty})\bigr ) -\widetilde{c}b,\end{aligned}\ ] ] from which the lemma follows . [ lem : uniqueeigenvalue ] let @xmath243 . there exists @xmath79 such that if @xmath258 and @xmath259 , then @xmath152 admits at most one eigenvalue below @xmath260 . fix @xmath261 with @xmath262 . by the lower bound , we see that there exist @xmath263 such that if @xmath264 , then @xmath265 . so we will restrict attention to @xmath113 s such that @xmath266 , with @xmath267 . suppose , to get a contradiction , that @xmath268 are eigenfunctions of @xmath174 corresponding to eigenvalues below @xmath260 . we write @xmath269 with @xmath270 as @xmath271 . so @xmath272 using the agmon estimates , this yields the following bound on normalized functions @xmath273 in @xmath274 . @xmath275 & \geq\int_1^{\infty } \bigl ( |v'(r)|^2 + \frac{1}{r^2}\bigl(\delta m - b \delta ( r-1 ) \bigr)^2 |v(r)|^2\bigr ) r\,dr + { \mathcal o}(b^{1/2})\\ & = \widetilde{\q}_m[v ] + { \mathcal o}(b^{1/2 } ) , \end{aligned}\ ] ] with @xmath276 = \int_1^{\infty } |v'(r)|^2 + \bigl(\delta m - b \delta ( r-1 ) \bigr)^2 |v(r)|^2\,dr.\ ] ] by translation and scaling @xmath277 is unitarily equivalent to ( the quadratic form of ) a de gennes operator ( see appendix [ sec : dg ] ) and therefore has spectrum given by @xmath278 only the first of these @xmath279counted with multiplicity is below @xmath195 ( for some values of @xmath280 ) , so we reach a contradiction if we have a subspace of dimension @xmath1 on which the quadratic form is small . [ lem : furtherrestrictiononm ] let @xmath281 . suppose @xmath152 admits an eigenvalue below @xmath282 . then there exists a constant @xmath283 such that @xmath284 by lemma [ lem : restrictiononm ] , @xmath285 . assuming that @xmath189 is the eigenfunction corresponding to the unique ( by lemma [ lem : uniqueeigenvalue ] ) eigenvalue @xmath286 below @xmath282 we can use the estimate in to find that @xmath250\geq \widetilde{\q}_m[u ] + \mathcal{o}(b^{1/2}),\ ] ] with @xmath277 as in . implementing the change of variable @xmath287 , we get ( here we write @xmath288 ) @xmath289=\delta b\int_0^{+\infty } \biggl(\rho-\xi_0+\xi_0-\frac{m-\phi b}{(\delta b)^{1/2}}\biggr)^2 de gennes operator ( see appendix [ sec : dg ] ) . by noticing that the first eigenvalue @xmath290 has a quadratic minimum @xmath291 at @xmath292 ( and using the bound on @xmath293 ) we find that there exists a positive constant @xmath294 such that @xmath289\geq \biggl[\theta_0\delta b + c_0\delta b\biggl(\xi_0-\frac{m-\phi b}{(\delta b)^{1/2}}\biggr)^2\,\biggr ] \| v \|^2.\ ] ] the second term above is bounded by some constant times @xmath295 according to the assumption . this in turn gives the existence of a positive constant @xmath244 such that holds . in the remainder of this section we will always restrict our attention to @xmath113 s satisfying the conclusion of lemma [ lem : furtherrestrictiononm ] . the strategy of the rest of the proof is as follows . we will construct an explicit trial state for the operator @xmath296 ( here we suppress the dependence on @xmath113 and @xmath8 for the simplicity of notation ) . this trial state will be constructed as the first terms of a formal expansion . by taking only finitely many terms ( for our purposes @xmath6 terms suffice ) and performing a localization one gets a well - defined trial state . in terms of the objects calculated below our explicit trial state will be as follows . let @xmath297 let furthermore , @xmath298 , with @xmath299 , and define ( with suitable @xmath300 , say @xmath301 ) @xmath302 then @xmath303 and @xmath304 by self - adjointness of @xmath305 we get that @xmath306 . since we by lemma [ lem : uniqueeigenvalue ] know that @xmath305 has at most one eigenvalue near @xmath307 , we can conclude that @xmath286 gives the first terms of the asymptotic expansion of that lowest eigenvalue of @xmath305 . we proceed to the termwise construction of the trial state . since ( by proposition [ prop : agmon_outside ] ) we have localization around @xmath66 , we implement unitarily the change of variables @xmath308 here , the @xmath51 is included for convenience . then @xmath309 here the estimate on the remainder should be understood in the following sense : we will only act with our operator on the function @xmath310 from which is localized near @xmath66 on the scale @xmath311 . so we may consider @xmath312 as a quantity of order @xmath195 . by lemma [ lem : furtherrestrictiononm ] the constant term @xmath313 vanishes to leading order . for reasons of expositions we will write @xmath314 and not insert the choice @xmath315 until later . recall that @xmath316 is bounded . integrating by parts , we find ( with @xmath288 ) @xmath317 we expand our operator @xmath305 as @xmath318 and obtain @xmath319 we make the ansatz @xmath320 equating order by order in the relation @xmath321 gives : [ [ order - b0 ] ] * order @xmath322 : * + + + + + + + + + + + + + + + + + + + + + + + + + + to leading order we find @xmath323 which is the eigenvalue problem for the de gennes operator discussed in appendix [ sec : dg ] . the optimal eigenvalue @xmath307 is attained for @xmath324 and @xmath325 . [ [ order - b-12 ] ] * order @xmath311 : * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + here we get @xmath326 by taking scalar product ( with measure @xmath327 ) , we find @xmath328 via the formulas we find @xmath329 in particular @xmath330 is independent of @xmath331 . moreover , since we can choose @xmath332 , we can let @xmath333 be the regularized resolvent @xmath334 of @xmath335 . this regularized resolvent is defined as the inverse of the operator @xmath336 restricted to the space @xmath337 . so we have , @xmath338.\end{aligned}\ ] ] [ [ order - b-1 ] ] * order @xmath339 : * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we get @xmath340 taking scalar product with @xmath341 again gives @xmath342 we will not calculate this expression in all detail . we are only interested in the dependence on @xmath331 . an inspection gives that it will be a polynomial of degree two . we will calculate the coefficient in front of @xmath343 to see that it is positive so that @xmath344 has a unique minimum with respect to @xmath331 . the term @xmath345 is easily calculated since @xmath346 contains one @xmath343 only . for the term @xmath347 we find one @xmath331 in @xmath348 and therefore also one in @xmath333 . the coefficient in front of @xmath331 in that term becomes @xmath349\bigr\rangle\\ = -4\bigl\langle ( \rho-\xi_0)\phi_{\xi_0 } , \bigl(\dg(\xi_0)-\theta_0\bigr)^{-1}_{\text{reg } } \bigl[(\rho-\xi_0)\phi_{\xi_0}\bigr]\bigr\rangle.\end{gathered}\ ] ] so , the coefficient in front of @xmath343 in @xmath344 will be ( see ) @xmath350\bigr\rangle = \xi_0\phi_{\xi_0}(0)^2>0.\ ] ] this means that we can write @xmath351 where @xmath89 and @xmath90 depend only on @xmath352 , @xmath51 , @xmath292 and @xmath353 ( but not on @xmath70 ) . we summarize these findings in a lemma . [ lem : expansion ] suppose @xmath354 with @xmath355 . then @xmath356 we have to control the asymptotic expansion in @xmath331 subject to the bound @xmath357 . define @xmath358 with @xmath359 being the constants from above and @xmath344 being the quadratic function of @xmath331 from . we also define an approximate eigenfunction by @xmath360 with @xmath361 , @xmath333 given by and @xmath362 being given by solving , i.e. @xmath363.\end{aligned}\ ] ] notice from the explicit form of the operators that @xmath333 depends linearly on @xmath331 and @xmath362 depends quadratically , so @xmath273 is normalized to leading order . also , by the mapping properties of @xmath334 each @xmath364 is a smooth , rapidly decreasing function ( see lemma 3.2.9 in @xcite ) . we can now estimate as follows @xmath365 ) v \|.\end{aligned}\ ] ] by the decay properties of @xmath273 , the last term is bounded by @xmath366 . our choice of @xmath273 gives that the first term is equal to @xmath367 + b^{-2 } ( \h_2 - \lambda_2 ) v_2 \| , \end{aligned}\ ] ] which is easily seen to be bounded by @xmath368 . using lemma [ lem : furtherrestrictiononm ] , theorem [ thm : extasymptot ] follows from lemma [ lem : expansion ] by the following argument . notice that the positive quadratic term in @xmath369 dominates the error term @xmath370 unless @xmath331 is bounded in which case the dependence on @xmath331 in the error term disappears . this finishes the proof of theorem [ thm : extasymptot ] . in this section we will indicate a similar calculation of the ground state eigenvalue in the case of the unit disc , thereby proving theorem [ thm : intasymptot ] , i.e. we work on @xmath371 and for a magnetic field satisfying assumption [ ass : magnetic ] . we mainly give the results of the calculations referring to the exterior case for details . we will have exponential localization estimate like the one of proposition [ prop : agmon_outside ] ( with domain of integration being @xmath372 , of course ) . therefore , also the rough ` localization ' of the relevant angular momenta lemma [ lem : restrictiononm]will hold in this case as well . so we can proceed to make a change of variable to the region near ( on the scale @xmath373 as before ) the boundary . the leading order terms in the expansion of the operator become very similar to the case of the exterior of the disc : @xmath374 the same calculations ( using the same ansatz ) as in the previous section show that ( with @xmath375 ) @xmath376 for some constants @xmath94 and @xmath95 , depending only on the spectral parameters and @xmath51 . thus , theorem [ thm : intasymptot ] follows from calculations / arguments completely analogous to the ones in section [ sec : section4 ] and we omit the details . using the results of theorem [ thm : extasymptot ] and [ thm : intasymptot ] it is now easy to prove theorem [ thm : main ] . we only consider the case of the disc , the complement of the disc being similar ( using theorem [ thm : extasymptot ] instead of theorem [ thm : intasymptot ] ) . assume first that @xmath377 denote by @xmath182 the function @xmath378 notice that @xmath379 is increasing for all large vaues of @xmath8 . choose a sequence @xmath380 such that @xmath381 , i.e. is a half - integer . let @xmath382 . choose @xmath383 . then , for all sufficiently large @xmath53 , @xmath384 . so @xmath385 and @xmath386 so we get from the eigenvalue asymptotics that @xmath387 \nonumber \\ & \quad+ \xi_0\phi_{\xi_0}(0)^2[(1/2 - \phi \varepsilon)^2 - 1/4 ] + o(1 ) \nonumber \\ & = \theta_0 \delta \varepsilon - \xi_0\phi_{\xi_0}(0)^2[\phi \varepsilon - \phi^2 \varepsilon^2 ] + o(1),\end{aligned}\ ] ] which is negative for small @xmath388 ( and for all sufficiently large @xmath53 ) since @xmath389 by assumption . suppose now that @xmath390 we restrict attention to the interval near infinity on which @xmath391 is increasing . here we can calculate the right - hand derivative @xmath392 so we see that for any @xmath393 there exists @xmath79 such that for all @xmath394 and all @xmath395 , @xmath396 we aim to prove monotonicity of @xmath72 , so it suffices to prove a positive lower bound on its right hand derivative @xmath397 , which exists by perturbation theory . perturbation theory yields , for any @xmath398 , @xmath399 here we completed the square and used the variational characterization of the eigenvalue in order to get the inequality . since @xmath400 , for some constant @xmath401 independent of @xmath8 , we can estimate , using the eigenvalue asymptotics and @xmath402 since @xmath403 were arbitrary , we get that @xmath404 in particular , @xmath72 is monotone increasing for large value of @xmath8 if is satisfied . in this section we will consider the case @xmath99 and a magnetic field @xmath75 satisfying assumption [ ass : magnetic ] with @xmath49 . we aim to prove theorem [ thm : wholeplane ] for @xmath49 . this , however , follows directly once the asymptotic expansion in theorem [ thm : r2deltapos ] is obtained , since then it follows that ( see ( * ? ? ? * section 2.3 ) ) @xmath405 the proof of theorem [ thm : r2deltapos ] follows the same idea as the proof of theorem [ thm : extasymptot ] . we use a localization of the ground state to restrict the situation to certain values of the angular momentum . then we show that if we find a trial state with low enough energy , it must be related to the ground state energy . finally we expand our operator formally and construct a trial state that has the correct energy . we start with a localization estimate valid for @xmath407 . for @xmath100 it gives the right length scale of the localization . [ prop : firstagmon ] suppose @xmath75 satisfies assumption [ ass : magnetic ] with @xmath407 . let @xmath23 be an eigenfunction of @xmath140 corresponding to an eigenvalue @xmath408 for some @xmath409 . then there exist positive constants @xmath244 and @xmath246 such that @xmath410 and @xmath411 if @xmath102 . by the localization estimates of proposition [ prop : firstagmon ] , the quadratic forms @xmath174 are well approximated by harmonic oscillators , whose ground state eigenvalues are simple . this implies simplicity of the low - lying eigenvalues of @xmath152 . [ lem : uniqueeigenvalue2 ] let @xmath101 . let @xmath412 . there exists @xmath79 such that if @xmath258 and @xmath259 , then @xmath152 admits at most one eigenvalue below @xmath413 . the proof of lemma [ lem : uniqueeigenvalue2 ] is similar to that of lemma [ lem : uniqueeigenvalue ] and will be omitted . let @xmath414 be a smooth cut - off function of the real variable @xmath415 satisfying @xmath416 and such that @xmath417 for all @xmath415 , and @xmath418 . next , let @xmath419 and @xmath420 be positive ( to determined below ) real numbers and define in @xmath45 the functions @xmath421 and @xmath422 via @xmath423 and @xmath424 . then there exists a constant @xmath107 such that @xmath425 next , for @xmath426 , let @xmath427 . then , pointwise in @xmath428 , it holds that @xmath429 as @xmath430 . moreover , @xmath431 is differentiable almost everywhere and if @xmath432 its gradient satisfies @xmath433 moreover , @xmath431 is bounded for all @xmath426 , so the function @xmath434 belongs to the form - domain of @xmath140 . with the ims formula , we find that @xmath435+\q[\chi_2\tpsi ] & \leq \bigl(2c_1m^2b^{2\alpha}+\lambda+16b^{2\sigma}\bigr)\|\tpsi\|^2\\ & \leq \bigl(2c_1m^2b^{2\alpha } + \delta b+\omega b^{1/2}+16b^{2\sigma}\bigr)\|\tpsi\|^2 . \end{aligned}\ ] ] using that the smallest dirichlet eigenvalue is greater than the smallest value of the magnetic field ( again , see @xcite ) , we find that @xmath436\geq \delta b\|\chi_1\tpsi\|^2\ ] ] and @xmath437\geq \bigl(\delta b+\frac{kb^{1 - 2\alpha}}{4m^2}\bigr)\|\chi_2\tpsi\|^2.\ ] ] inserting this into we find that @xmath438 which can be written @xmath439 choosing @xmath440 we find that all @xmath8s factor out , and hence @xmath441 with @xmath419 so small that the left parenthesis above becomes positive , we find that there exists a constant @xmath108 such that @xmath442 on the support of @xmath421 it holds that @xmath443 , and hence @xmath444 inserting this in above yields @xmath445 using monotone convergence we find that @xmath446 on the other hand , since @xmath443 on the support of @xmath421 it is clear that @xmath447 combining these two last inequalities we find with @xmath448 . to prove we essentially only have to reinsert the @xmath449-estimate in the previous calculations . by monotone convergence and the ims - formula , we have @xmath450 - \int |\nabla \phi_\ell|^2 |\psi|^2 \,dx.\end{aligned}\ ] ] the last term is negative , and we can estimate the first term using again the ims - formula and as @xmath451 \leq \q[\chi_1\psi ] + \q[\chi_2 \psi ] \leq ( \delta b + c_2 b^{1/2 } ) \| \psi \|^2\end{aligned}\ ] ] ( with @xmath452 and using @xmath453 ) . now follows from . with the help of proposition [ prop : firstagmon ] , we now get a first control of the involved angular momenta . [ lem : locangmomfirst ] let @xmath407 . suppose @xmath454 is an eigenfunction of @xmath140 with eigenvalue below @xmath413 . then @xmath455 the proof of lemma [ lem : locangmomfirst ] is similar to the one of lemma [ lem : restrictiononm]taking into account the weaker localization given by proposition [ prop : firstagmon]and will be omitted . by lemma [ lem : uniqueeigenvalue2 ] there is at most one eigenvalue of @xmath152 for sufficiently low energy . so it suffices to construct a trial state . the trial function ( and all its derivatives ) will be localized on the length scale @xmath311 near @xmath66 ( see for the explicit choice of trial state ) . also the function has support away from @xmath457 . the calculation is slightly different in different regimes of angular momenta @xmath113 . in this subsection , we consider angular momenta satisfying that latexmath:[\ ] ] in particular this expression is positive . for @xmath584 , we define the montgomery type operator @xmath585 as a self - adjoint operator on @xmath586 . let us denote by @xmath587 the eigenvalues of @xmath536 , with corresponding eigenfunctions @xmath588 , @xmath589 and so on . [ thm : non - degenerate ] the function @xmath590 has a unique minimum at @xmath551 . furthermore , the minimum is non - degenerate , i.e. @xmath591 we introduce the following notation , @xmath592 by the estimates in @xcite we know that @xmath593 a numerical value of @xmath106 , calculated by v. bonnaillie - nol , is @xmath594 . [ prop : secdiffzeropos ] it holds that @xmath595 perturbation theory . the eigenvalues of @xmath152 can be solved explicitly in terms of confluent hypergeometric functions , and plotted by the computer . below we include a figure with the lowest eigenvalue of the limit operator @xmath163 and the lowest eigenvalue of the annulus of inner radius @xmath158 and outer radius @xmath596 . ( dotted ) and the lowest eigenvalue @xmath168 ( solid ) for @xmath597 and @xmath158 . right : the lowest eigenvalue @xmath72 plotted for @xmath598 . the dotted lines are the lowest eigenvalue of @xmath152 for @xmath599 , @xmath158 and @xmath596 . ] ( dotted ) and the lowest eigenvalue @xmath168 ( solid ) for @xmath597 and @xmath158 . right : the lowest eigenvalue @xmath72 plotted for @xmath598 . the dotted lines are the lowest eigenvalue of @xmath152 for @xmath599 , @xmath158 and @xmath596 . ] sf was partially supported by the lundbeck foundation , the danish natural science research council and the european research council under the european community s seventh framework program ( fp7/20072013)/erc grant agreement 202859 .

the interaction of a charged particles beam with a plasma is an important subject of relevance for many fields of physics , such as inertial confinement fusion ( icf ) driven by ion or electron beams @xcite , high energy density physics @xcite , and related astrophysical phenomena @xcite . this interaction is also relevant , among others , for the fast ignition scenario ( fis ) @xcite , where the precompressed deuterium - tritium ( dt ) core of a fusion target is to be ignited by a laser - generated relativistic electron beam . in addition , a promising icf scheme has been recently proposed @xcite , in which the plasma target is irradiated simultaneously by intense laser and ion beams . within this scheme several experiments @xcite have been performed to investigate the interactions of heavy ion and laser beams with plasma or solid targets . an important aspect of these experiments is the energy loss measurements for the ions in a wide - range of plasma parameters . it is expected in such experiments that the ion propagation would be essentially affected by the parametric excitation of the plasma target by means of laser irradiation . this effect has been supported recently by particle - in - cell numerical simulations @xcite . motivated by the experimental achievements mentioned above , in this paper we present a study of the effects of intense radiation field ( rf ) on the interaction of projectile particles with an electron plasma . previously this has been a subject of great activity , starting with the pioneering work of tavdgiridze , aliev , and gorbunov @xcite . more recently the need of the comprehensive analysis of the complexity of the beam - matter interaction in the presence of strong laser fields has stimulated a number of theoretical studies . @xcite and @xcite have developed a time - dependent hamiltonian formulation to describe the effect of a strong laser field on energy losses of swift ions moving in a degenerate electron gas . it has been shown that the energy loss is reduced for ions at intermediate velocities , but is increased for slow ions , due to a resonance process of plasmon excitation in the target assisted by a photon absorption @xcite . for plasma targets , on the other hand , it has been obtained that the energy loss is reduced in the presence of a laser field when the projectile velocity is smaller than the electron thermal velocity @xcite . moreover , the reduction of the energy loss rate increases with decreasing the angle between the projectile velocity and the polarization vector of the laser field @xcite . next , an interesting effect has been obtained by @xcite considering the oscillatory motion of the light projectile particles in the laser field . in this case the energy loss depends on the projectile charge sign and mass and , therefore , results in a different stopping powers for electrons and positrons . in recent years special attention has been also paid to the regimes of high intensities of the laser field when the quiver velocity of the plasma electron exceeds the projectile velocity @xcite . it has been shown that the projectile particles can be even accelerated in a plasma due to the parametric effects excited by an intense laser field . furthermore , @xcite have studied the influence of a high - intensity laser field on the coulomb explosion and stopping power for a swift h@xmath3 cluster ion in a plasma target . they have demonstrated that the laser field affects the correlation between the ions and contributes to weaken the wake effect as compared to the laser - free case . all papers mentioned above deal with a nonrelativistic projectile particles interacting with a laser irradiated plasma . in the present paper we study more general configurations thus considering the interaction of relativistic particles with an electron plasma . our objective is then to study some aspects of the energy loss of relativistic particles which have not been considered previously . as an important motivation of the present paper is the research on the topic of the fis for inertial confinement fusion @xcite which involves the interaction of a laser - generated relativistic electron beam ( reb ) with a plasma . although the concept of the fis implies an overdense plasma and the propagation of a relativistic electron beam from the border of a precompressed target to the dense core occurs without crossing the laser beam , the target plasma is assumed to be parametrically excited by the rf through high - harmonics generation . in this situation , the presence of the rf as well as the relativistic effects of the beam can dramatically change the main features of the standard stopping capabilities of the plasma . the plan of the paper is as follows . in section [ sec:2 ] , we outline the theoretical formalism for the electromagnetic response of a plasma due to the motion of a relativistic charged particles beam in a plasma in the presence of an intense rf . the stopping of the relativistic beam is formulated in section [ sec:3 ] . in this section we calculate the effects of the rf on the mean energy loss ( stopping power ) of the test particle considering two somewhat distinct cases with a weak ( sec . [ sec:3 - 1 ] ) and intense ( sec . [ sec:3 - 2 ] ) radiation field as well as a wide range for the test particle energies , extending from nonrelativistic up to the ultrarelativistic energies . the dynamics of the electron plasma is treated by a system of fluid equations which results in a high - frequency approximation for the dielectric response function . the results are summarized in section [ sec:4 ] . the whole interaction process of the relativistic charged particles beam with plasma involves the energy loss and the charge states of the projectile particle and as an additional aspect the ionization and recombination of the projectile driven by the rf and the collisions with the plasma particles . a complete description of the interaction of the beam requires a simultaneous treatment of all these effects including , in particular , in the case of the structured particles the effect of the charge equilibration on the energy loss process . in this paper , we do not discuss the charge state evolution of the projectiles under study , but concentrate on the rf effects on the energy loss process assuming an equilibrium charge state of the projectile particle with an effective charge @xmath4 . this is motivated by the fact that the charge equilibration occurs in time scales , which are usually much smaller than the time of passage of the beam through target . in addition , the effects mentioned above are not important in the case of the relativistic electron beams . the problem is formulated using the hydrodynamic model of the plasma and includes the effects of the rf in a self - consistent way . the external rf is treated in the long - wavelength limit , and the plasma electrons are considered nonrelativistic . these are good approximations provided that ( i ) the wavelength of the rf ( @xmath5 ) is much larger than the debye screening length @xmath6 ( with @xmath7 the thermal velocity of the electrons and @xmath8 the plasma frequency ) , and ( ii ) the quiver velocity of the electrons in the rf @xmath9 ( where @xmath10 is the amplitude of the rf ) is much smaller than the speed of light @xmath11 . these conditions can be alternatively written as ( i ) @xmath12 , and ( ii ) @xmath13 , where @xmath14 is the intensity of the rf . as an estimate in the case of a dense plasma , with electron density @xmath15 @xmath16 , we get @xmath17 w/@xmath18 . thus the conditions ( i ) and ( ii ) are well above the values obtained with currently available high - power sources of the rf , and these approximations are well justified . the response of the plasma to the incoming relativistic beam is governed by the hydrodynamic equations for the density @xmath19 and the velocity @xmath20 of the plasma electrons as well as by the maxwell equations for the electromagnetic fields . thus , @xmath21 \right ] , \label{eq:2}\end{aligned}\]]where @xmath22 is the rf , and @xmath23 and @xmath24 are the self - consistent electromagnetic fields which are determined by the maxwell equations , @xmath25here @xmath26 is the induced current density , @xmath27 and @xmath28 are the charge and current densities for the beam , respectively , @xmath29 is the equilibrium plasma density in an unperturbed state . as mentioned above we consider a relativistic beam of charged particles and , therefore , the influence of the rf @xmath30 on the beam is ignored . consider now the solution of eqs . - for the beam - plasma system in the presence of the high - frequency rf . in an unperturbed state ( i.e. , neglecting the self - consistent electromagnetic fields @xmath23 and @xmath24 in eq . and assuming the homogeneous initial state ) the plasma velocity satisfies the equation @xmath31 which yields the equilibrium velocity for the plasma electrons , @xmath32 . here @xmath33 and @xmath34 are the quiver velocity and the oscillation amplitude of the plasma electrons , respectively , driven by the rf . next we consider the linearized hydrodynamic and maxwell s equations for the plasma and electromagnetic fields , respectively . for sufficiently small perturbations , we assume @xmath35 , @xmath36 , ( with @xmath37 and @xmath38 ) . thus , introducing the fourier transforms @xmath39 , @xmath40 , @xmath41 , and @xmath42 with respect to the coordinate @xmath43 , the linearized hydrodynamic equations read @xmath44 \delta n(\mathbf{k},t ) = -in_{e}\left ( \mathbf{k}\cdot \delta \mathbf{v}(\mathbf{k},t)\right ) , \label{eq:5 } \\ & & \left [ \frac{\partial } { \partial t}+i\left ( \mathbf{k}\cdot \mathbf{u}_{e}(t)\right ) \right ] \delta \mathbf{v}(\mathbf{k},t ) = -\frac{e}{m}\left [ \mathbf{e}(\mathbf{k},t)+\frac{1}{c}\left [ \mathbf{u}_{e}(t)\times \mathbf{b}(\mathbf{k},t)\right ] \right ] . \label{eq:6}\end{aligned}\]]in order to solve eqs . and it is convenient to introduce instead of the fourier transform @xmath45 a new unknown function @xmath46 via the relation @xcite @xmath47where @xmath48 , @xmath45 represents one of the quantities @xmath49 , @xmath50 , @xmath41 , and @xmath42 . substituting this relation for the quantities @xmath39 , @xmath50 , @xmath41 , and @xmath42 into eqs . and it is easy to see that the obtained equation for the unknown function @xmath46 constitutes an equation with periodic coefficients @xcite . therefore , we introduce the decompositions @xmath51where @xmath52 and @xmath53 are the corresponding amplitudes of the @xmath54th harmonic . it is also useful to find the connection between these amplitudes . this can be done using the fourier series representation of the exponential function @xmath55 as well as the summation formula @xmath56 for the bessel functions @xcite which yields @xmath57here @xmath58 is the bessel function of the first kind and of the @xmath59th order . using these results from eqs . and it is straightforward to obtain @xmath60where @xmath61 and we have neglected the term of the order of @xmath62 ( see the last term in eq . ) according to the approximation ( ii ) . using the transformation formulas and from the system of equations and it is easy to obtain @xmath63 next let us evaluate the induced current density which is given by @xmath64 $ ] in the coordinate space . employing eqs . , , and for the fourier transform of the induced current density we obtain @xmath65 \right\ } \label{eq:16 } \\ & & = \frac{i\omega _ { p}^{2}}{4\pi } \sum_{s , r=-\infty } ^{\infty } \frac{1}{\omega _ { r}}j_{\ell -r}(\zeta ) j_{s - r}(\zeta ) \left [ \mathbf{e}_{s}(\mathbf{k},\omega ) + \frac{(\ell -r)\omega _ { 0}}{\omega _ { r}}\frac{\mathbf{a}}{\zeta } \left ( \mathbf{k}\cdot \mathbf{e}_{s}(\mathbf{k},\omega ) \right ) \right ] , \nonumber\end{aligned}\]]where @xmath66 is the plasma frequency . from the maxwell equations we express the magnetic field through the electric field . in terms of the amplitudes of the @xmath54th harmonics [ see definition given by eqs . and ] the electric field is determined by the system of equations @xmath67 , \label{eq:17 } \\ & & \left ( \mathbf{k}\cdot \mathbf{e}_{\ell } ( \mathbf{k},\omega ) \right ) = 4\pi i\left [ e\delta n_{\ell } ( \mathbf{k},\omega ) -\delta _ { \ell 0}\rho _ { 0}(\mathbf{k},\omega ) \right ] . \label{eq:18}\end{aligned}\]]here @xmath68 and @xmath69 are the ordinary fourier transforms of the current and charge densities of the beam , respectively . note that the @xmath59th harmonics of these quantities are given by @xmath70 and @xmath71 . also the longitudinal part of the electric field is determined by the poisson equation . insertion of eq . into eq . results in @xmath72 \mathbf{e}_{\ell } ( \mathbf{k},\omega ) -\mathbf{k } ( \mathbf{k}\cdot \mathbf{e}_{\ell } ( \mathbf{k},\omega ) ) \nonumber \\ & & = -\frac{\omega _ { p}^{2}}{c^{2}}\sum_{s , r=-\infty } ^{\infty } \frac{(\ell -r)\omega _ { 0}}{\omega _ { r}}j_{\ell -r}(\zeta ) j_{s - r}(\zeta ) \left [ \mathbf{e}_{s}(\mathbf{k},\omega ) + \frac{\omega _ { \ell } } { \omega _ { r}}\frac{\mathbf{a}}{\zeta } \left ( \mathbf{k}\cdot \mathbf{e}_{s}(\mathbf{k},\omega ) \right ) \right ] \label{eq:19 } \\ & & + \frac{4\pi i\omega _ { \ell } } { c^{2}}\delta _ { \ell 0}\mathbf{j}_{0}(\mathbf{k},\omega ) , \nonumber\end{aligned}\]]where @xmath73 is the longitudinal dielectric function of a plasma . the obtained expression completely determines the electromagnetic response in the beam - plasma system in the presence of the rf . the resulting equation represents a coupled and infinite system of linear equations for the quantities @xmath74 ( with @xmath75 ) . the ( infinite ) determinant of this system determines the dispersion equation for the electromagnetic plasma modes in the presence of the rf . it should be also emphasized that the hydrodynamic description of a plasma is justified at high frequencies when @xmath76 . in particular , this condition requires that the velocity of the projectile particle should exceed the thermal velocity @xmath7 of the plasma electrons . equation can be further simplified excluding there the longitudinal part @xmath77 of the electric field by means of the poisson equation together with the induced charge density , eq . . in the first step we multiply both sides of eq . by @xmath78 and sum up over the harmonic numbers @xmath59 and @xmath79 . using the summation formula involving two bessel functions ( see above eq . ) this results in @xmath80 similarly repeating the first step for the latter expression for the longitudinal electric field one finally obtains @xmath81 this expression has been previously obtained and studied by many authors @xcite assuming nonrelativistic beam of charged particles . thus , inserting eq . into eq . we finally obtain @xmath72 \mathbf{e}_{\ell } ( \mathbf{k},\omega ) = \frac{4\pi i\omega } { c^{2}}\delta _ { \ell 0}\mathbf{j}_{0}\left ( \mathbf{k},\omega \right ) \nonumber \\ & & -\frac{\omega _ { p}^{2}}{c^{2}}\sum_{s , r=-\infty } ^{\infty } \frac{(\ell -r)\omega _ { 0}}{\omega _ { r}}j_{\ell -r}(\zeta ) j_{s - r}(\zeta ) \mathbf{e}_{s}(\mathbf{k},\omega ) \label{eq:22 } \\ & & -4\pi i\rho _ { 0}(\mathbf{k},\omega ) \sum_{r=-\infty } ^{\infty } ( -1)^{r}\frac{\boldsymbol{\chi } _ { \ell r}(\mathbf{k},\omega ) } { \varepsilon ( \omega _ { r})}j_{\ell -r}(\zeta ) j_{r}(\zeta ) , \nonumber\end{aligned}\]]where we have introduced the vector quantity @xmath82 \frac{\mathbf{a}}{c^{2}\zeta } . \label{eq:23}\ ] ] as mentioned above eq . as well as its equivalent eq . represent an infinite system of the coupled linear equations for the amplitudes @xmath83 which , in general , do not allow an analytic solution . next , we will make an _ ad hoc _ assumption neglecting the second term in the right - hand side of eq . which contains an infinite sum over the harmonics @xmath84 . in this case the solution of eq . is trivial and it is given by @xmath85 . \nonumber\end{aligned}\]]now let us examine the physical consequences of the neglect of the second term in eq . . for this purpose we insert the last expression into the second term of the right - hand side of eq . . then it is straightforward to see that this term corresponds to the erenkov radiation ( or absorbtion ) in a plasma by a moving charged particles beam . note that this effect is absent in the field - free case since the physical conditions of the erenkov radiation can not be fulfilled in this case . thus the rf essentially changes the dispersion properties of the plasma and the erenkov effect becomes now possible . in particular , an inspection of a general expression shows that in a long wavelength limit the emission of the transverse electromagnetic waves occurs at the frequencies @xmath86 for @xmath87 ( emission ) , and at @xmath88 for @xmath89 ( absorbtion ) , where @xmath90 and @xmath91 are the relativistic factor and the velocity of the beam , respectively . moreover , the ratio of the neglected term , related to the erenkov effect , to the contribution given by eq . is of the order of @xmath92 and the neglect of the second term in eq . is justified at the high - frequencies of the rf ( @xmath93 ) . in the next sections the further calculations will be done using the approximate expression . before closing this section let us consider briefly some limiting cases of eq . . for instance , in the electrostatic limit ( assuming that @xmath94 ) from eq . one obtains @xmath95which confirms eq . . in the limit of the vanishing radiation field with @xmath96 , eq . yields @xmath97 , where @xmath98 . \label{eq:26}\]]it is seen that in the last expression @xmath99 is the fourier transform of the electric field created in a medium by an external charge with the density @xmath100 and current @xmath68 @xcite . with the basic results presented in the previous sec . [ sec:2 ] , we now take up the main topic of this paper , namely the investigation of the stopping power of a relativistic charged particle beam in a plasma irradiated by an intense laser field . for further progress the charge and current densities of the beam in eq . should be specified . furthermore , we will assume that these quantities are given functions of the coordinates and time . in particular , for a beam of charged particles having a total charge @xmath4 and moving with a constant velocity @xmath101 , the charge and current densities are determined by @xmath102 and @xmath103 , respectively . here @xmath104 is the spatial distribution of the charge in the beam with the normalization @xmath105 . accordingly , the fourier transforms of these quantities are @xmath106q(\mathbf{k})\delta ( \omega -\mathbf{k}\cdot \mathbf{u})$ ] , @xmath107 , and @xmath108 is the fourier transform of the distribution @xmath104 ( form - factor of the beam ) . note that @xmath108 is normalized in a such way that @xmath109 . in particular , for the gaussian beam with a charge distribution @xmath110 , we obtain @xmath111 . here @xmath112 and @xmath113 are the characteristic sizes of the beam along and transverse to the direction of motion , respectively . from eqs . and it is straightforward to calculate the time average ( with respect to the period @xmath114 of the laser field ) of the stopping field acting on the beam . then , the averaged stopping power ( sp ) of the beam becomes @xmath115 \frac{(\mathbf{u}\cdot \mathbf{a})}{c^{2}\zeta } \right ] \right\ } , \nonumber\end{aligned}\]]where @xmath116 and @xmath90 is the relativistic factor of the beam . in eq . the symbol @xmath117 denotes an average with respect to the period of the laser field . hence , the sp depends on the beam velocity @xmath101 , the frequency @xmath118 and the intensity @xmath119 of the rf ( the intensity dependence is given through the quiver amplitude @xmath120 ) . moreover , since the vector @xmath121 in eq . is integrated over the angular variables , @xmath122 becomes also a function of the angle @xmath123 between the velocity @xmath101 , and the direction of rf polarization , represented by @xmath120 . in addition , it is well known that within classical description an upper cut - off parameter @xmath124 ( where @xmath125 is the effective minimum impact parameter or the distance of closest approach ) must be introduced in eq . to avoid the logarithmic divergence at large @xmath126 . this divergence corresponds to the incapability of the classical perturbation theory to treat close encounters between the projectile particles and the plasma electrons properly . for @xmath125 , we use the effective minimum impact parameter excluding hard coulomb collisions with a scattering angle larger than @xmath127 . the resulting cut - off parameter @xmath128 is well known for energy loss calculations ( see , e.g. , @xcite ; @xcite and references therein ) . here @xmath7 is the thermal velocity of the electrons . in particular , at high - velocities this cut - off parameter reads as @xmath129 . when , however , @xmath130 , the de broglie wavelength exceeds the classical distance of closest approach . under these circumstances we choose @xmath131 . equation contains two types of the contributions . the first one is proportional to the imaginary part of the inverse of the longitudinal dielectric function @xmath132 and is the energy loss of the beam due to the excitations of the longitudinal plasma modes . the second contribution comes from the imaginary part of @xmath133^{-1}$ ] , where @xmath134 has been introduced above and determines the energy loss of the beam due to the excitations of the transverse plasma modes . therefore , due to the resonant nature of the integral in eq . , we may replace the imaginary parts of the functions @xmath135^{-1}$ ] and @xmath133^{-1}$ ] using the property of the dirac @xmath136 function . however , as discussed in the preceding section , the terms containing the factor @xmath137 with zero harmonic number ( @xmath138 ) correspond to the usual erenkov effect which is absent here . neglecting such terms we finally arrive at @xmath139 \delta \left ( \varepsilon ( \omega _ { \ell } ) \right ) . \nonumber\end{aligned}\]]by comparison , the sp in the absence of the rf is given by @xcite @xmath140which after straightforward integration for the point - like particles ( with @xmath141 ) yields @xmath142 in the presence of the rf , the sp @xmath143 is modified and is given by the @xmath138 term in eq . ( no photon sp ) . it is evident that the rf decreases the sp @xmath143 by a factor of @xmath144 . finally , we would like to mention that the relativistic effects in eq . are given by the second term in the square brackets and by the dispersion function @xmath134 . in a nonrelativistic limit ( @xmath145 ) the former vanishes while the latter is replaced by @xmath146 . to illustrate the features of the stopping of the relativistic charged particles in a laser - irradiated plasma , we consider below two examples when the intensity of the rf is small but the angle between the vectors @xmath101 and @xmath120 is arbitrary ( sec . [ sec:3 - 1 ] ) . in the second example we assume that the polarization vector @xmath10 of the rf is parallel to the beam velocity @xmath101 ( sec . [ sec:3 - 2 ] ) . in this section we consider the case of a weak radiation field ( @xmath147 ) at arbitrary angle @xmath123 between the velocity @xmath101 of the beam and the direction of polarization of the rf @xmath120 . assuming a point - like test particle ( with @xmath141 ) , in eq . we keep only the quadratic terms with respect to the quantity @xmath120 ( or @xmath148 ) and for the stopping power @xmath122 we obtain @xmath149 \right . \label{eq:31 } \\ & & + \frac{|\omega _ { -1}|}{\omega _ { -1}}\frac{\delta \left ( \varepsilon ( \omega _ { -1})\right ) } { d(k,\omega ) } \left [ ( \mathbf{k}\cdot \mathbf{a})-\frac{\omega _ { p}^{2}}{\omega _ { -1}^{2}}\frac{(\mathbf{u}\cdot \mathbf{v}_{e})}{c^{2}\zeta } \right ] \nonumber \\ & & -\frac{2|\omega |}{\omega } \frac{(\mathbf{k}\cdot \mathbf{a})}{k^{2}}\delta \left ( \varepsilon ( \omega ) \right)\bigg\ } , \nonumber\end{aligned}\]]where @xmath150 and @xmath143 is the field - free sp given by eq . . the remaining terms are proportional to the intensity of the radiation field ( @xmath151 ) . note that due to the isotropy of the dielectric function @xmath152 as well as the dispersion function @xmath134 the angular integrations in eq . can be easily done . this results in @xmath153 , \label{eq:32}\]]where @xmath154 is the scaled quiver velocity and two functions have been introduced as follows @xmath155with @xmath156 , @xmath157 , @xmath158 . also @xmath159 is the relativistic factor of the beam with @xmath160 . in particular , in the case of the ultrarelativistic beam with @xmath161 from eqs . and we obtain @xmath162 , \label{eq:33a } \\ & & \phi _ { 2}(\nu ) = 2\ln ( 1 + \eta ^{2 } ) , \label{eq:34a}\end{aligned}\]]and in eq . the additional energy loss of the beam ( i.e. the term of eq . proportional to @xmath163 ) is mainly determined by the function @xmath164 . consider next the angular distribution of the sp at low - intensities of the rf . figure [ fig:1 ] shows the dimensionless quantity @xmath165/s_{0}$ ] ( the relative deviation of @xmath122 from @xmath143 ) as a function of the angle @xmath166 for the scaled laser intensity @xmath167 and for @xmath168 , @xmath169 , @xmath170 , where @xmath171 is the electron classical radius . the quantity @xmath172 corresponds to the density of the plasma @xmath15 @xmath16 . in addition , left and right panels of fig . [ fig:1 ] demonstrate @xmath173 for the stopping of the nonrelativistic ( with @xmath174 and @xmath175 ) and relativistic ( with @xmath176 , @xmath177 and @xmath178 ) charged particles , respectively . at nonrelativistic velocities the second term in eq . can be neglected and the angular distribution of the quantity @xmath173 has a quadrupole nature ( the angular average of @xmath173 vanishes ) . at @xmath179 the excitation of the waves with the frequencies @xmath180 leads to additional energy loss . at @xmath181 the energy loss changes the sign and the total energy loss decreases . when the particle moves at @xmath182 with respect to the polarization vector @xmath120 the radiation field has no any influence on the sp . at the relativistic velocities ( fig . [ fig:1 ] , right panel ) the field - free sp @xmath143 as well as the relative deviation @xmath173 are strongly reduced and the critical angle @xmath183 , which now depends on the factor @xmath184 , is shifted towards higher values . finally , at ultrarelativistic velocities ( fig . [ fig:1 ] , dotted line in the right panel ) , @xmath173 is positive for arbitrary @xmath123 and the radiation field systematically increases the energy loss of the particle ( see also eq . with eqs . and ) . let us now investigate the influence of the radiation field on the stopping process of the relativistic charged particle when its velocity @xmath101 is parallel to @xmath120 . it is expected that the effect of the rf is maximal in this case . in the case of the point - like particles after straightforward integrations in eq . we arrive at @xmath186 . \label{eq:35}\ ] ] we have used here the same notations as in sec . [ sec:3 - 1 ] . from eq . it is seen that no photon sp ( the term with @xmath138 ) oscillates with the intensity of the laser field . however , the radiation field suppresses the excitation of the plasma modes and this term by a factor of @xmath187 is smaller than the field - free sp @xmath143 . as follows from eq . at high - intensities of the rf the no photon sp vanishes when @xmath188 with @xmath189 , where @xmath190 are the zeros of the bessel function @xmath191 ( @xmath192 , @xmath193 , ) . then the energy loss of the charged particle is mainly determined by other terms in eq . and is stipulated by excitation of plasma waves with the combinational frequencies @xmath194 . in the case of the ultrarelativistic charged particle ( with @xmath161 ) eq . gets essentially simplified . in this equation neglecting the small corrections proportional to @xmath195 we obtain @xmath196with @xmath197 . equation can be further simplified neglecting the small quantity @xmath198 in the argument of the bessel function . the remaining infinite sum is evaluated in appendix [ sec : app1 ] [ see eq . ] which results in @xmath199thus , in the case of the ultrarelativistic charged particles the energy loss of the beam essentially increases with the intensity of the rf . let us recall , however , that eq . is only valid below relativistic intensities of the rf when @xmath200 . for treating precisely the resonance in eq . at @xmath201 we should go beyond the present long - wavelength approximation of the laser field . figure [ fig : inf ] demonstrates the quantity @xmath202 for three distinct values of the laser frequency @xmath1 as a function of the rf intensity @xmath197 , where the sp is given by eq . for an ultrarelativistic projectile particle . note that the quantity @xmath203 is independent on the plasma density @xmath29 . it is seen that the sp is indeed not sensitive to the variation of the laser frequency and the approximation is justified at @xmath200 . next let us examine an interesting regime of the high - intensities of the rf when @xmath204 . it is clear that such a regime requires nonrelativistic velocities of the charged particles beam . thus , at @xmath205 we consider here the relativistic effects as small corrections to the corresponding nonrelativistic expressions . using the asymptotic behavior of the bessel function @xcite in a general eq . it is straightforward to obtain @xmath206 \right\ } , \nonumber\end{aligned}\]]where we have introduced the following quantities @xmath207 in appendix [ sec : app1 ] we have found some approximate analytical expressions for these functions . recalling that @xmath208 the relativistic corrections ( @xmath209 ) can be easily deduced from eq . . let us consider briefly two particular cases . when @xmath210 ( where @xmath211 ) all terms in eq . have a positive sign and the function @xmath212 in eq . is maximal in this case due to the constructive interference of the excited waves with the frequencies @xmath194 . note that the relativistic correction given by @xmath213 vanishes in this case . from eq . it follows that @xmath214 and eq . can be evaluated explicitly . the result reads @xmath215}{\sinh [ \pi \gamma a(0)]}\right\ } , \label{eq:41a}\]]where @xmath216 , @xmath217 . in the second regime when @xmath218 , @xmath219 and the relativistic corrections given by @xmath220 again vanishes , @xmath221 . equation then yields a negative contribution to the energy loss with @xmath222}{\tanh [ \frac{1}{2}\pi \gamma a(\eta ) ] } \right\ } < 0 . \label{eq:41b}\]]the function @xmath223 is minimal in this case since the different harmonics cancel effectively each other . however , it should be emphasized that the contribution of the term containing the function @xmath223 to the sp can either be positive or negative depending on the dimensionless quantity @xmath224 . moreover , the absolute value of this contribution decreases with the parameter @xmath224 . the results of the numerical evaluation of the sp given by a general eq . are shown in figs . [ fig:2][fig:4 ] both for nonrelativistic and relativistic projectiles , where the ratios @xmath225 and @xmath226 are plotted as the functions of the rf intensity @xmath227 ( figs . [ fig:2 ] and [ fig:3 ] ) and the relativistic factor @xmath228 of the electron beam ( fig . [ fig:4 ] ) , respectively . here @xmath229 and @xmath230 are measured in units @xmath231 w/@xmath18 and @xmath232 m , respectively . also in figs . [ fig:2 ] and [ fig:3 ] the parameter @xmath233 varies in the interval @xmath234 to ensure the applicability of the long - wavelength approximation of the laser field . from fig . [ fig:2 ] it is seen that the sp of a nonrelativistic particle exhibits a strong oscillations . furthermore , it may exceed the field - free sp and change the sign due to plasma irradiation by intense ( @xmath235 ) laser field . these properties of a nonrelativistic sp have been obtained previously for a classical plasma @xcite as well as for a fully degenerated plasma @xcite . the effect of the acceleration of the projectile particle shown in fig . [ fig:2 ] occurs at @xmath236 ( with @xmath237 ) when the no photon sp nearly vanishes . it should be noted that in the laser irradiated plasma a parametric instability is expected @xcite with an increment increasing with the intensity of the radiation field . this restricts the possible acceleration time of the incident particle . the similar behavior of the sp is demonstrated in the right panel of fig . [ fig:2 ] , where however the plasma density is @xmath238 @xmath16 and two orders of magnitude exceeds the value adopted in the left panel of fig . [ fig:2 ] . we note that although the ratio @xmath203 is reduced with a plasma density , the field - free sp given by eq . is proportional to @xmath29 which in turn results in a strong enhancement of the sp @xmath239 with a plasma density . next , in fig . [ fig:3 ] , we consider weakly relativistic regimes of the electron beam with @xmath240 ( left panel ) and @xmath241 ( right panel ) . at smaller intensities of the rf , @xmath242 , the sp is described qualitatively by eq . and reaches its maximum at @xmath243 ( or at @xmath244 ) . this is because the electrons are driven by the laser field in the direction of the beam with nearly same velocity @xmath91 effectively enhancing the amplitude of the excited waves . the minimum or two maxima of the sp at @xmath245 in the left panel of fig . [ fig:3 ] are formed due to the constructive interference of the excited harmonics as discussed above ( see eq . ) . the other minima of the sp at @xmath246 in the left panel of fig . [ fig:3 ] are formed due to the destructive interference of the excited harmonics ( see eq . ) . let us note that at @xmath247 the effect of the enhancement of the sp of the test particle moving in a laser irradiated plasma is intensified at smaller frequency of the laser field @xmath248 ( see figs . [ fig:2 ] and [ fig:3 ] ) while at higher intensities , @xmath235 , the quantity @xmath249 increases with the frequency @xmath1 . finally , in fig . [ fig:4 ] , the quantity @xmath250 is shown as a function of the electron beam energy at various intensities of the laser field given here by a dimensionless parameter @xmath62 . it is seen that at the moderate energies of the electron beam the sp strongly increases with the laser intensity as has been also demonstrated in figs . [ fig:2 ] and [ fig:3 ] . comparing two panels of fig . [ fig:4 ] we may conclude that the position of the maximum of the sp depends only on the intensity of the rf , @xmath251 , while the maximum value of the sp is scaled as @xmath252 and is sensitive to both the intensity and frequency of the rf . furthermore , with increasing the relativistic factor @xmath228 of the electron beam one asymptotically arrives at the ultrarelativistic regime described by eqs . and . interestingly this occurs at the moderate energies of the electron beam with @xmath253 which , in particular , pertain to the fis relativistic electron beam in the typical 1- to 2-mev energy range of practical interest . in this paper , we have investigated the energy loss of a relativistic charged particles beam moving in a laser irradiated plasma , where the laser field has been treated in the long - wavelength approximation . the dynamics of the beam - plasma system in the presence of the rf is studied by the linearized fluid and maxwell equations for the plasma particles and the electromagnetic fields , respectively . the full electromagnetic response of the system is derived involving all harmonics of the rf . it has been shown that , in general , the excited longitudinal and transversal modes are parametrically coupled due to the presence of the rf . as a result , the erenkov radiation by the charged particles beam , which is absent in a plasma in the field - free case , becomes possible . however , in contrast to the usual erenkov effect in a laser - free medium , the spectrum of the radiation in the present case is discrete and the emitted frequencies are localized around the harmonics of the rf depending essentially on the energy of the incoming beam . in particular , at ultrarelativistic energies of the beam emitted frequencies are shifted towards higher harmonics of the rf . in the course of our study , we have derived a general expression for the sp , which has been also simplified in the limit of a low - intensity laser field . as in the field - free case , the sp in a laser irradiated plasma is completely determined by the dielectric function of the plasma which in the present context is given by the hydrodynamic approximation . explicit calculations have been done for two particular cases assuming ( i ) a low - intensity laser field but arbitrary orientation of the laser polarization vector @xmath10 with respect to the charged particles beam velocity @xmath101 , and ( ii ) an arbitrary ( but nonrelativistic ) intensity of the laser field when @xmath10 is parallel to @xmath101 . in the case ( i ) it has been shown that the rf increases the sp for small angles between the velocity of the beam and @xmath10 while decreasing it at large angles . furthermore , the relative deviation of the sp from the field - free value is strongly reduced with increasing the beam energy . in the case ( ii ) and at high - intensities of the rf two extrem regimes have been distinguished when the excited harmonics cancel effectively each other reducing strongly the energy loss or increasing it due to the constructive interference . as in the previous nonrelativistic treatments @xcite an acceleration of the projectile particle is expected at high - intensities of the rf when the quiver velocity of the plasma electrons exceeds the beam velocity @xmath91 . special attention has been paid to the relativistic effects of the beam . we have demonstrated that in general the energy loss increases with the beam energy forming a maximum at moderate relativistic factors @xmath254 which , however , are shifted towards higher energies of the beam with the laser intensity . in addition , the enhancement of the sp is more pronounced when the laser frequency approaches the plasma frequency in agreement with pic simulations @xcite . finally , it has been also shown that the sp of the ultrarelativistic electron beam increases systematically with the intensity of the rf exceeding essentially the field - free sp . going beyond the present model , which is based on several approximations , we can envisage a number of improvements . in particular , a simple possibility qualitatively involving relativistic intensities of the laser field is to replace the quiver velocity @xmath255 by its relativistic counterpart @xmath256 , where @xmath257^{-1/2}$ ] . the latter approximation can be easily obtained from a single - particle dynamics in a plane rf . in the course of our study we have also neglected the erenkov radiation which might be important in the total balance of the energy loss process . however , in a general eq . this effect is well separated by its spectral characteristics and can be treated independently . we intend to address this issue in our forthcoming investigations . the work of h.b.n . has been supported by the state committee of science of the armenian ministry of higher education and science ( project no . 13 - 1c200 ) . in this appendix we evaluate the infinite sum containing the square of the bessel functions and involved in eq . . also we will derive some analytical expressions for the functions @xmath223 and @xmath213 introduced in sec . [ sec:3 - 2 ] [ see eqs . and ] . our starting point is the neumann s formula @xcite @xmath258using this expression as well as the known relation @xcite @xmath259the desired summation can be performed easily @xmath260the latter formula is valid at @xmath261 . next we derive approximate but very accurate expressions for the functions @xmath223 and @xmath213 which determine the sp in eq . . for that purpose consider the following two functions @xmath262 where @xmath263 and @xmath264 are some arbitrary positive variables . note that @xmath265 at @xmath266 and @xmath267 . consider the derivatives of these functions with respect to the variable @xmath268 . using the known summation formulas involving the trigonometric functions @xcite we arrive at @xmath269}{\sinh(\pi u)}-\frac{1}{\pi u}\right\ } , \label{eq : ap4a } \\ & & \frac{\partial}{\partial a}g_{2}(a , b ) = 2a\sum_{\ell=1}^{\infty}\frac{(-1)^{\ell}}{\ell}\frac{\sin(2\alpha\ell)}{\ell^{2}+u^{2 } } = \frac{\pi a}{u^{2}}\cos\left [ \pi q(\alpha)\right ] \left\ { \frac { \sinh\left [ 2\pi up(\alpha)\right ] } { \sinh(\pi u)}-2p(\alpha)\right\ } \label{eq : ap5a}\end{aligned}\ ] ] with @xmath270 , @xmath271 $ ] , and @xmath272where @xmath273 and @xmath274 are the fractional and integer parts of @xmath275 , respectively . consequently , from eqs . and we obtain @xmath276}{\sinh ( x ) } dx -\ln\frac{u}{u_{0 } } , \label{eq : ap7 } \\ & & g_{2}(a , b ) = g_{2}(a_{0},b)+\pi\cos [ \pi q(\alpha ) ] \left\{\int_{\pi u_{0}}^{\pi u } \frac{\sinh [ 2xp(\alpha ) ] } { \sinh ( x)}\frac{dx}{x } -2p(\alpha ) \ln\frac{u}{u_{0}}\right\ } . \label{eq : ap8}\end{aligned}\ ] ] here @xmath277 and @xmath278 is an arbitrary initial value of @xmath268 . thus , the infinite sums in eqs . and have been represented in the equivalent integral forms with finite integration limits . assuming that @xmath279 let us apply eqs . and in a particular case with @xmath280 . taking into account the initial conditions ( see eqs . and ) @xmath281 from eqs . and we obtain @xmath282}{\sinh(x)}dx , \label{eq : ap9a } \\ & & g_{2}(a , b ) = \pi\cos[\pi q(\alpha)]\left\ { \int_{\pi b}^{\pi u}\frac{\sinh [ 2xp(\alpha ) ] } { \sinh ( x)}\frac{dx}{x } - 2p(\alpha)\ln\frac{u}{b } \right\ } . \label{eq : ap10a}\end{aligned}\ ] ] for further progress we note that at @xmath279 the hyperbolic functions in eqs . and can be replaced by an exponential function , @xmath283 , which yield the following approximations for the functions @xmath284 and @xmath285 @xmath286 -\exp\left [ -2\pi u q_{-}(\alpha)\right ] \right\ } \label{eq : ap9 } \\ & & + \frac{1}{2q_{+}(\alpha)}\left\{\exp\left [ -2\pi bq_{+}(\alpha)\right ] -\exp\left[-2\pi u q_{+}(\alpha)\right ] \right\ } , \nonumber \\ & & g_{2}(a , b ) \simeq\pi\cos[\pi q(\alpha)]\big\ { e_{1}\left[2\pi u q_{+}(\alpha ) \right ] -e_{1}\left[2\pi u q_{-}(\alpha ) \right ] \label{eq : ap10 } \\ & & + e_{1}\left [ 2\pi bq_{-}(\alpha)\right ] -e_{1}\left [ 2\pi bq_{+}(\alpha)\right ] -2p(\alpha)\ln\frac{u}{b } \big\ } , \nonumber\end{aligned}\ ] ] where @xmath287 is the exponential integral function and @xmath288 . let us now consider the opposite case of small @xmath289 . then at @xmath290 we may neglect the variable @xmath291 in eqs . and and represent these expressions in the approximate form @xmath292 here @xmath293 are the functions @xmath294 at @xmath295 and are evaluated in a similar way as approximate eqs . and . for that purpose consider again eqs . and with @xmath296 and @xmath295 , that is @xmath297}{\sinh ( x)}dx -\ln a , \label{eq : ap11 } \\ & & g_{2}(a,0 ) = g_{2}(1,0)+\pi\cos [ \pi q(\alpha ) ] \left\{\int_{\pi}^{\pi a}\frac{\sinh [ 2xp(\alpha ) ] } { \sinh ( x ) } \frac{dx}{x } -2p(\alpha ) \ln a\right\ } . \label{eq : ap12}\end{aligned}\ ] ] in the latter expressions replacing the hyperbolic sine functions by the exponential functions we arrive at the following approximate formulas for the quantities @xmath298 and @xmath299 @xmath300 -\exp\left[-2\pi aq_{-}(\alpha ) \right ] \right\ } \label{eq : ap13 } \\ & & + \frac{1}{2q_{+}(\alpha ) } \left\{\exp\left[-2\pi q_{+}(\alpha ) \right ] -\exp\left[-2\pi aq_{+}(\alpha ) \right ] \right\ } -\ln a , \nonumber \\ & & g_{2}(a,0 ) \simeq g_{2}(1,0)+\pi\cos\left[\pi q(\alpha ) \right ] \left\{e_{1}\left[2\pi q_{-}(\alpha ) \right ] -e_{1}\left[2\pi q_{+}(\alpha ) \right ] \right . \label{eq : ap14 } \\ & & \left . + e_{1}\left[2\pi aq_{+}(\alpha ) \right ] -e_{1}\left[2\pi aq_{-}(\alpha ) \right ] -2p(\alpha ) \ln a\right\ } . \nonumber\end{aligned}\ ] ] finally , for the quantities @xmath301 and @xmath302 it follows from eqs . and that @xcite @xmath303 , \label{eq : ap15 } \\ & & g_{2}(1,0 ) \simeq c_{1 } \sin ( 2\alpha ) -\frac{1}{2 } c_{2}\sin ( 4\alpha ) + \frac{2}{3}\pi ^{3}\cos \left [ \pi q(\alpha ) \right ] p(\alpha ) \left [ p^{2}(\alpha ) -\frac{1}{4}\right ] , \label{eq : ap16}\end{aligned}\]]where @xmath304 , @xmath305 . for approximate evaluation of the quantities @xmath301 and @xmath302 we have used the expansion @xmath306 at @xmath307 for the logarithmic function . then at @xmath289 the final analytical expressions for @xmath284 and @xmath285 are obtained from eqs . , , and , , , respectively . the relative accuracy of the derived approximations is less than @xmath308 in a wide range of the parameters . 99 akopyan , e.a . , nersisyan , h.b . & matevosyan , h.h . energy losses of a charged particle in a plasma in an external field allowing for the field action on plasma and particle motion . _ radiophys . _ * 40 * , 823826 . couillaud , c. , deicas , r. , nardin , ph . , beuve , m.a . , guihaume , j.m . , renaud , m. , cukier , m. , deutsch , c. & maynard , g. ( 1994 ) . ionization and stopping of heavy ions in dense laser ablated plasmas . e _ * 49 * , 15451562 . frank , a. , blaevi , a. , grande , p.l . , harres , k. , heling , th . , hoffmann , d.h.h . , knobloch - maas , r. , kuznetsov , p.g . , nrnberg , f. , pelka , a. , schaumann , g. , schiwietz , g. , schkel , a. , schollmeier , m. , schumacher , d. , schtrumpf , j. , vatulin , v.v . , vinokurov , o.a . & roth , m. ( 2010 ) . energy loss of argon in a laser - generated carbon plasma . e _ * 81 * , 026401 ( 16 ) . frank , a. , blaevi , a. , bagnoud , v. , basko , m.m . , brner , m. , cayzak , w. , kraus , d. , heling , th . , hoffmann , d.h.h . , ortner , a. , otten , a. , pelka , a. , pepler , d. , schumacher , d. , tauschwitz , an . & roth , m. ( 2013 ) . energy loss and charge transfer of argon in a laser - generated carbon plasma . lett . _ * 110 * , 115001 ( 15 ) . hoffmann , d.h.h . , tahir , n.a . , udrea , s. , rosmej , o. , meister , c.v . , varentsov , d. , roth , m. , schaumann , g. , frank , a. , blaevi , a. , ling , j. , hug , a. , menzel , j. , hessling , th . , harres , k. , gnther , m. , el - moussati , s. , schumacher , d. & imran , m. ( 2010 ) . high energy density physics with heavy ion beams and related interaction phenomena . _ contrib . plasma phys . _ * 50 * , 715 . oguri , y. , tsubuku , k. , sakumi , a. , shibata , k. , sato , r. , nishigori , k. , hasegawa , j. & ogawa , m. ( 2000 ) . heavy ion stripping by a highly - ionized laser plasma . methods phys . b _ * 161 - 163 * , 155158 . roth , m. , cowan , t.e . , key , m.h . , hatchett , s.p . , brown , c. , fountain , w. , johnson , j. , pennington , d.m . , snavely , r.a . , wilks , s.c . , yasuike , k. , ruhl , h. , pegoraro , f. , bulanov , s.v . , campbell , e.m . , perry , m.d . & powell , h. ( 2001 ) . fast ignition by intense laser accelerated proton beams . lett . _ * 86 * , 436439 . roth , m. , stckll , c. , s , w. , iwase , o. , gericke , d.o . , bock , r. , hoffmann , d.h.h . , geissel , m. & seelig , w. ( 2000 ) . energy loss of heavy ions in laser - produced plasmas . _ europhys . lett . _ * 50 * , 2834 . stckl , c. , frankenheim , o.b . , roth , m. , su , w. , wetzler , h. , seelig , w. , kulish , m. , dornik , m. , laux , w. , spiller , p. , stetter , m. , stwe , s. , jacoby , j. & hoffmann , d.h.h . ( 1996 ) . interaction of heavy ion beams with dense plasmas . _ laser part . beams _ * 14 * , 561574 . tabak , m. , hammer , j. , glinsky , m.e . , kruer , w.l . , wilks , s.c . , woodworth , j. , campbell , e.m . , perry , m.d . & mason , r.j . ignition and high gain with ultrapowerful lasers . plasmas _ * 1 * , 16261634 . tahir , n.a . , deutsch , c. , fortov , v.e . , gryaznov , v. , hoffmann , d.h.h . , kulish , m. , lomonosov , i.v . , mintsev , v. , ni , p. , nikolaev , d. , piriz , a.r . , shilkin , n. , spiller , p. , shutov , a. , temporal , m. , ternovoi , v. , udrea , s. & varentsov , d. ( 2005 ) . proposal for the study of thermophysical properties of high - energy - density matter using current and future heavy - ion accelerator facilities at gsi darmstadt . * 95 * , 035001 ( 14 ) . tavdgiridze , t.l . & tsintsadze , n.l . energy losses by a charged particle in an isotropic plasma located in an external high frequency electric field . _ zh . eksp . * 58 * , 975978 [ english translation : _ sov . . jetp _ * 31 * , 524525 ( 1970 ) ] . wang , g .- q . , song , y .- h . , wang , y .- n . & mikovi , z.l . ( 2002 ) . influence of a laser field on coulomb explosions and stopping power for swift molecular ions interacting with solids . a _ * 66 * , 042901 ( 111 ) . wang , g .- q . , e , p. , wang , y .- hu , z .- h . , gao , h. , wang , y .- c . , yao , l. , zhong , h .- y . , cheng , l .- h . , yang , k. , liu , w. & xu , d .- influence of a strong laser field on coulomb explosion and stopping power of energetic h@xmath3 clusters in plasmas . plasmas _ * 19 * , 093116 ( 15 ) .

the magnitude of finite metric spaces is a partially defined numerical invariant which was introduced by leinster via a category theoretic analogue of the euler characteristic of spaces , group and posets . if a finite metric space is scaled up sufficiently then the magnitude tends upwards to the number of points in the metric space and , typically , as the metric space is scaled down the magnitude tends to one ; this allows one to think of the magnitude as the ` effective number of points ' in the metric space . interestingly , magnitude has applications in measuring biodiversity ( see @xcite and @xcite ) . in previous papers @xcite and @xcite we speculatively studied the magnitude of _ infinite _ metric spaces using finite subsets as approximations . whilst we do not know that the magnitude can always be defined in that way it seems to give meaningful answers for many subsets of euclidean space , both numerically and theoretically . for instance , in @xcite it is shown using that approach that a closed , bounded subinterval of the real line can be assigned a magnitude equal to one plus half its length ; similarly magnitudes can be assigned to circles and to cantor sets . ( these results are corroborated here in section 1 using the measure theoretic approach of this paper ) . in @xcite the magnitudes of various simple subsets of @xmath1 and @xmath2 were calculated numerically and for various spaces considered it was observed that if the metric space was scaled up then the magnitude seemed to asymptotically depend on classical intrinsic volumes of the space such as volume and euler characteristic . for _ convex _ sets it appeared that this dependence held even non - asymptotically , so for a square it seems , numerically at least , that the magnitude is given by @xmath3 . in this paper we look at the magnitude of homogeneous riemannian manifolds , using a measure theoretic approach rather than the finite subset approach mentioned above . the magnitude is calculated explicitly for the @xmath0-sphere of radius @xmath4 with its intrinsic metric , leading to the following answer . r \bigl(\bigl(\frac{r}{n-1}\bigr)^2 + 1\bigr))\bigl(\bigl(\frac{r}{n-3}\bigr)^2 + 1\bigr ) \dots \bigl(\bigl(\frac{r}{2}\bigr)^2 + 1\bigr ) } { 1-e^{-\pi r}}\quad & \text{for $ n$ odd}. \end{cases}\ ] ] for a general homogeneous riemannian manifold @xmath6 we calculate the leading three terms in the magnitude as the manifold is scaled up : the first term is proportional to the volume , the second term vanishes and the third term is proportional to the total scalar curvature ( @xmath7 ) . writing @xmath8 for @xmath6 scaled up by a factor of @xmath9 and writing @xmath10 for the volume of the unit @xmath0-ball , we have @xmath11 this suffices to determine the asymptotics of the magnitude of a homogeneous surface , this asymptotic expression involving just the area and euler characteristic of the surface @xmath12 : @xmath13 this expression in the area and euler characteristic is precisely the penguin valuation of @xcite . there are a few things to note . the first thing to note is that although we use the measure theoretic approach rather than the finite approximation approach , we show that all the results of @xcite obtained using the finite approximations are all also obtained using the measure theoretic approach . the second thing to note is that in the cases that are known , namely the @xmath0-spheres and surfaces , the constant term is precisely the euler characteristic . this is particularly interesting as the magnitude arose from a category theoretic generalization of the notion of euler characteristic of various mathematical objects . indeed we might have named magnitude as `` the euler chacteristic of metric spaces '' if that were nt such a potentially confusing term . the final thing to note is that the intrinsic volumes , which are the terms so far identified in the asymptotic expansion all appear in the asymptotic expansion of the heat kernel @xcite and , as the definition of is slightly reminiscent of the heat kernel , this looks intriguing . before giving a more detailed synopsis of the contents of the paper , it is worth making some comments about the metric that we are considering on a riemannian manifold . there is potential for confusion here in that we are considering metric spaces and riemannian metrics , so the word ` metric ' is being used with two different senses . to reduce the likelihood of confusion , in this paragraph and the next paragraph we will talk about ` distance metric ' for the metric on a metric space and ` riemannian metric ' for an inner product on each tangent space of a smooth manifold : a distance metric is a global object , a riemannian metric is an infinitessimal object . in order to define the magnitude of a riemannian manifold we need a distance metric on the manifold , we will be considering the intrinsic distance metric which arises in the following way . a riemannian metric gives rise to an arc - length function and this gives rise to a distance metric by defining the distance between two points to be the infimum of the lengths of the paths between the two points . if there is a riemannian embedding of a riemannian manifold in euclidean space then there are two obvious distance metrics , the intrinsic metric and the subspace metric : in this paper we are primarily considering the intrinsic metric whereas in @xcite and @xcite we considered mainly the subspace metric . so , for instance , on the earth , the intrinsic distance metric is realized by travelling over the surface of the earth , but the subspace distance metric is realized by tunnelling _ through _ the earth . we compare the asymptotics for the magnitude of the @xmath0-sphere with the subspace metric and to that with the intrinsic metric ; they are seen to have the same leading term but different lower order terms . there now follows a more detailed synopsis of what is in this paper . in section 1 we define the magnitude of a metric space using a measure on the space . the key observation here is that if @xmath6 is a homogeneous metric space and @xmath14 is a non - zero homogeneous measure on @xmath6 , then the magnitude of @xmath6 is given by the following formula : for any @xmath15 , @xmath16 provided that the denominator is finite and non - zero . it is also shown in this section that the measure theoretic definition gives the same magnitude as the finite subset approach did for the circle , line segment and cantor set in @xcite . in section 2 we give a brief exposition on the notion of intrinsic volumes of riemannian manifolds . intrinsic volumes are defined for a wide class of subsets of euclidean space and include such invariants as the volume , the surface area of the boundary and the euler characteristic . in @xcite and @xcite we mainly considered these for polyconvex sets that is unions of convex sets following klain and rota s treatment @xcite , but here we consider a differential treatment , utilizing @xcite and @xcite . one key point for us is that for @xmath17 a closed submanifold of @xmath18 these intrinsic volumes are intrinsic in the sense of differential geometry that is to say they are independent of the embedding and depend only on the riemannian metric of @xmath17 . so for such a riemannian manifold there are @xmath19 intrinsic volumes @xmath20 , but it turns out that about half of these vanish because @xmath17 is closed , @xmath21 if @xmath22 is odd : the fact that @xmath23 can be explained by the fact that @xmath24 measures the surface of the boundary , which is of course trivial when @xmath17 is closed . the non - zero intrinsic volumes are also known in this case as killing - lipschitz curvatures and in a few cases are well - known invariants : @xmath25 is the volume of @xmath17 ; the next non - trivial intrinsic volume @xmath26 is the total scalar curvature ( tsc ) of @xmath17 ; and @xmath27 is the euler characteristic of @xmath17 . in general these can be obtained by integrating over @xmath17 a certain form defined in terms of the curvature . in section 3 we get to one of the main results which is an explicit formula for @xmath28 the magnitude of the @xmath0-sphere of radius @xmath4 equipped with its intrinsic metric . ( here it is slightly misleading to talk of the radius of the @xmath0-sphere as it is being thought of as an abstract manifold with the arclength distance metric and not as a subset of @xmath29 with the subspace distance metric ; however it is perhaps slightly less comfortable to talk of the sphere of curvature @xmath30 . ) the formula has many of the properties you might expect , such as @xmath31 as @xmath32 . asymptotically as @xmath33 , writing @xmath10 for the volume of the unit @xmath0-ball , we get @xmath34 this is certainly consistent with the asymptotics being described by the intrinsic volumes : for instance the alternate terms vanish and we pick up the volume , the total scalar curvature and the euler characteristic . in section 4 we see the other main result that for all homogeneous riemannian manifolds the first three terms in the asymptotic expansion are like those in the above expansion , further work will be needed to get more terms in the expansion . however , this is sufficient to show that for a homogeneous riemannian surface asymptotically the magnitude has the form conjectured for the asymptotic magnitude of metric subsets of euclidean space in @xcite and @xcite . in section 5 the magnitude of the @xmath0-sphere with its subspace metric is considered . a closed form is given for @xmath35 , but not for general @xmath0 ; however , for general @xmath0 it is shown that the leading term in the asymptotic expansion is the same as for the intrinsic metric , namely @xmath36 , but that the next non - trivial term is different . it is a pleasure to thank jonathan jordan , tom leinster and david speyer for their contributions to section 1 , as well as for other informative discussions . i would also like to thank mark meckes for comments on an earlier draft . in this section we first recall the definition of the magnitude of finite metric spaces in terms of weights and generalize it to infinite metric spaces using weight measures . we show that speyer s theorem for homogeneous spaces immediately generalizes and that this definition is consistent with previous calculations for intervals , circles and cantor sets . leinster introduced the notion of the euler characteristic of a finite category which is not always defined , but which generalizes many pre - existing notions of euler characteristic , such as the euler characteristic of a group or a poset or the cardinality of a set . then using lawvere s observation that a metric space can be viewed as an enriched category he defined the notion of the euler characteristic or cardinality of a finite metric space it was decided that these terms were already too overburdened , so we renamed the concept the magnitude of a metric space . this has the following definition . for @xmath6 a finite metric space , a _ weighting _ on @xmath6 is an assignment of a number @xmath37 to each point @xmath38 such that the _ weight equation _ is satisfied for each point @xmath15 : @xmath39 if a weighting exists for @xmath6 then the magnitude @xmath40 is defined to be the sum of the weights : @xmath41 note the following two things about the definition . firstly , the weights can be negative . secondly , if more than one weighting exists then the magnitude is independent of the choice of weighting ( a proof appears below in the more general measure theoretic setting ) . if you require a little intuition then one way of thinking about the weight equations is as follows . each point can be thought of as an organism , with the weight being the amount of ` heat ' being generated by that point ( a negative weight denoting some sort of refrigeration ) . each organism experiences the heat generated by the other organisms in a fashion which decays exponentially with the distance . the organisms each wish to experience unit heat and this is what is expressed by the weight equation . we can try to extend the idea of magnitude to non - finite metric spaces . two approaches present themselves . * use finite approximation to the metric space . * use a measure in the definition of a weighting . the first of these approaches was explored in @xcite and @xcite . in this approach a metric space @xmath6 is approximated by taking a sequence @xmath42 of finite subsets of @xmath6 , which converges to @xmath6 in the hausdorff topology , so @xmath43 as @xmath44 . one can then take the corresponding sequence of magnitudes @xmath45 and hope that they tend to a limit , and furthermore hope that the limit is independent of the choice of approximating sequence . this seems to work in many cases such as compact subsets of euclidean space . this approach has the advantage that it can be implemented on a computer to obtain approximations to the magnitude of some spaces as in @xcite . the second approach is the one that will be used in this paper . this was independently suggested to me by tom leinster and david speyer ( my own thoughts on the matter lay more in realm of currents and differential forms ) . the idea is to generalize the definition of magnitude using measures in the following way . if @xmath6 is a metric space then a _ weight measure _ on @xmath6 is a signed measure @xmath46 on @xmath6 such that for all @xmath15 @xmath47 if a weight measure @xmath46 exists then @xmath40 , the _ magnitude _ of @xmath6 , is defined to be the mass of @xmath6 with respect to @xmath46 : @xmath48 note that the magnitude , if it exists , is independent of the choice of weight measure . suppose that @xmath46 and @xmath49 are weight measures then @xmath50 it is worth giving a brief mention here to some connections between measure theory and riemannian geometry ( for more details see , for example , @xcite ) . if @xmath17 is an orientable riemannian manifold then there is a volume form @xmath51 , well - defined up to sign the choice of sign corresponding to the choice of orientation . the volume form is defined by smoothly picking an ordered orthonormal basis @xmath52 for each tangent space , taking the dual basis @xmath53 and defining @xmath54 . integrating with respect to this volume form does not quite give a measure because of the sign ambiguity think how on the real line we have @xmath55 . however , there is an associated measure @xmath56 , called the density , which obviates the sign problem . in the case of euclidean space this density is the usual lebesgue measure . as no confusion should occur , we will not distinguish between the volume form and the associated density . if @xmath17 is an orientable @xmath0-dimensional submanifold of euclidean space @xmath18 then there are two obvious measures that one might consider . firstly the riemannian metric on @xmath18 can be pulled back to @xmath17 to make it into an orientable riemannian manifold so we can use the associated density . alternatively , one could use the @xmath0-dimensional hausdorff measure on @xmath18 restricted to @xmath17 . provided that the hausdorff measure is normalized so that the @xmath0-dimensional cube @xmath57^n$ ] has unit measure , these two measures on @xmath17 coincide . one of the key tools of this paper is the following generalization of speyer s theorem on the magnitude of finite homogeneous spaces given in @xcite . recall that a homogeneous space is one for which there is a transitive group of isometries , or , in other words , one in which all of the points look the same . [ thm : speyer ] suppose that @xmath6 is a homogeneous metric space and @xmath14 is an invariant metric on @xmath6 then @xmath58 is independent of @xmath59 . if this quantity is non - zero and finite then a weight measure on @xmath6 is given by @xmath60 thus the magnitude is given by @xmath61 this follows immediately from the definition of magnitude . it is worth pointing out here that every _ compact _ homogeneous space has a unique - up - to - scaling , invariant , positive measure , thus has a well - defined magnitude . aside from the calculations for the interval and cantor set given below , this theorem is the only way i currently know for calculating the magnitude of non - finite metric spaces . we can now look at the spaces whose cardinality was defined using finite approximations in @xcite , and show that the same results are obtained using weight measures . these spaces are closed straight line segments , circles and cantor sets . we consider first @xmath62 the length @xmath63 closed line segment . the following was essentially told to me by tom leinster and david speyer independently . [ thm : intervalweightmeasure ] let @xmath14 be the lebesgue measure on @xmath62 the length @xmath63 line segment and let @xmath64 and @xmath65 be the dirac delta measures supported at the two respective end points . then a weight measure on @xmath62 is given by @xmath66 . hence the magnitude is simply : @xmath67 it is just necessary to show that the given measure satisfies the weight equation for every point @xmath68 . we can consider @xmath62 as the subset @xmath69\subset { \mathbb{r}}$ ] where @xmath70 , then for every @xmath68 @xmath71^y_a-[e^{y - x}]_y^b\right ) = 1.\end{aligned}\ ] ] thus the magnitude is given by @xmath72 as required . now turn to the ternary cantor set . the following lemma , pointed out to me by jonathan jordan , will be rather useful in calculating the magnitude . it says that if we have a weight measure on a subset of the line which on some subinterval is a multiple of the lebesgue measure then if the subinterval is removed then a weight measure is obtained for the resulting space by putting simple dirac delta measures at the end points of the removed subinterval . [ lemma : measureonhole ] if @xmath6 is a subset of @xmath73 with @xmath46 a weight measure on @xmath6 with @xmath69\subset x$ ] and @xmath74}=\frac12\mu|_{[a , b]}$ ] , then a weight measure on @xmath75 is given by @xmath76 so the magnitude satisfies @xmath77 we need the easy - to - verify identity @xmath78 to obtain @xmath79 . thus for @xmath80 @xmath81\\ & = \int_{x\in x}e^{-d(x , y)}{\text{\rm d}}\nu-\int_{x\in [ a , b]}e^{-d(x , y)}\tfrac12{\text{\rm d}}\mu\\ & \qquad+\tfrac12\tanh\bigl(\tfrac{b - a}2\bigr ) \int_{x\in x\setminus ( a , b)}e^{-d(x , y)}({\text{\rm d}}\delta_a+{\text{\rm d}}\delta_b)\\ & = 1+\tfrac12[e^{y - b}-e^{y - a}]+\tfrac12\tanh\bigl(\tfrac{b - a}2\bigr)\bigl(e^{y - a}+e^{y - b}\bigr)\\ & = 1.\end{aligned}\ ] ] the expression for the weight follows immediately . this can now be repeatedly to obtain the magnitude of the cantor set , and this agrees with the answer in @xcite . note that there the cantor set was approximated below via finite subsets , and here it is approximated above by subsets of @xmath82 $ ] with non - zero measure . the length @xmath63 ternary cantor set @xmath83 has magnitude given by @xmath84 start by defining @xmath85 to be a length @xmath63 line interval which is parametrize by the interval @xmath57 $ ] with the points in the unit interval being identified with their ternary exansion , so that a point a third of the way along @xmath85 has expansion @xmath86 . the interval @xmath85 is given the standard weight measure as in theorem [ thm : intervalweightmeasure ] , @xmath87 , which will be called @xmath88 . define @xmath89 and its weight measure @xmath90 inductively by first removing from @xmath91 the open middle third of each maximal subinterval ; this means removing points having expansions @xmath92 of the form @xmath93 the length of each removed open interval will be @xmath94 . by lemma [ lemma : measureonhole ] above we obtain a weight measure @xmath90 by restricting the weight measure @xmath95 on @xmath91 , except on the newly pruned end points where we have to put a dirac delta measure of mass @xmath96 . taking the limit , we obtain we obtain a measure @xmath46 on the ternary cantor set @xmath83 , this measure is the restriction of half the lebesgue measure plus a dirac delta on each point in the cantor set which has a finite ternary expansion the mass at a point with @xmath97 terms in the ternary expansion is @xmath96 , and there are @xmath98 such points , so the total mass of @xmath46 , i.e. , the magnitude is precisely @xmath99 as required . now finish this section by considering the circle . the cardinality of @xmath100 the circle of circumference @xmath63 is given by @xmath101 this follows immediately from the homogeneity theorem . in conclusion , the confidence expressed in @xcite when defining the magnitude in terms of subsets is now justified . in @xcite it was unclear that the answers given for the cantor set and circle were independent of the choice of approximating subsets ; but they agree with the answers given here which are independent of the choice of weight measure . this section is a brief exposition on intrinsic volumes for riemannian manifolds ; these invariants are also known as curvature measures , lipshitz - killing curvatures , generalized curvatures and quermassintegrale . i will mention weyl s tube formula and the connection with total scalar curvature and volumes of geodesic speres . this treatment is mainly obtained from the books @xcite and @xcite . this section also serves to fix some notation and conventions . in @xcite , following @xcite , we considered the intrinsic volumes @xmath102 as functions on polyconvex subsets of euclidean space @xmath103 . in fact the intrinsic volumes are defined for a much larger class of subsets of @xmath103 via the theory of curvature measures ( see @xcite ) . the intrinsic volumes can be defined for subsets which have an appropriate notion of unit normal bundle a so - called normal cycle . this class of subsets includes smooth submanifold with boundary and more generally sets of positive reach . in general , the basic properties of the intrinsic volumes are the following . additivity : : @xmath104 . homogeneity : : @xmath105 . normalization : : @xmath106^i)=1 $ ] . then @xmath107 is the usual @xmath108-dimensional volume of @xmath6 and @xmath109 is the euler characteristic of @xmath6 . for a closed smooth submanifold of @xmath103 there are many formulas for the intrinsic volumes . the simplest formulas are for when the submanifold in question is a hypersurface @xmath110 . in this case , at each point in @xmath111 there are @xmath0 principal curvatures @xmath112 , and the @xmath113th symmetrized curvatures @xmath114 is defined to be the @xmath113th - elementary symmetric function in the principal curvatures : @xmath115 . the intrinsic volumes are given , up to factor , by integrating these symmetrized curvatures over the submanifold . more precisely , recalling that @xmath116 denotes the volume of the unit @xmath113-sphere , for @xmath117 , @xmath118 from this one obtains , for instance , @xmath119 . this allows us to calculate the intrinsic volumes of @xmath120 the @xmath0-sphere of radius @xmath4 sitting in @xmath121 in the usual way . here all of the principal curvatures are the reciprocal of the radius , @xmath122 , so the symmetrized curvatures are given by @xmath123 and so , for @xmath117 , @xmath124 in the case of a submanifold of codimension greater than @xmath125 there are not principal curvatures , but the intrinsic volumes can be expressed in terms of the marginally more complicated lipschitz - killing curvatures . however it is still true that for @xmath126 a closed submanifold @xmath127 whenever @xmath22 is odd . from these curvature expressions , as weyl notes , a good undergraduate could derive the following tube formula . as an easy exercise , you may wish to verify this for the @xmath0-sphere using the formulas given above for the intrinsic volumes . suppose that @xmath128 is a closed , smooth submanifold of euclidean space , that @xmath129 is the set of points of distance at most @xmath130 from @xmath131 , and that @xmath132 is the volume of the unit @xmath97-ball , then @xmath133 weyl s key observation , however , was that the intrinsic volumes of @xmath6 can be expressed purely in terms of its _ intrinsic _ geometry we have already observed that @xmath134 and @xmath135 . so the intrinsic volumes only depend on the riemannian metric of @xmath6 and not on the way it is embedded , nor on the codimension in which it is embedded . for instance , the intrinsic volumes can be written as polynomial riemann curvature tensor . however , in this paper there is just one of these intrinsic volumes that we are interested in . after the @xmath0-volume , the next non - trivial intrinsic volume is proportional to the _ total scalar curvature _ : @xmath136 here @xmath137 is the _ scalar curvature _ which can be thought of as measuring the `` lack of stuff '' near @xmath138 . for instance , if you draw a circle on the surface of the earth then it has a circumference than a euclidean circle with the same radius this deficiency in the circumference is measured by the scalar curvature ( in the case of a surface the scalar curvature is twice the gaussian curvature ) . to make this precise , for @xmath38 and @xmath139 , let @xmath140 be the set of points at a distance of @xmath141 from @xmath138 ; for @xmath141 sufficiently small this will be an @xmath142-sphere , known as a _ geodesic sphere_. the scalar curvature @xmath143 at @xmath138 is uniquely determined by the second order term in the power series expansion of the volume of geodesic spheres centered at @xmath138 , in the following sense : @xmath144 it is in this way that the scalar curvature can be viewed as a measure of the `` lack of stuff '' near @xmath138 . the above formula will be used later in the calculation of sub - leading term in the asymptotics of the magnitude of homogeneous metric spaces . as an example , note that in an @xmath0-sphere of radius @xmath4 we have , for all @xmath138 , @xmath145 . so the scalar curvature is given by @xmath146 which you can check is consistent with the value calculated for @xmath147 above . this section consists of one of the main results of this paper which is an explicit formula for the magnitude of the @xmath0-sphere . the asymptotic behaviour is also considered and the euler characteristic is seen to be the constant term in the asymptotic expansion . we take @xmath148 the @xmath0-sphere with the standard metric of curvature @xmath149 ( which means that it isometrically embeds in @xmath29 with radius @xmath4 ) thought of as a homogeneous space . from speyer s homogeneous magnitude theorem ( theorem [ thm : speyer ] ) the magnitude is given by @xmath150 this can be calculated explicitly . [ thm : magnitudeofspheres ] for @xmath151 , the magnitude of the homogeneous @xmath0-sphere of radius @xmath4 is given by @xmath5 \dfrac{\pi r \bigl(\bigl(\frac{r}{n-1}\bigr)^2 + 1\bigr))\bigl(\bigl(\frac{r}{n-3}\bigr)^2 + 1\bigr ) \dots \bigl(\bigl(\frac{r}{2}\bigr)^2 + 1\bigr ) } { 1-e^{-\pi r}}\quad & \text{for $ n$ odd } , \end{cases}\ ] ] where for @xmath152 the numerator is interpreted as being @xmath153 . the idea is to find some explicit integrals to calculate magnitude , evaluate them to obtain @xmath154 and @xmath155 , show that the induction formula holds , @xmath156 and then apply induction . we first fix some point @xmath157 and define the coordinate @xmath158 on the @xmath0-sphere to be the distance from the point @xmath157 , so @xmath159 . the points at a distance @xmath158 from @xmath157 form an @xmath142-sphere @xmath160 of euclidean radius @xmath161 , see figure [ fig : localball ] ; @xmath162 thus , writing @xmath163 for the volume of the unit @xmath142-sphere , the volume of @xmath160 is given by @xmath164 . locally , as @xmath158 parametrizes distance , the volume form on @xmath165 is the product of @xmath166 and the volume form on @xmath160 . so @xmath167 where we made the substitution @xmath168 . this can be evaluated for the @xmath125-sphere and @xmath169-sphere : @xmath170_0^\pi } = \frac{\pi r}{1-e^{-\pi r}}\;;\\ \left|s^2_r\right| = \frac{\int_{0}^{\pi}\sin(r)\,{\text{\rm d}}r } { \int_{0}^{\pi } e^{-rr}\sin(r)\ , { \text{\rm d}}r } = \frac{2}{\text{im}\int_{0}^{\pi } e^{-rr}e^{ir}\ , { \text{\rm d}}r } = \frac{2(r^2 + 1)}{1+e^{-\pi r}}\;.\end{gathered}\ ] ] these will form the base cases for the induction . for the inductive step , write @xmath171 so that @xmath172 . we will show that @xmath173 from which it is immediate that @xmath174 , and the required result follows by induction . to prove the relation for @xmath175 we observe @xmath176^\pi_0+\int_{0}^{\pi}\left(-\sin(r)\right)\left(\tfrac1n \sin^{n}(r)\right){\text{\rm d}}r\\ & = k_n-0-\tfrac1n k_{n+2},\end{aligned}\ ] ] from which we deduce @xmath177 , as we wanted . to prove the relation for @xmath178 we perform integration by parts twice : @xmath179^\pi_0 -\int_{0}^{\pi } -\tfrac 1re^{-rr}(n+1)\cos(r)\sin^{n}(r)\ , { \text{\rm d}}r\\ & = 0 + \tfrac{(n+1)}{r}\left[-\tfrac 1r e^{-rr}\cos(r)\sin^{n}(r)\right]^\pi_0\\ & \qquad -\tfrac{(n+1)}{r}\int_{0}^{\pi } -\tfrac 1re^{-rr}\left(-\sin(r)\sin^{n}(r ) + n\cos(r)\cos(r)\sin^{n-1}(r)\right ) { \text{\rm d}}r\\ & = 0 + 0 + \tfrac{(n+1)}{r^2}\int_{0}^{\pi } e^{-rr}\left(-\sin^{n+1}(r ) + n\left(1-\sin^2(r)\right)\sin^{n-1}(r)\right ) { \text{\rm d}}r\\ & = -\tfrac{(n+1)^2}{r^2}i_{n+2 } + \tfrac{(n+1)n}{r^2}i_n.\end{aligned}\ ] ] from this we deduce that @xmath180 which suffices to complete the inductive step and hence we complete the proof . the first thing to note from this is that the formula for @xmath181 in the above theorem also works for @xmath182 provided that this notation is interpreted correctly . if @xmath182 is taken to mean the two points on the equator of @xmath183 the radius @xmath4 circle , then @xmath184 is the metric space consisting of two points a distance @xmath153 apart , and then it is immediate that the magnitude is given by @xmath185 the next thing to note is the @xmath31 as @xmath32 , so effectively , the @xmath0-sphere looks more and more like a point as it gets smaller and smaller . a third thing to note is that the two formulae can be combined into a single formula which clearly demonstrates the roots of the numerator lying at alternate integral points on the imaginary axis : @xmath186 where @xmath187 and @xmath188 is a constant which ensures that @xmath31 as @xmath32 . the roots of the numerator are also roots of the denominator , but i have no idea of the relevance of this . a fourth things worthy of note is that the denominators very quickly converge to @xmath125 . this means that the behaviour for @xmath4 away from zero , say for @xmath189 , is dominated by the numerator . we can define @xmath190 , for @xmath191 , to be the numerator : @xmath192 where the empty product is interpreted as @xmath125 for @xmath193 and @xmath152 . this polynomial in @xmath194 gives us the asymptotic behaviour of @xmath181 in the sense that @xmath195 as @xmath196 . previously for the closure of an open subset of euclidean space the asymptotics of the magnitude were conjectured to be controlled by certain intrinsic volumes . in the case of the spheres the leading term and constant term can be seen to behave in the same way . the leading term of @xmath197 is proportional to the volume of the @xmath0-sphere @xmath165 and the constant term is equal to its euler characteristic : writing @xmath10 for the volume of the unit @xmath0-ball , @xmath198 to deal with the euler characteristic first , note that the euler characteristic of a sphere just depends on the parity of the dimension : @xmath199 and @xmath200 . the constant term in the polynomial @xmath190 is immediately seen to have exactly this property . that the leading term has the required form can be proved inductively using the inductive formula @xmath201 together with the following relations ( see , for example , @xcite ) for @xmath10 the volume of the unit @xmath0-ball and @xmath202 the volume of the unit @xmath0-sphere : @xmath203 assuming that the result is true for @xmath204 and writing @xmath205 for the top coefficient of a polynomial , we have @xmath206 the result follows by induction after checking the base cases , @xmath152 and @xmath35 , using @xmath207 , @xmath208 , @xmath209 and @xmath210 . we will see in the next section that the leading term in the asymptotic expansion of the magnitude of any homogeneous is of this form . in @xcite and @xcite we gave a precise conjecture for the asymptotic behaviour of the magnitude of ` nice ' subsets of euclidean space and this involved the being asymptotically the same as the ` penguin valuation ' defined for @xmath6 a subset of euclidean space as @xmath211 we are here considering the sphere with its intrinsic metric so it is not a metric subspace of euclidean space , but the leading term and constant term we obtain are the same as for in the penguin valuation . however , the subdominant term is different . we would expect @xmath212 but we get the following : it can be proved inductively as the formula for the leading term was in the above theorem , it is also a consequence of the theorem in the next section on general homogeneous riemannian manifolds . the first non - trivial , subdominant term of @xmath197 is given by @xmath213 these results are compared with analogous ones for the magnitude of the @xmath0-sphere with the subspace metric in section 5 . in this section we will look at the three leading terms in the asymptotics of the magnitude of a homogeneous riemannian manifold , as the manifold is scaled up . the leading term will be seen to be proportional to the volume of the manifold , the second - leading term will be seen to be zero and the third - leading term will be seen to be proportional to its total scalar curvature , or in other words the top two leading terms are proportional to the top two intrinsic volumes . in particular , for a homogeneous surface @xmath214 , the magnitude is asymptotically @xmath215 , and this is precisely the evaluation of the functional from @xcite . by speyer s homogeneous magnitude theorem ( theorem [ thm : speyer ] ) we know that the magnitude of a homogeneous riemannian manifold @xmath6 is given by @xmath216 if we look at this as we scale up the space we should consider @xmath217 we wish to consider the asymptotics of the denominator @xmath218 . the relevant piece of asymptotic analysis we will require is a version of watson s lemma ( see for example @xcite ) . if @xmath219\to [ 0,\infty)$ ] is a bounded function which , for some @xmath220 , has an expansion @xmath221 then @xmath222 the idea behind watson s lemma is that as a function of @xmath223 $ ] , when @xmath224 gets larger and larger @xmath225 looks more and more like a function supported on a smaller and smaller neighbourhood of @xmath226 , so the integral will , up to exponentially small corrections , only depend on the germ of @xmath227 near @xmath226 . this can now be used to identify the leading asymptotics of the magnitude for homogeneous manifolds . [ thm : homogeneousasymptotics ] if @xmath6 is an @xmath0-dimensional homogeneous riemannian manifold and @xmath7 denotes total scalar curvature then as @xmath6 is scaled up the asymptotics of the magnitude are as follows : @xmath228 the idea is to pick a point @xmath229 and define the coordinate @xmath230 . let @xmath231 be the set of points at a distance @xmath141 from @xmath157 , then if @xmath232 denotes the scalar curvature of @xmath6 , which will be the same at every point , then we know @xmath233 because @xmath141 parametrizes distance , locally the volume form splits as @xmath234 . we can write @xmath235 for the diameter of @xmath6 . then we find , using watson s lemma appropriately , @xmath236 where in the last equality we have used the fact that @xmath237 . thus we get @xmath238 this gives the first description of the asymptotics , to get the second it is just necessary to recall that @xmath239 that @xmath240 and that @xmath241 . this theorem immediately gives the asymptotics of the magnitude of a homogeneous surface , by taking @xmath35 in the above and noting that @xmath242 is precisely the euler characteristic @xmath243 . for @xmath12 a homogeneous riemannian surface , the magnitude is asymptotically given in terms of the area and the euler characteristic by @xmath244 it is interesting to note that in the examples we know of , namely for spheres and surfaces , the constant term in the asymptotic expansion of the magnitude is precisely the euler characteristic . in this section we look at the metric on the @xmath0-sphere obtained by thinking of it as sub - metric space of @xmath29 via the standard embedding . an explicit formula is given for the magnitude of the @xmath169-sphere with this metric , whereas for @xmath245 just the leading few terms of the asymptotic expansion are given . this is sufficient to see that the leading term is the same as for the magnitude of the sphere with the intrinsic metric but that the lower order terms are not the same . in @xcite we showed that equipping the radius @xmath4 _ circle _ with the intrinsic metric and the subspace metric gave rise to different magnitudes , but that asymptotically the magnitudes differed by a vanishingly small piece : @xmath246 with the techniques so far developed in this paper it is not difficult to calculate the magnitude of the @xmath169 sphere with these two metrics and see that only the leading term in the asymptotic expansion are the same . for the @xmath169-sphere of radius @xmath4 with the subspace metric the magnitude is given by @xmath247 thus the sub - leading terms of the asymptotic behaviour are different for the two different metrics . the intrinsic metric result is in theorem [ thm : magnitudeofspheres ] , so it is just necessary to do the calculation for the subspace metric . this is analogous to the calculation for the intrinsic metric . by the homogeneous magnitude theorem @xmath248 using the substitution @xmath249 gives the following evaluation of the integral : @xmath250 the result follows immediately . whilst i do nt know how to get a general closed form for the magnitude of the sphere with the subspace metric , using the techniques of the previous section it is straight forward to calculate the first few terms in the asymptotic expansion of the magnitude and compare them with those for the intrinsic metric . for the @xmath0-sphere of radius @xmath4 with the subspace metric the magnitude is given asymptotically by @xmath251 for the @xmath0-sphere of radius @xmath4 with the intrinsic metric the magnitude is given asymptotically by @xmath252 the intrinsic metric result follows from theorem [ thm : homogeneousasymptotics ] together with the fact from section [ section : intrinsicvolumes ] that the scalar curvature of the @xmath0-sphere with radius @xmath4 is @xmath253 . for the subspace metric result , proceed as usual by using speyer s homogeneous magnitude theorem and writing @xmath202 for the volume of the unit @xmath0-sphere . @xmath254 again making the substitution @xmath249 whilst noting that @xmath255 and @xmath256 the integral can be manipulated as follows : @xmath257 substituting this back into the above formula gives the required result on recalling that @xmath258 . one can think of this result as saying that just as the intrinsic and subspace metrics are infinitesimally the same in a certain sense though i do nt know the correct way to formalize that so the corresponding magnitudes are asymptotically the same in a certain sense , namely @xmath259 as @xmath33 , although they are not the same in a stronger sense because , in general , @xmath260 as @xmath33 . h. donnelly , _ heat equation and the volume of tubes _ , inventiones mathematicae ( 1975 ) * 29 * ( 3 ) pp . 239 - 243 . a. gray , tubes , birkauser ( 2004 ) . d. klain and g .- c . rota , introduction to geometric probability , cambridge university press ( 1997 ) . t. leinster , _ a maximum entropy theorem with applications to the measurement of biodiversity _ , preprint http://arxiv.org/abs/0910.0906[arxiv:0910.0906 ] . t. leinster , _ an adventure in analysis _ , blog post at the @xmath0-category caf , november 2009 , http://golem.ph.utexas.edu/category/2009/11/an_adventure_in_analysis.html t. leinster and s. willerton , _ on the asymptotic magnitude of metric spaces _ , preprint http://arxiv.org/abs/0908.1582[arxiv:0908.1582 ] . p. d. miller , _ applied asymptotic analysis _ , graduate studies in mathematics 75 , american mathematical society , 2006 . morvan , generalized curvatures , springer - verlag ( 2008 ) . s. schanuel , _ what is the length of a potato ? an introduction to geometric measure theory _ , in categories in continuum physics ( buffalo , n.y . , 1982 ) , 118126 , lecture notes in mathematics , * 1174 * , springer , berlin , 1986 . a. solow and s. polasky , _ measuring biological diversity _ , environmental and ecological statistics ( 1994 ) * 1 * ( 2 ) 95103 . s. willerton , _ heuristic and computer calculations of the magnitude of metric spaces _ , preprint http://arxiv.org/abs/0910.5500[arxiv:0910.5500 ] . a. weyl , _ on the volume of tubes _ , american journal of mathematics ( 1939 ) * 61 * ( 2 ) pp . 461472

active galactic nuclei ( agns ) produce enormous luminosities in extremely compact volumes . it was found some local galaxies show similarities to those of bright agns , but with relatively weaker broad - line emission and multiwavelength radiative cores ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? these low - luminosity active galactic nuclei ( llagns ) are very common in the local universe , which may be the scaled - down luminosity version of bright agns ( see * ? ? ? * for a recent review ) . actually , the agn may only spend a small fraction of its lifetime in the highly luminous , qso - like phase , and much more time is spent in a weakly accreting state ( llagns or normal galaxies , e.g. , * ? ? ? both bright agns and llagns are believed to be powered by the matter falling onto a super massive black hole ( bh ) . the observational properties of llagns are quite different from those of bright qsos , which are thought to be driven by different accretion modes . the optical / uv bumps observed in qsos can be naturally interpreted as multi - temperature blackbody emission from a cold , optically thick , and geometrically thin standard disk ( ssd , * ? ? ? however , most of llagns lack the evident optical / uv bump as that of bright agns . the possible physical reason is that the standard thin accretion disk is absent in these llagns . a hot , optically thin , geometrically thick radiatively inefficient accretion flow model has been developed in the last several decades ( riaf , an ` updated ' version of the original advection dominated accretion flow model , adaf ; e.g. , ichimaru 1977 ; narayan & yi 1994 , 1995 ; abramowicz et al . 1995 ; see kato et al . 2008 ; narayan & mcclintock 2008 ; and yuan & narayan 2013 for reviews ) . the riaf model can successfully explain most observational features of nearby llagns ( e.g. , quataert et al . 1999 ; cao & rawlings 2004 ; nemmen et al . 2006 ; wu et al . 2007 ; wu & gu 2008 ; gu & cao 2009 ; yuan et al . 2009a ; yuan et al . 2009b ; xu & cao 2009 ; ho 2009 ; cao 2010 ; yu et al . 2011 ; nemmen et al . 2012 ; see yuan 2007 and ho 2008 for recent reviews ) . the agns are traditionally divided into radio loud ( rl ) and radio quiet ( rq ) , and the origin for this dichotomy is still a matter of debate . the distinction between the rl and rq objects is normally based on a radio - loudness parameter , which was defined as the ratio of the monochromatic flux density at 5 ghz and optical b band at 4400 @xmath1 ( @xmath2 ) . the loudness @xmath3 is usually taken as the division between rq and rl agns ( particularly in qso studies , e.g. , * ? ? ? @xcite introduced a new definition of the radio loudness parameter by comparing the 5 ghz radio luminosity to the 2 - 10 kev luminosity , @xmath4 , and proposed that @xmath5 would be the barrier separating rl and rq agns , which roughly corresponding to @xmath3 . the use of x - ray luminosity with respect the optical one should largely avoid extinction problems which normally occur in the optical band . however , most of the traditionally rq llagns become rl according to the criterion @xmath3 or @xmath5 if using their nuclear emission at radio and optical / x - ray waveband ( e.g. , * ? ? ? * ; * ? ? ? @xcite redefined the boundary for the rl and rq llagns as @xmath6 based on a sample of low - luminosity seyferts and fanaroff - riley type i radio galaxies ( fr is ) . in the last decade , there is great progress in estimating the bh mass in both normal galaxies and agns ( e.g. , * ? ? ? * ; * ? ? ? the measurements of bh mass help us to further understand the issue of radio loudness . @xcite found that the loudness parameter increases with decreasing of the eddington ratio . @xcite investigated the radio loudness of a total 199 sources which include broad - line radio galaxies ( blrgs ) , rl qsos , seyferts , low - ionization nuclear emission - line region galaxies ( liners ) , fr is , and palomar - green qsos ( pgqs ) , and found that there are two distinct , approximately parallel tracks on radio - loudness@xmath0eddington - ratio plane . they further proposed a quantitative definition for the radio - loudness parameter that is dependent on the eddington ratio ( see their equation 5 ) , where the rl sources include blrgs , rl qsos , fr is , while rq sources comprise seyferts , liners and pgqs ( hereafter we use this criteria to distinguish the rl / rq ) . the radio emission of rl agns is believed to originate in the relativistic jets , where many large - scale radio jets have been observed directly . compared with rl agns , the origin of nuclear radio emission from rq objects is still not well understood , since that rq agns are much fainter radio sources , and most of them only show a compact radio core . the radio variability of rq qsos confined the radio emission to @xmath70.1 pc ( e.g. , * ? ? ? * ) , and the high - resolution vlbi observations show that the region of the radio emission in rq llagns should be @xmath8 to @xmath9 pc ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? several possibilities have been proposed to explain the radio emission in these rq agns . the first explanation is that rq agns are scaled - down version of the rl agns where exist a weak , small - scale jet ( e.g. , * ? ? ? * ; * ? ? ? a small fraction of rq objects show core - jet or linear structures at pc - scale or sub - pc scale , which do support the weak - jet scenario ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the accretion - jet model is frequently used to model the multiwavelength spectral energy distribution ( sed ) of agns ( particularly in llagns , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? however , this scenario can not answer the bimodality of rl / rq distribution for agns ( e.g , * ? ? ? * ; * ? ? ? the second possibility is that the radio emission of the rq agns come from the hot optically thin plasma in the disk winds , where the plasma is completely ionized and has a density high enough for bremsstrahlung emission that make a significant contribution to the observed compact radio core ( e.g. , * ? ? ? * ; * ? ? ? * ) . in this model the mass - loss rate in the winds are significant and the observed luminosities of radio emission from quasars imply that they should accrete at super - eddington rates . however , most of rq agns are accreting at sub - eddington rates , where the radio core emission should not dominantly come from the disk winds . the third possibility is that the radio emission also come from the accretion flows ( e.g. , riaf or disk corona ) . @xcite found that the tight correlation between the radio and x - ray luminosities in rq qsos is similar to that of active stars ( @xmath10 ) . they , therefore , proposed that both the radio and x - ray emission from the nuclei of rq agns may dominated by the magnetic reconnection heated corona . @xcite found that the radio emission of most nearby llagns is higher than that predicted by radiation from the pure thermal electrons in the riaf , and their radio emission should have other origin ( e.g. , nonthermal electrons in riaf or jet ) . the radio spectrum of sgr a * , a supermassive bh in our galaxy , consists of two components , which dominate below and above 86 ghz , respectively . @xcite found that the component above 86 ghz can be well explained by the thermal electrons from the riaf , while low - frequency radio spectrum can be explained if there exist a small fraction of nonthermal electrons in the riaf ( see also * ? ? ? * ) . it is natural that both the thermal and nonthermal electrons exist in the hot plasma ( riaf or disk corona ) , since that the turbulence , magnetic reconnection , and weak shocks can accelerate electrons and generate a nonthermal tail at high energies in the distribution function of thermal electrons . the sgr a * provide an excellent observational evidence for the existence of hybrid of thermal and nonthermal electrons in accretion flows ( e.g. , riaf ) , since that no evident radio jet was observed up to now and , therefore , its radio emission may be dominated by other mechanisms . it is interesting to note that there is also independent observational evidence for nonthermal electrons in accretion flows . @xcite reported a high - energy tail in the hard state of the x - ray binary , cygnus x-1 , extending from 50 kev to @xmath115 mev . the data at mev energies , collected with the comptel instrument of the compton gamma - ray observatory , can be explained by a power - law distribution of nonthermal electrons in the riaf / corona ( e.g. , * ? ? ? * ) . in this work , we try to explore whether the nonthermal elctrons in riaf can explain the radio emission of nearby llagns , which always have only compact radio cores . lccccccccl name & @xmath12 & @xmath13 & @xmath14 & telescope & @xmath15 & telescope & @xmath16 & @xmath17 & refs . + & mpc & @xmath18 & @xmath19 & & @xmath19 & & & & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) + ngc 1097 & @xmath20 & 8.10 & @xmath21 & vla & @xmath22 & c & 1.1@xmath23 & 0.03% & 1,2,3 + ngc 2787 & @xmath24 & 8.07 & @xmath25 & vlba & @xmath26 & c & 3.0@xmath27 & 10.0% & 4,4,5 + ngc 3147 & @xmath28 & 8.60 & @xmath29 & vlba & @xmath30 & c & 5.3@xmath23 & 0.5% & 4,4,6 + ngc 3169 & @xmath31 & 7.79 & @xmath32 & vlba & @xmath33 & c & 6.0@xmath23 & 3.0% & 4,4,6 + ngc 3226 & @xmath34 & 8.07 & @xmath25 & vlba & @xmath35 & c & 1.1@xmath23 & 3.0% & 4,4,6 + ngc 3227 & @xmath36 & 7.44 & @xmath37 & vlba & @xmath38 & x & 2.1@xmath39 & 0.4% & 4,4,7 + ngc 3414 & @xmath40 & 8.42 & @xmath41 & merlin & @xmath42 & c & 4.2@xmath27 & 0.3% & 8,8,5 + ngc 3718 & @xmath43 & 7.85 & @xmath44 & vlba & @xmath45 & c & 1.4@xmath23 & 3.0% & 4,4,9 + ngc 3941 & @xmath46 & 7.55 & @xmath47 & vla & @xmath48 & x & 5.5@xmath27 & 0.5% & 10,10,7 + ngc 3998 & @xmath49 & 8.72 & @xmath50 & vlba & @xmath51 & x & 2.3@xmath23 & 3.0% & 4,4,11 + ngc 4138 & @xmath43 & 7.51 & @xmath52 & vla & @xmath53 & x & 1.5@xmath39 & 0.1% & 10,10,7 + ngc 4143 & @xmath43 & 8.25 & @xmath54 & vlba & @xmath55 & c & 5.0@xmath27 & 3.0% & 4,4,6 + ngc 4168 & @xmath56 & 7.98 & @xmath57 & vlba & @xmath58 & x & 3.8@xmath27 & 4.0% & 4,4,12 + ngc 4203 & @xmath59 & 7.77 & @xmath60 & vlba & @xmath61 & c & 6.0@xmath27 & 5.0% & 4,4,13 + ngc 4477 & @xmath56 & 8.00 & @xmath62 & vla & @xmath63 & x & 4.95@xmath27 & 0.04% & 10,10,7 + ngc 4565 & @xmath59 & 7.46 & @xmath64 & vlba & @xmath65 & c & 7.0@xmath27 & 7.0% & 4,4,6 + ngc 4594 & @xmath66 & 8.46 & @xmath67 & vla & @xmath68 & c & 3.2@xmath27 & 15.0% & 14,15,5 + ngc 4639 & @xmath56 & 6.68 & @xmath69 & vla & @xmath70 & c & 4.5@xmath23 & 4.0% & 10,10,13 + ngc 4698 & @xmath56 & 7.41 & @xmath71 & vla & @xmath72 & c & 3.5@xmath27 & 3.0% & 10,10,5 + ngc 4736 & @xmath73 & 6.98 & @xmath74 & vla & @xmath75 & c & 5.7@xmath27 & 3.0% & 1,4,16 + to examine whether the radio emission of nearby llagns with only compact cores can be explained by the nonthermal electrons in the riaf , we search the literatures for high - resolution radio and x - ray data to ensure that their emission being from the nuclei of sources . in the radio band , we only selected the sources that have been observed by high - resolution radio telescopes ( e.g. , vla , merlin , vlba and vlbi ) , which are mainly chosen from @xcite , @xcite and @xcite , where all these works tried to give a complete radio imaging survey of all nearby llagns given in palomar spectroscopic survey @xcite . for purpose of our study , we exclude the sources that observed with radio jet or even linear radio structures where the radio emission may mainly originate in the jets / outflows . the 5 ghz nuclear radio luminosities are shown in table ( 1 ) , where two sources observed at other waveband ( ngc 1097 at 8.4 ghz and ngc 4736 at 15 ghz ) was converted to 5 ghz by assuming @xmath76 as found in @xcite for llagns . in the x - ray waveband , we only selected the sources have compact x - ray cores , which are observed by high - resolution telescopes of @xmath77 and/or @xmath78 . several sources with the high - resolution optical data from @xmath79 are also included in building the seds ( * ? ? ? * and references therein ) . @xcite gave lower limits for the intrinsic agn optical emission for a sample of liners , which were constrained from their optical variabilities . we also include the lower limits of the intrinsic agn optical emission for three sources ( ngc 3998 , ngc 4594 and ngc 4736 , * ? ? ? we estimate the bh mass from the velocity dispersion @xmath80 of the host bulges for llagns in our sample using the @xmath81 relation given by @xcite . most of velocity dispersions are selected from the hyperleda database with one exception , ngc 1097 , from @xcite . we find that the velocity dispersion of most llagns in our sample is consistent with that reported in @xcite . our sample include 20 llagns ( see table 1 ) . in this work , we consider the riaf model proposed by @xcite , which can be considered as an updated version of the original adaf ( e.g. , * ? ? ? * ) , where both outflows and the possible existence of nonthermal electrons are considered . here , we briefly summarize the model as follows . the more accurate global structure and dynamics of the accretion flow is calculated numerically to obtain the ion and electron temperature , density at each radius . in particular , we employ the approach suggested by @xcite for calculating the structure of the riaf in general relativistic frame , which allows us to calculate the structure of the accretion flow surrounding either a spinning or a nonspinning bh . in this work , we calculate the global structure of the accretion flows surrounding massive schwarzschild black holes . in stead of using a constant accretion rate , we assume that the accretion rate is a function of radius , e.g. , @xmath82 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , where @xmath83 is the outer radius of the riaf and @xmath84 is the accretion rate at @xmath83 . the global structure of the riaf can be calculated with proper outer boundaries , if the parameters @xmath85 , @xmath86 , @xmath87 , and @xmath88 are specified ( see * ? ? ? * for more details ) . all radiation processes ( synchrotron , bremsstrahlung and compton scattering ) are included self - consistently in the calculations of the riaf structure . the parameter @xmath89 is dimensionless accretion rate , where the eddington accretion rate is defined as @xmath90 . for the viscosity parameter @xmath86 and magnetic parameter @xmath91 [ defined as the ratio of the gas to the total pressure ( sum of gas and magnetic pressure ) ] , we adopt typical values of @xmath92 and @xmath93 , which are widely used in riaf models . there is obviously a degeneracy between @xmath16 and @xmath94 when the accretion rate at the innermost region of the riaf is concerned . we adopt @xmath95 in this work , which is well constrained from the observation of our galactic center black hole , sgr a * ( * ? ? ? * but see also @xcite for slightly higher value of @xmath96 from simulations ) . we fix the outer boundary at @xmath97 ( @xmath98 is gravitational radius ) , which is roughly consistent with the prediction of the disk - corona evaporation model for the llagns with very low eddington ratios ( e.g. , @xmath99 , * ? ? ? the most poorly constrained parameter is @xmath88 , which describes the fraction of the turbulent dissipation that directly heats the electrons in the flow . @xcite found that the parameter @xmath88 may be in range of @xmath100 based on the simulations , depending on the model details . in this work , we adopt @xmath101 because we find that the x - ray spectra of most llagns in our sample can be well fitted with this fixed @xmath88 value . following @xcite , we assume that the injected energy in the nonthermal electrons is equal to a fraction @xmath17 of the energy in thermal electrons , where we take @xmath17 to be independent of the radius . the thermal electrons are assumed to have relativistic maxwell - boltzmann distribution , while the nonthermal electrons are assumed to be the power - law tail of the thermal electrons , which can be described by the parameter @xmath102 [ @xmath103 , where @xmath104 is the electron lorentz factor ] . the number density of the nonthermal electrons can be obtained if @xmath17 is given . @xcite proposed that the parameters , @xmath102 and @xmath17 , for describing the nonthermal electrons are degenerate in the radio waveband in our model . therefore , we adopt a fiducial value of @xmath105 , and allow the parameter @xmath17 to be free . after determining the distribution of thermal and nonthermal electrons , we calculate their radiation . the synchrotron emission from both thermal and nonthermal electrons are calculated ( e.g. , * ? ? ? the comptonization of seed photons from both thermal and nonthermal electrons are considered , which was calculated by the method proposed by @xcite . in spectral calculations , the gravitational redshift effect is considered , while the relativistic optics near the black hole is neglected . more details about spectrum calculation can be found in ( * ? ? ? * and references therein ) . in summary , our model have only two free parameters , @xmath16 and @xmath17 in fitting the sed of llagns . we present the typical spectrum of the riaf with a population of hybrid electrons in figure [ f1 ] ( top panel ) , where the short - dashed , long - dashed and solid lines represent the emission from the thermal electrons , nonthermal electrons and sum of them respectively , where typical parameters @xmath106 , @xmath107 , and @xmath108 are adopted for given outer radius @xmath109 . in this model , the x - ray emission mainly come from the comptonization of seed photons that produced by both thermal electrons and nonthermal electrons . we find that the low - frequency radio emission is dominated by the self - absorbed synchrotron emission from the nonthermal electrons , which is normally 2 - 3 order of magnitude higher than that from thermal electrons in riaf . our results are similar for other parameters if the typical parameter @xmath110 several percent is adopted as that used in sgr a * @xcite . the high - energy x - ray emission dominantly come from the inner region of the accretion flows ( e.g. , within several tens gravitational radii ) . however , we find the region of the radio emission is much larger . the dotted lines in figure [ f1 ] ( top panel ) show the spectrum from the nonthermal electrons with outer boundary @xmath111 , @xmath112 and @xmath113 respectively for above given other parameters . we find that the radiation region of the low - frequency radio emission at several ghz can up to several thousands gravitational radii , where the radio emission at @xmath114 is negligible . the bottom panel of figure [ f1 ] show the spectrum of nonthermal electrons with @xmath115 , @xmath116 ( solid line ) ; @xmath105 , @xmath108 ( dashed line ) and @xmath117 , @xmath118 ( dotted line ) , where the other parameters keep the same as above . it can be found that @xmath102 and @xmath17 are degenerate for radio emission ( e.g. , @xmath119 ghz , see also * ? ? ? therefore , it should be impossible to determine @xmath102 and @xmath17 from the observational low - frequency radio emission alone . , @xmath107 , @xmath108 and @xmath105 at given @xmath120 , where the short - dashed , long - dashed and solid lines are the radiation from the thermal electrons , nonthermal electrons and sum of them respectively . the dotted lines represent the radiation of the nonthermal electrons in the riaf with different outer radius at @xmath121 , @xmath112 , and @xmath113 ( from top to bottom ) respectively . the bottom panel show the spectrum from the nonthermal electrons with @xmath115 , @xmath116 ( solid line ) ; @xmath105 , @xmath108 ( dashed line ) , and @xmath117 , @xmath118 ( dotted line ) respectively , where the other parameters are the same as above . , width=340 ] our calculations show that the x - ray emission from the riaf is almost proportional to the bh mass for a given @xmath85 , and it scales as @xmath122 with @xmath123 for a given bh mass ( see also * ? ? ? * ; * ? ? ? * ) . in this work , we further explore the possible relation between the radio luminosity and bh mass / accretion rate for riaf model with both thermal and nonthermal electrons , where the radio emission dominantly originate in the nonthermal electrons . figure [ f2 ] ( top panel ) shows the relation between radio luminosity and bh mass for @xmath124 and @xmath125 with @xmath108 respectively , where we find that the radio emission is roughly proportional to @xmath126 ( solid lines ) . the bottom panel of figure [ f2 ] shows the relation between the radio luminosity and dimensionless accretion rate for given bh masses @xmath127 and @xmath128 respectively , where we find that the radio emission is roughly proportional to @xmath129 with @xmath130 for both cases . to explore the radio emission for a sample of llagns with different bh masses as a whole , we define a mass - corrected radio - loudness as @xmath131 , according to the dependence of the radio and x - ray emission on the bh mass in our model , where @xmath132 . and @xmath133 for two given accretion rates @xmath124 and @xmath125 with @xmath108 , where the solid points represent @xmath134 , @xmath135 , @xmath136 , @xmath137 , and @xmath138 respectively , and the solid lines are their best fittings . the bottom panel present the relation between @xmath139 and @xmath16 for two given bh masses @xmath140 and @xmath141 , where the solid points represent @xmath16=@xmath142 , @xmath143 , @xmath144 , @xmath145 , and @xmath146 respectively , and solid lines are their best fittings . , width=340 ] we present the relation between the mass - corrected loudness , @xmath147 , and the eddington ratio , @xmath148 , for all llagns in figure [ f3 ] , where both quantities roughly do not depend on the bh mass . it can be found that there is an anti - correlation between @xmath147 and @xmath148 for llagns . we further demonstrate the theoretical relation between @xmath147 and @xmath148 in our model for different values of @xmath17 from 0 to 10% at given @xmath149 ( see figure [ f3 ] ) . it can be found that our theoretical model also predict that @xmath147 is anti - correlate with @xmath148 as observational features of llagns . this is because the loudness @xmath147 of our model scales @xmath150 , while @xmath151 . we find that the radio emission of most llagns can be explained by our model if @xmath152% of the electron energy is ejected into the nonthermal electrons ( see figure [ f3 ] ) . however , pure riaf without power - law electrons ( @xmath153 , see figure [ f3 ] ) always underpredicts the radio emission by several orders of magnitude compared the observations . , and the eddington - scaled x - ray luminosity , @xmath148 , where @xmath4 . the solid points represent the llagns in our sample , and the solid lines represent the model prediction with the parameter @xmath17= 10% , 1% , 0.1% , 0.01% , and 0% ( from top to bottom ) respectively.,width=340 ] we further fit the radio to x - ray spectrum of all sources in our sample using the riaf model with hybrid electrons . the results are shown in figures [ f4]-[f5 ] , and the adopted parameters , @xmath16 and @xmath17 , of each source are listed in table ( 1 ) . with exception of the optical / uv waveband in several sources , the model can successfully fit the sed of these sources from the radio to hard x - rays with only two free parameters . the x - ray emission dominantly come from the thermal electrons of the riaf for all sources in our sample , while the observed radio emission is always 2 - 3 orders of magnitude higher than that predicted by the purely thermal riaf model . the radio emission can be well described by the synchrotron radiation from the nonthermal electrons in the riaf if a small fraction of the electron thermal energy resides in the power - law tail , where the @xmath17 range from 0.03% to 15% , with a typical value of @xmath113% , for the sources in our present sample ( see table 1 ) . in this work , we explore the hybrid thermal - nonthermal synchrotron emission from the hot accretion flows of the riaf . such hot accretion flows are expected to exist in sources with low - mass accretion rates ( e.g. , llagns ) . we calculate the spectrum of the global riaf model with hybrid electrons , and find that synchrotron emission from the nonthermal electrons can be up to 2 - 3 orders of magnitude greater than that from the purely thermal electrons even with only several percent of electron thermal energy injected into the nonthermal electrons . this model has explained the multi - wavelength spectrum of sgr a * very well @xcite , and we further extend this model to other nearby llagns , which have only a compact radio core and have no evidence of the jet . our model can roughly reproduce the observed anticorrelation between the mass - corrected radio loudness and eddington ratio as found in the llagns . we further perform the detailed modeling of the sed for 20 llagns in our sample from the radio to x - ray waveband . we find that the sed of llagns can be well described by our model , where the x - ray emission is produced predominantly by the inverse compton scattering of the seed synchrotron photons produced by both thermal and nonthermal electrons in the riaf , while the radio emission mainly come from the nonthermal electrons . it is now established that rq agns are not radio - silent , and do emit the radio emission at some level . @xcite found that 85% of the nearby seyfert nuclei show the nuclear radio emission at 5 ghz , and their typical radio morphology is a compact core ( either unresolved or slightly resolved ) . @xcite observed six nearby llagns with high - resolution vlba , and found that the radio emission is still unresolved even at milliarcsecond scale , which roughly corresponds to several to ten thousands gravitational radii . therefore , contrary to the rl agns which have been observed with jet or core - jet structures directly , the physical origin of the radio emission in rq agns is much unclear . the most popular model is that there may exist a scaled - down version of the jet in rq agns . the coupled riaf - jet model has been explored to fit the multiwavelength spectrum of llagns ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? we note that it is reasonable to apply the riaf - jet model to the sources with observed jet structures ( e.g. , fr is ) , while it is still controversial whether the same model can be used to the llagns with only compact radio cores , where these sources have no evident jet structure even at scale of several thousand gravitational radii . another possibility is that the radio emission in rq agns also originate in the hot plasma ( e.g. , riaf in llagns and disk - corona in bright agns ) , where the turbulence , weak shocks and/or magnetic reconnection events may occasionally accelerate a fraction of the electrons to a harder power - law tail . the synchrotron emission from these nonthermal power - law electrons may account for the radio emission from the compact cores of these llagns . the hybrid thermal - nonthermal synchrotron emission from the riaf have well reproduced the two components of the radio spectrum and also the x - ray flares observed in sgr a * ( e.g. , * ? ? ? we confirm the result that the pure riaf always underpredicts the radio emission for nearby llagns @xcite . however , the low - frequency radio emission of the compact cores in llagns can be accounted if only a small fraction of the viscous dissipation energy in the accretion flow goes into accelerating electrons to a nonthermal power - law distribution ( see figures [ f4]-[f5 ] ) . it is well established that magnetic reconnection should be unavoidable and play an important role in converting the magnetic energy into the particles ( e.g. , * ? ? ? * ; * ? ? ? . both the magnetic reconnection itself ( e.g. , * ? ? ? * ) or the diffusive shock caused by the violent plasma motions in the magnetic reconnection region ( e.g. , * ? ? ? * ) can naturally accelerate a small fraction of the electrons to be the power - law distribution . a small fraction of the power - law electrons presented in the riaf will not affect the global structure of the riaf ( e.g. , * ? ? ? therefore , our work provide the possibility that the radio emission of nearby llagns may originate in the nonthermal electrons of the hot accretion flow . it may be the similar case for other bright rq agns , where the radio emission is dominated by the nonthermal electrons in the corona . @xcite found the similarities of the radio / hard x - ray correlation in rq agns and magnetically active stars , which support that the radio emission of rq agns also dominantly come from the corona as that of active stars . we will further test this issue in a subsequent work for bright rq agns considering the possible physical mechanism for the production of thermal and nonthermal electrons in the corona above the ssd . @xcite found a strong anticorrelation between the radio loudness and eddington ratio for the llagns ( see also * ? ? ? * ; * ? ? ? * ; * ? ? ? we investigate the dependence of the mass - corrected radio loudness , @xmath131 , on the eddington ratio , @xmath154 , where the scaling of @xmath155 is derived from our model as that @xmath156 and @xmath157 . we find that the mass - corrected radio loudness is still anticorrelate with the eddington ratio , where both quantities are not affected by the mass term . it is interesting to note that the dependence of the radio emission on the bh mass , @xmath157 , in our model is similar to that of jet model ( e.g. , * ? ? ? * ) and that derived from the observed fundamental plane of bh activity @xcite . therefore , it should be cautious to get the conclusion that the origin of the radio and x - ray emission is similar for the stellar - mass bh x - ray binaries ( xrbs ) and the supermassive bh agns based only on the fundamental plane equation of the bh activity , since that the dependence of the radio emission on the bh mass is similar for our model and the jet model . we find that the relation between the radio luminosity and x - ray luminosity is @xmath158 at a given bh mass in our model based on @xmath159 and @xmath160 . our model prediction on the radio@xmath0x - ray correlation is much shallower than that observed in xrbs ( e.g. , * ? ? ? * ; * ? ? ? * ) or that predicted by accretion - jet model ( e.g. , * ? ? ? * ) , where @xmath161 . our results provide a diagnostic that can distinguish the possible origin of the radio emission in llagns by investigating the radio@xmath0x - ray correlation for llagns with the same or similar bh masses . however , our present sample is still limited , which prevent us to further test this issue . from a theoretical perspective , there are still some uncertainties in the current riaf model with hybrid thermal and nonthermal electrons . the dependence of the spectrum on the model parameters have been explored in former works ( e.g. , * ? ? ? * ; * ? ? ? the riaf spectrum is not sensitive to some parameters ( e.g. , @xmath86 , @xmath87 ) , we adopt the typical values @xmath92 and @xmath93 , which are constrained from observations and/or simulations . wind parameter @xmath95 is adopted directly , which is constrained from observations of sgr a*. we adopt the radius @xmath109 as the outer boundary . we find that the x - ray emission mainly come from the innermost region of the accretion flow , which is less affected by the outer boundary condition if it is larger than several tens gravitational radii . however , we find that the low - frequency radio emission come from the much larger regions from several tens to several thousand gravitational radii . the llagns in our sample have very low eddington ratios ( @xmath162 ) , and the whole accretion may be through riaf . @xcite proposed that the truncation radius should be around one thousand to several tens of thousands gravitational radii ( if the possible outer ssd is present ) for @xmath163 as adopted in fitting the sed of the llagns in our sample . therefore , the assumption of the outer boundary @xmath109 should be reasonable , and will not affect our main results . the region of the radio emission from the nonthermal electrons in riaf is more or less consistent with the size of the compact radio cores as observed by the vlbi for some nearby llagns ( e.g. , * ? ? ? we neglect the contribution of the emission from the possible outer ssd ( if present ) in our sed fittings , since that we still do nt know whether the outer ssd still exists in the llagns with extreme low eddington ratios or not , and the optical emission also easily suffer the contamination from the host galaxies or contaminate by the nuclear starbursts . for example , @xcite reported an evidence of recent starburst formation in the nucleus of ngc 1097 , and its optical / uv emission may dominated by the stellar contribution . we find that our model prediction on the optical emission is slightly smaller than the observed emission but higher than the lower limits of the possible intrinsic agn optical emission that derived from optical variabilities . therefore , the observed optical emission of llagns is roughly consistent with our model predictions even without considering the possible contribution from the outer ssd ( e.g. , figures [ f4]-[f5 ] ) . the parameter @xmath88 , describing the fraction of the turbulent dissipation that directly heats the electrons in the flow , is still unclear , which may be in the range of several percent up to several tens percent based on recent simulations ( e.g. , * ? ? ? * ) and/or model fitting of the sgr a * @xcite . we find that most of llagns in our sample can be better fitted if @xmath101 is adopted ( see figures [ f4]-[f5 ] ) . it should be noted that the adopted @xmath101 is slightly smaller than that used in sgr a * ( @xmath164 , * ? ? ? * ) , which may caused by different microphysics since that the accretion rate in sgr a * much smaller than that used in our work . our results will be roughly unchanged even if we use @xmath164 as that used in sgr a * ( e.g. , * ? ? ? * ) , but the model fitting of the x - ray emission is less perfect as that with @xmath101 . the values of @xmath102 and @xmath17 that describing the nonthermal electrons is not strongly constrained , but which are degenerate for low - frequency radio emission ( e.g. , * ? ? ? we fix the typical value of @xmath102=3.0 , and then adjust the parameter @xmath17 to fit the radio emission of llagns in our sample . we find that only a small fraction of the energy in nonthermal electrons is sufficient to produce the radio emission at low frequencies . the value of @xmath17 can even be smaller if the parameter @xmath102 is also smaller . our fitting results show that the value of @xmath17 in these llagns of our sample is more or less consistent with that constrained for sgr a * ( e.g. , @xmath111% in * ? ? ? * ; * ? ? ? * ) and that derived from the gamma - ray background ( e.g. , @xmath114% in * ? ? ? we appreciate feng yuan , xinwu cao and the anonymous referee for helpful comments and suggestions . we also thank the members of hust astrophysics group for useful discussions . this work is supported by the national basic research program of china ( 2009cb824800 ) , the nsfc ( grants 11103003 , 11133005 and 11173011 ) , the doctoral program of higher education ( 20110142120037 ) , the fundamental research funds for the central universities ( hust : 2011ts159 ) . cappi , m. et al . 2006 , , 446 , 459 coppi , p. s. , & blandford , r. d. 1990 , , 245 , 453 ding , j. yuan , f. , & liang , e. 2010 , , 708 , 1545 eracleous , m. , shields , j. c. , chartas , g. , & moran , e. c. 2002 , , 565 , 108 eracleous , m. , hwang , j. a. , flohic , & h. m. l. g. 2010 , , 187 , 135 filho , m. e. , barthel , p. d. , & ho , l. c. 2006 , , 451 , 71 corbel , s. , nowak , m. a. , fender , r. p. , tzioumis , a. k. , & markoff , s. 2003 , , 400 , 1007 gallo , e. , fender , r. p. , & pooley , g. g. 2003 , , 344 , 60 giroletti , m. , & panessa , f. 2009 , , 706 , 260 goodman , j. , & uzdensky , d. 2008 , , 688 , 555 gonzalez - martin , o. , masegosa , j. , marquez , i. , guainazzi , m. , & jimenez - bailon , e. 2009 , , 506 , 1107 gu , m. , & cao , x. 2009 , , 399 , 349 hawley , j. f. , & balbus , s. a. 2002 , , 573 , 738 heinz , s. , & sunyaev , r. a. 2003 , , 343 , 59 ho , l. c. , greene , j. e. , filippenko , a. v. , sargent , w. l. w. 2009 , , 183 , 1 ho , l. c. 2009 , , 699 , 626 ho , l. c. 2008 , ara&a , 46 , 475 ho , l. c. 2002 , , 564 , 120 ho , l. c. , et al . 2001 , , 549 , 51 ho , l. c. , & ulvestad , j. s. 2001 , , 133 , 77 ho , l. c. 1999 , , 516 , 672 ho , l. c. , filippenko , a. v. , & sargent , w. l. w. 1997 , , 487 , 568 hopkins , p. f. , & hernquist , l. 2006 , , 166 , 1 hummel , e. , van der hulst , j. m. , dickey , j. m. 1984 , , 134 , 207 ichimaru , s. 1977 , , 214 , 840 igumenshchev , i. v. , & abramowicz , m. a. 1999 , , 303 , 309 inoue , y. , totani , t. , & ueda , y. 2008 , , 672 , 51 kaspi , s. , maoz , d. , netzer , h. , peterson , b. m. , vestergaard , m. , & jannuzi , b. t. 2005 , , 629 , 61 kato , s. , fukue , j. , & mineshige , s. 2008 , black - hole accretion disks : towards a new paradigm ( kyoto : kyoto univ . press ) kellermann , k. i. , sramek , r. a. , schmidt , m. , green , r. f. , & shaffer , d. b. 1994 , , 108 , 1163 laor , a. , & behar , e. 2008 , , 390 , 847 leipski , c. , falcke , h. , bennert , n. , & huttemeister , s. 2006 , , 455 , 161 li , s .- l . , cao , x. 2009 , , 400 , 1743 liu , b. f. , & taam , r. e. 2009 , , 707 , 233 mcconnell , m. l. , et al . 2000 , , 543 , 928 merloni , a. , heinz , s. , & di matteo , t. 2003 , , 345 , 1057 miller p. , rawlings s. , & saunders r. , 1993 , , 263 , 425 nagar , n. m. , falcke , h. , & wilson , a. s 2005 , , 435 , 521 nagar , n. m. , falcke , h. , wilson , a. s. , & ulvestad , j. s. 2002 , , 392 , 53 narayan , r. & mcclintock , j. e. 2008 , new astronomy reviews , 51 , 733 narayan , r. , & yi , i. 1995 , , 452 , 710 narayan , r. , & yi , i. 1994 , , 428 , 13 nemmen , r. , storchi - bergmann , t. , & eracleous , m. 2012 , submitted to apj , arxiv:1112.4640 nemmen , r. s. , storchi - bergmann , t. , yuan , f. , eracleous , m. , terashima , y. , & wilson , a. s. 2006 , , 643 , 652 @xmath166zel , f. , psaltis , d. , & narayan , r. 2000 , , 541 , 234 panessa , f. , barcons , x. , bassani , l. , cappi , m. , carrera , f. j. , ho , l. c. , & pellegrini , s. 2007 , , 467 , 519 panessa , f. , bassani , l. , cappi , m. , dadina , m. , barcons , x. , carrera , f. j. , ho , l. c. , & iwasawa , k. 2006 , , 455 , 173 ptak , a. , terashima , y. , ho , l. c. , & quataert , e. 2004 , , 606 , 173 quataert , e. , di matteo , t. , narayan , r. , & ho , l. c. 1999 , , 525 , 89 quataert , e. , & narayan , r. 1999 , , 520 , 298 shakura , n. i. , & sunyaev , r. a. 1973 , , 24 , 337 sharma , p. , et al . 2007 , , 667 , 714 sikora , m. , stawarz , l. , & lasota , j .- 2007 , , 658 , 815 spruit , h. c. 1988 , , 194 , 319 satyapal , s. , dudik , r. p. , ohalloran , b. , & gliozzi , m. 2005 , , 633 , 86 stone , j. m. , pringle , j. e. , & begelman , m. c. 1999 , , 310 , 1002 stone , j. m. , & pringle , j. e. 2001 , , 322 , 461 steenbrugge , k. c. , jolley , e. j. d. , kuncic , z. , & blundell , k. m. 2011 , , 413 , 1735 storchi - bergmann , t. , nemmen , r. s. , spinelli , p. f. , eracleous , m. , wilson , a. s. , filippenko , a. v. , livio , m. 2005 , , 624 , 13 terashima , y. & wilson , a. s. 2003 , , 583 , 145 thean , a. , pedlar , a. , kukula , m. j. , baum , s. a. , & odea , c. p. 2000 , , 314 , 573 wrobel , j. m. & ho , l. c. 2006 , , 646 , 95 wu , q. & gu , m. 2008 , , 682 , 212 wu , q. , yuan , f . , & cao , x. 2007 , , 669 , 96 wu , q. , & cao , x. 2006 , pasp , 118 , 1098 wu , q. , & cao , x. 2005 , , 621 , 130 xu , y .- d . & cao , x 2009 , raa , 9 , 401 yuan , f. , & narayan , r. 2013 , ara&a , in preparation yuan , f. , wu , m. , bu , d. 2012 , submitted to apj yuan , f. , bu , d .- f . 2010 , , 408 , 1051 yuan , f. , yu , z. , & ho , l. c. 2009a , , 703 , 1034 yuan , f. 2007 , asp conf . 373 , the central engine of active galactic nuclei , ed . l. c. ho and j .- m . wang ( san francisco ) , 95 yuan , f. , shen , z .- q . , & huang , l. 2006 , , 642 , 45 yuan , f. , & cui w. 2005 , , 629 , 408 yuan , f. , quataert , e. , & narayan , r. 2003 , , 598 , 301 yuan , y .- f . , cao , x. , huang , l. , & shen , z .- q . 2009b , , 699 , 722

discrete nonlinear models play a very important role in many physical applications @xcite . an important class of these models consists of a few completely integrable lattice equations , such as the toda lattice @xcite , the ablowitz - ladik lattice @xcite , and the integrable sine - gordon lattice @xcite . the fact that these lattices possess exact soliton solutions demonstrates that , in principle , discreteness of the host media does not preclude the propagation of localized coherent structures . moreover , the mobility of soliton - like excitations in discrete media is a key issue in many physical contexts ; for instance , the mobility of dislocations , a kind of topological solitons , is of paramount importance in the physics of plastic deformation of metals and other crystalline bodies @xcite . a prototypical class of discrete models , relevant to a variety of applications , consists of the klein - gordon dynamical lattices @xcite . one of the main representatives of this family of models is the so - called @xmath0 model @xcite , which features a cubic nonlinearity . this simple power law nonlinearity renders this model a ripe testbed for studying the existence and stability of nonlinear solutions , and comparing their properties in continua and lattices . in the @xmath1-dimensional continuum framework ( and in the absence of spatially dependent external potentials ) , a solution can be shifted arbitrarily along @xmath2 by any @xmath3 ( @xmath2 is the spatial coordinate and @xmath4 ) , due to the existing translational invariance . on the other hand , in the discrete system , translational invariance is generically lost and equilibrium static solutions exist only for a discrete rather than for a continuum set of @xmath3 @xcite . some of these equilibrium solutions correspond to energy maxima and are unstable , while others , corresponding to energy minima , are stable . the difference between such maxima and minima of the energy is typically referred to as the peierls - nabarro barrier ( pnb ) . it is of particular interest to develop discretizations that do not feature such barriers . in such cases , one might expect that the ensuing models would be more faithful representations of their continuum counterparts , regarding both symmetry properties and travelling solution features ( even though there are some notes of caution that should be made ; see e.g. the discussion of @xcite ) . in that vein , recently , a number of non - integrable discrete klein - gordon equations free of the peierls - nabarro barrier ( pnb - free ) has been systematically constructed . the first set of such models which were , by construction , hamiltonian ones , was obtained by speight and co - workers @xcite using the bogomolnyi argument @xcite . a second successful attempt led to the construction of momentum - conserving discretizations @xcite . it was then demonstrated , surprisingly , that the pnb - free models of that kind conserving both energy and classical momentum do not exist @xcite . new pnb - free @xmath0 lattices were derived by barashenkov _ . however , we note that for the lattices derived in @xcite , only the kink - type solution has been considered , while it is well - known that the continuum klein - gordon equation can support a number of other solutions @xcite . on the other hand , a general approach to the construction of the pnb - free lattices was recently reported in @xcite . this approach is based on the use of the _ discretized first integral _ ( dfi ) of the corresponding static field equation , and the integration constant that enters dfi generates the one - parameter solution space . the dfi is in fact a two - point nonlinear algebraic equation from which the exact static solutions of the three - point pnb - free discrete models can be found . in this work we systematically use the dfi approach and derive the hamiltonian and non - hamiltonian pnb - free klein - gordon lattices supporting the one - parameter space of solutions , generalizing the lattices offered in @xcite . we then focus on the @xmath0 field and discuss in detail the energy - conserving pnb - free model proposed very recently in @xcite , as well as the momentum - conserving model of @xcite ( the two are identical in the static case ) . for these models we describe the full solution space of the underlying static problem that also includes the solutions derived earlier in @xcite . our setting is the following : we consider the hamiltonian of the klein - gordon field , @xmath5 , with the kinetic and potential energy functionals respectively defined as , @xmath6 @xmath7dx\,,\ ] ] where @xmath8 is the unknown field and @xmath9 is a given potential function . the corresponding equation of motion is @xmath10 where @xmath11 . equation ( [ kg ] ) will be discretized on the lattice @xmath12 , where @xmath13 , and @xmath14 is the lattice spacing . we would like to construct a nearest - neighbor discrete analog to eq . ( [ kg ] ) of the form @xmath15 such that in the continuum limit @xmath16 we have @xmath17 _ and _ that the solution to the three - point static problem corresponding to eq . ( [ kleingordondiscrete1 ] ) , @xmath18 can be found from a reduced two - point problem @xmath19 if this reduction is achieved , then the exact static solutions can be constructed upon solving the algebraic eq . ( [ twopointstatic ] ) iteratively , starting from arbitrary admissible value of @xmath20 or @xmath21 . arbitrariness in the choice of the initial condition implies that the static solution can be placed anywhere with respect to the lattice and , for that reason , such lattices are called translationally invariant . discretization eq . ( [ kleingordondiscrete1 ] ) may result in a non - hamiltonian model due to the non - potential nature of the background forces . in the latter setting , the absence of an energy functional renders ambiguous the definition of the pn barrier , hence we clarify this point in what follows . suppose that we have two equilibrium solutions , @xmath22 and @xmath23 ( often , in hamiltonian models with pnb , the second one corresponds to a linearly unstable energy maximum while the first one to a linearly stable energy minimum ) . the work done by the inter - particle and background forces to move the @xmath24th particle from the position @xmath22 to the position @xmath23 is @xmath25 and the total work performed to `` transform '' the solution @xmath22 to @xmath23 is @xmath26 . the pn barrier is defined to be equal to @xmath27 . for hamiltonian models this definition is equivalent to the classical one because @xmath27 is equal to the energy difference between the second ( possibly higher energy unstable ) configuration and the first ( possibly lower energy stable ) configuration . for non - hamiltonian models , @xmath28 will depend on the path connecting the initial and final configurations of particles . to calculate @xmath27 one , therefore , has to specify such a path . while the height of the pn barrier in a non - hamiltonian lattice depends on the path , the pnb - free , non - hamiltonian lattice can be unambiguously defined as the translationally invariant lattice where the quasi - static transformation between the two configurations of interest can be done continuously through the set of _ equilibrium _ configurations . along the path through the equilibrium configurations , forces acting on particles are zero and thus , @xmath29 for any @xmath24 which , in turn , results in @xmath30 . this notion of the pnb - free lattice is also applicable to the hamiltonian lattices and will serve as our definition of the pnb - free models ; notice , however , that we do not overlook the mathematical subtleties involved in this definition , including the question of whether @xmath27 is zero along all realizable paths connecting @xmath22 and @xmath23 in the non - hamiltonian models , among equilibrium configurations , as is the case for their hamiltonian siblings . these and related questions , including an appropriate modified definition of the relevant quantities for models such as those of the discrete nonlinear schrdinger type , will be left for a future publication . our scope in what follows is to generalize the approach developed in @xcite to show how to construct _ all possible _ exact static solutions of the models of @xcite and @xcite ( see also @xcite ) , including those derived in @xcite . this will also lead us to introduce a number of solutions ( both localized and extended ones ) that , to the best of our knowledge , have not been discussed / analyzed previously , such as the ones that will be termed `` inverted '' ( see below ) . we will also discuss the stability of certain solutions among the obtained ones , for each of the models of interest ( since their stability properties are different ) . we also note in passing that while our presentation will be geared towards the @xmath0 models , our results regarding the existence of solutions can equally well be applied to discrete equations of the nonlinear schrdinger ( nls ) type @xcite , such as e.g. the ablowitz - ladik model . in particular , let us consider equations of the form : @xmath31 ( where the overbar denotes complex conjugate ) with @xmath32 and @xmath33 . then , looking for standing wave solutions of the form @xmath34 , with @xmath21 real , one would obtain the klein - gordon static problem for the standing wave spatial profile @xmath21 . hence all the discussion given below for the existence of such solutions can be appropriately translated in the existence of standing waves of the corresponding discrete nls models . the reader should be cautioned however that the stability properties in the latter context may differ [ a relevant example will be discussed later in the text ] . our presentation will be structured as follows . in section [ sec : pnpfreekg ] , we develop the dfi approach and derive pnb - free klein - gordon models supporting one - parameter space of static solutions . in section [ sec : phi4 ] , we present some pnb - free @xmath0 lattices and describe some of their basic properties . in section [ exactsolutionsdiscrete ] , we discuss the details of the construction of the general exact static solutions of the models both in the localized ( hyperbolic function ) and in the extended ( general elliptic function ) form , and for both signs of the nonlinearity prefactor @xmath35 . @xmath36 corresponds to the so - called defocusing case , while @xmath37 corresponds to the focusing case in the standard terminology of nls equations . examples of the solutions are then given in section [ analysis ] . in section [ linearstability ] , we analyze the stability of the obtained solutions . in section [ dynamics ] , slow kink dynamics in the pnb - free models is compared numerically with that in the classical @xmath0 model . in section [ continuumsolutions ] we give a complete list of bounded and unbounded exact solutions to the continuum @xmath0 field . in section [ conclusions ] , we summarize our findings and present our conclusions . following the lines of the dfi approach of ref . @xcite , we start from the first integral of the static eq . ( [ kg ] ) , @xmath38 where @xmath39 is the integration constant . the first integral can also be taken in modified forms @xcite , e.g. , as @xmath40 = 0 \,,\end{aligned}\ ] ] or as @xmath41 = 0 \,,\end{aligned}\ ] ] where @xmath42 and @xmath43 are continuous functions and @xmath44 . note that @xmath45 and @xmath46 are equivalent only if @xmath47 . we will consider the case of @xmath48 , @xmath47 , i.e. , the unchanged form of the first integral , eq . ( [ firstintu ] ) , together with the case of @xmath48 , @xmath49 , for which we have the two possibilities @xmath50 and @xmath51 of which only the first one will be discussed below . we then construct the dfis corresponding to eq . ( [ firstintu ] ) and eq . ( [ firstintv ] ) , which are respectively given by @xmath52 @xmath53 where we demand that @xmath54 in the continuum limit ( @xmath55 ) . first we construct a pnb - free klein - gordon lattice using the unchanged form of the first integral , eq . ( [ firstintu ] ) , and corresponding dfi , eq . ( [ twopointmapu ] ) . calculating @xmath56 and multiplying the result by @xmath57 , we find @xmath58 discretizing the left - hand side of eq . ( [ d1 ] ) we obtain the lattice klein - gordon equation @xmath59 whose static solutions can be found from the two - point problem , eq . ( [ twopointmapu ] ) , and thus , the lattice is pnb - free . the lattice of this type was first derived in @xcite where it was also demonstrated that it conserves the momentum @xmath60 the integration constant @xmath39 that appears in eq . ( [ twopointmapu ] ) cancels out in eq . ( [ d1discr ] ) . this means that all kinds of static solutions derived from eq . ( [ twopointmapu ] ) for different @xmath39 ( kink solution corresponds to @xmath61 ) will be the static solutions to one and the same klein - gordon lattice , eq . ( [ d1discr ] ) , since it does not depend on @xmath39 . the right - hand side of eq . ( [ d1discr ] ) becomes non - singular when the discretization of the potential @xmath9 is a polynomial function having symmetry @xmath62 . the most general expression of this type was given in @xcite . to construct hamiltonian pnb - free lattices we discretize not the equation of motion , eq . ( [ kg ] ) , but the hamiltonian , eq . ( [ kinen ] ) and eq . ( [ poten ] ) . we now use the modified first integral in the form of eq . ( [ firstintv ] ) with the upper sign and rewrite the potential energy functional , eq . ( [ poten ] ) , as follows , @xmath63 ^ 2 + 2\phi_x\sqrt{2v(\phi)-c } \right)dx\,,\ ] ] where we omitted the constant term . discretizing the kinetic energy , eq . ( [ kinen ] ) , and the potential energy , eq . ( [ potenfi ] ) , we obtain the discrete hamiltonian @xmath64 ^ 2 \nonumber \\ & + & 2\frac{\phi_n-\phi_{n-1}}{h}\sqrt{2v(\phi_{n-1},\phi_n)-c}\big\}\,.\end{aligned}\ ] ] if the background potential is discretized as suggested in @xcite , @xmath65 then the last term of the hamiltonian eq . ( [ hamiltoniandiscr ] ) reduces to @xmath66 $ ] and it disappears in the telescopic summation . with the choice of eq . ( [ pnpfreecondition ] ) , the equations of motion derived from eq . ( [ hamiltoniandiscr ] ) are @xmath67 where , according to eq . ( [ twopointmapv ] ) and eq . ( [ pnpfreecondition ] ) , @xmath68 this lattice conserves the hamiltonian ( total energy ) @xmath69 ^ 2\big\}\,.\end{aligned}\ ] ] obviously , static solutions to the lattice eq . ( [ hamiltpnpfree ] ) can be found from the two - point dfi @xmath70 and it is also clear that the hamiltonian eq . ( [ hamiltoniandiscr1 ] ) is pnb - free . the energy - conserving lattice involves the integration constant @xmath39 ( through the function @xmath71 ) and this is different from what we had for the momentum - conserving model . the special case of @xmath61 yields the original model by speight @xcite , which supports a kink solution . models with @xmath72 describe solutions different from the kink solution . suppose that we use the same function @xmath73 in the dfis eq . ( [ twopointmapu ] ) and eq . ( [ twopointmapv ] ) to construct different @xmath74 terms . then equations @xmath75 and @xmath76 are equivalent . linear combination of those terms can be used to write the following pnb - free klein - gordon model @xmath77 where the constant coefficients satisfy the continuity constraint @xmath78 assuming that @xmath79 can depend on @xmath14 . static solutions of eq . ( [ kleingordondiscrall ] ) can be found iteratively from eq . ( [ twopointmapu ] ) or eq . ( [ twopointmapv ] ) . the model of eq . ( [ kleingordondiscrall ] ) can be generalized in a number of ways . for example , one can append terms which disappear in the continuum limit and ones that vanish upon substituting @xmath75 . let us call such terms as @xmath80-terms . furthermore , any term @xmath81 can be modified by multiplying by a continuous function @xmath82 , which never vanishes and whose continuum limit is unity . such multiplication will not change either the continuum limit , or the static solutions of the model . particularly we will study the pnb - free model of the form @xmath83,\end{aligned}\ ] ] which is the lattice eq . ( [ d1discr ] ) modified by a multiplier @xmath84 and augmented with the @xmath80-term with arbitrary coefficient @xmath85 . this @xmath80-term was constructed to fit to the @xmath86 invariant offered in @xcite for the discretization of the @xmath0 field . finally we note that expressions similar to eq . ( [ d1 ] ) can be derived from derivatives @xmath87 and @xmath88 and they can produce new pnb - free models . to give examples of klein - gordon lattices , we will discretize the well - known @xmath0 field with the potential @xmath89 where the parameter @xmath90 . the corresponding equation of motion is @xmath91 the above , so - called , @xmath0 equation supports moving periodic solutions that can be expressed in terms of the jacobi elliptic functions . the latter can be reduced to localized hyperbolic function solutions ( when the elliptic modulus @xmath92 ) . bounded solutions of this sort were discussed in the context of structural phase transitions @xcite and were used as the starting point for derivation of exact solutions to a discrete @xmath0 model @xcite . in section [ continuumsolutions ] we complete the list of the solutions presented in @xcite by providing also the set of unbounded solutions of the present model . the simplest discretization of eq . ( [ kleingordon ] ) is of the form @xmath93 and it possesses a pn barrier . however , as mentioned above , one can construct pnb - free discrete klein - gordon models by discretizing the nonlinear term @xmath94 typically over three neighboring points , @xmath95 , in contrast to the classical discretization @xmath96 . the three - point discretization may be physically meaningful in some settings @xcite , but is also interesting from the more fundamental point of view of developing pnb - free discretizations and obtaining analytically ( or semi - analytically ) explicit waveforms of their solutions . we discretize the potential of eq . ( [ potential ] ) as follows @xmath97 then , the dfis , eqs . ( [ twopointmapu ] ) and ( [ twopointmapv ] ) , become respectively , @xmath98 @xmath99 it has been demonstrated that at @xmath61 eq . ( [ firstintdiscrete ] ) and eq . ( [ twopointmapvkg ] ) support kink solutions @xcite . for @xmath100 they support solutions different from kink . let us denote @xmath101 substituting eq . ( [ firstintdiscrete ] ) and eq . ( [ twopointmapvkg ] ) into eq . ( [ bop ] ) with @xmath102 we arrive at the following pnb - free @xmath0 lattice @xmath103 equation ( [ barashenkov1 ] ) gives a one - parameter @xmath104 family of pnb - free models . for the special case of @xmath61 and @xmath36 we get one of the lattices derived in @xcite , @xmath105 , and this lattice supports the kink solution . lattices with @xmath106 support solutions different from the kink . for @xmath107 we get from eq . ( [ barashenkov1 ] ) the @xmath39-independent model @xmath108 this non - hamiltonian pnb - free @xmath0 model conserves the momentum eq . ( [ mom ] ) and it will be referred to as the momentum - conserving ( mc ) model . substituting eq . ( [ firstintdiscrete ] ) into eq . ( [ bop ] ) with @xmath109 and @xmath107 we obtain another @xmath39-independent pnb - free model discovered in @xcite , @xmath110 this model will be called the energy - conserving ( ec ) model because it possesses the hamiltonian @xcite @xmath111 , \label{calh}\end{aligned}\ ] ] where the potential @xmath112 is given by @xmath113 ) , @xmath112 , defined by eq . ( [ hamsaxena ] ) , for @xmath114 and @xmath115 ( solid line ) and @xmath116 ( dashed line ) . for @xmath117 the potential is smooth and it has one minimum at @xmath118 and two maxima at @xmath119 . for @xmath120 the potential has two minima at @xmath119 and a maximum at @xmath118 ; as @xmath121 the potential @xmath122 . ] in fig . [ figure1 ] we plot the potential @xmath112 for @xmath114 ( i.e. , @xmath123 ) in both cases @xmath36 ( solid line ) and @xmath37 ( dashed line ) . for @xmath117 the potential is smooth and it has one minimum at @xmath118 and two maxima at @xmath119 . for @xmath120 the potential has two minima at @xmath119 and a maximum at @xmath118 ; note that in the limit @xmath121 , the potential @xmath122 . it is not possible to plot an analog of fig . [ figure1 ] for the mc model since the relevant background forces are of many - body type ( i.e. , involve nearest neighbors ) and are non - potential . pnb - free models given by eq . ( [ barashenkov1 ] ) , eq . ( [ physicad ] ) , and eq . ( [ saxena ] ) have exactly same static solutions derivable from dfi eq . ( [ firstintdiscrete ] ) . to find _ all _ static solutions to the pnb - free models of eq . ( [ barashenkov1 ] ) , eq . ( [ physicad ] ) , and eq . ( [ saxena ] ) we solve the dfi of eq . ( [ firstintdiscrete ] ) : @xmath124 where @xmath21 and @xmath20 can be interchanged due to the symmetry of the equation . starting from any admissible `` initial '' value @xmath125 , by iterating eq . ( [ pulse ] ) and its counterpart written as @xmath126 , one can construct recurrently the static solution to both the mc model of eq . ( [ physicad ] ) and the ec model of eq . ( [ saxena ] ) , or to a linear combination thereof . arbitrariness in the choice of @xmath125 implies the absence of pnb in these models , which has been also demonstrated in @xcite . as can be seen from eq . ( [ pulse ] ) , once the values of @xmath127 and @xmath128 are fixed , there are certain restrictions on the choice of the values of @xmath125 . in particular , inadmissible initial values are those for which the denominator becomes zero , i.e. , @xmath129 for @xmath120 . an exceptional case is that of @xmath130 , @xmath131 when an arbitrary sequence of @xmath132 is a solution of eq . ( [ firstintdiscrete ] ) . inadmissible values of @xmath125 are also ones for which @xmath133 . the condition @xmath134 leads to a biquadratic algebraic equation determining the borders of admissible region ; the roots of this equation are : @xmath135 let us introduce the following notations for these roots : @xmath136 in the nonlinear map eq . ( [ pulse ] ) for different values of @xmath127 at @xmath137 . there are three inadmissible regions marked with `` no '' and two inadmissible lines @xmath138 ( horizontal dashed lines ) . on these lines the on - site potential @xmath112 diverges ( see fig . [ figure1 ] ) . @xmath139 ( @xmath140 ) designate different branches of the borders of the admissible regions [ see eq . ( [ borders ] ) and eq . ( [ rootnotation ] ) ] . roots @xmath141 and @xmath142 merge at @xmath131 and at @xmath143 . roots @xmath142 and @xmath144 merge at @xmath145 . vertical dashed lines separate regions with different jacobi elliptic function solutions . vertical dotted line , situated ( for the chosen parameters ) at @xmath146 , divides the region of the @xmath147 solution into two portions corresponding to two roots of the first equation in eq . ( [ p1m1p1d ] ) . ] the admissible regions for the values of @xmath148 of the nonlinear map eq . ( [ pulse ] ) are shown in fig . [ figure2 ] for different values of @xmath127 at @xmath137 . the corresponding result for @xmath149 is presented in fig . [ figure3 ] . these graphs present a road map for constructing the various possible solutions of the above models . the symmetry of eq . ( [ firstintdiscrete ] ) suggests that the topology of the admissible regions is such that once started from an admissible value of @xmath148 , one can not leave the admissible region iterating eq . ( [ pulse ] ) , so that the static solution will surely be constructed for the whole chain . this is so because eq . ( [ pulse ] ) serves for calculating both back and forth points of the map , and if one is admissible , the other one is also admissible . equation ( [ pulse ] ) possesses two roots , which means that for an admissible initial condition one can construct two different solutions , e.g. , a kink and an antikink . when iterating , to keep moving along the same solution , one must take @xmath150 different from @xmath151 [ if the roots of eq . ( [ pulse ] ) are different ] . indeed , setting in the three - point static problem , eq . ( [ physicad ] ) , @xmath152 , we find @xmath153 comparing this with eq . ( [ pulse ] ) , it is readily seen that the equilibrium in the three - point equation in the case of @xmath152 can be achieved only if @xmath134 and the two roots coincide . as mentioned above , the latter requirement is equivalent to the condition that @xmath21 is on the border of the admissible region . as it can be deduced from eq . ( [ borders ] ) , the topology of the admissible regions presented in fig . [ figure2 ] and fig . [ figure3 ] does not significantly change among different but positive and among different but negative values of @xmath128 respectively ; of course , one has to exclude the particular case of @xmath130 and also the continuum limit ( see sec . [ continuumsolutions ] ) . here we will not discuss in detail the case of extremely high discreteness @xmath154 , even though the analysis of this case does not present any additional difficulties . to summarize , inside the admissible region , @xmath155 ( shown in figs . [ figure2 ] and [ figure3 ] ) , the nonlinear map of eq . ( [ pulse ] ) generates static solutions to the mc model of eq . ( [ physicad ] ) and to the ec model of eq . ( [ saxena ] ) . but for @xmath149 . roots @xmath141 and @xmath142 merge at @xmath131 , and roots @xmath142 and @xmath144 merge at @xmath145 . the @xmath156 solution , eq . ( [ n0n0p1d ] ) , is defined for @xmath157 . the @xmath158 solution , eq . ( [ n0p1n0d ] ) , is defined for @xmath159 , and this region is divided into two parts , @xmath160 and @xmath161 , each corresponding to a particular root of the first equation in eq . ( [ n0p1n0d ] ) . the border between these two regions is shown by the dotted line situated ( for the chosen parameters ) at @xmath162 . ] static solutions to the discrete pnb - free models of eq . ( [ physicad ] ) and eq . ( [ saxena ] ) have been reported @xcite in the form of the jacobi elliptic functions , @xmath163 and @xmath156 @xcite . below , we will derive such solutions . the general form of the solutions is @xmath164 where @xmath165 is the modulus of the jacobi elliptic functions , @xmath166 and @xmath167 are the parameters of the solution , and @xmath3 is the arbitrary initial position . finally the integers @xmath168 specify a particular form of the solution . in the limit of @xmath92 , eq . ( [ ellipticdiscrete ] ) reduces to the hyperbolic function form @xmath169 and , in the limit of @xmath170 , to the trigonometric function form @xmath171 substituting the ansatz eq . ( [ ellipticdiscrete ] ) into eq . ( [ firstintdiscrete ] ) and equalizing the coefficients in front of similar terms we find that it can be satisfied for a limited number of combinations of integer powers @xmath168 . solving the equations for the coefficients we relate the parameters of the solution eq . ( [ ellipticdiscrete ] ) , @xmath167 and @xmath166 , to @xmath128 and @xmath172 and also find the relation between @xmath39 , that enters eq . ( [ firstintdiscrete ] ) , and @xmath172 . for some of the sets @xmath173 , e.g. , for @xmath174 and for @xmath175 , we obtain imaginary amplitude @xmath166 in the whole range of parameters . essentially different , real amplitude solutions described by eq . ( [ ellipticdiscrete ] ) , have the following form and are characterized by the following parameters : the @xmath176 solution , @xmath177 , @xmath178 the @xmath179 solution , @xmath180 , @xmath181 the @xmath156 solution , @xmath182 , @xmath183 the @xmath184 solution , @xmath185 , @xmath186 the @xmath187 solution , @xmath188 , @xmath189 the @xmath147 solution , @xmath190 , @xmath191 it is worth making the following remarks : the solutions shown in eq . ( [ n0p1n0d ] ) and eq . ( [ n0n0p1d ] ) have real amplitudes for @xmath192 while the others for @xmath193 . the solutions should be interpreted in the following form . for a given @xmath128 , one can find @xmath167 by solving the first equation in eqs . ( [ p1n0n0d]-[p1m1p1d ] ) , and then @xmath166 from the second one . substituting these values in eq . ( [ ellipticdiscrete ] ) results in the static solutions of the original discrete model . the expressions for @xmath127 in eqs . ( [ p1n0n0d]-[p1m1p1d ] ) link the elliptic jacobi function solutions and the solution in the form of the nonlinear map , eq . ( [ pulse ] ) . as for the other free parameter of the solutions eq . ( [ p1n0n0d]-[p1m1p1d ] ) , the arbitrary shift @xmath3 , its counterpart in the nonlinear map , eq . ( [ pulse ] ) , is effectively the initial value @xmath125 . the solutions of eqs . ( [ p1n0n0d]-[p1m1p1d ] ) can be expressed in a number of other forms using the well - known identities for the jacobi elliptic functions @xcite . for example , shifting the argument by a quarter period , one can transform the @xmath176 solution to the form of @xmath194 , or , applying the ascending landen transformation , to the form of @xmath195 . mathematically , these three expressions look as different members of eq . ( [ ellipticdiscrete ] ) , but physically they are indistinguishable . in what follows , we analyze the static solutions derivable from eq . ( [ pulse ] ) discussing their relation to eqs . ( [ p1n0n0d])-([p1m1p1d ] ) . since the topology of the admissible regions is different for positive and negative @xmath128 , these two cases are studied separately . in the @xmath176 solution , ( [ p1n0n0d ] ) , when the elliptic modulus increases from its smallest value @xmath170 to its largest value @xmath92 , the amplitude @xmath166 increases from @xmath196 to @xmath197 , since @xmath198 [ see eq . ( [ borders ] ) and eq . ( [ rootnotation ] ) ] . as a result , the parameter @xmath127 monotonically decreases from @xmath199 to @xmath196 . thus , the @xmath176 solution is defined in the portion of the @xmath200-plane , @xmath201 and @xmath202 ( see fig . [ figure2 ] ) . the @xmath203 solution , eq . ( [ m1n0n0d ] ) , is defined for @xmath165 , and it is complementary to the @xmath176 solution since it is also valid for @xmath204 , but for @xmath205 ( see fig . [ figure2 ] ) . the @xmath206 solution , eq . ( [ n0m1n0d ] ) , is defined for @xmath165 and it occupies the region @xmath207 , @xmath205 ( see fig . [ figure2 ] ) . the @xmath147 solution , eq . ( [ p1m1p1d ] ) , is defined for unlimited @xmath125 in the range @xmath208 ( see fig . [ figure2 ] ) . this solution is only valid for @xmath209 , where , for fixed @xmath35 , @xmath210 is an increasing function of @xmath14 and @xmath211 . for @xmath212 the first expression in eq . ( [ p1m1p1d ] ) does not have solutions for @xmath167 . for @xmath209 , the equation has two roots , @xmath213 . for the limiting value , @xmath214 , one can find the corresponding amplitude @xmath215 from the second expression of eq . ( [ p1m1p1d ] ) , and then @xmath216 , from the last expression . for the case of @xmath137 presented in fig . [ figure2 ] , we find @xmath217 and @xmath146 . it should be noticed that we have not found a solution of the form of eq . ( [ ellipticdiscrete ] ) valid in the range of @xmath218 ( portion marked with the question mark in fig . [ figure2 ] ) . it is likely that static solutions in this range can not be expressed in terms of the jacobi elliptic functions because they do not survive in the continuum limit ( see sec . [ continuumsolutions ] ) . however , the solution can easily be constructed from the nonlinear map in eq . ( [ pulse ] ) . let us now discuss further several particular examples of the above solutions . first of all , in the limit @xmath219 [ see eq . ( [ hyperbolicdiscrete ] ) ] , the @xmath176 solution , eq . ( [ p1n0n0d ] ) , reduces to the kink solution @xcite , @xmath220 , \label{kinkcooper}\end{aligned}\ ] ] while the @xmath203 solution , eq . ( [ m1n0n0d ] ) , to the solution called hereafter the `` inverted '' kink , @xmath221}. \label{invertedkink}\end{aligned}\ ] ] in eqs . ( [ kinkcooper ] ) and ( [ invertedkink ] ) , @xmath3 is the ( arbitrary ) position of the solution and @xmath222 . this limiting case corresponds to @xmath131 , for which a heteroclinic connection is possible between the fixed points @xmath223 and @xmath224 ( see fig . [ figure2 ] ) , giving rise to the kink or the inverted kink . in this case , eq . ( [ pulse ] ) assumes the following simple form @xmath225 where one can choose either the upper or the lower signs . the kink , eq . ( [ kinkcooper ] ) , and the inverted kink , eq . ( [ invertedkink ] ) , can be derived from eq . ( [ kink ] ) taking initial values from @xmath226 and @xmath227 , respectively . in fig . [ figure4 ] we show ( a ) the kink and ( b ) the inverted kink solutions taking @xmath228 and @xmath229 , respectively , for @xmath137 . : ( a ) kink and ( b ) inverted kink at @xmath137 . the solutions can be found from the nonlinear map eq . ( [ kink ] ) with the initial conditions @xmath228 and @xmath229 for ( a ) and ( b ) , respectively . the kink and the inverted kink are also given by eqs . ( [ kinkcooper ] ) and ( [ invertedkink ] ) respectively , for @xmath230 . horizontal dashed lines , @xmath231 , show positions of singular points of the potential @xmath112 ( see fig . [ figure1 ] ) . ] close to 0 , at @xmath137 , obtained from the map of eq . ( [ pulse ] ) : ( a ) @xmath232 , @xmath233 and ( b ) @xmath232 , @xmath234 . the solution in ( a ) is the @xmath235 solution , eq . ( [ p1n0n0d ] ) , and the solution in ( b ) is the @xmath147 solution , eq . ( [ p1m1p1d ] ) . horizontal dashed lines , @xmath231 , show positions of singular points of the potential @xmath112 ( see fig . [ figure1 ] ) . ] in fig . [ figure5 ] , and for @xmath137 , we present two examples of solutions for positive and negative @xmath127 close to @xmath196 . in particular , for @xmath236 , taking initial value @xmath232 , we obtain from the map of eq . ( [ pulse ] ) the solution presented in panel ( a ) . in fact , it is the @xmath176 solution close to the hyperbolic function limit having the form of a periodic chain of kinks and anti - kinks . on the other hand , for @xmath234 and @xmath232 , we obtain from the map of eq . ( [ pulse ] ) the solution shown in panel ( b ) . this is the @xmath147 solution close to the hyperbolic function limit and has a form of a chain of kinks and inverted anti - kinks . in fig . [ figure6 ] , and for @xmath137 , we present the @xmath237 solution , eq . ( [ m1n0n0d ] ) , ( a ) close to the hyperbolic and ( b ) at the trigonometric limits . the solution shown in panel ( a ) is a chain of inverted kinks and inverted anti - kinks . these solutions are obtained from the nonlinear map eq . ( [ pulse ] ) setting @xmath238 , @xmath239 for ( a ) , and @xmath240 , @xmath241 for ( b ) . in summary , we have shown that the well - established ( hyperbolic and elliptic function ) solutions of the model correspond to a very narrow region of the two - parameter ( @xmath242 ) admissible space . a natural question is what is the typical profile outcome stemming from the use of other pairs of @xmath125 and @xmath127 in eq . ( [ pulse ] ) . generically , upon testing the different regions of the admissible regime we have observed that arbitrary choices may lead to seemingly erratic solutions with very large amplitudes . a different sign choice in the right hand side of eq . ( [ pulse ] ) may , however , lead to a periodically locked tail structure . a simple example of such a solution is given below . solution , eq . ( [ m1n0n0d ] ) , ( a ) close to the hyperbolic and ( b ) at the trigonometric limits . we set @xmath137 . the solution in ( a ) consists of a chain of inverted kinks and inverted anti - kinks . to obtain these solutions from the nonlinear map eq . ( [ pulse ] ) we set @xmath238 , @xmath239 for ( a ) , and @xmath240 , @xmath241 for ( b ) . ] for negative @xmath128 , we have the @xmath179 solution , eq . ( [ n0p1n0d ] ) , and the @xmath156 solution , eq . ( [ n0n0p1d ] ) . let us first start from the latter : when @xmath172 increases from @xmath196 to @xmath197 in eq . ( [ n0n0p1d ] ) , the parameter @xmath39 decreases monotonically from @xmath196 to @xmath199 , i.e. , this solution is defined in the `` dn '' portion of fig . [ figure3 ] . on the other hand , the @xmath179 solution of eq . ( [ n0p1n0d ] ) is only valid for @xmath209 , where @xmath210 is an increasing function of @xmath14 and @xmath211 . for @xmath212 the second expression of eq . ( [ n0p1n0d ] ) does not have solutions for @xmath167 . for @xmath209 , the equation has two roots , @xmath213 . for the limiting value , @xmath214 , one can find the corresponding amplitude @xmath215 from the second expression of eq . ( [ n0p1n0d ] ) and then @xmath216 from the last expression . when @xmath172 increases from @xmath214 to @xmath197 in eq . ( [ n0p1n0d ] ) , the parameter @xmath127 of the nonlinear map eq . ( [ pulse ] ) corresponding to the root @xmath160 ( @xmath161 ) increases from @xmath216 ( decreases from @xmath216 ) to @xmath199 ( to @xmath244 ) . thus , the @xmath158 solution occupies the rest of the admissible region for the case of @xmath243 ( see fig . [ figure3 ] ) . in the case of @xmath149 presented in fig . [ figure3 ] , we find @xmath245 and @xmath162 . both @xmath179 and @xmath156 solutions , in the limit @xmath246 ( @xmath247 ) , reduce to a homoclinic to @xmath196 pulse solution ( see also figs . [ figure1 ] and [ figure3 ] where this possibility is illustrated ) . this solution has the form : @xmath248 , \label{pulsecooper}\end{aligned}\ ] ] where @xmath249 this is illustrated by fig . [ figure7 ] where , taking @xmath149 , we show the solution obtained from the map eq . ( [ pulse ] ) at ( a ) @xmath250 ( @xmath156 solution ) and ( b ) @xmath251 ( @xmath179 solution ) , for initial value of @xmath252 . the figure clearly illustrates the two limits ( to the left and to the right of @xmath145 ) and their correspondence to pairs of pulse pulse solutions and ones of pulse anti - pulse solutions , as one enters the two different regimes `` dn '' and `` cn '' of fig . [ figure3 ] . solutions of this special form can be expressed as @xmath253 with constant @xmath166 and @xmath254 . for the sake of simplicity , we set @xmath255 and substitute the ansatz into eq . ( [ physicad ] ) or eq . ( [ saxena ] ) to find the zigzag solution @xmath256 to obtain this solution from the map eq . ( [ pulse ] ) , one has to set @xmath257 , which is the general condition for getting @xmath258 . substituting here @xmath166 from eq . ( [ zigzag ] ) , we get @xmath259 . for the case of @xmath137 presented in fig . [ figure2 ] , we have @xmath260 . the zigzag solution is an exceptional one , as @xmath261 for any @xmath24 . however , as shown above , this is only possible when eq . ( [ pulse ] ) has multiple roots , which means that the two - point static problem , eq . ( [ firstintdiscrete ] ) , is factorized . this solution does not have a counterpart in the continuum limit . at ( a ) @xmath250 and ( b ) @xmath262 , for initial value of @xmath252 . since @xmath127 is close to @xmath263 , corresponding to @xmath92 , the dn solution in ( a ) and the cn solution in ( b ) look like chains of separated pulses given by eq . ( [ pulsecooper ] ) . ] the static solutions of the discrete models eq . ( [ physicad ] ) and eq . ( [ saxena ] ) are exactly the same , but the dynamical properties of the two models are different . as the corresponding stability analysis takes into account the dynamical form of each model , it will be carried out separately for each of the two . let us first consider the mc model of eq . ( [ physicad ] ) . introducing the ansatz @xmath264 ( where @xmath265 is an equilibrium solution and @xmath266 is a small perturbation ) , we linearize eq . ( [ physicad ] ) with respect to @xmath267 and obtain the following equation : @xmath268 for the small - amplitude phonons , @xmath269 , with frequency @xmath270 and wave number @xmath271 , around the uniform steady states @xmath272 ( @xmath120 ) , eq . ( [ linphysicad ] ) is reduced to the following dispersion relation : @xmath273 while the spectrum of the vacuum solution for @xmath117 , @xmath274 , is @xmath275 for an arbitrary stationary solution @xmath265 , stability is inferred in the mc model if the eigenvalue problem obtained from eq . ( [ linphysicad ] ) by replacing @xmath276 with @xmath277 has only non - negative solutions @xmath278 . recalling that the mc model is a non - hamiltonian one , the resulting eigenvalue problem is a non - self - adjoint one involving a non - symmetric matrix . similarly , we obtain analogous expressions for the ec model of eq . ( [ saxena ] ) . the linearized equation reads @xmath279 ^ 2 } \varepsilon_{n } , \label{linsaxena}\end{aligned}\ ] ] and the corresponding dispersion relation for the linear phonon modes around @xmath280 has the form : @xmath281 while that of vacuum solution @xmath118 ( for @xmath117 ) is @xmath282 the stationary solution @xmath265 is stable in the ec model of eq . ( [ saxena ] ) if the self - adjoint eigenvalue problem obtained from eq . ( [ linsaxena ] ) by replacing @xmath276 with @xmath277 has only non - negative solutions @xmath278 . we note that all stable and unstable static solutions of the mc and the ec models , except for the zigzag solution eq . ( [ zigzag ] ) , possess a zero - frequency mode . this is a consequence of the effective translational invariance of the discrete pnb - free models ; this is related also to the freedom of selecting the free parameter @xmath3 in the corresponding solution expressions . our aim here is not to study the whole bunch of solutions in the whole range of parameters , but rather to demonstrate the existence of stable solutions and also provide some examples where solutions , being stable in one model , may be unstable in another . the results of stability analysis for some characteristic solutions are summarized in table [ table ] . first , we consider the kink and inverted kink solutions given in eq . ( [ kinkcooper ] ) and eq . ( [ invertedkink ] ) respectively : in a numerical experiment , these solutions were placed in the middle of a chain of @xmath283 particles with clamped boundary conditions . we have found that for the chosen set of parameters , the kink is stable in both mc and ec models , while _ the inverted kink is stable in the ec and unstable in the mc model_. in fig . [ figure8 ] we show the spectra of the kink and the inverted kink for their different positions with respect to the lattice , @xmath3 . in the figure , the horizontal lines show the borders of the phonon band of the vacuum @xmath119 and the dots show the frequencies of the kink s internal modes lying outside of the band . in all cases , the spectra are shown as functions of @xmath3 , even though , admittedly , the kink in the mc model presented in ( a ) demonstrates very weak sensitivity of its spectrum to variations of @xmath3 . let us now consider the periodic solutions depicted in fig . [ figure5](b ) ( @xmath147 ) and in fig . [ figure6](a ) ( @xmath203 ) . for these classes of solutions we used periodic boundary conditions and the length of the lattice was commensurate with the period of the solutions containing a number of periods . close to the hyperbolic function limit , these solutions are stable in the ec model but unstable in the mc model , following the stability of the building blocks ( inverted kinks ) associated with these structures . on the other hand , the @xmath203 solution becomes unstable in both models at the trigonometric limit depicted in fig . [ figure6](b ) . nonetheless , we find it remarkable that some of these `` apparently non - smooth '' solutions ( or maybe more correctly , solutions apparently containing a non - smooth continuum limit ) may be potentially stable in the discrete setting . it would be wortwhile to analyze the stability of such structures in more detail and as a function of their elliptic modulus @xmath172 . next , we consider the zigzag solution , eq . ( [ zigzag ] ) , which exists for @xmath284 . it can be shown that this solution can be stable in the ec model : indeed , substituting eq . ( [ zigzag ] ) into the relevant eigenvalue problem , we obtain the dispersion relation for the linear phonon modes , @xmath285 $ ] . the solution is stable when @xmath278 are non - negative , i.e. , when @xmath286 . combining this with the existence condition we find that the zigzag solution exists and it is stable in the ec model for @xmath287 . note that this exceptional solution does not possess the zero frequency mode , as it can be seen from the above dispersion relation . the amplitude of the solution , @xmath288 , is always greater than 1 . the zigzag solution is always unstable in the mc model , a result that can similarly be demonstrated . , @xmath137 ( @xmath114 ) . the inverted kink in mc model is unstable for the chosen parameters . horizontal lines show the borders of the phonon bands . ] the @xmath176 , @xmath179 , and @xmath156 solutions close to the hyperbolic limit were found to be generically unstable in both the mc and ec models . the instability of the pulse solution of eq . ( [ pulsecooper ] ) can be analytically justified in the present setting . it can be easily inferred from the invariance of the solution with respect to @xmath3 that the eigenvector leading to the zero eigenvalue of fig . [ figure8 ] is proportional to @xmath289 ( where @xmath290 is the pulse solution profile ) . this is an anti - symmetric eigenvector , given the pulse s symmetric ( around its center ) nature . but then , from sturm - liouville theory for discrete operators @xcite , there should be an eigenvalue such that @xmath291 with a symmetric ( i.e. , even ) eigenvector , resulting in the instability of the relevant pulse solution . our numerical simulations fully support this conclusion . the pulse solution was constructed by iterating eq . ( [ pulse ] ) for @xmath243 , @xmath145 , and @xmath292 , or directly from eq . ( [ pulsecooper ] ) . we found that for @xmath37 the pulse solution is unstable in both models and over a wide range of the discreteness parameter @xmath14 , including the case of rather small @xmath293 . we have also confirmed that the instability mode is similar to the pulse profile ( i.e. , of even parity ) . both the cn and dn solutions ( see , e.g. , fig . [ figure7 ] ) are also found to be unstable for the different parameter values that we have used in both the mc and ec models . this is rather natural to expect given that their `` building blocks '' , namely the pulse - like solutions are each unstable ( hence their concatenation will only add to the unstable eigendirections ) . finally , as far as the sn solutions are concerned , two basic instability modes were revealed . one resembles the instability of the pulse solution , when the whole structure , being waved around the top of the background potential , tends to slide down to one of the potential wells . the other instability modes can be understood if one regards the sn solution as a set of alternating kinks and anti - kinks [ see fig . [ figure5](a ) ] interacting with each other by means of overlapping tails . such a system is unstable because the kinks tend to annihilate with the anti - kinks . it is interesting to note that these stability results suggest a disparity with the earlier numerical findings of @xcite where the pulse - like solution was generically reported to be stable and analogous statements were made for the cn- and dn - type solutions . at this point , it is worth discussing a connection with nls - type models mentioned in the introduction . in the case of such models , the aforementioned unstable eigendirection ( with an eigenvector similar in profile to the pulse itself ) is `` prohibited '' by the additional conservation law of the @xmath294 norm , resulting in a separate neutral eigendirection with zero eigenfrequency . as a consequence , it is an interesting twist that the instability reported for such pulse - like solutions in scalar @xmath0 models would be absent in their discrete nls counterparts . .results of stability analysis [ table ] [ cols="<,^,^ " , ] in @xcite , we have compared the spectra and the long - term dynamics of the kinks in the two pnb - free models , eq . ( [ physicad ] ) and eq . ( [ saxena ] ) , with that of the classical discrete model , eq . ( [ kleingordondiscrete ] ) ( see fig . 1 of that paper ) . the case of very small kink velocities was not studied there . here we would like to focus on this case to demonstrate the qualitatively different behavior for slowly moving kinks in the pnb - free and classical models . to boost the kink in the pnb - free models eq . ( [ physicad ] ) and eq . ( [ saxena ] ) we used the following dynamical solution corresponding to the multiple eigenvalue @xmath295 , @xmath296 , where @xmath265 is a static kink solution , @xmath267 is the normalized translational eigenmode and @xmath297 is the amplitude playing the role of kink velocity . this construction yields a more accurate approximate solution for very small @xmath297 when linearized equations eq . ( [ linphysicad ] ) and eq . ( [ linsaxena ] ) are accurate . increasing @xmath297 leads to a decrease in the accuracy of the dynamical solution used for boosting . on the other hand , we note in passing that there are techniques developed recently for klein - gordon @xcite and even nonlinear schrdinger lattices @xcite that also allow the construction of finite speed , numerically exact travelling solutions in these classes of models . in fig . [ figure9 ] we present the lowest frequency normalized eigenmodes for the inter - site kinks in the three models , for the case of @xmath114 , @xmath36 ( @xmath137 ) . both pnb - free models , eq . ( [ physicad ] ) and eq . ( [ saxena ] ) , have the same translational eigenmodes with @xmath298 , shown by dots connected with solid lines . one can show by straightforward algebra that the pnb - free model modified by a non - vanishing multiplier @xmath299 as described in sec . [ sec : generalizations ] has the same translational eigenmode as the original model . assuming that the two models are respectively : @xmath300 where the function @xmath301 , then the steady state solutions @xmath265 will satisfy @xmath302 , for all @xmath24 , while the corresponding linearization equations ( using @xmath303 ) are respectively of the form @xmath304 . \label{extra4}\end{aligned}\ ] ] hence , when solving the corresponding eigenvalue problem ( again substituting @xmath305 by @xmath306 ) , for @xmath298 , the eigenvalue problems become identical , both satisfying @xmath307 hence the coincidence of the corresponding eigenvectors . we are extremely grateful to an anonymous referee for this remark and its proof . we do note , in passing , also that while this justifies the coincidence of the zero eigenfrequency modes of the pnb - free models in fig . [ figure9 ] , it is also in tune with the results of fig . [ figure8 ] for nonzero eigenfrequencies @xmath270 . the latter are not identical between the different models , as indicated by the left and middle panels of that figure . this is due to the differences between the corresponding eigenvalue problems of eqs . ( [ extra3 ] ) and ( [ extra4 ] ) , when @xmath308 . for the classical model , eq . ( [ kleingordondiscrete ] ) , the lowest - frequency mode has @xmath309 , and it is shown in fig . [ figure9 ] by open circles and dashed lines . actually , this mode is not a translational mode ( since , strictly speaking , there is no translational invariance ) but we use it to boost the kink . one can say that this will _ become _ the translational mode for this model in the continuum limit . we define the kink center of mass as @xmath310 the evolution of the kink coordinates is shown in fig . [ figure10 ] . kinks were boosted with two different amplitudes of the normalized lowest - frequency eigenmodes , @xmath311 and @xmath312 . results for the pnb - free models practically coincide , as depicted by the dashed and solid lines for the mc and ec models , respectively . it is readily seen that kinks in the pnb - free lattices propagate with roughly constant velocities . the oscillatory trajectories in fig . [ figure10 ] correspond to the kink in the classical discrete @xmath0 model . the faster kink , boosted with @xmath312 , propagates along the lattice but its velocity gradually decreases . the kinetic energy of the translational motion is partly lost to the excitation of the kink internal mode with @xmath313 , lying below the phonon frequency band . higher harmonics of this mode interact with the phonon spectrum producing radiation , which also slows the kink down . an even more dramatic difference is observed for the slower kinks , boosted with @xmath311 . here , the classical kink can not overcome the pn barrier and can not propagate , oscillating near the stable inter - site configuration . on the other hand , the kinks in the pnb - free models are not trapped by the lattice and propagate due to the absence of the pn barrier . alternatively , one can say that such waveforms can be accelerated by arbitrarily small external fields . ) and eq . ( [ saxena ] ) , have the same translational eigenmode with @xmath298 , shown by dots connected with solid lines ( see also the relevant discussion in the text ) . for the classical model , eq . ( [ kleingordondiscrete ] ) , the lowest - frequency mode has @xmath314 ( shown by open circles and dashed lines ) . results for @xmath114 , @xmath36 ( @xmath137 ) . ] , as a function of time for kinks propagating in the pnb - free mc ( dashed line ) and ec ( solid line ) models , as well as in the classical discrete @xmath0 model ( oscillatory line ) . the kinks were boosted with two different amplitudes of the normalized lowest - frequency eigenmodes , @xmath311 and @xmath312 . the faster classical kink is able to propagate , while the slower one is not , since it can not overcome the pn barrier . kinks in pnb - free models are not trapped by the lattice and can therefore propagate . ] in the continuum limit , @xmath55 , the borders of the admissible region , eq . ( [ borders ] ) , become @xmath315 field for ( a ) @xmath316 and ( b ) @xmath37 obtained as the continuum limits ( @xmath55 ) of those presented in fig . [ figure2 ] and fig . [ figure3 ] , respectively . in each panel there is one inadmissible region marked with `` no '' . ] in fig . [ figure11 ] we plot the admissible regions for ( a ) @xmath36 and ( b ) @xmath37 . the topology of the admissible regions for the @xmath0 field is simpler than the one pertaining to the discrete models . in the continuum limit there exists a sole inadmissible region for both cases @xmath90 , while three inadmissible regions exist for the discrete models in the case @xmath193 . one more simplification is that the domains of the @xmath147 and @xmath179 solutions do not split into two parts since the smaller root @xmath160 disappears in the continuum limit . particularly we note that the region marked with the question mark in fig . [ figure2 ] , @xmath317 , disappears in the continuum limit . that might be the reason why we failed to find a jacobi elliptic function expression for the discrete static solutions in this case . static solutions eqs . ( [ p1n0n0d]-[p1m1p1d ] ) obtained for the discrete models have their continuum counterparts as the traveling solutions to the @xmath0 field , eq . ( [ kleingordon ] ) . the general form of the solutions is @xmath318 where @xmath165 is the modulus of the jacobi elliptic functions , @xmath166 and @xmath167 are the parameters of the solution , @xmath3 is the arbitrary initial position and @xmath319 is the velocity of the solution . the integers @xmath168 once again specify the particular form of the solution . continuum analogues of eqs . ( [ p1n0n0d])-([p1m1p1d ] ) have the following form and are characterized by the following parameters : the @xmath176 solution , @xmath177 , @xmath320 the @xmath179 solution , @xmath180 , @xmath321 the @xmath156 solution , @xmath182 , @xmath322 the @xmath203 solution , @xmath185 , @xmath323 the @xmath206 solution , @xmath188 , @xmath324 the @xmath147 solution , @xmath190 , @xmath325 the above six solutions can be rewritten in a great variety of forms using the properties of the jacobi elliptic functions @xcite . however , we believe that they may constitute the full list of physically different solutions to the continuum @xmath0 equation since they fill the whole two - parameter space , @xmath326 , obtained as the continuum limit of corresponding space of the discrete models . all solutions are conveniently parameterized by a single parameter @xmath327 for @xmath120 , and @xmath328 for @xmath117 , as it is presented in fig . [ figure11 ] . the solutions in eqs . ( [ p1n0n0])-([n0n0p1 ] ) are bounded while the other ones are unbounded . the solutions in eq . ( [ n0p1n0 ] ) and eq . ( [ n0n0p1 ] ) are defined for @xmath329 while the others for @xmath120 . the solutions in eq . ( [ n0p1n0 ] ) and eq . ( [ p1m1p1 ] ) are valid for @xmath330 , the solution in eq . ( [ n0m1n0 ] ) is valid for @xmath331 , while the other solutions for @xmath165 . in the present paper we have shown that the reduction of the static problem of a discrete klein - gordon ( and by extension of the standing wave problem of a discrete nls ) equation to a two - point problem is a powerful tool for obtaining all possible static solutions of the corresponding model . we have applied this general idea to a momentum conserving @xcite and an energy conserving @xcite discretization of the @xmath0 model analyzing the full two - parameter plane of solutions and giving a natural parametrization for it . in particular , we have examined the admissible regions of the field value at a given point and of the constant entering the two - point function pertinent to the model . we have specifically illustrated how to use different choices of these two parameters to obtain not only the well - known hyperbolic function solutions and the established elliptic function solutions , but also novel classes of solutions including the inverted ( non - monotonic ) kink solutions presented herein and multi - kink generalizations thereof . we performed such computations both for the attractive ( focusing ) and for the repulsive ( defocusing ) types of nonlinearity . the presented methodology has the significant advantage over earlier work that by introducing the integration constant in the discretization of the first integral of the static part of the pde , it allows one to construct the _ full _ family of the static solutions of the corresponding model . earlier work had implicitly set this additional free parameter to @xmath196 ; this choice was sufficient for obtaining the important hyperbolic function solutions , but the present formalism systematically illustrates how to generalize the latter . we have also derived the continuum analogues of the discrete solutions and found that they fill the whole space of parameters obtained in the limit of @xmath332 from the parameter space of the discrete models . in addition , we have examined some of the key stability features of the solutions obtained in the various models , illustrating the different stability properties of the models considered herein ( even if their static solutions are identical ) . we have obtained interesting stability properties , including , e.g. , the counter - intuitive stability of the inverted kink and of some of its periodic generalizations for the ec model . the present study generates a variety of interesting questions . for instance , it would be relevant to examine whether the solutions presented herein ( including the inverted ones ) have counterparts in the `` standard '' discrete @xmath0 model and to analyze their respective stability . it would also be relevant to examine how the stability ( and existence since some of them may not survive the continuum limit ) of such solutions depend on the lattice spacing @xmath14 . finally , these and related ( e.g. stability ) questions in the context of discrete nls lattices ( see @xcite ) would also be equally or even more ( given the multitude of relevant applications of the latter model ) interesting to answer . such studies are currently in progress and will be reported elsewhere . the authors would like to thank d. e. pelinovsky for a number of stimulating and insightful discussions . we are also grateful to an anonymous referee for offering a proof that the mc and ec models have the same translational eigenmodes . pgk gratefully acknowledges support from nsf - dms-0204585 , nsf - dms-0505663 and nsf - career .

in view of various applications such as the study of two dimensional melting @xcite , investigations of mesoscale structure formation @xcite or engineering of colloidal crystals on spherical surfaces @xcite , the self assembly of sub-@xmath0 m colloidal particles at water air or water oil interfaces has gained much interest in recent years . these particles are trapped at the interface if the colloid is only partially wetted by both the water and the oil . this configuration is stable against thermal fluctuations and appears to be the global equilibrium state , as it is observed experimentally that the colloids immersed in the bulk phases are attracted to the interface @xcite ( see also sec . [ sec:1coll ] ) . the mutual interaction between the trapped colloids at distances close to contact , i.e. , within the range of molecular forces , is dominated by strong van der waals attraction . in order to avoid coagulation due to this attraction , the colloids can be stabilized sterically with polymers or with charges such that the colloids repel each other . variants of charge stabilization may include the coverage with ionizable molecules which dissociate in water , or the labeling of colloids with charged fluorescent markers . for charge stabilized colloids at large distances , the resulting repulsive force at water interfaces stems from a dipole dipole interaction as shown theoretically for point charges on surfaces @xcite and verified experimentally for polystyrene ( ps ) spheres on water oil interfaces @xcite . nonetheless , charged colloids at interfaces also show attractions far beyond the range of van der waals forces . the corresponding experimental evidence can be roughly classified as follows . ( i ) according to refs . @xcite , ps spheres ( radii @xmath1 @xmath0 m ) on flat water air interfaces using highly deionized water exhibit spontaneous formation of complicated metastable mesostructures . they are consistent with the presence of an attractive , secondary minimum in the intercolloidal potential at distances @xmath2 with a depth of a few @xmath3 . the use of water slightly contaminated by ions seems to move the minimum further out and to reduce its depth @xcite . ( ii ) in ref . @xcite , pmma spheres with radius @xmath4 @xmath0 m were trapped at the interface of water droplets immersed in oil . here , the secondary minimum has been measured at a distance @xmath5 and is reported to be surprisingly steep . the tentative explanation of these findings given in ref . @xcite invokes an analogue of long ranged flotation or capillary forces which decay @xmath6 . this interpretation was criticized in ref . @xcite ( with which the authors of ref . @xcite agreed @xcite ) and in ref . @xcite which both concluded that possible capillary forces in this system are much shorter ranged , i.e. , @xmath7 , but the authors of these references disagree with respect to the sign of this shorter ranged force . in yet another twist of the story , after completion of our work we encountered the very recent ref . @xcite in which the authors claim that long ranged capillary forces @xmath6 caused by the colloidal charges persist for sub-@xmath0 m particles . this conclusion is based on measurements of the meniscus shape around single glass spheres with radii 200 300 @xmath0 m floating at water oil and water air interfaces . motivated by the experimental data summarized above and the still incomplete theoretical understanding , here we undertake a quite general analysis of capillary interactions between two spherical colloids trapped at fluid interfaces . we characterize the system by a general stress field which acts on the interface , e.g. , due to a discontinuous electrostatic field at the interface , and by forces on the colloid of gravitational and/or electrostatic origin . special attention will be given to the role of a restoring force fixing the interface , e.g. , due to gravity or interface pinning . in sec . [ sec:1coll ] we present a free energy model for a single colloid trapped at an interface in the limit of small stresses and forces . the general solution for the interface deformation will be used in sec . [ sec:2coll ] in order to determine the effective potential between two colloids within the superposition approximation . in view of the differing theoretical results in the literature , the derivation of the interface deformation and the resulting effective potential is presented in detail in order to reveal properly the subtleties involved . in sec . [ sec : discussion ] we apply the results to the case of charged polymeric spheres on water interfaces . it turns out that the constraint of approximate mechanical isolation of the experimental systems renders the capillary interaction basically short ranged . long ranged attractive forces @xmath6 only arise through a restoring force acting on the interface ( which is , however , weak ) or in the presence of an external electric field . we shall discuss the relation to previous theoretical results , especially in view of the experimental results reported in refs . directions for further research will be pointed out . in sec . [ sec : summary ] we summarize our results . in this section we consider as a first step the equilibrium state of a single colloid of radius @xmath8 at the interface between two fluid phases denoted as 1 and 2 . the contact angle formed by the interface and the colloid surface is given by young s law , @xmath9 where @xmath10 is the surface tension between phases 1 and 2 , and @xmath11 is the surface tension between the colloid and phase @xmath12 . as a reference state , with respect to which changes in free energy will be measured , we take a planar meniscus configuration with the colloid at such a height @xmath13 that young s law is satisfied ( fig . [ fig : ref ] ) . this corresponds to the equilibrium configuration of an uncharged colloid at the interface if its weight can be neglected which for generic cases is a safe approximation for @xmath14 @xmath0 m @xcite ( see also sec . [ sec : discussion ] ) . we model the colloid as a smooth sphere so that the system is invariant under rotations around the colloid axis perpendicular to the reference planar meniscus . the presence of charges induces a shift of the system ( colloid and interface ) with respect to the reference state . here we neglect corresponding changes in the surface tensions @xmath10 and @xmath11 ; this approximation is expected to be valid provided the concentration of charges is sufficiently small . this shift is characterized by the meniscus profile @xmath15 relative to the planar configuration and by the height @xmath16 of the colloid center . in the reference configuration , the charge distribution is assumed to be already in equilibrium . in the following we do not consider the degree of freedom `` charge density field '' explicitly but take it to be fixed to that of the reference configuration . this amounts to neglecting the feedback of the interface displacement on the charge distribution . it turns out to be useful to introduce the radius @xmath17 of the three phase contact line and the angle @xmath18 as auxiliary variables ( see fig . [ fig : pert ] ) . in this subsection we formulate a free energy functional for the degrees of freedom @xmath15 and @xmath16 . as shown later , for the cases of interest here the deviations from the reference configuration are small enough to justify a perturbative treatment , and the free energy can be expanded up to quadratic order in @xmath19 and @xmath15 . we denote with @xmath20 the vertical force per unit area acting on the meniscus surface in the reference configuration . in the case of a charged colloid , @xmath20 is given by the @xmath21component of the difference of the maxwell stress tensor right above and below the meniscus , plus the pressure difference acting across the meniscus ( incluiding an imbalance in osmotic pressure due to the different concentration of ions just above and below the meniscus ) @xcite . we introduce the following dimensionless parameter to measure the relative strength of this force : @xmath22 where the integral extends over the flat meniscus @xmath23 ( the plane @xmath24 with a circular hole of radius @xmath25 ) . we assume @xmath26 with @xmath27 , so that the integral converges . in the same spirit we introduce the total vertical force @xmath28 acting on the charged colloid in the flat reference configuration , which includes gravity , the electrostatic force , and the total ( i.e. , hydrostatic and osmotic ) pressure exerted by the fluids . this leads to the definition @xmath29 these two dimensionless parameters will appear naturally in the course of the calculations . if @xmath30 and @xmath31 vanish , the reference configuration _ is _ the equilibrium state ; it is the global minimum of the free energy functional given in eqs . ( [ eq : fdef ] ) and ( [ eq : f ] ) below if @xmath32 , @xmath33 . the aforementioned perturbative expansion of the free energy can be rephrased as an expansion in terms of the small parameters @xmath30 and @xmath31 . the free energy @xmath34 of the colloid expressed relative to the aforementioned reference configuration consists of five terms : @xmath35 in the following we discuss each contribution : * _ fluid contact of the colloid_. with @xmath36 denoting the surface area of the colloid which is in contact with phase @xmath12 , the surface free energy of the colloid due to its exposure to the phases 1 and 2 is @xmath37 in appendix [ sec : fcont ] we express this contribution as a function of @xmath38 and @xmath15 . the final result , valid up to corrections of at least third order in @xmath30 or @xmath31 , reads @xmath39 ^ 2 + \pi \gamma ( r_0 ^ 2 - r_{0,\rm ref}^2 ) . \ ] ] * _ change of the meniscus area_. the free energy contribution due to variations in the meniscus area relative to the reference state reads @xmath40 where @xmath41 is the surface of the fluid interface projected onto the plane @xmath24 ( in which the reference interface is located ) , and @xmath42 is the gradient operator on the flat reference interface . for small slopes ( @xmath43 ) one obtains : @xmath44 since the @xmath45dependent term is of second order in @xmath45 , we have approximated the integration domain @xmath41 by @xmath46 ; the corrections are at least of third order in the small parameters @xmath30 or @xmath31 . the first term in this expression represents the change in the area of the meniscus which is cut out by the colloid , and in eq . ( [ eq : fdef ] ) it cancels the second term of eq . ( [ eq : fcont_iso ] ) . * _ volume forces on the fluids_. we consider the case that the only volume force acting on the fluid phases is gravity . the electrostatic forces are active only at the surfaces , where a net charge can accumulate . if the spherical colloid is replaced by a long cylinder of radius @xmath25 ( fig . [ fig : cylinder ] ) the change in gravitational potential energy relative to the reference state due to the displacements of volumes of the fluid phases can be calculated easily : @xmath47 assuming that the mass density of phase 2 is larger than the one of phase 1 , i.e. , @xmath48 ; @xmath49 denotes the capillary length @xmath50 in terms of the acceleration @xmath51 of gravity . for a spherical colloid @xmath52 in eq . ( [ eq : fvol ] ) has to be corrected to account for the specific dependence of the displaced fluid volume on the colloid height @xmath16 , yielding a slightly cumbersome expression @xcite . this correction is usually numerically small and we neglect it for reasons of simplicity . as it will be shown in subsec . [ sec : lambda ] this is justified because the results of interest here are insensitive to the precise form of @xmath53 in the limit @xmath54 ( @xmath55 mm , which is much larger than any other relevant length scale of the system ) . * _ force on the fluid interface_. the aforementioned surface force density @xmath20 acts on the fluid interface between phase 1 and phase 2 . the free energy change due to the ensuing displacements of the meniscus is @xmath56 here @xmath20 is the surface force in the _ reference _ configuration ( @xmath57 corresponds to a force pointing upward ) . changes in the force induced by meniscus deformations and colloidal displacements contribute terms of higher orders in @xmath30 or @xmath31 in eq . ( [ eq : finter ] ) . the replacement of @xmath58 by @xmath23 in eq . ( [ eq : finter ] ) introduces terms of higher order , too . * _ contribution from the colloid_. the free energy change due to a vertical displacement of the colloid is @xmath59 where @xmath60 is the vertical force on the colloid in the _ reference _ configuration . like for @xmath61 , changes of @xmath60 due to deviations from the reference configuration contribute to higher order terms . in conclusion , by adding eqs . ( [ eq : fcont_iso ] ) , ( [ eq : fmen ] ) , ( [ eq : fvol ] ) , ( [ eq : finter ] ) , and ( [ eq : fcoll ] ) we obtain the following approximate expression for the total free energy , which is correct up to second order in @xmath30 or @xmath31 , and which is a function of @xmath38 and a functional of @xmath15 : @xmath62 + \pi \gamma [ u_0 - \delta h ] ^2 - f \delta h , \ ] ] where @xmath63 . the equilibrium configuration of the system minimizes the free energy expression ( [ eq : f ] ) . the minimization proceeds in two stages . first we seek the minimum with respect to @xmath38 at fixed @xmath15 : @xmath64 where we have used the definition ( [ eq : epsf ] ) . using repeatedly the definitions of @xmath16 and @xmath17 in terms of the auxiliary angle @xmath18 ( fig . [ fig : pert ] ) we can compute the change of the contact radius : @xmath65 plus corrections of second order in @xmath30 , @xmath31 . in the second step , we minimize eq . ( [ eq : f ] ) with respect to @xmath15 at fixed @xmath38 . this is a problem of variations with a free boundary condition at @xmath66 @xcite . variation with respect to @xmath67 yields a second order ordinary differential equation , @xmath68 while variation with respect to @xmath69 provides a boundary condition : @xmath70 where the last equality follows from eq . ( [ eq : h_eq ] ) . the second boundary condition one has to impose on eq . ( [ eq : younglaplace ] ) follows from the requirement that the meniscus is asymptotically flat far from the colloid ( assuming that @xmath71 ) : @xmath72 equation ( [ eq : younglaplace ] ) describes mechanical equilibrium of the interface such that the laplace pressure balances the forces acting on the interface . the boundary condition ( [ eq : u0 ] ) expresses mechanical equilibrium of the colloidal particle : at the contact line ( i.e. , the circle with radius @xmath17 at @xmath73 ) the interface exerts a force onto the colloid which has a non vanishing contribution only in @xmath74direction with the magnitude @xmath75 \approx 2 \pi \gamma r_{0,\rm ref } u'_0 $ ] . this contact line force is balanced by the total force @xmath60 . the solution of eqs . ( [ eq : younglaplace][eq : u_infty ] ) can be written in terms of the modified bessel functions of zeroth order : @xmath76 , \ ] ] where the integration constant @xmath77 is determined by the boundary condition ( [ eq : u0 ] ) . for typical values of the parameters , @xmath49 is of the order of millimeters and therefore much larger than any other length scale occurring for experiments with sub - micrometer colloids . in order to study the intermediate asymptotics ( @xmath79 ) of @xmath15 as given by eq . ( [ eq : solbessel ] ) , we insert the asymptotic expansions of the bessel functions @xcite as @xmath78 and retain those terms which do not vanish in this limit . assuming that @xmath80 decays sufficiently fast , eq . ( [ eq : solbessel ] ) reduces to @xmath81 where @xmath82 and @xmath83 is euler s constant . in eq . ( [ eq : sol1 ] ) , we have expressed the integration constant @xmath77 appearing in eq . ( [ eq : solbessel ] ) in terms of @xmath31 by using the boundary condition ( [ eq : u0 ] ) . the first term in eq . ( [ eq : sol1 ] ) is a solution of the homogeneous part of eq . ( [ eq : younglaplace ] ) ( with @xmath84 in the equation ) , demonstrating that the limit @xmath78 is singular as long as @xmath85 . the second term corresponds to a particular solution of the inhomogeneous differential equation . if the surface force @xmath20 decays algebraically , @xmath86 , this term decays like @xmath87 ( since we have assumed @xmath27 ) , so that the logarithmic contribution is dominant . if @xmath88 , the asymptotic behavior is no longer given by eq . ( [ eq : sol1 ] ) but a different dependence on @xmath49 arises . at distances @xmath89 of the order of @xmath49 , the expression ( [ eq : sol1 ] ) is not valid and a crossover to the exact solution ( [ eq : solbessel ] ) takes place in order to satisfy the boundary condition at infinity ( eq . ( [ eq : u_infty ] ) ) . figure [ fig : u(r ) ] sketches the behavior of the meniscus profile @xmath15 . as can be checked directly in the differential equation ( [ eq : younglaplace ] ) , one has @xmath90 asymptotically as @xmath91 , which expresses the balance between the surface force @xmath20 on the meniscus and the gravitational force . there is , however , another intermediate regime @xmath92 but not too large , in which @xmath15 decays @xmath93 and which corresponds to the solution of eq . ( [ eq : younglaplace ] ) with @xmath94 set to zero , i.e. , when gravity is balanced by the laplace pressure induced by the meniscus curvature . these conclusions , as well as the functional dependence of @xmath15 on @xmath20 , are robust and independent of the details of the boundary condition at @xmath95 . in order to support this statement , we consider two cases which implement the distant boundary condition differently : * _ pinned interface_. in the absence of gravity the interface is assumed to be pinned at a finite distance @xmath96 from the colloid . this corresponds to setting @xmath84 in eq . ( [ eq : younglaplace ] ) and replacing eq . ( [ eq : u_infty ] ) by the new boundary condition @xmath97 . in this case the solution of eq . ( [ eq : younglaplace ] ) is given by @xmath98 where @xmath99 in analogy to eq . ( [ eq : epspi ] ) . in the intermediate asymptotic regime ( @xmath100 ) , the meniscus profile is then given by @xmath101 assuming that @xmath20 decays sufficiently fast so that the integrals in eqs . ( [ eq : sol_pin ] ) and ( [ eq : epspi_pin ] ) converge as @xmath102 . equation ( [ eq : sol2 ] ) resembles eq . ( [ eq : sol1 ] ) with @xmath103 replaced by @xmath96 . * _ pinned curved reference interface_. in some experiments the interface between phases 1 and 2 is in fact closed , so that the colloidal particle lies at the surface of a large nonvolatile spherical droplet of phase 2 which is immersed in phase 1 @xcite ( fig . [ fig : droplet ] ) and fixed by certain means ( e.g. by a glass plate ) . + the free energy functional ( [ eq : f ] ) has to be modified to account for the curvature of the reference interface as well as of the constraint that the droplet volume remains unchanged under deformation . to determine the interface deformation , we minimize the functional and employ the boundary condition that the droplet is fixed at some point far from the colloid . the mathematical details and the corresponding solution for @xmath15 are presented in appendix [ sec : droplet ] . here we quote only the intermediate asymptotic behavior ( @xmath104 , where @xmath105 is the radius of the undeformed droplet ) : @xmath106 with @xmath107 a numerical constant given by eq . ( [ eq : tildec ] ) . again eq . ( [ eq : sol3 ] ) closely resembles eqs . ( [ eq : sol1 ] ) and ( [ eq : sol2 ] ) . the physical reason for the occurrence of the singularity in the limit @xmath108 is that a `` restoring force '' far from the particle is required to yield a well defined unperturbed interface which allows one to determine the deformation @xmath15 unambiguously . for example , if one takes the limit @xmath109 in the young laplace eq . ( [ eq : younglaplace ] ) , it is inconsistent to impose the boundary condition @xmath110 in the corresponding solution . the special case @xmath111 , however , is not singular ; this will be discussed in sec . [ sec : discussion ] . furthermore , the comparison of eqs . ( [ eq : sol1 ] ) , ( [ eq : sol2 ] ) , and ( [ eq : sol3 ] ) demonstrates that the functional form of the intermediate asymptotic behavior is independent of how the restoring force is implemented . this corresponds to the so called intermediate asymptotic behavior of the second kind @xcite , characterized by the following features : 1 . there is a length scale ( @xmath49 , @xmath96 , or @xmath105 ) which is much larger than the other length scales of the system under consideration and which seems at first sight to be irrelevant . 2 . nevertheless , this length scale determines the dominant logarithmic ( or more generally , the power law ) dependence . 3 . the detailed physical origin of this length scale ( in the examples we have considered , gravity , pinning of a reference flat or curved interface ) does not matter . well known examples of this kind of asymptotic behavior are critical phenomena in phase transitions @xcite . in that case , it is the microscopic length scale given by the amplitude of the correlation length which can not be set to zero although it is much smaller than the correlation length itself . this microscopic length scale is required to formulate the power law behavior of certain properties of the system , but its detailed physical origin is unimportant for the universal decay exponents of the power laws . in this section we consider the equilibrium state of two identical colloids floating at the interface at a fixed lateral distance @xmath112 and compute the effective interaction potential @xmath113 generated by the meniscus . the free energy can be derived along the same lines leading to eq . ( [ eq : f ] ) , but with due account for the fact that in the presence of two colloids the meniscus slope no longer exhibits rotational symmetry . nonetheless @xmath114 depends only on the distance @xmath112 between the centers of the two spheres . the reference configuration is that of two colloids floating on a planar interface with the corresponding reference free energy being independent of @xmath112 . in this case one has for the free energy relative to that of the reference configuration a contribution @xmath115 from the meniscus and a contribution of the form @xmath116 from each colloid . ( the total free energy includes also the direct interaction between the colloids ; this contribution will be considered in subsec . [ sec : elec ] . ) from eqs . ( [ eq : fcont_aniso ] ) , ( [ eq : fmen ] ) , ( [ eq : fvol ] ) , ( [ eq : finter ] ) , and ( [ eq : fcoll ] ) one obtains : @xmath117 + \sum_{\alpha=1,2 } \left\ { \frac{\gamma}{2 r_{0 , \rm ref } } \oint_{\partial s_\alpha } \!\!\ ! d\ell \ ; [ \delta \hat{h}_\alpha - \hat{u}]^2 - \hat{f } \delta \hat{h}_\alpha \right\ } .\ ] ] here , @xmath118 is the meniscus profile in the presence of two colloids , @xmath119 are the corresponding heights , @xmath120 is the vertical force per unit area on the meniscus in the reference configuration , and @xmath121 is the force on each colloid . by symmetry , one has @xmath122 and @xmath123 . @xmath124 are the circular disks delimited by the contact lines of the colloids in the reference configuration , @xmath125 are the contact lines , with the convention that we trace them counterclockwise , and @xmath126 ( see fig . [ fig:2coll_top ] ) . the equilibrium configuration is the minimum of the free energy given by eq . ( [ eq : f2 ] ) . the minimization procedure follows closely subsec . [ sec : minf ] . first , minimizing with respect to @xmath127 at fixed meniscus height @xmath118 leads to the height of the colloids , @xmath128 where @xmath129 is the mean height of the contact line . in the next step , we minimize with respect to @xmath118 at fixed @xmath119 . variation in the interior of the domain @xmath130 provides a second order partial differential equation , @xmath131 while variation at the boundary @xmath132 provides the following _ transversality conditions _ @xcite : @xmath133 where the last equality follows from eq . ( [ eq : h_eq2 ] ) . in this expression , @xmath134 is the derivative in the outward normal direction of @xmath125 . ( in this way , the triad ( @xmath135 ) is right - handed , where @xmath136 is the unit vector in the outward normal direction , @xmath137 is the unit vector in the counterclockwise tangent direction , and @xmath138 is the unit vector in the positive @xmath74direction . ) when applying the transversality condition , in the context of gauss theorem one must keep in mind that the boundary of the region @xmath130 consists of the contours @xmath125 traced in _ clockwise direction _ with the normals directed towards the interior of @xmath124 . ( the boundary at @xmath139 does not contribute due to eq . ( [ eq : u_infty2 ] ) . ) in the special case of a single colloid , rotational invariance reduces eq . ( [ eq : trans ] ) to eq . ( [ eq : u0 ] ) . finally , one has the additional boundary condition @xmath140 the solution of eq . ( [ eq : younglaplace2 ] ) with the boundary conditions given by eqs . ( [ eq : trans ] ) and ( [ eq : u_infty2 ] ) is a difficult task . ( we are only aware of the already very involved solution of the homogeneous eq . ( [ eq : younglaplace2 ] ) , i.e. , @xmath141 , with the simplified boundary condition @xmath142 @xcite . ) for the present purpose , one can use the so - called _ superposition approximation _ @xcite , which yields the correct solution in the asymptotic limit of large separation @xmath143 between the colloids . let @xmath144 denote the equilibrium meniscus profile as if colloid @xmath145 was alone , with @xmath146 and @xmath147 denoting the corresponding forces . the superposition approximation then reads : @xmath148 notice that the fields @xmath149 and @xmath150 are defined in the domain @xmath151 , while the fields @xmath152 and @xmath153 are defined in the smaller domain @xmath130 . equations ( [ eq : younglaplace2 ] ) and ( [ eq : u_infty2 ] ) are fulfilled by this approximate solution , but the boundary condition ( [ eq : trans ] ) is violated : using eqs . ( [ eq : superposition ] ) and the boundary condition in eq . ( [ eq : u0 ] ) for the single colloid solution , one obtains @xmath154_{{\bf r } \in \partial s_1 } \simeq \left [ \frac{\partial { u}_2 ( { \bf r})}{\partial n_1 } - \frac{1}{r_{0 , \rm ref } } \left\ { { u}_2 ( { \bf r})- \frac{1}{2 \pi r_{0 , \rm ref } } \oint_{\partial s_1 } \!\!\ ! d\ell \ ; { u}_2 \right\ } \right]_{{\bf r } \in \partial s_1 } , \ ] ] and a similar expression for the other colloid with the indices 1 and 2 interchanged . in general , this expression does not vanish as required by eq . ( [ eq : trans ] ) . if @xmath112 is large , eq . ( [ eq : trans1 ] ) can be evaluated by expanding @xmath155 into a taylor series around the center of colloid 1 , yielding to lowest order ( @xmath156 is the outward directed normal unit vector of @xmath157 ) @xmath158_{{\bf r } \in \partial s_1 } \simeq \frac{1}{4 } r_{0 , \rm ref } [ { \bf e}_1 { \bf e}_1 : \nabla \nabla { u}_2 ( d ) + \nabla^2 { u}_2 ( d ) ] \ ] ] in dyadic notation . inserting the single colloid solution given in eq . ( [ eq : sol1 ] ) , one finds that this expression decays like @xmath159 if @xmath160 , and like @xmath161 if @xmath162 . the superposition solution can be used to determine the vertical displacement according to eq . ( [ eq : h_eq2 ] ) : @xmath163 where @xmath164 is the relative vertical displacement of an isolated colloid ( eq . ( [ eq : h_eq ] ) ) and @xmath165 is the average of the single colloid meniscus height at the contact line of the other colloid . the meniscus induced effective potential between the two colloids ( without their direct interaction ) is defined as @xmath166 where @xmath167 is the free energy of the two colloid equilibrium configuration ( eq . ( [ eq : f2 ] ) ) while @xmath168 is the single colloid equilibrium free energy ( eq . ( [ eq : f ] ) ) . as noted before the energy of the reference configuration is independent of the separation @xmath112 and drops out from eq . ( [ eq : veff_a ] ) . we insert eqs . ( [ eq : superposition ] ) , ( [ eq : z_superposition ] ) , and ( [ eq : meanu ] ) into eq . ( [ eq : veff_a ] ) and exploit the invariance of the free energy under exchange of the colloids , i.e. , the symmetry under exchanging indices @xmath169 . after carrying out some analytic manipulations one finds the following expression for the effective potential : @xmath170 \nonumber \\ & & \mbox { } - \int_{s_1 } \!\!\ ! da \ ; \left [ \gamma |\nabla { u}_2|^2 + \frac{\gamma}{\lambda^2 } { u}_2 ^ 2 - 2 { \pi}_2 { u}_2 \right ] \nonumber \\ & & \mbox { } + \frac{\gamma}{r_{0 , \rm ref } } \oint_{\partial s_1 } \!\!\ ! d\ell \ ; [ \bar{{u } } - { u}_2]^2 - 2 f \bar{{u}}\;.\end{aligned}\ ] ] the first integral accounts for the change in surface energy , gravitational potential energy , and surface stress potential energy of the meniscus due to the overlap of the meniscus deformations caused by the two colloids . the second integral is the corresponding change due to the fact that the interface is reduced by an amount @xmath171 compared to the single colloid case because of the presence of the second colloid . the third integral is the change in surface free energy of one colloid due to the extra meniscus deformation induced by the second colloid . the last term is the change in energy due to the vertical displacement of one colloid by this extra meniscus deformation . for the mathematical manipulations to follow , it is suitable to rewrite eq . ( [ eq : vmen1 ] ) by applying gauss theorem to the integrals involving @xmath172 and by using the fact that the functions @xmath173 fulfill eq . ( [ eq : younglaplace ] ) individually , e.g. , @xmath174 = \nonumber \\ & & \mbox { } - \gamma \oint_{\partial s_1 } \!\!\ ! d\ell \ ; \frac{\partial ( { u}_1 { u}_2)}{\partial n_1 } + \int_{\mathbb{r}^2\backslash(s_1 \bigcup s_2 ) } \!\!\!\!\!\ ! da \ ; \left [ { \pi}_1 { u}_2 - \frac{\gamma}{\lambda^2 } { u}_1 { u}_2 \right ] . \nonumber \\ \end{aligned}\ ] ] thus one obtains from eq . ( [ eq : vmen1 ] ) @xmath175 ^ 2 - 2 f \bar{{u } } .\end{aligned}\ ] ] in this form all the integrals , except the first one , are performed over bounded domains . this allows one to carry out a taylor expansion which yields a uniformly valid asymptotic expansions of the terms as @xmath176 . in the following calculations , the integrals are evaluated in polar coordinates with the origin at the center of @xmath124 and @xmath177 is the position vector with respect to this origin . in particular , @xmath178 is the center of colloid 1 with respect to the center of colloid 2 , and @xmath179 is the center of colloid 2 with respect to the center of colloid 1 ( fig . [ fig:2coll_top ] ) so that , e.g. , @xmath180 . 1 . in order to compute the leading asymptotic behavior of the first term in eq . ( [ eq : vmen2 ] ) as @xmath176 we distinguish two cases : 1 . @xmath85 : in this case , @xmath181 ( eq . ( [ eq : sol1 ] ) ) while @xmath182 ( @xmath27 ) as @xmath183 . asymptotically the main contribution stems from the region near @xmath171 and @xmath184 employing the definition ( [ eq : epspi ] ) . it will turn out that there is no need to compute also the next to leading term in order to obtain @xmath185 as the leading term will be part of the leading contribution to @xmath114 and it will not be cancelled by other contributions . @xmath186 : in this case @xmath187 if @xmath188 . the regions contributing mainly to the integral are the neighborhoods of @xmath171 and @xmath189 . we employ the taylor expansion of the integrand ( which is valid up to a maximum order depending on how fast @xmath94 decays ) to evaluate the leading and the next to leading contributions if @xmath190 ( using dyadic notation ) : @xmath191_{{\bf r}_2 = d { \bf e}_x } \nonumber \\ & & \mbox { } + \int_{\mathbb{r}^2\backslash s_2 } \!\!\!\!\!\ ! da \ ; { u}_2 \left[{\pi}_1 + \dots \right]_{{\bf r}_1 = - d { \bf e}_x } \nonumber \\ & \simeq & 2 \pi \gamma r_{0 , \rm ref}\ ; \varepsilon_\pi \;{u}(d ) + \frac{1}{2 } \pi \nabla^2 { u } ( d ) \int_{r_{0,\rm ref}}^\infty dr\ ; r^3\,\pi(r ) \nonumber \\ & & \mbox { } + { \pi}(d ) \int_{\mathbb{r}^2\backslash s_2 } \!\!\!\!\!\ ! da \ ; { u}_2 + \dots \ ; . \end{aligned}\ ] ] we can simplify the result further by evaluating the last integral by repeated partial integration with the explicit solution for @xmath192 given in eq . ( [ eq : sol1 ] ) : @xmath193 .\ ] ] one finally obtains @xmath194 { \pi } ( d ) + \dots \nonumber \end{aligned}\ ] ] where we have employed eq . ( [ eq : younglaplace2 ] ) ( with @xmath195 for simplicity ) . convergence of the integral of @xmath192 imposes the more stringent constraint @xmath190 on the decay of @xmath196 ( see eq . ( [ eq : int_uda ] ) ) . + the validity of formulae ( [ eq : pi1u2_a ] ) and ( [ eq : pi1u2_b ] ) has been checked explicitly by comparing their predictions with the numerical evaluation of the integral for a surface stress of the form @xmath197 with @xmath190 . 2 . the second term in eq . ( [ eq : vmen2 ] ) can be estimated in the limit @xmath176 by expanding the integrand into a taylor series , which is uniformly valid in the integration domain : @xmath198_{{\bf r}_2 = d { \bf e}_x } \simeq \pi r_{0 , \rm ref}^2 \ , { \pi } ( d ) \ , { u } ( d ) + \dots \ ; . \end{aligned}\ ] ] 3 . the third term in eq . ( [ eq : vmen2 ] ) reads : @xmath199_{r_1=r_{0 , \rm ref } } \\ & \simeq & \int_0^{2 \pi } \!\!\ ! d\varphi \ ; \left\ { ( { \bf r}_1 \cdot \nabla ) \left [ { u}_1 \left ( \phantom { \frac{1}{2 } } \!\!\ ! { u}_2 + { \bf r}_1 \cdot ( \nabla { u})_2 + \right.\right.\right.\nonumber \\ & & \qquad \qquad \left . \left . \left . \frac{1}{2 } { \bf r}_1 { \bf r}_1 : ( \nabla \nabla { u})_2 + \dots \right)_{{\bf r}_2=d { \bf e}_x } \right ] \right\}_{r_1=r_{0 , \rm ref } } \nonumber \\ & \simeq & 2 \pi r_{0 , \rm ref } { u}(d ) \left . \frac{d { u}}{d r } \right|_{r = r_{0 , \rm ref } } + \frac{\pi}{2 } r_{0 , \rm ref } \nabla^2 { u}(d ) \left . \frac{d ( r^2 { u})}{d r } \right|_{r = r_{0 , \rm ref } } + \dots \nonumber \\ & \simeq & 2 \pi r_{0 , \rm ref } \varepsilon_f { u}(d ) - \frac{\pi}{2\gamma } r_{0 , \rm ref}^3 \left(\varepsilon_f + \frac{2{u}_0}{r_{0 , \rm ref } } \right ) { \pi}(d ) + \dots \end{aligned}\ ] ] using the boundary condition for @xmath192 in eq . ( [ eq : u0 ] ) . 4 . the fourth term in eq . ( [ eq : vmen2 ] ) is evaluated analogously : @xmath200_{r_1=r_{0 , \rm ref } } \\ & \simeq & \int_0^{2 \pi } \!\!\ ! d\varphi \ ; \left\ { { \bf r}_1 \cdot \left [ ( \nabla { u}^2)_2 + { \bf r}_1 \cdot ( \nabla \nabla { u}^2)_2 + \dots \right]_{{\bf r}_2=d { \bf e}_x } \right\}_{r_1=r_{0 , \rm ref } } \nonumber \\ & \simeq & \pi r_{0 , \rm ref}^2 \nabla^2 { u}^2 ( d ) + \dots \ ; . \end{aligned}\ ] ] 5 . the fifth term in eq . ( [ eq : vmen2 ] ) is given by @xmath201 ^ 2 & = & \int_0^{2 \pi } \!\!\ ! d\varphi \ ; { u}_2 ^ 2 - 2 \pi \bar{{u}}^2 \nonumber \\ & \simeq & \frac{1}{2 } \pi r_{0 , \rm ref}^2 \left [ \nabla^2 { u}^2 ( d ) - 2 { u } \nabla^2 { u } ( d ) \right ] + \dots \nonumber \\ & \simeq & \frac{1}{2 } \pi r_{0 , \rm ref}^2 \left [ \nabla^2 { u}^2 ( d ) + \frac{2}{\gamma } { \pi } ( d ) \ , { u } ( d ) \right ] + \dots \;. \end{aligned}\ ] ] 6 . in order to evaluate the sixth term in eq . ( [ eq : vmen2 ] ) , we expand @xmath202 in the definition given by eq . ( [ eq : meanu ] ) : @xmath203 the asymptotic behavior of the effective potential is finally obtained from eq . ( [ eq : vmen2 ] ) by collecting all terms . there are two qualitatively different cases : * : the limit @xmath204 is singular and the single colloid meniscus profile exhibits a logarithmic dependence . one finds that the leading contribution is provided by the terms proportional to @xmath205 : @xmath206 which represents a long - ranged attractive effective potential , irrespective of the sign of the forces @xmath60 and @xmath94 acting on the system . * : the single colloid meniscus profile decays like @xmath207 if @xmath161 . the leading contribution is proportional to @xmath208 , because the terms proportional to @xmath205 cancel each other : @xmath209 this correspond to a shorter ranged effective interaction , which in principle can be either attractive or repulsive depending on the precise form of the function @xmath210 . in the particular case that @xmath210 decays monotonically to zero , e.g. , @xmath211 , it is easy to check that @xmath114 amounts to a repulsive force , because the sign of @xmath192 is opposite to the sign of @xmath196 . we have seen that the error committed by the superposition approximation in satisfying one of the boundary conditions , eq . ( [ eq : correction_sup ] ) , decays like @xmath208 , too . this suggests that the corrections to the superposition approximation could modify the precise value of the constant factor acting as an amplitude in eq . ( [ eq : shortrangev ] ) , and _ a priori _ it can not be excluded that there are cancellations leading to a vanishing amplitude , and therefore to an even faster decay for large @xmath112 . thus the superposition approximation might not be reliable enough for calculating @xmath114 if @xmath186 . if one traces back the origin of the dominant contributions to @xmath212 , one finds that in both cases only the first , third , and sixth term in eq . ( [ eq : vmen2 ] ) are relevant . they correspond physically to the effect of the overlap of the two single colloid meniscus profiles and the change of the colloid height ( first integral and the term @xmath213 in eq . ( [ eq : vmen1 ] ) , respectively ) . equations ( [ eq : logv ] ) and ( [ eq : shortrangev ] ) describe the asymptotic behavior of the meniscus induced effective intercolloidal potential and thus represent a central result of our analysis . they provide the explicit functional dependence on an arbitrary stress field @xmath20 which decays sufficiently fast . the assumptions entering their derivation are ( i ) that the deviations of the meniscus profile from the reference configuration are small , allowing one to confine the analysis to a free energy expression which is quadratic in the deviations , and ( ii ) the superposition approximation , which expresses the two colloid equilibrium state in terms of the single colloid state . the analysis shows that the limit @xmath214 is non singular only in the case @xmath162 . equation ( [ eq : logv ] ) can be used to determine the flotation force . there are no stresses acting at the meniscus , @xmath215 , while the force @xmath60 on the colloids is due to their weight corrected by the buoyancy force . accordingly , eq . ( [ eq : logv ] ) reduces to the flotation potential @xmath216 where @xmath217 is known as the capillary charge @xcite , by analogy with two dimensional electrostatics . the asymptotic form for @xmath218 originates from a potential proportional to the modified bessel function @xmath219 ( see eq . ( [ eq : solbessel ] ) ) @xcite . the order of magnitude of the capillary charge is @xmath220 . for a typical value at room temperature and for colloids with a mass density of the order of , the prefactor of the logarithm can be estimated as @xmath221 therefore , compared with the thermal energy , the flotation force is negligible for submicrometer sized colloids . another application is the case of electrically charged colloids . if one of the liquid phases is water , the charge of the colloid is screened ( the debye length of pure water is @xmath222 @xmath0 m and smaller in the presence of an electrolyte ) , and the effective electric field is that of a dipole oriented perpendicularly to the fluid interface @xcite . the electrostatic field decays as @xmath223 and the stress on the meniscus as @xmath224 . both the electrostatic stress and the osmotic ionic pressure decay in the same manner @xcite . thus the total intercolloidal potential at intermediate distances is @xmath225 if gravity is neglected both as a force on the colloid and as a restoring force for the interface @xmath226 , then one can indeed show that the ensuing condition of mechanical isolation ( no net force on the system ) leads to @xmath186 , i.e. , precisely the situation for which the limit @xmath78 is non singular . to see this , we consider the total stress tensor @xmath227 which consists of the maxwell stress tensor ( due to the electrostatic field ) and a diagonal osmotic pressure tensor ( due to the electrolytes ) @xcite . at interfaces @xmath227 can be discontinuous . the total volume @xmath228 of the system is divided into volumes @xmath229 , @xmath230 , and @xmath231 ( see fig . [ fig : forcebalance ] for the explanation of the notation in the following equation ) . the total force reads ( the superscript @xmath232 denotes evaluation on the positive(negative ) side of the oriented surface , i.e. , the side the arrows in fig . [ fig : forcebalance ] point to ( do not point to ) ) @xmath233 \nonumber \\ & = & \int_{s_{\rm men , ref } } \!\!\!\!\!\ ! da \ ; \pi \ , { \bf e}_z + f { \bf e}_z \ ; . \ ] ] in the first line , we have applied gauss theorem with due account of the possible discontinuities of the tensor @xmath227 accross the interfaces . in the second line , @xmath234 in the fluid phases @xmath229 and @xmath230 , because the counterion distribution of the present reference configuration is the equilibrium distribution and thus locally force free . this distribution is considered to be fixed . the second term in the second line is the total force on the meniscus ( which can have only a normal component ) , while the terms in square brackets sum up to the force @xmath60 acting on the colloid . thus the vertical component of the total force is @xmath235 if it vanishes , as it is the case for an isolated system , then @xmath236 . according to eq . ( [ eq : logv ] ) this implies that the long ranged logarithmic contribution to @xmath114 is absent and thus the limit @xmath78 is regular . physically this means that there is no need for a restoring force acting on the fluid fluid interface when the deformation is created by localized internal stresses . instead , according to eq . ( [ eq : shortrangev ] ) , one obtains a potential @xmath237 for the present case of a dipolar electric field ( see above ) . this shorter ranged potential can not counterbalance the direct electrostatic dipolar repulsion @xmath238 . such a counterbalance would be needed for a straightforward explanation of the aforementioned experimentally observed attractions . the line of argument to explain the absence of the logarithm contribution to @xmath114 was already put forward in refs . @xcite and @xcite , where exclusively the case @xmath239 has been addressed . our detailed analysis complements and generalizes these contributions . in ref . @xcite , @xmath240 is estimated by taking into account only the degree of freedom `` meniscus profile '' , @xmath241 , and considering only the change in meniscus area due to the superposition of the dimples . this corresponds to retaining only the term @xmath242 in expression ( [ eq : vmen1 ] ) . although the type of @xmath112dependence obtained that way is correct , the sign of the force turns out to be wrong ( attraction instead of repulsion ) . the reason is that the contributions @xmath243 and @xmath244 in eq . ( [ eq : vmen1 ] ) are equally important as the retained term . this is taken into account in the more detailed analysis of ref . @xcite , where the limiting case @xmath245 ( point particle ) is considered from the outset , so that `` height of the colloid '' is not an independent degree of freedom , i.e. , @xmath246 . correspondingly , @xmath60 is set to zero ( @xmath247 ) , and its effect is modelled by a dirac delta contribution to the stress @xmath20 such that @xmath248 . the potential @xmath249 calculated this way corresponds to keeping the three terms mentioned above which are relevant ( the first and third term in the first integral of eq . ( [ eq : vmen1 ] ) and the last term of eq . ( [ eq : vmen1 ] ) ) and to setting @xmath250 , since this limit is regular when @xmath251 . our analysis has shown that the terms which are dropped in the limit @xmath252 ( second and third integral in eq . ( [ eq : vmen1 ] ) , second , fourth , and fifth terms in eq . ( [ eq : vmen2 ] ) ) yield indeed a subdominant contribution to eq . ( [ eq : shortrangev ] ) . however , the integral appearing as a prefactor in eq . ( [ eq : shortrangev ] ) is divergent for @xmath252 , so that in ref . @xcite a short distance cutoff @xmath253 has been introduced which is expected to be of the order of @xmath8 . the precise value of @xmath253 depends on the details of the implementation of this cutoff ; for the example used in ref . @xcite , the application of eq . ( [ eq : shortrangev ] ) yields @xmath254 . in the presence of gravity mechanical isolation is violated and as we have shown @xmath255 . ( the force balance argument ( eq . ( [ eq : forcebalance ] ) ) can be easily generalized to include the effect of the gravitational volume force . ) in the case of a curved reference interface ( corresponding to the experiment reported in ref . @xcite using colloids trapped at the interface of water droplets in oil ) , eq . ( [ eq : forcebalance ] ) is replaced by @xmath256 where @xmath257 is the radial unit vector of the unperturbed spherical droplet . force balance in the vertical direction yields a factor @xmath258 ( see fig . [ fig : droplet ] ) in the integral , which can be expanded for small angles in the limit @xmath259 . this leads to a curvature induced correction @xmath260 even for mechanical isolation . thus we see that independently of the details of the implementation of the distant boundary conditions the logarithmic term in the potential @xmath114 ( eq . ( [ eq : logv ] ) ) has a strength which is proportional to ( colloid size / system length)@xmath261 , which is nevertheless too weak to explain the reported attractive total interaction . mechanical isolation is violated in the presence of an external homogeneous electric field @xmath262 . for the experimental setup used in ref . @xcite it can not be ruled out that such an external field may have distorted the measurements @xcite . since this is the only experiment which provides quantitative information about the secondary minimum in @xmath263 , we discuss the case of an external field in more detail . in this experiment position recording measurements were performed on a hexagonal configuration of seven trapped colloids on a water droplet immersed in oil @xmath264 m ) . the latter was confined between two glass plates and the droplet stuck on the upper glass plate with a contact angle close to @xmath265 . residual charges might have resided on the upper plate @xcite . the measurements yielded the position of the secondary minimum , @xmath266 , and the curvature at the minimum , @xmath267 . with regard to the latter value we remark that systematic corrections can be estimated to lower @xmath268 by a factor 2 to 3 . these systematic corrections include ( i ) multiparticle effects and ( ii ) center of - mass movement of the droplet . we have estimated the effect of ( i ) by carrying out langevin simulations of the seven particle system using the intercolloidal potential eq . ( [ eq : vtot1 ] ) below . as for the effect of ( ii ) , any shape deformation induced by the moving colloids changes the center of mass position of the droplet since the droplet is fixed to the upper glass plate . the corresponding change in the gravitational energy of the droplet translates into a weak confining potential for the colloids which limits the stochastic movement of the center of mass of the seven colloids . this effect might be part of an explanation for the absence of center of mass movement observed in ref . @xcite . according to eqs . ( [ eq : logv ] ) and ( [ eq : vtot ] ) , the total intercolloidal potential in an external field is @xmath269 using the aforementioned experimental data for @xmath270 and @xmath271 , one obtains from eq . ( [ eq : vtot1 ] ) @xmath272 , @xmath273 and @xmath274 which is surprisingly deep , even if reduced by a factor 2 to 3 due to the systematic corrections mentioned above . furthermore , from eq . ( [ eq : logv ] ) we deduce @xmath275 and with @xmath276 n / m we find @xmath277 . the long ranged meniscus deformation ( see eq . ( [ eq : sol1 ] ) ) is on the scale of nanometers . the short ranged meniscus deformation near the colloid can only be evaluated with a specific microscopic model for @xmath20 . for a rough estimate of @xmath30 , we consider the colloid charge to be concentrated on the surface . the asymptotic behavior of the stress tensor in this case is given by @xmath278 @xcite . if we assume this form of the stress tensor to hold for all @xmath89 , we find @xmath279 for the values of @xmath10 and @xmath253 given above . for not too small contact angles , this _ a posteriori _ justifies the perturbative approach which we have adopted . if the system has a net charge @xmath280 , then @xmath281 ) and ( [ eq : logv ] ) ) . using the values for @xmath282 and @xmath10 as given above we find @xmath283 for the value @xmath284 quoted in ref . @xcite , this yields a relatively small electric field @xmath285 , indicating how sensitive the system can be to spurious external fields . alternatively , an electric field @xmath286 is sufficient for the meniscus induced logarithmic potential to have a depth of the order @xmath287 . thus the external field offers the possibility to tune easily the capillary long ranged attraction and to manipulate the structures formed by the colloids at the interface . in ref . @xcite , experimental results for the meniscus deformation around glass particles of radii 200 300 @xmath0 m trapped at water air or water oil interfaces are reported . the data for the meniscus slope @xmath288 at the contact circle imply ( eq . ( [ eq : u0 ] ) ) @xmath289 ( water oil interface ) , about fifteen times larger than the corresponding @xmath290 caused by gravity alone . furthermore , the reported meniscus shape for one sample contains a logarithmic part which is consistent with @xmath291 ( see eq . ( [ eq : sol1 ] ) ) . as in the experiment analyzed in the previous subsection , the experimental observations could be understood within the framework of the theoretical model we have developed in terms of an external electric field violating mechanical isolation . yet independently of us , the authors of ref . @xcite have developed a theoretical model based on the same physical hypotheses as our approach which , they claim , explains the observed long ranged meniscus deformation . here we would like to note two important errors which flaw their analysis : 1 . ( 3.15 ) in ref . @xcite , in terms of which they interpret their observations , can be obtained by inserting the large distance ( @xmath292 ) asymptotic behavior of @xmath20 in our eq . ( [ eq : sol1 ] ) ( equivalent to their eq . ( 3.14 ) up to an additive constant ) both into the integral term _ and into the definition of @xmath30 _ ( eq . ( [ eq : epspi ] ) ) . since the dominant contribution to the integral defining @xmath30 stems from points @xmath293 , this procedure is clearly inadmissible . as a consequence , they obtain a wrong , nonvanishing prefactor of the logarithm , in spite of their explicit consideration of mechanical isolation ( see their eq . ( 6.6 ) ) . 2 . in order to calculate the intercolloidal effective potential within the superposition approximation , the formula relating @xmath294 to the meniscus deformation @xmath295 _ in the presence of gravity alone _ ( i.e. , eq . ( [ eq : flotv ] ) ) is applied even though @xmath296 . thus , an additional term contributing to the leading logarithm is not included ( compare their eq . ( 3.16 ) and our eq . ( [ eq : logv ] ) ) . in ref . @xcite also a numerical analysis is carried out . a detailed study of the relation between the results of this numerical analysis and our theoretical predictions will be published elsewhere @xcite . it may be possible that the presence of an external field is consistent with the data from refs . @xcite , as well as with the presence of the secondary potential minimum observed in the experiments using planar troughs , in particular in the cases @xmath297 , which fall into the intermediate asymptotic regime considered here . but this still remains as an open problem . given that already nanometer distortions of the meniscus produce noticeable attractions , the surface topography of colloids might be relevant . in ref . @xcite , colloidal surface roughness is proposed as an explanation of the attraction . the meniscus contact line is assumed to be pinned at defects on the colloid surface caused by surface roughness . this imposes a different boundary condition for the meniscus at contact , which is then deformed even in the absence of electrostatic forces . the corresponding analysis in ref . @xcite is concerned only with the term @xmath242 in the expression ( [ eq : vmen1 ] ) , leading to the conclusion that @xmath113 decays as @xmath298 and corresponds to an attractive potential with a strength of @xmath299 for meniscus deformations of the order of @xmath300 . this conclusion , however , can not be simply carried over to the case of charged colloids . as we have shown the contributions of the terms @xmath301 and @xmath244 in eq . ( [ eq : vmen1 ] ) are relevant and can change the qualitative behavior of @xmath113 even in the limit of point colloids . it would be worthwhile to generalize the analysis of ref . @xcite along the line of arguments presented here in order to assess the importance of surface roughness . this should be complemented by more precise experimental information about the actual colloidal topography . recently , an explanation based on a contaminated interface has been advocated @xcite . the air water interface would be actually a two dimensional emulsion consisting of hydrophylic ( water ) patches and hydrophobic ( silicon oil ) patches . the colloidal particles ( hydrophobic in character according to ref . @xcite ) would cluster in the hydrophobic patches . thus , confinement of the colloids by finite size hydrophobic patches would give the impression of an effective intercolloidal attraction . at present , this explanation is only of qualitative nature . thermal fluctuations of the interface position around its mean value @xmath302 also induce an effective interaction between the colloids which confine these fluctuations ( casimir like force ) . using a gaussian model of the fluctuating interface in analogy to the procedure employed in ref . @xcite for calculating fluctuation induced forces between rods in a membrane , one finds for uncharged colloids a fluctuation induced potential @xmath303 , which is too small and falls off more rapidly than the intercolloidal dipole repulsion . here , the generalization to the charged case might give hints for the effective attractions between _ very _ small particles trapped at interfaces . concerning particle sizes well below the debye length , one should modify our analysis to account for the overlap of the screening ionic clouds ( this would affect , e.g. , the superposition approximation for the stress field @xmath94 , eq . ( [ eq : superposition ] ) ) . finally , in ref . @xcite the attraction of particles trapped at a nematic air interface is reported and an explanation in terms of a logarithmic meniscus deformation is proposed which parallels the explanation given in ref . @xcite : in this case , the deformation would be caused by the elastic distortion induced by the particles on the nematic phase @xcite . our detailed theoretical treatment shows that no long range ( logarithmic ) meniscus distortion can arise on an interface if the system is mechanically isolated and the excess free energy of the perturbed interface is correctly described by an expression like eq . ( [ eq : f ] ) . thus it appears that the simple explanation of the observed colloidal attractions given in ref . @xcite is not correct . however , it is not clear whether the free energy of a distorted nematic air interface is equivalent to that of a simple fluid interface due to the long ranged interactions in the nematic bulk caused by defects . thus , a generalization of our theory to interfaces involving nematic phases would be desirable in order to assess the possibility of long ranged colloidal attractions in more detail . we have analyzed the effective force induced by capillary deformation between two smooth spherical colloids floating at a fluid interface . the relevant degrees of freedom are the meniscus deformation and the height of the colloids ( fig . [ fig : pert ] ) , whose equilibrium values are given by minimization of a free energy functional . this functional was derived assuming small deviations from a reference configuration ( fig . [ fig : ref ] ) . it incorporates the surface tensions of the three interfaces involved ( `` colloid fluid phase 1 '' , `` colloid fluid phase 2 '' , and `` fluid phase 1 fluid phase 2 '' ) , the potential energy of the colloids under the action of a force @xmath60 , the potential energy of the fluid interface in an arbitrary surface stress field @xmath304 , and the potential energy due to a restoring force acting on the interface ( eqs . ( [ eq : fdef])([eq : f ] ) ) . the effective intercolloidal potential ( eq . ( [ eq : veff_a ] ) ) was calculated in the limit of large separations by using the superposition approximation ( eq . ( [ eq : superposition ] ) ) . we have shown in this limit that the contribution to the effective potential by the interfaces `` colloid fluid phases '' is subdominant . if the total force acting on the system , consisting of the two colloids and the meniscus , does not vanish , the presence of the restoring force acting on the fluid interface is essential although its precise form does not matter ( subsec . [ sec : lambda ] ) . in this case , the effective interaction is long ranged and attractive ( eq . ( [ eq : logv ] ) ) . if the total force vanishes , the restoring force is irrelevant altogether , the effective interaction is shorter ranged ( eq . ( [ eq : shortrangev ] ) ) , and it can not be computed reliably within the superposition approximation . as an application , we have considered the case of like charged , micrometer sized particles when the capillary deformation is due to the ensuing electrostatic field . we have discussed how one can tune the long ranged attractive interaction by an external electric field , but we conclude that the experimentally observed attraction in an isolated system can not be explained within the present model . possible directions for future research such as colloidal surface roughness and fluctuations of the interface have been discussed . we kindly acknowledge helpful discussions on the subject of the manuscript with m. nikolaides , a. bausch , k. danov , and p. kralchevsky . in this appendix we determine the contribution to the free energy due to the exposure of the colloid to the two fluid phases . here we consider the general case of no rotational invariance , so that the meniscus height @xmath305 depends on the distance @xmath89 from the axis through the colloid center and on the angle of revolution @xmath306 around this axis ( fig . [ fig : contactline ] ) . accordingly , the auxiliary variables @xmath307 and @xmath308 also depend on the revolution angle and one has @xmath309 and @xmath310 where @xmath311 is the meniscus height at contact . the surface areas of the colloid in contact with phase 1 and phase 2 are @xmath312\ ] ] and @xmath313 respectively . the expression for the free energy in eq . ( [ eq : fcont ] ) is based on these formulae for the special case @xmath314 ( fig . [ fig : ref ] ) and on young s law ( eq . ( [ eq : young ] ) ) : @xmath315 = \nonumber \\ & = & \frac{1}{2 } \gamma r^2 \int_0^{2 \pi } \!\!\!\ ! d \varphi \ , [ \cos \xi(\varphi ) - \cos \theta]^2 + \frac{1}{2 } \gamma r^2 \int_0^{2 \pi } \!\!\!\ ! d \varphi \ , [ \cos^2 \theta - \cos^2 \xi(\varphi ) ] \nonumber \\ & = & \frac{1}{2 } \gamma \int_0^{2 \pi } \!\!\!\ ! d \varphi \ , [ u(r_0(\varphi ) , \varphi ) - \delta h]^2 + \frac{1}{2 } \gamma \int_0^{2 \pi } \!\!\!\ ! d \varphi \ , [ r_0 ^ 2 ( \varphi ) - r_{0,\rm ref}^2 ] .\end{aligned}\ ] ] the second term , which arises upon completing the square , represents the change of the meniscus area which is cut out by the colloid . since @xmath45 and @xmath38 are already of first order in @xmath30 or @xmath31 , one can replace @xmath307 by @xmath25 in the first term and obtains @xmath316 ^ 2 + \frac{1}{2 } \gamma \int_0^{2 \pi } \!\!\!\ ! d \varphi \ , [ r_0 ^ 2 ( \varphi ) - r_{0,\rm ref}^2 ] , \ ] ] plus corrections of at least third order in @xmath30 or @xmath31 . in the special case of rotational invariance , this expression reduces to eq . ( [ eq : fcont_iso ] ) . for the purpose of sec . [ sec:2coll ] one can rewrite eq . ( [ eq : almostfcont_aniso ] ) as the line integral @xmath317 ^ 2 + [ r_0 ^ 2 - r_{0,\rm ref}^2 ] \right\ } , \ ] ] where @xmath318 is a circular disk of radius @xmath319 , here centered at the origin , and @xmath320 denotes its circumference . in this appendix we calculate the deformation of the interface of a spherical droplet due to a floating colloidal particle . the reference configuration corresponds to a colloid floating at the surface of the spherical droplet such that eq . ( [ eq : young ] ) is fulfilled . here we will consider only axially symmetric configurations ( see fig . [ fig : droplet ] ) . in terms of the polar angle @xmath321 $ ] , the distance @xmath89 measured along the unperturbed spherical droplet surface is @xmath322 . ( note that in this appendix we consider @xmath323 and @xmath324 as functions of @xmath325 instead of @xmath89 . ) we define @xmath326 and we introduce the angle @xmath327 with @xmath328 at which another boundary condition ( to be specified below ) holds reflecting the fact that the droplet is fixed in space . with this additional boundary condition we model closely the actual experimental setup of ref . @xcite , where the droplet was attached to a glass plate with a contact angle less than @xmath265 @xcite . * in spherical coordinates , @xmath329 reads @xmath330 where @xmath331 is the differential of the solid angle , and @xmath332 is the gradient operator on the unit sphere . as before , this functional is now expanded for small deviations from the reference configuration , and we obtain @xmath333 , \end{aligned}\ ] ] where we have neglected terms evaluated at @xmath334 which vanish in the limit @xmath259 . * the term @xmath61 in eq . ( [ eq : finter ] ) has to be amended by the extra work upon deformation which is caused by the pressure difference in the reference configuration between the interior and the exterior of the droplet , @xmath335 : @xmath336 \nonumber \\ \label{eq : finter1 } & \simeq & - 2 \pi \gamma \int_{\psi_{0}}^{\psi_1 } \!\!\ ! d\psi \ ; \sin\psi \ , \left [ \frac{r_{\rm drop}^2}{\gamma } \pi \ , u + ( 2 r_{\rm drop } u + 2 u^2)\right ] . \end{aligned}\ ] ] since @xmath337 is of order @xmath338 , we have to keep also terms @xmath339 in this expression for consistency . for simplicity we have neglected the contribution due to the change of @xmath41 during the deformation . such a term would affect the behavior of the meniscus only near @xmath334 and vanishes in the limit @xmath259 . * the change in volume of the perturbed droplet reads @xmath340 where @xmath341 is a contribution due to the spherical shape of the colloid ; it depends on @xmath38 and @xmath342 but can be safely neglected . ( this is the same kind of term encountered in calculating @xmath53 , see eq . ( [ eq : fvol ] ) . ) expanding @xmath343 for small deformations @xmath45 up to linear order yields @xmath344 physically , the deformation of the interface occurs under the constraint that the droplet volume remains unchanged . this constraint ( @xmath345 ) will be implemented in the free energy functional by means of a lagrange multiplier @xmath0 which itself turns out to be linear in the parameters @xmath30 and @xmath31 . this justifies keeping only the linear term in eq . ( [ eq : deltav ] ) . the contribution @xmath346 ( see eq . ( [ eq : fcont_iso ] ) ) is concerned only with quantities at @xmath334 , the colloid surface , and thus in the limit of large droplet radii this contribution are the same as in the planar case . the free energy related to the work done upon moving the colloid , @xmath347 ( see eq . ( [ eq : fcoll ] ) ) , remains also unchanged . for simplicity we also neglect gravity and set @xmath84 . in conclusion , the free energy functional to be minimized is the sum @xmath348 plus the constraint term @xmath349 ( where @xmath0 is a dimensionless lagrange multiplier ) : @xmath350 + \pi \gamma [ u_0 - \delta h ] ^2 - f \delta h .\ ] ] the quantity @xmath351 can be interpreted as a homogeneous pressure field enforcing the constant volume constraint . we note that the terms linear in @xmath45 present in eqs . ( [ eq : fmen1 ] ) and ( [ eq : finter1 ] ) have cancelled in the total free energy . minimization with respect to @xmath38 yields the same result as in eq . ( [ eq : h_eq ] ) . subsequent minimization with respect to @xmath352 leads to @xmath353 and ( compare with eq . ( [ eq : u0 ] ) ) @xmath354 the solution must satisfy the constant volume constraint @xmath355 finally , the second boundary condition mentioned at the beginning expresses the physical requirement that the droplet is fixed in space . ( otherwise , the application of a localized force at the interface would shift the droplet as a whole without deforming it . ) as an example , we assume that the droplet interface is pinned at @xmath327 ( another example , force balance by suitable localized counterstresses , is studied in ref . @xcite ) : @xmath356 the general solution of the inhomogeneous legendre equation ( [ eq : spheryl ] ) is @xmath357 \pi(\sigma ) , \ ] ] where @xmath77 and @xmath358 are integration constants and @xmath359 are solutions of the homogeneous legendre equation . for notational simplicity , evaluation of a function at @xmath334(@xmath327 ) will be denoted by the subscript @xmath360 . from eq . ( [ eq : u_1 ] ) one obtains @xmath361 from the other boundary condition ( [ eq : u_0 ] ) it follows that @xmath362 where we have defined the auxiliary function @xmath363 .\ ] ] finally , in order to impose the integral constraint ( [ eq : constraint ] ) we employ the following identities : @xmath364 \pi(\sigma ) = \\ \int_{\psi_0}^{\psi_1 } \!\!\!\ ! d\sigma \ , \sin \sigma \ ; \pi(\sigma ) \ ; \int_{\psi_0}^{\sigma } \!\!\!\ ! d\psi \ ; \sin \psi \ , [ p(\sigma ) q(\psi ) - p(\psi ) q(\sigma ) ] = \nonumber \\ \mbox { } - \int_{\psi_0}^{\psi_1 } \!\!\!\ ! d\sigma \ , \sin \sigma \ ; \pi(\sigma ) \ ; \frac{1}{2 } \left[1-\frac{w(\sigma)}{w_0}\right ] \;. \nonumber \end{aligned}\ ] ] the first equality follows from interchanging the order of integration . the last equality follows most easily by noticing ( i ) that the @xmath325integral in the second line is , as a function of @xmath365 , a solution of the differential equation ( [ eq : spheryl ] ) with @xmath45 replaced by this integral , @xmath366 replaced by @xmath367 , @xmath368 , and the analogues of the boundary conditions @xmath369 , and ( ii ) that the function @xmath370 is the solution to the same differential equation with the same boundary conditions . with these identities eq . ( [ eq : constraint ] ) turns into @xmath371 \;,\ ] ] where we have introduced the numerical constants @xmath372 \nonumber \\ { \cal q } & : = & \int_{\psi_0}^{\psi_1 } \!\!\!\ ! d\psi \ , \sin \psi \ ; [ q(\psi ) - q_1 ] .\end{aligned}\ ] ] we note that the constants @xmath77 and @xmath358 are determined by the two linear eqs . ( [ eq : ab1 ] ) and ( [ eq : ab2 ] ) . these results permit us to study the intermediate asymptotic regime @xmath373 . in the expressions derived above , we have to take the limit @xmath259 while keeping fixed @xmath327 , @xmath374 and @xmath375 ( so that @xmath376 ) . in this limit , the constants @xmath377 and @xmath378 are finite , while @xmath379 and @xmath380 . the integral over @xmath94 in eq . ( [ eq : ab1 ] ) simplifies as follows ( provided that the stress field @xmath381 decays sufficiently fast ) : @xmath382 where we have used @xmath383 . the integrals over @xmath94 in eqs . ( [ eq : ab2 ] ) and ( [ eq : sol_drop ] ) can be simplified in the same manner . from eq . ( [ eq : ab1 ] ) we obtain @xmath384 and eq . ( [ eq : ab2 ] ) yields @xmath385 one finds indeed that the lagrange multiplier @xmath0 determined by eq . ( [ eq : mu ] ) is linear in @xmath386 . finally , the general solution ( [ eq : sol_drop ] ) simplifies to @xmath387 which renders eq . ( [ eq : sol3 ] ) with the constant @xmath388 } .\ ] ] this constant is finite and non vanishing provided @xmath327 is not close to @xmath265 . the case @xmath389 is pathological because @xmath390 for a smooth profile , so that the boundary condition ( [ eq : u_1 ] ) would overdetermine the problem . r. aveyard , b. p. binks , j. h. clint , p. d. i. fletcher , t. s. horozov , b. neumann , v. n. paunov , j. annesley , s. w. botchway , d. nees , a. w. parker , a. d. ward , and a. n. burgess , phys . lett . * 88 * , 246102 ( 2002 ) .

neutrals play an important role in the evolution of the weakly ionized solar chromosphere where the number density of neutrals can vastly exceed the number density of protons . therefore , modeling the neutral - ion interactions and studying the effect of neutrals on the ambient plasma properties is an important task for better understanding the observed emission lines and the propagation of disturbances from the photosphere through the chromosphere and transition region into the solar corona . previous theoretical analysis and recent spectral analysis of raster scans with iris data , for instance in mg ii h 2803- chromospheric spectral line , commonly show the occurrence of double - peaks in the spectroscopic lines , which could be explained only with the presence of neutrals @xcite . the solar chromosphere and transition region regulate the mass , energy and momentum transfer from the underlying photosphere and convection zones into the upper atmospheric layer , known as the corona . it is well - known that the complexity of the chromosphere is related to its partially ionized plasma , which does not fulfill local thermodynamical equilibrium . due to chemical reactions such as impact ionization , the hydrogen gas in the chromosphere varies from predominantly neutral to predominantly ionized , which has important consequences on the fluid behavior of the plasma and can significantly affect the magneto - thermodynamics descriptions . as we follow the decreasing plasma density and pressure gradients through the chromosphere the plasma behavior changes from highly collisional to weakly collisional and the system transitions from gas - pressure dominated to magnetically driven . these observed plasma properties affect the propagation of magneto - hydrodynamic waves through the chromospheric layer and influence the deposit of energy and related plasma heating there . up to now , the numerical modeling of plasma field properties and wave propagation within the solar chromosphere extensively includes different sets of multi - dimensional mhd simulations @xcite as well as improved mhd models . the latter include partial ionization effects by including an ambipolar diffusion term in the generalized ohm s law and in the energy equations , as considered by @xcite . previously , the effects of partial ionization and their consequence on the chromospheric magneto - thermodynamics have been considered by using generalized ohm s law in extended mhd approach @xcite or in a radiative - mhd models , based for example on the bifrost code @xcite which combines mhd equations with non - grey and non - lte radiative transfer and thermal conduction along magnetic field lines @xcite . adding the effects of partial ionization leads to additional chromospheric heating , which depends on the geometry of the model chromospheric magnetic field , as the ambipolar diffusion term is proportional to the perpendicular current @xcite . alternatively , the multi - fluid simulations which are discussed in this paper more rigorously account for partial ionization effects by considering neutrals as a separate species , which relaxes the possible limitations of state - of - the - art improved mhd approximations . recently , more attention has being drawn to the consequences of ion - neutral interactions on the propagation of mhd waves throughout the chromosphere @xcite . analytical calculations have suggested that the dominant presence of neutrals in the chromosphere would lead to an over - damping of the alfn waves there , which would change their commonly expected energy deposit in the corona @xcite . nevertheless , there are no rigorous calculations of partial ionization effects on the propagation and damping of alfvn , fast and slow magnetosonic in a reactive gravitationally stratified collisional media and their interaction with the surrounding plasma . in this respect , the present model provides a first attempt to consider such interactions within chemically reactive multi - fluid simulations . to pursue this goal we have developed two - dimensional two - fluid simulation setups to study the interaction between charged particles and neutrals in a reactive gravitationally stratified collisional media . the model is a variation of the two - fluid approach recently developed by @xcite to study chromospheric reconnection in a reactive media , where the gravitational force has been neglected and only constant temperature and densities have been considered . in this work we have extended the model by @xcite to include the observed temperature and density stratification . in addition , considering pure hydrogen plasma , we have removed the additional charge exchange terms introduced by @xcite , which in the absence of heavy ions simply translate into elastic collisions @xcite . separate mass , momentum and energy conservation equations are considered for the ions and the neutrals . furthermore the interaction between the two fluids is determined by the elastic collisions and the chemical reactions ( e.g. impact ionization and radiative recombination ) , which are provided as additional source terms . to initialize the system we consider an ideal gas equation of state with equal initial temperatures for the electrons , ions and the neutrals , but different density profiles for the charged and neutral particles species . the initial temperature and density profiles are height - dependent and follow vernazza - avrett - loeser ( hereafter val c ) atmospheric model @xcite for the solar chromosphere . to avoid unphysical outflows and artificial heating we have search for a chemical and collisional equilibrium between the ions and the neutrals , eliminating the initial hydrodynamic pressure imbalances . next , we have considered the ion - neutral interactions in a partially ionized magnetized plasma with an initial magnetic profile , corresponding to a homogeneous magnetic flux tube . the magnetic field profile is not force - free and introduces additional pressure gradients in the upper chromosphere . finally we include an external velocity driver and simulate the propagation of magnetosonic waves through the photosphere and the chromosphere . in this section , we describe in details our two - fluid model , where the protons and electrons are described within a single - fluid mhd description neglecting the electron inertia terms . a separate second fluid is used to describe the evolution of neutrals , which interact with the plasma via elastic collisions , impact ionization and radiative recombination . the numerical schemes , the model equations and the initial conditions are explained in the subsections below . the 2d two - fluid numerical simulations are carried out using a fully coupled multi - fluid / maxwell solver @xcite which has been implemented within coolfluid @xcite . the latter is an open source platform for high - performance scientific computing @xcite , featuring advanced computational models / solvers for tackling re - entry aerothermodynamics @xcite , simulation of experiments in high - enthalpy facilities @xcite , ideal magnetohydrodynamics for space weather prediction @xcite , radiation transport by means of ray tracing and monte carlo @xcite algorithms , etc . the temporal evolution of the multi - fluid equations is based on an implicit second - order three - point backward euler scheme , while the spatial derivatives are calculated using a state - of - the - art second - order finite volume method for unstructured grids . the algebraic system resulting from a newton linearization is solved by means of a generalized minimal residual ( gmres ) algorithm complemented by a parallel preconditioner ( additive schwartz method ) as provided by the petsc library @xcite . the maxwell solver is fully relativistic and includes a divergence - cleaning correction @xcite to ensure that the magnetic field remains divergence - free . the discretization of the convective fluxes are based on a upwind advection upstream splitting method ( ausm+up scheme ) for the multi - fluid equations and on a modified steger - warming scheme for the maxwell counterpart . the numerical solver is explained in great details in @xcite . for the purpose of studying wave excitation by photospheric drivers we use a subsonic inlet at the lower boundary with prescribed electromagnetic fields , ion and neutral velocities . the value of the temperature at the lower boundary is imposed and the pressure for both plasma and neutrals is extrapolated to ensure zero pressure gradient at the lower boundary . for the side boundaries and at the top we use supersonic outlet with open boundary conditions , which allows the generated perturbation to propagate out of the simulation domain . here we present the model equations for the two - fluid treatment , with mhd protons and electrons and a separate fluid description for the neutrals . the continuity , momentum and energy equations for the reactive and resistive viscous plasma and the neutrals are presented below . the change of plasma and neutral mass density is governed by the interplay of ionization and recombination rates : @xmath0 where the @xmath1 and @xmath2 are the mass density of protons and neutrals , @xmath3 and @xmath4 are the corresponding velocities , and @xmath5 and @xmath6 are the ionization and recombination rates , which satisfy @xmath7 and @xmath8 . the explicit formulas for the ionization rates and the corresponding cross - sections for the chemical reactions are presented below in a subsection devoted to describe the initial chemical equilibrium . the momentum equations for the plasma and the neutral fluids are given by : @xmath9 where @xmath10 and @xmath11 are the corresponding pressures for the ions and the neutral fluid . the first equation describes the momentum transfer of the mhd plasma , including both charged particles ( protons and electrons ) , therefore the gradient of the electron pressure @xmath12 is included in the plasma flux . in this study the electron inertia terms are neglected . furthermore we assume charge neutrality @xmath13 and equal temperatures between the two charged particles @xmath14 . the stress tensor for both ions and neutrals follows the standard expression @xcite : @xmath15,\ ] ] where @xmath16 is the dynamic viscosity coefficient . for the ions we select @xmath17 $ ] and for the neutrals @xmath18 $ ] . the current in the lorentz force is given by the generalized ohm s law presented below and the gravity force in the chromospere is well approximated by a constant solar acceleration in vertical direction @xmath19 . the elastic collisions between the ions and neutrals are described for example in @xcite @xmath20 where @xmath21 is the center of momentum mass . @xmath22 is the collisional frequency in hz which depends on the average temperature @xmath23 and the collisional cross - section @xmath24 . recent estimates , which are based on atomic physics calculations , show that the temperature dependence of the collisional cross - sections in the selected chromospheric region is insignificant @xcite . therefore for the purpose of our study we have selected constant cross - sections @xmath25 $ ] as considered in @xcite . for the purely electron - hydrogen plasma which is considered here with no heavy ions , the effect of the charge exchange is reduced to regular elastic collisions @xcite and therefore it is not taken into account separately , as done for example in @xcite . the momentum transfer due to inelastic collisions is discussed in more details in the next subsection . the corresponding energy conservation laws for the charged and neutral fluids follow the equations : @xmath26 @xmath27 where , in eq . [ energyplasma ] , electrons are considered to move at the ion speed @xmath28 . for the momentum and energy evolution , an ideal gas equation of state is assumed with @xmath29 . the adiabatic index for all species is considered equal @xmath30 . the kinetic and internal energies for the protons and neutrals are given by @xmath31 the plasma heat flux vector includes contributions from both electrons and protons and takes into account conductivities parallel and perpendicular with respect to the direction of the external magnetic field @xcite . assuming equal number density and temperature for the ions and electrons the plasma heat flux becomes @xcite : @xmath32 where the @xmath33 is the unit vector in the direction of the magnetic field . the parallel and perpendicular heat conductivities @xmath34 and @xmath35 are calculated based on the classical transport theory @xcite : @xmath36 n_it_i,\end{aligned}\ ] ] where @xmath37 and @xmath38 accordingly are the temperature and density dependent collisional times for electrons and protons , @xmath39 and the corresponding cyclotron frequencies and @xmath40 . the heat conduction for the neutral gas is unaffected by the magnetic field : @xmath41 where the isotropic heat conductance coefficient is set to @xmath42 $ ] as considered in @xcite . the electromagnetic fields above are self - consistently evolved according to the full maxwell equations , which are complemented with the divergence cleaning electric @xmath43 and magnetic field @xmath44 potentials . the full maxwell solver allows for propagation of high - frequency waves , which are neglected by the standard mhd approximation . the hyperbolic divergence cleaning procedure guarantees that the magnetic field remains divergence - free . any deviations are instantaneously propagated out with the corresponding characteristic velocities @xmath45 and @xmath46 , where @xmath47 is the speed of light , and @xmath48 and @xmath49 are positive parameters as described in @xcite : @xmath50 in order to compare our results to existing mhd simulations of chromospheric wave propagation , in this study the current is computed by a simplified version of the generalized ohm s law , where we neglect the secondary effects of electron inertia @xmath51 , the hall term @xmath52 , the battery term @xmath53 , and the electron - neutral collisions @xmath54 . under these assumptions the resistive ohm s law takes the simplified form : @xmath55 to compute the electrical resistivity , @xmath56 , we calculate both the collisional frequencies of electrons with ions and electrons with neutrals , @xmath57 , following the expressions @xcite @xmath58 where @xmath59 is the center of mass between electrons and ions , and @xmath60 , @xmath61 are the average temperatures . the collisional cross - section between ions and electrons is computed from braginskii s expressions @xcite , @xmath62 , where @xmath63 is the debye length and @xmath64 is the approximate value of the coloumb logarithm in the chromosphere . the collisional cross - section for electron - neutral collisions is taken to be @xmath65 $ ] , although quantitatively similar results are obtained when a smaller collisional cross - secion is used with @xmath66 $ ] . the conductivity is inversely proportional to the electrical resistivity @xmath67 . in order to model the reactive inhomogeneous chromospheric plasma , we initialize the code with a height - dependent temperature profile , given by the original val c chromospheric model @xcite . the val c model , as well as the more recent follow - up chromospheric models @xcite , do not distinguish between the temperature of molecular hydrogen and protons , and provide a single temperature profile for the single ionized hydrogen and the neutrals . the lack of observational data for the separate temperature profiles of neutrals and ions poses a problem for a realistic non - lte multi - fluid modeling . therefore to best relate to the existing global atmospheric models within our two - fluid description we have assumed initially isothermal plasma with @xmath68 . the initial density profiles for the neutrals has also been adopted from val c model @xcite . the initial proton density follows a modified val c profile , which assumes initial chemical equilibrium . given the fact that our model is based on initially isothermal single ionized proton - electron plasma with neutral hydrogen and currently does not take into account chemical reactions which involve heavier species , adopting the val c atmospheric density profiles for the protons and neutrals results in a chemical imbalance between the ionization and the recombination rates . this imbalance could be resolved if different temperature profiles for the protons and neutrals are assumed . however such different temperature profiles are currently not provided by observations and including them would make the comparison to existing ( generalized ) mhd models more difficult . the ionization and the recombination coefficients in @xmath69 $ ] for hydrogen plasma used in our study follow the expressions in @xcite @xmath70 where @xmath71 is a dimensionless parameter , which takes into account the temperature dependence of the ionization and recombination rates . in this expression , @xmath72 is the electron temperature in ev , @xmath73ev is the ionization potential for hydrogen atoms and @xmath74 is a constant , which takes into account the influence of heavy ions and has been introduced to match the expected ionization rates @xmath75 , as predicted by the val c model @xcite . we should note that despite the different functional form for the selected temperature range these ionization and recombination coefficients are practically identical to the ones proposed by voronov and smirnov @xcite , which have been used in the multi - fluid modeling by @xcite. the ionization and recombination rates which account for the mass , momentum and energy transfer in the multi - fluid system read : @xmath76 in order to build a solid physical understanding of the reactive multi - fluid problem and for the sake of comparison to results from existing fluid theories , we have initialized the code assuming a chemical equilibrium , where the ionization and the recombination rates initially balance each other @xmath77 . therefore , if we assume an initially isothermal plasma with temperature and neutral density profiles following the tabulated values from val c model @xcite , in order to maintain a chemical equilibrium for the neutrals we need to slightly modify the ion density profile . figure [ fig1:init_temp ] represents a 1d cut with the initial mass density and temperature profiles used in the simulations . all plasma parameters vary with the heliocentric distance and the neutral density changes by 5 orders of magnitude . the ions and neutrals are initially considered in thermal equilibrium with @xmath78 . the modified plasma density in our model has similar shape to the inferred electron density presented in the chromospheric model by @xcite . the initial profiles capture the photospheric temperature minimum around 500 km and the consequent temperature increase in the upper chromosphere . the ionization level increases , as the neutral and ion densities become more comparable in the upper chromosphere . the 2d contour plots of the initial temperature profile and its variation throughout the simulations are presented in the results section in figure [ fig3:initt ] . + in the next section we present in details the results from the simulations and discuss their relation to previous models in the available literature . our multi - fluid simulations are performed based on dimensional quantities and all results are provided in si units . where necessary , an appropriate conversion from cgs or other units has been introduced to correct for the different expressions of the source terms and the transport coefficients as provided in the literature . for visualization purposes , we use cartesian grid , where the heliocentric distance is represented by the vertical @xmath79 axis , and the horizontal variations are given in @xmath80 direction . the physical size of the box is @xmath81 mm by @xmath81 mm and we have used a moderate spacial resolution of 200 grid points in each direction . we should note that increasing the resolution does not significantly affect the wave propagation , but increases initial force imbalance due to sharper gradients and results in stronger velocity outflows . in order to study the evolution of sound and magnetosonic waves in the chromosphere , we have adopted a simple gaussian magnetic field profile and a photospheric foot - point velocity driver used in mhd and improved partially ionized mhd models with ambipolar diffusion terms , as presented for example in @xcite : @xmath82 , title="fig:",scaledwidth=30.0%],title="fig:",scaledwidth=30.0%],title="fig:",scaledwidth=30.0% ] the magnetic field strength has been set to @xmath83 g , the flux tube is placed in the center of the domain at @xmath84 mm , and the width of the initial magnetic field foot - point has been selected to @xmath85 km as in @xcite . the magnitude of the velocity driver is set to @xmath86 m/s as assumed in @xcite . the period of the driver was set to @xmath87s , so that the corresponding frequency is @xmath88hz . the time step used for the simulations varies between @xmath89ms and @xmath90s . we should note that the selected initial magnetic field profile is divergence - free , but not force - free and it does carry some current . due to the low magnitude of the magnetic field considered here , the lorentz force associated with it is rather small . i.e. of the order of @xmath91n . nevertheless this force becomes significant in the upper chromosphere , where the pressure gradient and the gravity are substantially reduced due to the depleted plasma density in this region . these currents cause plasma expansion due to the current - related horizontal outflows in the upper chromosphere , resulting in plasma cooling , as described later in the paper . to compare our results to previous mhd and generalized mhd simulations we have used several different initial simulation setups . in order to generate acoustic and magnetosonic waves in the photosphere , we have perturbed the horizontal and vertical velocity of the ions , as well as the horizontal and vertical velocities of the neutrals . the main difference between applying the velocity driver on the ions and the neutrals is in the amplitude of the excited perturbations . since ions are much less abundant than neutrals , while perturbing the ion velocity we input less kinetic energy into the system : this generates smaller pressure gradients and results in less intense waves . however , apart from their amplitude , all physical quantities ( e.g. wave properties , velocity and temperatures profiles ) remain the same . therefore for a better representation , in the remaining of the paper we will concentrate on the case of velocity driver applied on the neutrals . figure [ fig2:mag_field ] describes the magnetic field profiles at two different times . the initial magnetic field is given on the left panel and has a gaussian profile , following eq . [ eqn : bfield ] . it has a maximum magnitude of @xmath92 g inside the flux tube and zero value outside the tube . the right panel describes the evolution of the magnetic field at a later stage at @xmath93s . the evolution shows strong diffusion of the magnetic field at the location of the temperature minimum in the photosphere . since the resistivity in our study is defined within the classical transport theory @xcite , it has a strong dependence on the temperature @xmath94 . therefore the observed diffusion of the magnetic field is related to resistive dissipation and occurs in the region where the resistivity is strongest . the magnetic field diffusion in the upper part of the chromosphere is related to ohmic dissipation in the region where the lorentz force becomes comparable to the otherwise dominant pressure gradient . the lower chromosphere is dominated by the pressure gradients , gravity and collisions . in the upper chromosphere the current associated with the initial magnetic field profile leads to strong horizontal flows ( with maximum velocities of the order of 1 - 2 km / s ) , which result in expansion of the flux tube . towards the end of the simulations the flux tube opens up and starts to resemble the observed magnetic funnels . as mentioned above in the upper chromosphere at @xmath95 mm , the expansion caused by the horizontal flows and the horizontal currents leads to associated cooling and density depletion of the plasma inside the flux tube . as the plasma cools , it becomes heavy . inside the flux tube there is no current to push the plasma outwards and the matter starts to fall back towards the lower chromosphere . this leads to an extended region with decreased plasma temperature and depleted density inside the flux tube . figure [ fig3:initt ] illustrates the initial vertical temperature profile ( left panel ) with homogeneous horizontal distribution and the onset of inhomogeneous horizontal profile ( right panel ) , due to the described expansion and related plasma cooling inside the magnetic flux tube . at @xmath93s the velocity driver has already generated sound and fast magnetosonic waves and we observe symmetric oscillations in all physical quantities . the vertical driver mixes the cooler plasma from the photosphere with the hotter regions in the mid and upper chromosphere , forming clear wave patterns . although shock waves are never formed within our study , as the waves propagate upward some compressional heating at the wave fronts is observed . within the flux tube the compressional heating is overcome by the cooling caused by the expansion and the plasma there remains cooler . some additional cooling is also observed near the temperature minimum in the photosphere where the magnetic field diffuses due to the enhanced resistivity . we should note that the cooling in the middle of the flux tube is a mhd effect and does not exist in pure hydrodynamic description , as proven by analogous test simulations performed without magnetic field . as we explained above the lorentz force at the upper chromosphere becomes dominant over the gravity and the pressure gradient and the associated current results in horizontal outflows , which cause expansion and result in related cooling of the confined plasma inside the magnetic flux tube . reactive hydrodynamic multifluid simulations with exactly the same plasma conditions except for the electromagnetic fields show no cooling at in the upper chromosphere and no significant change in temperature over all , apart from some cooling at the photospheric level , close to the initial temperature minimum . we should note that in the extremely dense photospheric and chromospheric conditions the collisional frequency between ions and neutrals is very big . as a result the collisional time is much shorter than any other dynamical timescales in the system and all initial differences between the different particle species practically immediately balance out , leading to a single dynamics for the evolution of their fluid velocities and temperature . in this respect the effect of the electromagnetic fields on the electrons and ions is transported to the neutrals through frequent collisions , so that their temperatures and velocities become practically the same . therefore in this plot and throughout the paper we will show single plots referring to the temperature and velocities for both species . figure [ fig : rhot0top ] shows the observed plasma cooling and the density depletion for both ions and neutrals at the top of the chromosphere , @xmath96 mm . the figure shows the initial constant values of the densities and temperature ( the constant straight lines ) and their depletion and decrease inside the flux tube in the course of evolution . as expected the minimum of the densities and the temperature is in the center of the flux tube . we should note that although originally only the ions are affected by the horizontal currents , due to the strong collisionality the neutrals follow the ions and their density and temperatures are also reduced . the ion and the neutral densities decrease up to one order of magnitude inside the flux tube and their value outside the flux tube oscillates in time with the propagation of the magnetosonic waves . within 178 seconds the temperature and the densities outside the flux tube are higher than the initial values and the temperature inside the tube is reduced by two thousands kelvin from @xmath97k to as low as @xmath98k . the figure shows the temperature and density profiles within the first 6 wave periods , before the waves reach the upper boundary . afterwards the sound and the fast waves result in density and temperature oscillations with combination of additional heating / cooling and higher amplitude density fluctuations outside the flux tube . regardless of the waves , the cooling and the density depletion inside the flux tube remain until the end of the simulations and are also present in simulations without initial driver . we should note that the heating at the boundary of the flux tube might be related to ion - neutral interactions as modeled by generalized mhd models , where the heating due to ambipolar diffusion term is proportional to the perpendicular current . such scenarios are discussed for example by @xcite . figure [ fig : uv ] shows an example of the horizontal and vertical velocity profiles at @xmath93s . this time moment is not unique and has been randomly selected to illustrate the wave behavior as visible in the two velocity components . both horizontal and vertical velocity components show symmetric contours typical for sound waves and fast magnetosonic waves propagating along the magnetic field lines . the vertical velocity component is higher than the horizontal velocity . we can also see the change in the plasma velocity related to the distortion of the wave fronts along the flux tube , which becomes clearly visible in the upper chromosphere as we approach the alfvnic point where @xmath99 . in the upper chromosphere the wave propagation is strongly affected by the magnetic field , and we observe a transition from fast to slow magnetosonic waves . we should note that in both magnetized and hydrodynamic multi - fluid cases the initial val c atmosphere is slightly convectionally unstable and results in horizontal and vertical flows . one of the reasons for that is that the local thermodynamical equilibrium assumption in their atmospheric model is not fully satisfied . an additional reason is the fixed viscosity coefficients used in our model , which result in under - viscous plasma in the photosphere . within physical time of 700 seconds in both hydro and magnetized cases without initial wave driver , the maximum velocity amplitude is of the order of @xmath100km / s for the horizontal flows and 2 - 2.5km / s for the vertical flows . typically the horizontal flows are within @xmath101 m/s . due to the induced horizontal currents , the applied initial magnetic field profile enhances the horizontal flows in the upper chromosphere where the lorentz force dominates over the plasma and neutral gas pressure gradients and the plasma @xmath102 becomes close to unity . still without initial driver the unstable convective flows are at least an order of magnitude smaller than the estimated sound speed throughout the photosphere and chromosphere , as well as the alfvn , slow ( tube ) and fast speeds ( calculated based upon the total ion and neutral density ) inside the magnetic flux tube . figure [ fig : vs ] shows the vertical profile of the sound speed @xmath103 and its variation in time in the center of the flux tube . the different times have been selected to qualitatively represent the sound speed variations as a function of the initial velocity driver , whose period was set to @xmath104 seconds . within the first four wave periods the sound speed decreases in the upper chromosphere , mainly due to the cooling and density depletion caused by the horizontal currents . within six wave periods the sounds speed starts to oscillate with the amplitude of the driver throughout the entire chromosphere . as the driven waves reach the @xmath99 region in the upper chromosphere above @xmath105 mm the period of the sound waves oscillations changes due to mode conversion . in addition the density variations in the upper chromosphere increase as the waves reach the upper boundary , which results in higher amplitude sound waves at @xmath106s . figure [ fig : va ] illustrates the temporal evolution of the alfvn speed in the center of the magnetic flux tube based on the total plasma and neutral density , @xmath107 , where @xmath108 is the total density . at @xmath109 the aflvn speed profile is fully determined by the initial magnetic field and density profiles . as the magnetosonic waves excited by the initial photospheric velocity driver propagate up in the chromosphere , they induce density variations which result in variations of the alfvn speed over time . additionally the magnetic field diffuses both in the photosphere where the spitzer resistivity is highest and in the upper chromosphere , where the cross - section of the flux tube increases as the initial vertical flux tube starts to open into a magnetic funnel . this opening leads to diffusion of the magnetic field and results in reduction of the alfvn speed over time . we should note that the deliberate choice of plotting the alfvn speed based on the total density is made in order to compare the model results and expectations to existing mhd and generalized mhd theories , where the effect of the neutrals is treated as additional terms in the energy and momentum equations of the plasma , but the neutral density is not evolved separately and the plasma density is governed by the charged particles , protons and electrons . as expected due to the huge number density difference between the ions and neutrals , the alfvn speed based on the ion density inside the flux tube is much bigger and compared to it the plasma flows remain practically sub - alfvnic at all times in the entire simulation box . apart from the sound and the alfvn speed we can also calculate the velocity of the slow magnetosonic waves or the so - called tube speed . due to the high plasma density in the photosphere and the low chromosphere , the tube speed calculated in the middle of the magnetic flux tube at the lower boundary is only @xmath110 m/s . similar to the alfvn speed it increases with height and at the top of the chromosphere it becomes more than 40km / s and becomes comparable to the sound speed . while the sound speed inside the flux tube decreases in the time due to the observed cooling , the tube speed based on the total density gradually increases in height at all times . the tube speed based on the ion density remains quasi conserved throughout the magnetic flux tube with a magnitude of approximately 40km / s throughout the chromosphere . figure [ fig : va_to_vs ] shows the ratio of the alfvn speed divided by the sound speed , which corresponds to the inverse square root of the plasma @xmath102 in the center of the magnetic flux tube . the left panel shows the ratio , when the magnitude of the alfvn speed is calculated based on the total density including the contribution of the plasma and the neutrals . the right panel shows the variation of the @xmath111 ratio when the alfvn speed is calculated based solely on the plasma density . clearly , if we consider only the density of the magnetized plasma , the entire chromosphere as well as most of the photosphere will be in a very low plasma @xmath102 regime , as @xmath112 . on one hand the neutrals are not directly affected by the magnetic field and therefore they should not enter in the calculation of the alfvn speed . on the other hand the high collisional rate between ions and neutrals forces the two species to move together and within an mhd picture only the total mass density plays a role . if we look at the @xmath111 ratio , when the alfvn speed is calculated based on the total density the plasma @xmath102 remains high in the entire photosphere and above until approximately @xmath105 mm in the upper chromosphere , where @xmath102 becomes of the order of unity . in the uppermost chromosphere @xmath113 , therefore in this region the magnetic field plays a role and the plasma is no longer governed by hydrodynamics . the ratio of alfvn to sound speed on the left panel is very low close to the solar surface and increases with heliocentric distance . at the bottom of the chromosphere in the center of the magnetic flux tube the alfvn speed is approximately two orders of magnitude slower than the sound speed . at 1 mm it varies between 0.25 and 0.28 , depending on the simulation time . due to the magnetic field diffusion , the alfvn speed slightly descreases in time , which results in a lower @xmath111 ratio . at 1.5 mm the ratio increases to 0.5 - 0.8 and close to @xmath114 mm the two speeds match , which in mhd models marks the threshold of plasma @xmath102 = 1 . the highest ratio of the two speeds is reached at the top of the chromosphere and is of the order of 3.2 , which would correspond to a plasma @xmath115 in the single fluid description . we should note that in the estimates above the alfvn speed is calculated based on the total mass density , which includes ions and neutrals . when we consider the classical alfvn speed which was invented for fully ionized media and is based on the plasma fluid density only ( ions and electrons ) , then the flow becomes supersonic at much lower heights , approximately above @xmath116 km . figure [ fig : kinenneutrals ] depicts the altitudinal variation of the kinetic energy density of neutrals at the longitudinal position of the photospheric driver as a function of the driver period . the left and the right panels represent different instances of time , which are proportional to the period of the initial driver . at @xmath109 there are no flows in the system and the kinetic energy for both species is zero . within the first six wave periods the kinetic energy in the photosphere and the lower chromosphere remains in phase with the period of the driver and the kinetic energy gradually decreases with distance as the neutral density gets depleted . after the waves reach the upper chromosphere and cross the @xmath117 region the period of the velocity oscillations changes , which results in out - of - phase fluctuations for the kinetic energy of the neutrals . figure [ fig::kinenions ] shows the kinetic energy density of ions as a function of height and the period of the initial driver . on the left panel once again we can see that it takes about 6 periods for the waves to reach the upper chromosphere . the drop in the kinetic energy around @xmath86 km is due to the trough in the initial density profile at this location , see figure [ fig1:init_temp ] . the currents induced by the lorentz force in the upper chromosphere cause some additional flows and different frequency velocity oscillations , which quickly become in phase with the initial driver as the driven oscillations reach the given height . the right panel shows that at a later stage ( 358 seconds approximately correspond to 12 periods of the initial driver ) the ion kinetic energy fluctuations are fully in phase in the lower and middle chromosphere , and similar to the sound speed their period changes in the upper chromosphere above @xmath105 mm where the plasma @xmath102 becomes of the order of unity and below . another aspect of the initial velocity driver is the effect of the induced oscillations on the chemical reactions in the system figures [ fig : gammaion ] and [ fig : gammarec ] show 2d contour plots with the change of the ionization and the recombination rates induced by the initial driver . the ionization becomes affected by the waves quicker than the recombination , which results in local loss of chemical equilibrium . finally figure [ fig : ionizrate ] shows the change of ionization fraction @xmath75 caused by the initial driver inside the magnetic flux tube . due to the several orders of magnitude difference between the number density of ions and neutrals in the photosphere and the lower chromosphere the ionization fraction there remains almost unchanged by the driven oscillations . for heights @xmath118 1 mm where the ion mass density reaches a local maximum we observe the onset of related fluctuations in the ionization rate . the period of the oscillations changes in time and the relative number density of ions decreases in time in the upper chromosphere above @xmath105 mm , where the loss of chemical equilibrium results in recombination rate prevailing over the ionization . within the fully 3d coolfluid framework we have used a two - fluid numerical simulation setup , which treat the charged particles within a generalized mhd approximation and considers a separate fluid of neutrals . the two fluids are coupled through collisions and chemical reactions , such as impact ionization and radiative recombination . the newly developed simulation module includes the effects of gravitational force imposed on the initial stratified density and temperature profiles , given by a modified val c chromospheric model . our simulations take into account braginskii collisional transport coefficients with anisotropic heat conduction , spitzer - type resistivity and accounts for the basic chemical reactions in a hydrogen plasma . we have initialized the simulations with a simple gaussian magnetic flux tube , which diffuses in time and opens starting to form a magnetic funnel . to compare to existing mhd and generalized mhd models we have imposed initial thermal and chemical equilibrium . we have included a photospheric velocity driver at the foot - point of the magnetic field , and have followed the evolution of the system in the presence of the driven sound and fast magnetosonic waves . within 5 minutes of simulated real time the initial driver induces loss of chemical equilibrium with dominant recombination in the upper chromosphere and gradual reduction of the ionization fraction in time . regardless of the selected low magnitude of the magnetic field , we find significant heating of the higher chromospheric layers outside the magnetic flux tube with some overall cooling at low to mid - low altitudes . significant cooling is also observed in the upper chromospheric layers inside the flux tube , where the temperature can be more than @xmath119k degrees cooler than the surrounding plasma outside the flux tube . this cooling is partially counter - balanced by the heating due to the slightly unstable convecting atmosphere . the plasma heating observed at the wave fronts of the propagating fast magnetosonic fluctuations driven by the initial velocity driver as well as the heating outside the magnetic flux tube in the upper chromosphere account for a temperature increase of @xmath86k up to @xmath120k . the observed heating strongly depends on the initial driver : the vertical velocity driver heats the chromospheric plasma more than the horizontal one . the horizontal velocity driver at the selected frequency in our case does not excite waves and does not lead to significant heating . this comparison between the vertical and horizontal drivers significantly differs from the results of previous single fluid mhd studies of chromospheric wave propagation , where slow magnetosonic waves have been excited by the horizontal velocity driver @xcite . with both drivers we observe depletion of the number densities of ions and neutrals in the upper region of the magnetic flux tube , where the lorentz becomes important as the plasma @xmath102 and @xmath121 ratio increase . as the alfvn speed becomes faster than the sound speed in the case of the vertical velocity driver we observe a mode conversion with transformation from fast to slow magnetosonic waves . within our model most of the net plasma heating outside the flux tube in the upper chromosphere is wave - based and is related to kinetic to thermal energy conversion as the wave fronts of the fast magnetosonic waves steepen . in the current plasma - neutral interactions we do not observe strong heating along the perpendicular component of the current , as expected from mhd models with partial ionization effect relying on ambipolar diffusion terms @xcite . within the current multi - fluid approach the plasma heating due to the partial ionization effects , if present , is either more isotropic and not bound to the direction perpendicular to the external magnetic field profile , or it is fully overcome by the dominant wave - based heating . we should note that the partial ionization effects in our study are not very strong and the main energy source for heating are the excited magnetosonic fluctuations . the role of the partial ionization could perhaps become more prominent for stronger magnetic fields . although this paper does not discuss in details the redistribution of the different energy types and the conversion from kinetic to thermal energy , we have to mention that there is energy transport from the photosphere throughout the chromosphere . apart from kinetic energy transported through the chromosphere in the form of waves , there is also an associated electromagnetic energy transfer . the estimated poynting flux related to the fast magnetosonic waves is much stronger in the horizontal than in vertical direction . however , we expect that to change whenever slow magnetosonic or alfvn waves are excited in the system , as already suggested by some preliminary model results . to further improve our model and make it better suitable for longer wave propagation studies we would consider implementing additional wave damping layer , similar to what had been proposed and implemented for mhd simulations by @xcite . this work was supported by project grant c 90347 ( esa prodex 9 ) and fwo 12k1416n postdoctoral fellowship at cmpa , ku leuven . this work was partially co - funded by projects goa/2015 - 014 ( ku leuven ) and g.0a23.16n ( fwo - vlaanderen ) . the research leading to these results has also received funding from the european commission s seventh framework programme ( fp7/2007 - 2013 ) under the grant agreements solspanet ( project n 269299 , www.solspanet.eu ) , eheroes ( project n 284461 , www.eheroes.eu ) and swiff ( project n 263340 , www.swiff.eu ) . a. alvarez laguna is grateful to iwt phd fellowship for the support during this research . avrett , e. h. , & loeser , r. 2008 , , 175 , 229 - 276 moore , r. l. , & fung , p. c. w. 1972 , , 23 , 78 cox , d. p. , & tucker , w. h. 1969 , , 157 , 1157 vernazza , j. e. , avrett , e. h. , & loeser , r. 1981 , , 45 , 635 de pontieu , b. , rouppe van der voort , l. , mcintosh , s. w. , et al . 2014 , science , 346 , 1255732 de pontieu , b. , title , a. m. , lemen , j. r. , et al . 2014 , , 289 , 2733 peter , h. , tian , h. , curdt , w. , et al . 2014 , science , 346 , 1255726 pereira , t. m. d. , carlsson , m. , de pontieu , b. , & hansteen , v. 2015 , , 806 , 14 carlsson , m. , hansteen , v. h. , gudiksen , b. v. , leenaarts , j. , & de pontieu , b. 2016 , , 585 , a4 martnez - sykora , j. , de pontieu , b. , hansteen , v. h. , & gudiksen , b. 2016 , , 817 , 46 martnez - sykora , j. , de pontieu , b. , hansteen , v. , & carlsson , m. 2015 , philosophical transactions of the royal society of london series a , 373 , 20140268 braginskii , s. i. 1965 , transport processes in a plasma , reviews of plasma physics , 1 , 205 . brandenburg , a. & zweibel , e. g. 1994 the formation of sharp structures by ambipolar diffusion . apjl , 427 , l91l94 . ( doi:10.1086/187372 ) cheung , m. c. m. & cameron , r. h. 2012 magnetohydrodynamics of the weakly ionized solar photosphere . apj , 750 , 6 . ( doi:10.1088/0004 - 637x/750/1/6 ) fedun , v. , erdlyi , r. , & shelyag , s. 2009 , , 258 , 219 fedun , v. , shelyag , s. , & erdlyi , r. 2011 , , 727 , 17 fontenla , j. m. , avrett , e. h. , & loeser , r. 1993 , , 406 , 319 gent , f. a. , fedun , v. , mumford , s. j. , & erdlyi , r. 2013 , , 435 , 689 golding , t. p. , carlsson , m. & leenaarts , j. 2014 , 784 , 30 ( doi:10.1088/0004 - 637x/784/1/30 ) gudiksen , b. v. , carlsson , m. , hansteen , v. h. , et al . 2011 , , 531 , a154 hansteen , v. , de pontieu , b. , carlsson , m. , lemen , j. , title , a. , boerner , p. , hurlburt , n. , tarbell , t. d. , wuelser , j. p. et al . 2014 the unresolved fine structure resolved : iris observations of the solar transition region . science , 346(6207 ) . ( doi:10.1126/science.1255757 ) hansteen , v. h. , carlsson , m. & gudiksen , b. 2007 3d numerical models of the chromosphere , transition region , and corona . in the physics of chromospheric plasmas ( eds p. heinzel , i. dorotov@xmath122c & r. j. rutten ) , vol . 368 of astronomical society of the pacific conference series , p. 107 . hansteen , v. h. , de pontieu , b. , rouppe van der voort , l. , van noort , m. & carlsson , m. 2006 , dynamic fibrils are driven by magnetoacoustic shocks . apj , 647 , l73l76 . ( doi:10.1086/507452 ) khomenko , e. , collados , m. , & felipe , t. 2008 , , 251 , 589 khomenko , e. , & collados , m. 2006 , , 653 , 739 khomenko , e. & collados , m. 2012 , , 747 , 87 ( doi:10.1088/0004 - 637x/747/2/87 ) shelyag , s. , khomenko , e. , de vicente , a. , & przybylski , d. 2016 , , 819 , l11 khomenko , e. , daz , a. , de vicente , a. , collados , m. , & luna , m. 2014 , , 565 , a45 leake , j. e. and lukin , v. s. and linton , m. g. and meier , e. t. , multi - fluid simulations of chromospheric magnetic reconnection in a weakly ionized reacting plasma , apj , vol . 760 , pg 109 , ( dec . 2012 ) meier , e. t. and shumlak , u. , a general nonlinear fluid model for reacting plasma - neutral mixtures , physics of plasmas , vol . 19 , 7 ( 2012 ) leake , j. e. and lukin , v. s. and linton , m. g. , magnetic reconnection in a weakly ionized plasma , arxiv e - prints , 1302.3287 ( feb . 2013 ) voronov , g. s. , atomic data and nuclear data tables , 65 , 1 , ( 1997 ) smirnov , b. m. , physics of atoms and ions , ( 2003 ) l. spitzer , physics of fully ionized gases , interscience , new york , ( 1956 ) soler , r. , ballester , j. l. , & zaqarashvili , t. v. 2015 , , 573 , a79 alvarez laguna a. , lani a. , mansour n. n. , kosovichev a. , poedts s. , a two - fluid computational model to study reconnection in reactive plasmas under chromospheric conditions , wccm xi , eccm v , ecfd vi , ( 2014 ) alvarez laguna a. , lani a. , deconinck h. , mansour n. n. , poedts s. , a fully - implicit finite - volume method for multi - fluid reactive and collisional magnetized plasmas on unstructured meshes , j. comp . ( 2016 ) lani a. , quintino t. , kimpe d. , deconinck h. , vandewalle s. , poedts s. , the coolfluid framework : design solutions for high - performance object oriented scientific computing software , in : v. s. sunderan , g. d. van albada , p. m. a. sloot , j. j. dongarra ( eds . ) , computational science iccs 2005 , vol . 1 of lncs 3514 , emory university , springer , atlanta , ga , usa , 2005 , pp . 281286 . lani a. , quintino t. , kimpe d. , deconinck h. , vandewalle s. , poedts s. , reusable object - oriented solutions for numerical simulation of pdes in a high performance environment , scientific programming . special edition on poosc 2005 14 ( 2 ) , 111139 ( 2006 ) lani a. , villedieu n. , bensassi k. , kapa l. , vymazal m. , yalim m. s. , m. panesi , coolfluid : an open computational platform for multi - physics simulation and research , in : aiaa 2013 - 2589 , 21th aiaa cfd conference , san diego ( ca ) ( 2013 ) panesi m. , lani a. , magin t. , pinna f. , chazot o. , deconinck h. , numerical investigation of the non equilibrium shock - layer around the expert vehicle , aiaa paper 2007 - 4317 , 38th aiaa plasmadynamics and lasers conference , miami ( florida ) ( 2007 ) garicano mena j. , pepe r. , lani a. , deconinck h. , assessment of heat flux prediction capabilities of residual distribution method : application to atmospheric entry problems , commun . 17 ( 3 ) , 682702 ( 2015 ) zhang w. , lani a. , chew h. b. , panesi m. , `` modeling of non - equilibrium plasmas in an inductively coupled plasma facility '' , 45 th plasmadynamics and laser conference , aiaa 2014 - 2235 , atlanta ( ga ) ( 2014 ) yalim m. s. , vanden abeele d. , lani a. , quintino t. , deconinck h. , a finite volume implicit time integration method for solving the equations of ideal magnetohydrodynamics for the hyperbolic divergence cleaning approach , jcp 230 ( 15 ) , 61366154 ( 2011 ) c .- d . , ommes p. , and schneider r. , a three - dimensional finite - volume solver for the maxwell equations with divergence cleaning on unstructured meshes , _ computer physics communications _ , 130:83117 ( 2000 ) balay s. , abhyankar s. , adams m. f. , brown j. , brune p. , buschelman k. , dalcin l. , eijkhout v. , gropp w. d. , kaushik d. , knepley m. g. , mcinnes l. c. , rupp k. , smith b. f. , zampini s. , and zhang h. , petsc web page , http://www.mcs.anl.gov/petsc ( 2015 )

two bodies kept at different temperatures and separated by a vacuum gap experience a radiative heat transfer mediated by photons . this energy exchange is limited by the well known stefan - boltzmann s law in the far field , i.e. when the distance separating the bodies is large compared to the thermal wavelength @xmath0 , of the order of @xmath1 m at ambient temperature . the pioneering works of rytov @xcite , polder and van hove @xcite first showed that this limit can be surpassed in the near - field regime , as a result of the tunneling of evanescent waves . in particular , the heat transfer can exceed the one between two blackbodies ( i.e. the ideal far - field scenario predicted by stefan - boltzmann s law ) even of several orders of magnitue when the materials support surface resonances , such as plasmons in metals ( typically lying in the ultraviolet range of frequencies ) and phonon - polaritons in dielectrics ( typically in the infrared ) @xcite . since the contribution of each field mode to radiative heat transfer is weighted by the planck thermal distribution , negligible in the ultraviolet range at ordinary temperatures , dielectrics supporting surface resonances are typically the best candidates to maximize the heat flux . stimulated by the theoretical developments , several applications have been proposed for radiative heat transfer , ranging from thermophotovoltaic @xcite or solar thermal @xcite energy conversion , to heat - assisted data storage @xcite , nanoscale cooling @xcite , and the recent emerging field of thermotronics @xcite . on the experimental side , the theoretical predictions have been verified during the last decade both in the plane - plane and sphere - plane configuration , for a wide range of distances , going from some nanometers to several microns @xcite . recently , the idea of manipulating the heat flux through the manipulation of some external parameters have attracted a remarkable attention . in fact , both the control of the overall value of the flux and its spectral properties can be extremely relevant for several applications , and in particular for energy conversion . in the spirit of a manipulation thourgh geometrical properties , structured surfaces have been the topic of many theoretical investigations . more specifically , by considering both 1d and 2d periodic gratings , both the radiative heat transfer @xcite and the casimir force at and out of thermal equilibrium @xcite have been studied by employing a variety of theoretical approaches . it has been shown that gratings represent indeed a tool to modify , both by reducing and amplifying , radiative heat transfer , as well as to influence its spectral properties . concerning the casimir force , it must be mentioned that the force between a sphere and a dielectric @xcite or a metallic @xcite grating has been recently measured and theoretically investigated @xcite . in the domain of radiative heat transfer , metallic gratings deserve ideed a special attention . in fact , although , as mentioned above , surface resonances for a metal typically have ultraviolet frequencies , the presence of a periodically structured pattern can result in the presence of new surface resonances , referred to a spoof plasmons @xcite , whose frequency can be adjusted as a function of the grating parameters and can be brought , for realistic values of the grating lenghtscales , in the infrared region , thus contributing to the flux . based on this behavior , the authors of ref . have recently theoretically predicted a high enhancement of the flux between two identical gold gratings . a similar study has been performed for 1d and 2d metallic gratings in refs . , and more recently in ref . . in these papers the reflection upon each grating has been described by means of the fourier modal method ( fmm ) @xcite , equipped with the factorization rule introduced in ref . . it is worth stressing that , since the radiative heat transfer is calculated as an integral with respect to both frequency and wavevector , the choice of the method used to derive the grating reflection operators can drastically affect the computational time as well as the results . this paper is precisely devoted to the study of radiative heat transfer between two gold gratings . we present and discuss a modification to the fmm technique , known as adaptive spatial resolution @xcite ( we will in the following refer to this modified method as asr ) , a technique specifically introduced to accelerate the convergence . this method has also been shown to overcome the known instabilities appearing for metallic gratings @xcite . with respect to previous works , we extend here this technique to deal with arbitrary conical incidence . based on this approach , we provide a detailed study of the influence of the grating depth on the overall value of the flux as well as on its spectral properties . we also compare our results to part of the works mentioned above on heat transfer between metallic gratings . we show that ( 1 ) the asr technique produces a dramatic increase of the convergence rate ( thus implying a drastic reduction of computational time ) and ( 2 ) the numerical results for heat transfer considerably differ from results previously obtained using the standard fmm . the paper is structured as follows . in sec . [ secintro ] we present our physical system , describing also our notation and main definitions . in sec . [ secmethods ] we discuss in detail the adaptive spatial resolution . then , in sec . [ secnumerics ] we present our numerical results , by studying the radiative heat transfer between two gold gratings as a function of the grating height . we finally give in sec . [ secconcl ] some conclusive remarks . the system we are going to address consists of two 1d lamellar gratings separated by vacuum , as shown in fig . [ figgeometry ] . the two gratings are structured along the @xmath2 axis , translationally invariant along the @xmath3 axis , and separated by a vacuum gap of thickness @xmath4 along the @xmath5 axis . the two gratings share the same period @xmath6 , while they can have in general different filling fractions @xmath7 ( see fig . [ figgeometry ] ) . finally , they have different grating heights @xmath8 , while the substrate below each structured region is assumed to be infinitely thick for both gratings . . the gratings , in general made of different materials , are infinite in the @xmath9 plane , and periodic in the @xmath2 direction with the same period @xmath6 . they have corrugation depths @xmath10 ( @xmath11 ) , infinite thicknesses below the grating region , and lengths of the elevated part of the grating @xmath12 . this defines the filling factors @xmath13.,width=321 ] the two gratings , labeled with indexes 1 and 2 , are kept by some external heat source at constant temperatures @xmath14 and @xmath15 . among the numerous theoretical techniques developed to calculate the radiative heat transfer between them , we employ an approach based on the knowledge of the individual scattering operators of the bodies involved , recently introduced for systems involving two @xcite and three @xcite bodies . this method is based on a plane - wave mode decomposition of the fields , each mode @xmath16 being identified by the direction of propagation @xmath17 along the @xmath5 axis , the polarization index @xmath18 [ assuming the values @xmath19 which respectively correspond to transverse electric ( te ) and transverse magnetic ( tm ) modes ] , the frequency @xmath20 and the transverse wavevector @xmath21 . in this description , the @xmath5 component of the wavevector @xmath22 is a dependent variable defined as @xmath23 in virtue of this mode decomposition , we define the trace of an operator @xmath24 as @xmath25 assuming that the external environment is thermalized with body 1 , the energy this body receives per unit surface and time is given by @xcite @xmath26 , \label{defphi}\ ] ] where @xmath27 ( @xmath28 ) is the reflection operator of body 1 ( 2 ) for an incoming wave propagating in direction @xmath29 ( @xmath30 ) and we have defined @xmath31 moreover , in eq . we have defined the population differences @xmath32 , with @xmath33 the intracavity operator @xmath34 and the projection operators @xmath35 where @xmath36 is the kronecker delta and being @xmath37 [ @xmath38 the projector on the propagative ( @xmath39 ) [ evanescent ( @xmath40 ) ] sector . we remark that , as discussed in ref . , the periodicity on the @xmath2 axis makes it natural to replace the mode variable @xmath41 with @xmath42 with @xmath41 taking values in the first brillouin zone @xmath43 $ ] and @xmath44 assuming all integer values . based on eq . , the calculation of the flux @xmath45 is now reduced to the problem of describing the field reflection upon each grating . the details about the asr technique employed in this paper are discussed in detail in the next section . the physical system we consider is shown in fig . [ figgrating ] : it is a lamellar grating of period @xmath6 , invariant along the @xmath3 axis , with relative dielectric permittivity @xmath46 . the grating is inlayed between two homogeneous media whose permittivities are @xmath47 ( input medium ) and @xmath48 ( output medium ) . a monochromatic plane wave with frequency @xmath20 and parallel wavevector @xmath21 illuminates the structure . throughout the calculation , the vacuum wave number is denoted by @xmath49 and we assume a time dependence of the form @xmath50 . the particularity of the asr @xcite is the use of a new coordinates system @xmath51 in which the coordinate in the @xmath2 direction is stretched around the permittivity discontinuities . this allows to better describe the permittivity jump , which is crucial in the case of metallic structures . below , we derive the metric tensor and write the maxwell s equations associated to the coordinates change . then we solve them in the three regions of space : @xmath52 , @xmath53 and @xmath54 . after , we derive the boundary conditions and obtain the fields amplitudes . based on the geometry depicted in fig . [ figgrating ] , the reflection operator we are going to calculate is @xmath55 for a vacuum - grating inferface located at @xmath56 . this operator will allow us to easily deduce @xmath27 and @xmath28 as described in detail below . and @xmath57 having uniform permittivity ( @xmath47 and @xmath48 , respectively ) and @xmath58 , having a periodic @xmath46.,width=321 ] we start by observing that , in general , the presence of a lamellar grating naturally divides the period region @xmath59 $ ] by a set of discontinuity points @xmath60 with @xmath61 , with @xmath62 and @xmath63 . in view of the change of coordinates from @xmath2 to @xmath64 , we analogously define a set of points @xmath65 given by @xmath66 , i.e. uniformly distributed between @xmath67 and @xmath68 . for the explicit definition of the coordinate transformation @xmath69 we follow ref . and write @xmath70 in each interval @xmath71 $ ] , with @xmath72 , where @xmath73 in these expressions @xmath74 being a control parameter , for which we take the value @xmath75 through all the numerical calculations presented in this paper . the corresponding metric tensor is diagonal and reads @xmath76 , with @xmath77 . thus maxwell s equations @xmath78 and @xmath79 become : @xmath80 @xmath81 with @xmath82 , @xmath83 being the impedance of vacuum . we now derive from eqs . and @xmath84 and @xmath85 and plug them into the other equations , obtained the following system for the tangential components of the fields : @xmath86 @xmath87 with @xmath88 and @xmath89 . next , following the procedure discussed more in detail e.g. in ref . , we rewrite this systems of equations in the fourier space and truncate the series up to the truncation order @xmath90 , obtaining @xmath91 @xmath92 where @xmath93 $ ] , @xmath41 being defined in the first brillouin zone @xmath43 $ ] . in eq . we have gathered in the column vectors @xmath94 and @xmath95 the @xmath96 fourier components of the fields @xmath97 and @xmath98 . moreover , we have introduced the toeplitz matrices @xmath99 ( resp . @xmath100 and @xmath101 ) defined in terms of the fourier coefficients of the function @xmath102 [ resp . @xmath103 and @xmath104 . we can further simiplify the notation by introducing the following definitions @xmath105 which allow us to write @xmath106 in this equation , @xmath107 is a diagonal matrix containing the eigenvalues of @xmath108 and @xmath109 is the matrix of associated eigenvectors . then the solution of eq . can be expressed under the form @xmath110 and consequently [ from eq . ] : @xmath111 where @xmath112 and @xmath113 are the amplitudes of the different waves traveling in the @xmath30 and @xmath29 directions , respectively . the calculation presented here for the grating region , characterized by a periodic @xmath46 , can be directly applied to the homogeneous regions @xmath114 and @xmath57 as well . in this case , the framework can be simplified by observing that @xmath115 and @xmath116 , where for the input ( output ) region @xmath117 ( @xmath118 ) . this directly leads to the introduction of the four additional matrices @xmath119 , @xmath120 , @xmath121 , and @xmath122 . in order to simplify the boundary conditions , the solution in the output medium is modified as follows @xmath123 by means of the introduction of a phase factor . having solved maxwell s equations in each region , we can now write down the boundary conditions at the interfaces @xmath56 and @xmath124 : @xmath125 and , denoting @xmath126 , @xmath127 in terms of the @xmath128-matrix algorithm , this gives @xmath129 and @xmath130 in the last two equations the unknowns are the amplitudes @xmath131 and @xmath132 of the incoming waves , @xmath133 and @xmath134 of the reflected and transmitted waves , and the amplitudes @xmath112 and @xmath113 of the field in the grating region , in which we have absorbed the factor @xmath135 . this is irrelevant for our purposes , since we are only interested in the field amplitudes in the homogeneous media . the final part of the calculation is straightforward ( see e.g. ref . ) . we define a chained @xmath128 matrix as @xcite @xmath136 having introduced the associative operation @xmath137 , which for three square matrices @xmath138 , @xmath139 and @xmath140 of dimension @xmath141 is defined as @xmath142 where each matrix have been decomposed in four square blocks of dimension @xmath143 . this @xmath128 matrix satisfies the relation @xmath144 thus its upper - left @xmath143 block , relating the reflected field @xmath133 to the incident one @xmath131 in the upper region , can be identified as the @xmath145 operator we are looking for in the @xmath146 reference system . m , period @xmath147 m , filling fraction @xmath148 , infinite thickness below the grating region , and temperatures @xmath149k and @xmath150k . the flux is calculated using both the fmm ( red dashed line ) and the asr ( black solid line ) , for different truncation orders @xmath90 . the black dashed line corresponds to the asymptotic value obtained using the asr for @xmath151 , while the gray dot - dashed line is associated to a 10% error with respect to this asymptotic value.,width=302 ] the steps described in the last section lead to the derivation of a reflection matrix @xmath145 , expressed in the transformed reference system @xmath146 . actually , the operator @xmath55 needed for the calculation of the flux given by eq . has to be expressed not only in the standard @xmath152 cartesian reference , but also with respect to the basis of two polarization unity vectors @xmath153}_{\text{te}}^{\phi}(\mathbf{k}_n,\omega)&=\frac{1}{k_n}(-k_y\hat{\mathbf{x}}+k_{x , n}\hat{\mathbf{y}}),\\ \hat{\bbm[\epsilon]}_{\text{tm}}^{\phi}(\mathbf{k}_n,\omega)&=\frac{c}{\omega}(-k_n\hat{\mathbf{z}}+\phi k_{z , n}\hat{\mathbf{k}}_n ) , \end{split}\label{polvect}\ ] ] where @xmath154 and we have defined @xmath155 and @xmath156 . thus , the final reflection matrix @xmath157 takes the form @xmath158 where @xmath159 is the matrix associated with the transformation from @xmath146 to @xmath152 , whereas @xmath160 accounts for the transition from the ( te , tm ) basis to the canonical @xmath152 basis for fields propagating in the @xmath161 direction ( note that we have @xmath30 for the incident field and @xmath29 for the reflected one , as manifest from fig . [ figgrating ] ) . in order to derive @xmath159 , we start observing that we have , labeling throughout this section with a prime the fields in the @xmath146 coordinate system , @xmath162 and @xmath163 . this means that the @xmath159 matrix will be block - diagonal and take the form @xmath164 moreover , since the coordinate stretch leaves the period unchanged , one just needs to express the @xmath96 vectors @xmath165 and @xmath166 as a function of @xmath167 and @xmath168 for a given value of @xmath20 and @xmath41 in the first brillouin zone , as @xmath169 and @xmath170 . starting from the fourier decompositions @xmath171 we easily get @xmath172_{n , m}=\frac{1}{d}\int_0^ddu\,e^{i[k_{x , m}u - k_{x , n}x(u)]},\ ] ] and in an analogous way @xmath173_{n , m}=\frac{1}{d}\int_0^ddu\,f(u)e^{i[k_{x , m}u - k_{x , n}x(u)]}.\ ] ] concerning the second transformation matrix @xmath160 , it represents a basic change of basis and its action is written as @xmath174 where the ( te , tm ) basis ( and thus the matrix @xmath160 ) depends on the propagation direction @xmath161 . using eq . , the transformation matrix reads @xmath175 as stated at the beginning of our calculation , the @xmath55 operator we calculated is relative to a grating having its interface with vacuum at @xmath56 . as a consequence , in order to deduce the @xmath28 relative to grating 2 , we have to include a phase shift taking into account the fact that its vacuum - grating inferface coincides with the plane @xmath176 . as described e.g. in ref . , the matrix elements of this modified operator are given by @xmath177{\displaystyle{\langle p,\mathbf{k},n,\omega|}}\mathcal{r}^-{\displaystyle{|p',\mathbf{k}',n',\omega'\rangle}}.\nonumber \end{split}\ ] ] we now focus on the issue of deriving @xmath27 from the known operator @xmath55 . for simplicity , in the following we will always consider two identical gratings , i.e. having the same height @xmath8 and filling factor @xmath178 . in this case , when evaluating the flux given by eq . , it is convenient to exploit the symmetry of the configuration and calculate ( for each frequency and wavevector ) only once the grating reflection matrix . nevertheless , one must keep in mind that in our formalism the unit polarization vector associated to te polarization defined in eq . is independent of the propagation direction @xmath161 along the @xmath5 axis , while the @xmath2 and @xmath3 components of the tm polarization vector are proportional to @xmath161 . since the @xmath2 and @xmath3 components are the ones which are conserved in the scattering process , this implies that the reflection matrix @xmath27 coincides with @xmath55 in the diagonal blocks ( te , te ) and ( tm , tm ) , while the non - diagonal blocks ( te , tm ) and ( tm , te ) undergo an overall sign change . this sign - change issue is irrelevant , of course , in the case of a planar slab , since this system does not mix the two polarizations , but it must indeed be taken into account when dealing with bodies producing a coupling between te and tm modes . and the one for @xmath179 , i.e. the one between two planar slabs . the two identical gratings have period @xmath147 m , filling fraction @xmath148 , infinite thickness below the grating region , and temperatures @xmath180k and @xmath181k . our results ( black solid line ) are compared to the results of ref . ( red points ) . the blue dot corresponds to a calculation using the asr but without inverting the sign of non - diagonal blocks of the reflection matrix ( see text for more details ) . the inset shows a zoom of the same curve in linear @xmath8 scale.,width=321 ] we are now ready to discuss our first numerical results . these concern a couple of gratings having the same features used in ref . . the two gratings have filling factor @xmath148 , period @xmath147 m , temperatures @xmath180k and @xmath181k , and are placed at a distance @xmath182 m . both gratings are made of gold , for which we have used a drude model @xmath183 where the plasma frequency and the dissipation rate are respectively equal to @xmath184ev and @xmath185mev . in order to highlight the features of the asr technique , we start by calculating the flux for a given grating height @xmath186 m . as discussed in detail in ref . , a crucial point when using the fmm technique is the choice of the truncation order @xmath90 , i.e. the number of diffraction orders taken into account in the fourier decomposition of the field , going from @xmath187 to @xmath90 . the same issue applies of course to the asr method as well . we show in fig . [ figfmmasr ] the value of the flux @xmath45 as a function of the truncation order , both using the fmm ( red dashed line ) and the asr ( black solid line ) . it is manifest that in both cases the first truncation orders give very different results , suggesting that @xmath90 has to be further increased . nevertheless , starting from @xmath90 of the order of 5 , the two curves are dramatically different . the curve corresponding to the asr technique very quickly converges to a stable result , and the points corresponding to @xmath188 are basically undistinguishable . the dashed black line in the plot corresponds to the value obtained with the asr for @xmath151 , while the dot - dashed line is associated to a 10% error with respect to this result . the results obtained with the fmm show a very different behavior . first of all , up to @xmath189 , the obtained flux is non - monotonic , and has a quasi - oscillatory behavior . moreovoer , even when becoming monotonic as a function of @xmath90 , the flux converges very slowly to its asymptotic value . more specifically , for @xmath190 the error is of the order of 10% , while for @xmath191 ( the value used in ref . ) it is of the order of 3% . m , period @xmath192 m , filling fraction @xmath193 , infinite thickness below the grating region , and temperatures @xmath194k and @xmath195k . the black solid line corresponds to our calculation using the asr , while the blue dashed line gives the result obtained without inverting the sign of non - diagonal blocks of the reflection matrix ( see text for more details ) . the red points are extracted from ref . .,width=321 ] based on this discussion of the convergence , we use , in the following , a truncation order @xmath196 for the asr , with an associated relative error of the order of 1% on the integrated flux . using these parameters , as in ref . , we study the heat transfer @xmath197 as a function of the grating height @xmath8 , and discuss the amplification factor @xmath198 with respect to the case @xmath179 , i.e. to the case of a planar slab . the results are shown in the main part of fig . [ figh ] for a very wide range of values of @xmath8 , going from @xmath179 to 1 mm , and compared with the ones ( red points ) calculated in ref . from @xmath179 to 6@xmath199 m . while for @xmath200 nm the amplification factor is close to 1 , increasing @xmath8 produces a huge amplification of the heat flux . in particular , the ratio @xmath198 grows monotonically with @xmath8 , and reaches a horizonal asymptotote around @xmath201 m , with a value slightly above 34 . in the inset of fig . [ figh ] , we show the same curve in linear @xmath8 scale from 0 to @xmath202 m . as manifest from the plot , in spite of a qualitative agreement , the two curves are quantitatively different , with a disagreement going up to around 50% for @xmath203 m . for this value of the grating height we have calculated the flux using fmm with @xmath191 and obtained a result in agreement within around 1% with the one coming from the asr with @xmath196 . we conclude that the truncation order @xmath191 is large enough to assure a good convergence for the integrated flux using fmm . thus , the difference between our results and the ones of ref . does not seem to originate from an insufficient truncation order . it is now instructive to investigate the evolution of the spectral properties of the heat flux as a function of the grating depth @xmath8 . this behavior is illustrated in fig . [ figgrid ] . panel ( a ) shows the spectral flux @xmath204 for @xmath179 , i.e. in a slab - slab configuration . it is well known that in presence of surface modes , while approaching the near - field regime the heat flux becomes more and more monochromatic at the resonance frequency of the mode @xcite . even if the distance @xmath182 m does not fully lie in this distance regime , a signature of the existence of such surface modes is already present . nevertheless , this is not the case for two gold slabs , simply because the plasma frequency for gold is located well outside the window of frequencies contributing to the flux , defined by the planck function @xmath205 . as a result , the only non - monotonic behavior for the flux is observed in the low - frequency region @xmath206\,$]rad / s , shown in the inset of fig . [ figgrid](a ) . the spectral flux tends to increase at small frequencies as a result of the divergence of @xmath207 for @xmath208 , this growth being at some point compensated by the fact that each mode of the field carries an energy @xmath209 , tending to 0 for @xmath208 . let us now focus on fig . [ figgrid](b ) , where the grating heights @xmath210 m and @xmath211 m are taken into account . we first remark that the non - monotonic behavior at low frequencies is still present , but with a decreased maximum value . this feature remains basically unchanged for all the higher values of @xmath8 shown in fig . [ figgrid ] . the main feature of fig . [ figgrid ] is anyway the appearance of two peaks in the spectral flux , sign of the existence of two surface modes , whose frequency depends on the grating height , coherently with the fact that these are indeed spoof plamons . the participation of these new modes produces an overall flux amplification equal to 2 and 2.7 , respectively , for @xmath210 m and @xmath211 m . an analysis of panel ( c ) of the same figure shows that increasing @xmath8 produces both a decrease of the frequency of spoof - plasmon modes and an amplification of the peak height . as shown in fig . [ figgrid](c)-(d ) the scenario becomes even more interesting for higher values of @xmath8 . in this case more and more resonances enter the window of frequencies relevant for radiative heat transfer . this feature has been already discussed in ref . by analyzing the heat - flux transmission coefficient for a given value of the wavevector @xmath212 . we basically observe two resonances for @xmath213 m and @xmath214 m , while for @xmath215 m and @xmath216 m not only we start observing a comb of resonance frequencies , but we clearly see that the fact that they approach each other produces a constructive interference between them , increasing the overall value of @xmath204 in the frequency range considered . this is even more evident in fig . [ figgrid](f ) , where for @xmath217 m the frequency - comb behavior is manifest , as well as how this eventually produces a smooth asymptotic @xmath204 ( reached for @xmath218 mm ) which does not show any abrupt change with respect to @xmath20 . this is the profile giving the asymptotic flux amplification close to 34 discussed above and shown in fig . [ figh ] . in order to compare further our results with previous works , we focus on a configuration studied in ref . , namely the heat transfer between two identical gratings having height @xmath219 m , period @xmath220 nm and filling factor @xmath193 , placed at distance @xmath182 m . the two chosen temperatures are @xmath194k and @xmath195k . in ref . , the authors show the spectrum associated with this heat transfer ( calculated using the standard fmm ) from which we have extracted some points , shown in red in fig . [ figspectrum ] . we have applied our numerical scheme to this scenario and obtained the black curve shown in the same figure . it is manifest that , while sharing the same qualitative behavior , the two curves considerably differ quantitatively , in particular in the region between the two main peaks . in order to try to explain this disagreement we have noticed that in one of their recent papers on this topic ( namely , ref . ) , the authors clearly state that the fact that the two structures coincide implies that the two reflection operators @xmath27 and @xmath28 are equal . as discussed above , this is actually not correct because of the fact that the unit polarization vector in tm polarization depends on the propagation direction @xmath161 . nevertheless , we have performed a new calculation for the same structure , using the asr with @xmath196 , but omitting the sign change we need to deduce @xmath27 from @xmath221 . the result so obtained is given by the blue dashed curve in fig . [ figspectrum ] and clearly shows an impressively increased agreement with the red points . thus , we interpret the difference between the two results as due to this sign change , missing in ref . and related works . moreover , it must be stressed that the results presented in ref . also show an oscillatory behavior in frequency , in particular in the region between the two main peaks , which is completely absent in the results obtained by exploiting the asr , both with and without the sign change . in our opinion , this oscillations are a result of the well - known instabilities of the fmm when dealing with metals . with the aim of further trying to explain the discrepancy with ref . shown in fig . [ figh ] , we have recalculated using the asr one of the points of fig . [ figh ] , the one having @xmath222 m , without the mentioned sign change . the result is the blue dot in fig . [ figh ] which , even if approaching the result of ref . , still does not show a good agreement . we have addressed the caclulation of the radiative heat flux between two gold gratings . to this aim we have made use of the asr , a modified version of the fmm introduced to deal with the high dielectric contrast typical of metals . we have shown that this technique produces a striking increase of the convergence rate , allowing us to obtain the heat flux at a distance of 1@xmath223 m with a truncation order as low as @xmath196 . this implies a remarkable gain in computational time : to give an idea , for the grating parameters relative to fig . [ figfmmasr ] , the calculation of the spectral flux at frequency @xmath224rad / s takes approximately 1 minute using the asr with @xmath196 on a parallel 3.4ghz 16-core machine , while 1.5 hours are needed using the fmm with @xmath191 . since the required truncation order increases when decreasing the distance , the asr would allow to explore smaller separations , which would be prohibitive using the fmm . by using this improved numerical method , we have made a detailed study of the heat - flux amplification as a function of the grating depth . by studying both the flux ratio and the spectral properties , we have proved for our structure an amplification factor going up to 34 , explained in terms of the appearance of a comb of spoof - plasmon resonances . we have then compared our results to some previous works , and highlighted a quantitative disagreement both in the integrated flux and in its spectral distribution . in some cases , we have argued that this disagreement is due to the incorrect assumption that the reflection matrices of the two gratings coincide . our results show that the physics behind metallic gratings makes them ideal candidates for the manipuation of both the overall heat flux and its frequency components , both possibilities being very relevant for several applications such as thermophotovoltaic energy conversion . on a more technical side , our discussion proves that the fmm can provide slowly - converging and unstable results in the presence of metals , while the asr represents a much more reliable approach even for high dielectric contrasts . 99 s. m. rytov , academy of sience press of ussr , moscow ( 1953 ) . d. polder and m. van hove , phys . b * 4 * , 3303 ( 1971 ) . k. joulain _ et al . _ , rep . * 57 * , 59 ( 2005 ) . r. s. dimatteo , p. greiff , s. l. finberg , k. a. young - waithe , h. k. h. choy , m. m. masaki , and c. g. fonstad , appl . * 79 * , 1894 ( 2001 ) . a. narayanaswamy and g. chen , appl . . lett . * 82 * , 3544 ( 2003 ) . m. laroche , r. carminati , and j .- j . greffet , j. appl . phys . * 100 * , 063704 ( 2006 ) . s. basu , y .- b . chen , and z. m. zhang , int . j. energy res . * 31 * , 689 ( 2007 ) . m. francoeur , m. p. meng , and r. vaillon , appl . . lett . * 93 * , 043109 ( 2008 ) . s. basu and z. zhang , j. appl . * 105 * , 093535 ( 2009 ) . r. m. swanson , science * 324 * , 891 ( 2009 ) . c. j. chen , _ the physics of solar energy _ ( john wiley & sons , inc . , hoboken , 2011 ) . w. srituravanich , n. fang , c. sun , q. luo , and z. zhang , nano lett . * 4 * , 1085 ( 2004 ) . b. guha , c. otey , c. b. poitras , s. fan , and m. lipson , nano letters * 12 * , 4546 ( 2012 ) . p. ben - abdallah and s .- a . biehs , aip advances * 5 * , 053502 , ( 2015 ) . a. kittel , w. mller - hirsch , j. parisi , s .- a . biehs , d. reddig , and m. holthaus , phys . rev . lett . * 95 * , 224301 ( 2005 ) . l. hu , a. narayanaswamy , x. chen , and g. chen , appl . . lett . * 92 * , 133106 ( 2008 ) . a. narayanaswamy , s. shen , and g. chen , phys . b * 78 * , 115303 ( 2008 ) . e. rousseau , a. siria , g. joudran , s. volz , f. comin , j. chevrier , and j .- j . greffet , nature photon . * 3 * , 514 ( 2009 ) . s. shen , a. narayanaswamy , and g. chen , nano letters * 9 * , 2909 ( 2009 ) . t. kralik , p. hanzelka , v. musilova , a. srnka , and m. zobac , rev . instrum . * 82 * , 055106 ( 2011 ) . r. s. ottens , v. quetschke , s. wise , a. a. alemi , r. lundock , g. mueller , d. h. reitze , d. b. tanner , and b. f. whiting , phys . lett . * 107 * , 014301 ( 2011 ) . p. j. van zwol , l. ranno , and j. chevrier , phys . * 108 * , 234301 ( 2012 ) . p. j. van zwol , s. thiele , c. berger , w. a. de heer , and j. chevrier , phys . * 109 * , 264301 ( 2012 ) . t. kralik , p. hanzelka , m. zobac , v. musilova , t. fort , and m. horak , phys . lett . * 109 * , 224302 ( 2012 ) . et al . _ , nature * 528 * , 387 ( 2015 ) . b. song _ et al . _ , nature nanotechnology * 10 * , 253 ( 2015 ) . r. st - gelais , l. zhu , s. fan , and m. lipson , nature nanotechnology * 11 * , 515 ( 2016 ) . k. kloppstech _ et al . _ , preprint arxiv:1510.06311 ( 2015 ) . j. i. watjen , b. zhao , and z. m. zhang , appl . lett . * 109 * , 203112 ( 2016 ) . s .- a . biehs , f. s. s. rosa , and p. ben - abdallah , appl . . lett . * 98 * , 243102 ( 2011 ) . j. lussange , r. gurout , f. s. s. rosa , j .- greffet , a. lambrecht , and s. reynaud , phys . b * 86 * , 085432 ( 2012 ) . r. gurout , j. lussange , f. s. s. rosa , j .- hugonin , d. a. r. dalvit , j .- j . greffet , a. lambrecht , and s. reynaud , phys . b * 85 * , 180301(r ) ( 2012 ) . x. l. liu and z. m. zhang , appl . . lett . * 104 * , 251911 ( 2014 ) . j. dai , s. a. dyakov , and m. yan , phys . b * 92 * , 035419 ( 2015 ) . j. dai , s. a. dyakov , and m. yan , phys . b * 93 * , 155403 ( 2016 ) . j. dai , s. a. dyakov , s. i. bozhevolnyi , and m. yan , phys . rev . b * 94 * , 125431 ( 2016 ) . y. yang and l. wang , phys . . lett . * 117 * , 044301 ( 2016 ) . a. lambrecht and v. n. marachevsky , phys . * 101 * , 160403 ( 2008 ) . chiu , g. l. klimchitskaya , v. n. marachevsky , v. m. mostepanenko , and u. mohideen , phys . b * 81 * , 115417 ( 2010 ) . p. s. davids , f. intravaia , f. s. s. rosa , and d. a. r. dalvit , phys . a * 82 * , 062111 ( 2010 ) . f. intravaia , p. s. davids , r. s. decca , v. a. aksyuk , d. lpez , and d. a. r. dalvit , phys . a * 86 * , 042101 ( 2012 ) . j. lussange , r. gurout , and a. lambrecht , phys . a * 86 * , 062502 ( 2012 ) . r. gurout , j. lussange , h. b. chan , a. lambrecht , and s. reynaud , phys . a * 87 * , 052514 ( 2013 ) . n. graham , phys . a * 90 * , 032507 ( 2014 ) . j. wagner and r. zandi , phys . rev . a * 90 * , 012516 ( 2014 ) . a. noto , r. messina , b. guizal , and m. antezza , phys . a * 90 * , 022120 ( 2014 ) . r. messina , p. a. maia neto , b. guizal , and m. antezza , phys . a * 92 * , 062504 ( 2015 ) . h. b. chan , y. bao , j. zou , r. a. cirelli , f. klemens , w. m. mansfield , and c. s. pai , phys . * 101 * , 030401 ( 2008 ) . y. bao , r. gurout , j. lussange , a. lambrecht , r. a. cirelli , f. klemens , w. m. mansfield , c. s. pai , and h. b. chan , phys . lett . * 105 * , 250402 ( 2010 ) . a. a. banishev , j. wagner , t. emig , r. zandi , and u. mohideen , phys . lett . * 110 * , 250403 ( 2013 ) . f. intravaia et al . , nat . comm . * 4 * , 2515 ( 2013 ) . j. b. pendry , l. martn - moreno , and f. j. garcia - vidal , science * 305 * , 847 ( 2004 ) . f. j. garca - vidal , l. martn - moreno , and j. b. pendry , j. opt . a : pure appl . opt . * 7 * , s97 ( 2005 ) . h. kim , j. park , and b. lee , _ fourier modal method and its applications in computational nanophotonics _ ( crc press , boca raton , 2012 ) . g. granet , b. guizal , j. opt . a * 13 * , 1019 ( 1996 ) . g. granet j. opt . a * 16 * , 2510 ( 1999 ) . b. guizal , h. yala , and d. felbacq , opt . express * 34 * , 2790 ( 2009 ) . r. messina and m. antezza , europhys . lett . * 95 * , 61002 ( 2011 ) . r. messina and m. antezza , phys . a * 84 * , 042102 ( 2011 ) . r. messina , m. antezza , and p. ben - abdallah , phys . * 109 * , 244302 ( 2012 ) . r. messina and m. antezza , phys . rev . a * 89 * , 052104 ( 2014 ) . t. vallius , and m. honkanen opt . express * 10 * , 24 ( 2002 ) . g. granet , and b. guizal j. opt . a * 13 * , 1019 ( 1996 ) . j. dai , ph.d . thesis , school of information and communication technology , kth royal institute of technology , 2016 .

in their comment , @xcite delaney and greer ( dg ) claim that the unphysical result ( current @xmath2 ) presented by us in ref . ( hereafter , i ) as evidence against their variational approach @xcite is due to the fact ( i ) that our constraints ( electronic populations ) prevent a broken time - reversal symmetry , and ( ii ) that we used real orbitals , which concomitantly constrain incoming and outgoing electrons . further , dg accept that our idea to constrain populations `` is an interesting alternative to the wigner function '' ( wf ) , but ( iii ) argue that to get @xmath3 it is essential to use complex orbitals , like those obtained by applying periodic boundary conditions ( pbcs ) for electrodes . by using the superscript @xmath4 and by referring to their work,@xcite which we amply criticized @xcite they aim to convey the false impression that they used such complex orbitals previously . responding briefly , claim ( i ) is incorrect , claim ( ii ) has absolutely no real basis , while claim ( iii ) does not mender their variational approach ; the complex orbitals discussed by them solely represent a particular case of the general case of i and do not in the least change the unphysical linear response current @xmath2 . dg argue that the unphysical result ( @xmath2 ) deduced in i is the consequence of our unsuitable choice of constraints , namely the particle number operators . they ( incorrectly ) claim that constraining _ real _ particle number operators does not break time - reversal symmetry , and for this very reason the current vanishes even beyond the linear response limit examined in i. we shall immediately show that this is not true : our constraints _ allow _ a broken time - reversal symmetry . the fact that this symmetry is _ not _ broken is the _ result _ of the defective dg variational ansatz ( constrained energy minimization at zero temperature ) . below , we shall only discuss the particular constraint @xmath5 mentioned in the comment . let us consider a general many - body state , @xmath6 , expanded in terms of the complete set of the real eigenstates of the hamiltonian @xmath7 of the finite transport cluster without external bias ( @xmath8 ) , @xmath9 . notice that this _ exact _ expansion is just in the spirit of the dg approach , which is based on a configuration interaction ( ci ) , albeit approximate expansion [ cf . ( 1 ) of ref . ] . the expansion coefficients @xmath10 _ are _ allowed to be complex . accepting the dg challenge , we consider now the general response instead of the linear limit of i. we do no more single out the ground state @xmath11 , unlike e. g. , eq . ( 10 ) of i , and consider the general matrix elements of the relevant operators [ external bias hamiltonian , @xmath12 , and current operator at site @xmath13 , @xmath14 @xmath15 where @xmath16 is an arbitrary fixed site . ( unless otherwise specified , we use throughout the notations and definitions of i and ignore the spin for simplicity . ) notice that all the above calligraphic symbols denote real quantities . to determine the steady - state current , the dg variational ansatz prescribes a constrained energy minimization @xmath17 = \min , \ ] ] where @xmath18 , @xmath19 , and @xmath20 are real lagrange multipliers . the constraints are ( see i ) : @xmath21 the minimization with respect to @xmath10 and @xmath22 of eq . ( [ eq - min ] ) yields @xmath23 = 0\ ] ] and @xmath24 = 0 , \ ] ] respectively . notice the reversed sign of the last term in the square parentheses of eqs . ( [ eq - an ] ) and ( [ eq - an - ast ] ) . it is due the antisymmetry expressed in eq . ( [ eq - j - q ] ) , which is related to the fact that the matrix elements the hermitian current operator are purely imaginary , i. e. , @xmath25 are real . rephrasing , this reflects the fact that if the state @xmath26 corresponds to a current @xmath27 , @xmath28 corresponds to a current @xmath29 . to determine the expansion coefficients ( and the lagrange multipliers ) , one must solve eqs . ( [ eq - q - k ] ) ( [ eq - an - ast ] ) , and then the position - independent current can be computed @xmath30 as visible in eqs . ( [ eq - an ] ) and ( [ eq - an - ast ] ) , @xmath10 and @xmath22 obey _ different _ equations . so , the @xmath10 s are _ allowed _ to be complex , hence @xmath31 is also _ allowed _ to be complex . constraining populations , eq . ( [ eq - q - k ] ) , _ allows _ a broken time - reversal symmetry . to see whether the @xmath10 s _ are _ indeed complex ( thence whether a current @xmath32 , eq . ( [ eq - j ] ) , _ can _ indeed flow ) , one has to solve the set of eqs . ( [ eq - q - k ] ) ( [ eq - an - ast ] ) . beyond linear response , eqs . ( [ eq - q - k ] ) ( [ eq - an - ast ] ) are coupled nonlinear equations , and the solution is _ not _ necessarily unique . in i , we do _ solve _ this problem for linear response , show that the solution _ is _ unique , and the unphysical result @xmath2 is the _ outcome _ of these calculations . throughout our critical analysis of the dg approach , @xcite to be on the safe side , we confined ourselves to the linear response limit , wherein the solution is _ unique _ and its lamentable failure can be unambiguously stated . it _ is _ possible that the dg current vanishes beyond the linear response limit,@xcite but we can not _ safely _ state this . but _ if _ this were the case , it would have nothing to do with the dg claim : as seen above , constraining populations does not prevent a broken time - reversal symmetry . whether the solution of the nonlinear problem corresponds to @xmath2 or not did and does not represent our concern once the linear response limit is incorrect . this would be yet another unphysical dg prediction , as we already stated _ explicitly _ : see the last but one paragraph of sec . vi in ref . . let us now see what is wrong with the dg argumentation yielding the `` conclusion '' that constraining populations is unphysical . there are two reasons why their conclusion is incorrect : on one hand , the dg argumentation eliminates an important aspect and , on the other hand , it introduces an unjustified assumption . dg seem to have realized that current conservation is a serious stumbling block for their argumentation ; noteworthy , the steps 16 listed in the beginning of the comment omit the current conservation . they first discuss the minimization _ ignoring _ the current conservation . this amounts to exclude the last term of the square parentheses of eqs . ( [ eq - min ] ) , ( [ eq - an ] ) , and ( [ eq - an - ast ] ) above . this _ almost _ suffices to argue that eqs . ( [ eq - q - k ] ) ( [ eq - an - ast ] ) automatically imply real coefficients @xmath10 . `` almost '' but not _ really _ , because the unknown @xmath10 can be complex ( @xmath33 ) despite the fact that all the other quantities entering these equations are real ( remember @xmath34 ) . then , to more `` convincingly '' argue that our constraints do not break the time - reversal symmetry , dg supplement their argumentation with an _ unjustified _ supposition . namely , they simply _ postulate _ that `` the variational approach assumes a unique solution @xmath35 '' the dg variational approach is a well - defined mathematical problem . needless to say , whether a solution _ exists _ and is furthermore _ unique _ can not be `` assumed '' in any mathematical problem . this should be _ demonstrated _ , an imperious requirement particularly in the case of such an approach , whose shortcomings are so amply documented . @xcite in the last part of their argumentation , dg claim that their deduction that @xmath35 is real remains true if they impose current conservation . as discussed above , this unjustified assertion is contradicted by our eqs . ( [ eq - an ] ) and ( [ eq - an - ast ] ) . as a matter of fact , it is just the opposite sign of the last term in the square parentheses of eqs . ( [ eq - an ] ) and ( [ eq - an - ast ] ) that is related to the broken time - reversal symmetry expressed by the transformation @xmath36 , @xmath37 ( @xmath38 in dg s notation ) noted above . dg also mention this transformation but fail ( or avoid ) to correctly include the current conservation as a constraint in the energy minimzation that has to be done . to summarize , constraining populations is very possible and does not in the least exclude a broken time - reversal symmetry . were our choice to constrain electronic populations inadequate , which are used in the most well - established approaches for the uncorrelated and correlated transport [ like those based on boltzmann s equations , nonequilibrium green s functions ( negf ) , master equations , etc ] , all these approaches would fail , but this is not the case . they all employ the fermi distributions ( fds ) @xmath39 to express open boundary conditions for incoming electrons . before ending this section , we still make the following remark . most important for the invalidity of the variational dg approach demonstrated in i is that it leads ( even with the constraints most justifiable physically , namely @xmath40 ) to an _ unphysical _ result . we paid and pay little attention to the fact that the deduced current vanishes . by chance , the current could have been nonvanishing , as could fortuitously ( but must not ) happen when using the dg variational ansatz with wigner constraints ( e. g. , fig . 3 of ref . ) . there are certainly _ many _ possibilities to obtain nonvanishing currents by constraining ad hoc complex hermitian operators , since all these generally yield @xmath41 : see eqs . ( 18)(20 ) of i for @xmath42 ; one needs not be too `` careful '' for this , contrary to the impression which dg attempt to convey . essential to deduce results which are physically relevant is to be careful in selecting from the very broad class @xmath43 mentioned in the comment those hermitian operators which are associated to observables able to express the physical reality at the boundaries . and , according to the present community s wisdom , a choice better justifiable physically than the electronic momentum distributions ( fds ) is not known ; dg themselves have to admit that they represent `` an interesting alternative to the '' wf ( cf . last paragraph of the comment ) . attempting to give a physical basis to their criticism , dg switch to the single - particle description and claim that we can not constrain only incoming electrons , as appropriate for transport , because our single - particle wave functions ( orbitals ) are real , and outgoing electrons become concomitantly constrained . in i ( as well as in sec . [ sec - time - reversal ] ) we need not orbitals at all , nor constrain orbitals , so it is not true what dg write that we `` have decided to constrain the occupation of _ real _ states of the left electrodes of approximate form @xmath44 '' . these ( approximate ) wave functions pertain to an _ isolated _ electrode ( i. e. , uncoupled to a device ) with open boundary conditions ( obcs ) . here , _ open _ is meant not in the sense of transport theories ( of a system exchanging particles and energy with environment ) , but in that used , e. g. , in numerical exact diagonalization or dmrg ( density matrix renormalization group ) studies , of a system with open ends whose wave function vanishes at the boundaries [ in dg s notation , @xmath45 . our boundary conditions ( bcs ) for the isolated electrodes [ cf . ( 6 ) of i ] are _ general _ ( see sec . [ sec - complex ] ) . nevertheless , to understand why the dg claim is incorrect let us suppose in this section that we would have imposed obcs . the dg s analysis related to the functions @xmath46 explains nothing but the trivial fact that a current can not flow in an isolated electrode . how could a current flow in an electrode uncoupled to device ? what we constrained is clearly visible in eq . ( 23 ) of i : that the populations of the incoming electrons in two _ many - body _ states ( @xmath47 for @xmath8 and @xmath48 for @xmath49 ) pertaining to the _ total _ system ( electrodes _ coupled _ to the device ) are equal . we did not constrain noninteracting electrons confined within an isolated electrode described by the wave function denoted by dg by @xmath50 . this would be meaningless physically . that our constraints refer to _ incoming _ electrons was already indicated in the step ( ii ) of sec . ii of i. our erratum,@xcite which dg attempt to misinterpret to gain more credibility , merely emphasizes this fact ; it neither corrects an error nor retracts a statement . in fact , as already noted above and discussed in detail below , our results apply to more general cases than that analyzed in the comment , but let us still remain in the latter framework . dg calculations would also be possible by using single - electron wave functions @xmath51 to build all slater determinants needed to exhaus the total electrode - device hilbert space and express the general many - body state @xmath35 required for the prescribed minimization . ( notice that our demonstration of i is exact and does not rely upon any approximate ci expansion , contrary to ref . ) . the physical content of the orbitals @xmath52 with @xmath53 [ these are the dg s @xmath54 is that incoming ( @xmath55 ) and outgoing ( @xmath56 ) electrons move symmetrically in the isolated left electrode . if we formulated in i the dg variational ansatz in terms of orbitals , the transport approach would have _ allowed _ a broken @xmath55(coming)-@xmath56(going)symmetry when the electrodes are coupled to a device [ @xmath57 for a nonvanishing bias @xmath49 . a current @xmath32 can flow only through a device _ coupled _ to electrodes . these @xmath51 s would have been used throughout to express all averages needed in i [ including the rhs of eq . ( 23 ) , that is , at @xmath8 for electrodes _ coupled _ to a device ] and _ determined _ self - consistently from the dg variational ansatz , just like our @xmath10 s of i or above in sec . [ sec - time - reversal ] . whether this symmetry , _ allowed _ to be broken , _ is _ indeed broken and a current indeed flows ( @xmath3 ) should be determined by performing calculations within the dg variational ansatz , and this is exactly what we did within the many - body formalism of i. _ if _ the dg variational ansatz were valid , the minimization would yield wave functions @xmath58 different from @xmath59 and @xmath41 , in the same way in which it would have led to im@xmath60 in i and in sec . [ sec - time - reversal ] above . of course , this cumbersome approach based on the first quantization is equivalent to the second quantization formalism employed in i , and what is important is the unphysical _ outcome _ ( @xmath2 ) . it clearly demonstrates that the dg transport ansatz is invalid . dg chose the obcs as a stumbling block for the demonstration of i , but , because that case turned out to be more subtle , ended becoming trapped themselves in this pitfall . by contrast , the case of pbcs indicated by them as a way - out solution is really much easier to analyze and allows to easily understand the correctness of the result @xmath2 of i. as already noted , our derivation of the unphysical result @xmath2 for the linear response limit in i is very general . to give a flavor on how general are the situations where the dg variational ansatz fails , we referred to a rather broad class of uncorrelated and correlated models , which are mostly used in nanotransport , although our demonstration applies well beyond that class . in i we gave a formal general analytical demonstration and needed not bother to do expensive numerical calculations for certain device models nor become involved by giving a particular type of the bcs for the isolated electrodes or explicitly indicating which labels @xmath61 refer to single - particle states of incoming electrons and which to outgoing electrons , although they can be obviously specified ( see below in this section and sec . [ sec - el - h ] ) . by inspecting the second and third lines of eq . ( 6 ) of i , one can immediately realize that our isolated electrodes can be described by _ general _ boundary conditions . just for not to bother the reader with an unnecessary lengthy discussion of the bcs , we avoided , e. g. , to specify the site indices at the ends of the electrodes opposite to the device , and wrote @xmath62 and @xmath63 in the second and third lines of eq . ( 6 ) of i. the message of our refs . is unambiguous and leaves no hope : the dg approach is completely incorrect and can not be rescued . having demonstrated this , we doubted and doubt on the usefulness or necessity of a too detailed analysis of an approach so incorrect . the results of i apply for the aforementioned obcs ( sec . [ sec - real - orbitals ] ) , but also for boundaries that are , e. g. , periodic ( pbcs ) , anti - periodic ( moebius ) or general twisted ( see , e. g. , ref . and citations therein ) . to this , one should keep , e. g. , @xmath64 sites in the right electrode and impose , e. g. , that the annihilation operators satisfy @xmath65 , @xmath66 , or @xmath67 ( @xmath68 is a real constant phase ) , respectively . the left and right electrodes can even be described by different types of boundaries . in their comment , dg concede that our idea of constraining populations `` is an interesting alternative to the wigner function '' , but claim that it is essential to choose complex single - particle wave functions associated to pbcs . accepting also this dg s challenge now , let us explicitly work out in detail just this case where the pbcs are applied to the isolated electrodes and where the orbitals are complex , which dg suggest as a way out to rescue their variational ansatz . from now on , we shall always assume pbcs unless otherwise specified . with pbcs , and assuming homogeneous hopping integrals @xmath69 in electrodes for simplicity , one can write down explicit analytical formulas instead of the general unspecified ones of i , because the transformation matrix @xmath70 which diagonalizes @xmath71 entering eq . ( 22 ) of i is nothing but the well - known fourier transformation @xmath72 throughout , we use @xmath73 ; @xmath74 ; @xmath75 ; @xmath76 , and assume even @xmath77 for specificity . with eq . ( [ eq - fourier ] ) , the hamiltonians @xmath78 , @xmath79 , and @xmath80 of eq . ( 6 ) of i become ( @xmath81 , and @xmath82 ) @xmath83 \alpha_{k}^{\dagger } \alpha_{k } , \nonumber \\ h_{r } & = & \sum_{p } \left[\mu_r - 2 \mathcal{t}_r \cos(2 \pi p / m_r ) \right ] \alpha_{p}^{\dagger } \alpha_{p } , \nonumber \\ h_{d , e } & = & - \tau_l m_l^{-1/2}\sum_{k } \left(\alpha_{k}^{\dagger } a_{q_l } + a_{q_l}^{\dagger } \alpha_{k } \right ) \nonumber \\ & & - \tau_r m_r^{-1/2}\sum_{p } \left(\alpha_{p}^{\dagger } a_{q_r } + a_{q_r}^{\dagger } \alpha_{p } \right ) . \label{eq - h}\end{aligned}\ ] ] further operators affected by the transformation ( [ eq - fourier ] ) entering the relevant equations of i are @xmath84 representing the hamiltonian of the external bias @xmath12 and the current operators @xmath85 at the contacts , respectively . the sites within the device @xmath86 are not affected by eq . ( [ eq - fourier ] ) , and the corresponding operators ( device s hamiltonian @xmath87 and the current within the device @xmath88 ) can be found in i. by inspecting eq . ( [ eq - h ] ) , one can immediately see that , in spite of the fact that the single - particle wave functions @xmath89 and @xmath90 of eq . ( [ eq - fourier ] ) are complex and degenerate ( @xmath91 and @xmath92 correspond to the same energy ) , _ all _ the parameters of the hamiltonian @xmath93 are real . consequently , _ all _ its many - body eigenstates @xmath94 are ( can be chosen ) real . all the parameters entering @xmath12 are real , so the matrix elements @xmath95 [ eq . ( 13 ) of ref . ] are again real . as in the general case of i , the matrix elements of the current operator are purely imaginary ( i. e , real @xmath96 ) . all the matrix elements of the particle number operators @xmath97 ( @xmath98 ) remain real @xmath99 ( @xmath100 ) . ( 24 ) and ( 25 ) of i remain unaltered . whether imposing current conservation ( as we did ) or not ( as dg incorrectly claim that one could do@xcite , as if their method were so good to automatically include current conservation @xcite ) , the completely unphysical result ( @xmath2 ) follows as the ineluctable conclusion of applying the dg variational ansatz . for these pbcs , the labels of the single - particle states of incoming _ electrons _ can be explicitly given ( see also sec . [ sec - el - h ] ) : @xmath101 with @xmath102 and @xmath103 with @xmath104 . one can now _ explicitly _ see that only these incoming electrons can and are to be constrained in eqs . ( 12 ) and ( 23 ) of i or in the present eq . ( [ eq - q - k ] ) . they do differ from the outgoing electrons , whose labels are @xmath105 and @xmath106 , and one can convince oneself _ explicitly _ that outgoing electrons ( can ) remain _ un_constrained . so , these constraints correspond to fig . 7c of ref . and not to fig . 7a and 7b , contrary to what the comment claims . by simple algebraic manipulations of eqs . ( 23)(25 ) and ( 15 ) of i one can easily _ deduce _ that the distributions of the outgoing and incoming electrons are equal , @xmath107 and @xmath108 . ( of course , this is not at all surprising in view of the unphysical result @xmath2 . ) one can now _ explicitly _ see that the equality of these distributions is the _ outcome _ of the calculations of the _ defective _ dg variational ansatz , and is by no means ( not even implicitly ) assumed from the very beginning through an inappropriate choice . the constraints used by us within the calculations based on the dg variational ansatz _ allowed _ a broken symmetry between incoming and outgoing electrons . whether this symmetry is broken or not remained an _ open _ result , which emerged from the dg transport calculations . the _ result _ is that this symmetry is not broken , @xmath2 , demonstrating the _ incorrectness _ of the dg variational ansatz , and this also holds true for the pbcs , contrary to what the comment argues . from the above analysis of the pbcs it is also clear that the electrodes size @xmath77 can be arbitrary large ( which is impossible within the ab initio dg calculations @xcite ) . therefore one can also understand that the unphysical prediction @xmath2 is not limited to pbcs but holds for _ any _ other bcs as well , since otherwise , e. g. , the entire philosophy of solid - state physics to apply pbcs would break down . dg mention ( as we also did ) that , by constraining the wf , a nonzero current is possible . in refs . , we discussed that , _ luckily _ , mathematically this may be possible . why did dg constrain the wf ? only because in ref . they claimed that the fds can not be used for correlated transport . this claim does not at least apply for uncorrelated systems . at least there , fd - constraints are possible , and our critique of the dg variational ansatz from i obviously applies . in fact , in i we explained that open boundary conditions can also be formulated by means of fds even for correlated electronic devices , because they should be imposed in electrodes , wherein electrons are uncorrelated . it is the fd which has a precise physical meaning , and not the wf , which has a physical content _ only _ when it is a good approximation for the fd . according to the comment s original philosophy , a current flow is possible only to the extent to which the wf does differ from the fd . the inappropriateness of the wigner constraints could not be immediately recognized only because , luckily , the matrix elements of the fano operator are generally complex . they are generally complex no matter whether the single - electron functions are real or complex ( see refs . ) ; this is not the result of any `` careful '' choice of certain complex functions , as incorrectly claimed by dg . let us show that this is also the case when the comment s continuous space description is used instead of the discrete one of refs . . the fano operator reads @xmath109 where @xmath110 and @xmath111 the electron field creation and destruction operators . its general matrix element for two arbitrary @xmath112-body states @xmath48 and @xmath113 corresponding to the multielectronic wave functions @xmath114 ( @xmath115 is the vacuum ) and @xmath116 can be easily expressed as ( @xmath117 ) @xmath118 this matrix element is generally complex irrespective of whether the wave functions @xmath119 and @xmath120 entering eq . ( [ eq - fano - matrix ] ) are real or complex . ( [ eq - fano - matrix ] ) is general and holds whatever the employed single - particle wave functions ( which need _ not _ be specified there ) . the complex exponential entering eq . ( [ eq - fano - matrix ] ) belongs to the definition of the fano operator , eq . ( [ eq - fano - general ] ) , and has nothing to do with the employed orbitals . whatever the latter , it can be artificially split as @xmath121 \times \exp[-i p ( x + r/2)]$ ] . this trivial splitting and the notation @xmath122 for these factors is obviously done by dg in their eq . ( 7 ) merely for conveying the false impression that complex exponentials would represent a key point , which they would have `` carefully '' exploited in their work.@xcite let us express the population constraints for the incoming electrons ( @xmath123 ) of the right electrode ( @xmath124 ) , eq . ( 23 ) of i , making use of the fano operator @xcite and eqs . ( [ eq - fourier ] ) @xmath125 instead of constraining the above sums ( integrals ) over @xmath126 of the fano operators ( populations ) , dg s discretionary constraint of a single term in either electrode ( namely , @xmath127)@xcite involves a quantity which does not possess a physical meaning . emerging from such an _ ad hoc mathematical _ constraint , it is not at all surprising that the currents predicted by the original dg approach,@xcite whether they vanish or not , are completely unphysical , as demonstrated in refs . . to conclude , _ mathematically _ the wf - constraints used in conjunction with the dg variational ansatz ( can ) yield a nonvanishing @xmath32 ( without _ any _ physical relevance , cf refs . ) even when using real wave functions , in spite of the claimed `` warning bells '' that `` real wavefunctions carry no current''.@xcite in sec . [ sec - complex ] , to specify the labels of incoming and outgoing electrons , we have made the intuitive ( or , better , naive ) _ assumption _ that the single - particle states with positive ( negative ) wave numbers @xmath130 and @xmath129 correspond to right- ( left- ) motion . however , they are related via eqs . ( [ eq - fourier ] ) to _ quasi_-momenta and not necessarily to physical momenta . especially for later purposes , it is important to _ demonstrate _ that their sign is indeed related to the direction of the motion in the _ real _ world . let us consider the isolated left electrode . its ground state ( not to be confounded with that of the coupled electrode - device system , @xmath47 , cf . [ sec - real - orbitals ] ) is the fermi sea @xmath131 , where the fermi wave vector @xmath132 is determined by the number of electrons . using eq . ( [ eq - fourier ] ) one can straightfordwardly demonstrate that at any position @xmath133 within the ( left ) electrode , the average of the electron number current @xmath134 vanishes , @xmath135 . [ notice the opposite signs of @xmath136 and the _ electric _ current @xmath137 of sec . [ sec - time - reversal ] for electrons . ] let us also consider the states ( @xmath138 , @xmath139 ) @xmath140 they represent states with one extra electron ( @xmath141 ) and hole ( @xmath142 ) in the fermi sea , respectively . straightforward calculations using eqs . ( [ eq-1p-1h ] ) and ( [ eq - fourier ] ) yield @xmath143 the sign of the @xmath136-average does _ express _ the real direction of electron quantum - mechanical motion . therefore , eqs . ( [ eq - j-1p ] ) and ( [ eq - j-1h ] ) demonstrate that the sign of the wave vectors belonging to the brillouin zones of sec . [ sec - complex ] ( symmetric around zero ) specifies the direction of electron motion , and that electrons and holes with a given wave vector move in opposite directions . the latter result can also be seen by performing the general particle - hole transformation , @xmath144 ( charge conjugation ) and @xmath145 @xmath146 @xmath147 ( @xmath148 ) . using eq . ( [ eq - fano - general ] ) , one easily gets @xmath149 in view of the aforementioned , one can conclude that incoming and outgoing _ electrons _ correspond to the wave vectors @xmath150 & & \mbox{\hspace*{37ex } ( for electrons ) } \nonumber \\[-2ex ] \label{eq - el - out } & & -m_l/2 < k_{out } < 0 ; 0 < p_{out } < m_r/2 , \end{aligned}\ ] ] while for incoming and outgoing _ holes _ @xmath151 & & \mbox{\hspace*{37ex } ( for holes ) } \nonumber \\[-2ex ] \label{eq - h - out } & & 0 < k_{out } < m_l/2 ; -m_r/2 < p_{out } < 0 .\end{aligned}\ ] ] the fact that the above electron and hole descriptions are equivalent is trivial in general , but not in the context of transport approaches , wherein incoming _ charge carriers _ are to be constrained.@xcite if the charge carriers are _ electrons _ ( @xmath128-type conduction ) , the constraints should be imposed to incoming electrons , eq . ( [ eq - el - in ] ) . in this case , the intuitive assumption of sec . [ sec - complex ] is justified . however , if the charge carriers are _ holes _ ( @xmath129-type conduction ) , one should constrain the incoming holes , eq . ( [ eq - h - in ] ) ; that is , the labels @xmath61 in the above eq . ( [ eq - q - k ] ) and in eqs . ( 12 ) and ( 23 ) of i are those given by eq . ( [ eq - h - in ] ) and not by eq . ( [ eq - el - in ] ) . the analysis of this section and of secs . [ sec - real - orbitals ] and [ sec - complex ] makes it now clear why we preferred to consider the general case in i and not to enter in unnecessary involved details : they are absolutely not necessary to understand the unphysical prediction @xmath2 of the dg variational approach , and hence its lamentable failure . but because the incorrect dg claims in the comment brought us to enter such details , we can show another shortcoming of the original dg approach related to them , which we did not present so far . in their work,@xcite dg did not examine at all whether the molecule they considered , bdt ( benzenedithiolate ) , exhibits an @xmath128- or a @xmath129-type conduction . uncritically , they merely constrained @xmath152 and @xmath153 . even if their variational ansatz were correct , and even if these wfs were true distribution functions , these constraints would be appropriate only if the charge carriers were electrons [ conduction mediated by lumo ( lowest unoccupied molecular orbital ) ] . in reality , in bdt the majority charge carriers are holes ( @xmath129-type conduction ) , as clearly demonstrated by the recent , accurate experiment of ref . . by inspecting now eqs . ( [ eq - fano - h ] ) ( noting the reversed sign of @xmath0 in the lhs and rhs ) , ( [ eq - el - in ] ) , and ( [ eq - h - out ] ) , one is amazed to see that what dg constrained in ref . are in fact the _ outgoing _ majority carriers , and _ not _ the incoming ones . it is certainly too simplistic to describe the conduction through bdt merely as a process mediated by homo ( highest occupied molecular orbital ; @xmath129-type conduction ) instead of accounting for several / numerous ionization and electroaffinity levels , but the fact that the constraints of majority carriers ( holes ) are unphysical in ref . is a clear demonstration that uncritically using wigner boundaries is completely unjustified . to conclude , even if all the other dg ingredients were correct ( what is obviously not the case @xcite ) , this very reason irrefutably demonstrates that the results of refs . have absolutely no physical meaning . what would be the appropriate constraints in the case of ambipolar conduction , where both electrons and holes contribute to the current , is an issue,@xcite which we do not discuss here . in our first work @xcite that challenged the dg approach with wigner constraints,@xcite we considered uncorrelated and correlated quantum dots modeled by a single level whose energy offset from the electrodes fermi level @xmath154 is @xmath155 . the results of the dg calculations presented there ( e. g. , in figs . 25 and 7 ) are for @xmath156 , that is , the dot s level plays the role of a lumo ( @xmath128-type conduction ) . the charge carriers are electrons , and our constraints [ corresponding to the above eq . ( [ eq - el - in ] ) ] refer to incoming electrons . both the uncorrelated and the correlated models of ref . are described by hamiltonians @xmath157 possessing a particle - hole ( or charge conjugation ) symmetry ( see , e. g. , ref . and citations therein ) around @xmath158 : @xmath159 . that is , the zero - bias conductance @xmath160 ( as well as other relevant properties not considered in ref . , e. g. , the whole current - voltage characteristics ) should be identical irrespective whether the level is located above ( @xmath161 ) or below ( @xmath162 ) the electrodes fermi level @xmath163 . noteworthy , the charge carriers are electrons for positive @xmath155 and holes for negative @xmath155 . as a test for numerical calculations , we checked that dg calculations for the lumo case ( @xmath164 ) constraining the incoming electrons [ eq . ( [ eq - el - in ] ) ] and for the homo case ( @xmath165 ) constraining the incoming holes [ eq . ( [ eq - h - in ] ) ] yield the same , albeit completely unphysical linear conductance . [ the electric current operator has the same sign both in the electron and the hole representation , cf . ( [ eq - j - h ] ) . ] as clearly demonstrated,@xcite the dg - conductance computed in this way is completely unphysical , but it is still positive , @xmath166 both for positive and negative @xmath155 . that is , this ( modified ) dg approach can still `` predict '' that electrons flow from the lower potential to the higher potential , and holes flow from the higher potential to the lower potential . if we drew the curves of figs . 3 , 5 , and 7 of ref . also for @xmath165,@xcite by blindly computing the dg conductance using _ exactly _ the dg prescribed constraints,@xcite [ i. e. , eq . ( [ eq - el - in ] ) ] we could have shown a funny `` prediction '' of the dg approach @xcite , namely , that the linear conductance can be negative , @xmath167 ! that is , holes should have to flow from the lower potential to the higher potential . this results from the fact that the blind constraints of @xmath168 and @xmath169 erroneously constrain in fact the outgoing carriers ; this situation corresponds to fig . 7d , and not to fig . 7c of ref . . indeed , these dg wigner constraints break the time - reversal symmetry and yield a nonvanishing current , but what is the physical relevance ? as a matter of fact , it is just such an unphysical imbalance , which is shown in fig . 1 ( bottom ) of ref . ( the counterpart of fig . 7d of ref . and not of fig.7c , as incorrectly claimed in the comment ) , that breaks the time - reversal symmetry in ref . . in ref , we did not show this conductance @xmath170 because the demonstration of the severe failure of the dg approach with wigner constraints was sufficiently convincing even without mentioning this `` prediction '' , and we preferred to couch the discussion in terms as sober as possible . however , we have noted it above , since , in spite of the clear evidence of refs . , dg still continue to uncritically refer to their work @xcite in the comment . in the comment , dg claim that we `` doubt the validity of using the wigner function constraints to apply open system boundary conditions '' . as we repeatedly emphasized,@xcite we did not challenge in any of our works published so far @xcite the imposition of wigner constraints _ in general _ ; we irrefutably _ demonstrated _ that applying wigner constraints in the _ specific context of the dg variational approach _ yields completely incorrect results . another inaccurate assertion of dg is that `` bldea and kppel accept that our _ [ i . e. , dg s ] _ wigner constraints have non - zero linear response current . '' in reality , we demonstrated ( i ) that even if , _ by chance _ , this is possible,@xcite the dg current is completely unphysical , and ( ii ) that the dg conductance vanishes _ just _ in on - resonance cases , where the exact conductance attains its maximum ( unitary limit).@xcite to refute the false impression , which dg attempt to convey by using the superscript @xmath4 , we emphasize that they did never impose pbcs for the @xmath171-clusters , which mimic their electrodes of refs . . for this , it suffices to inspect eqs . ( 1 ) and ( 2 ) of ref . . still , let us emphasize for completeness that , even if they imposed such wigner pbcs , the results of the dg approach would have been incorrect : in sec . v of ref . , we demonstrated that the dg conductance computed with wigner constraints vanishes ( @xmath172 ) also with pbcs for electrodes , and this occurs just in a typical , well - known physical situation corresponding to the unitary limit , wherein the true conductance reaches the maximum value ( @xmath173 ) . our ref . is the only publication wherein the dg approach , as originally proposed ( i. e. , with wf - constraints ) , was worked out using plane wave orbitals @xmath174 in electrodes with pbcs . dg used nowhere the plane waves @xmath174 mentioned in the comment or other orbitals pertaining to electrodes with pbcs , which they denote by @xmath175 in the comment . or , more precisely , they should have denoted them so [ i. e. , @xmath176 , because in fact their expressions , like @xmath177 in the last but one paragraph of the comment , do not comply with textbooks quantum mechanics . concerning the operators denoted by @xmath178 , they are useless : these operators are not defined at all . most significantly , the comment contains equations and assertions , which defy more than eight decades of using the second quantization formalism . e. g. , dg incorrectly state that the action of the operator @xmath179 on a one - electron wave function @xmath180 is `` straight - forward '' . this operator can not act at all on @xmath180 , not even on a many - electron wave function @xmath181 in the coordinate space , as expressed in their incorrect eqs . ( 1 ) and ( 2 ) , respectively . it acts on many - body states @xmath182 belonging to the abstract fock vector space . further , it is incorrect what they write that the operators @xmath183 and @xmath184 create and destroy the _ eigen_states of @xmath78 . likewise , these operators act on _ any _ many - body fock states @xmath185 , wherein they create and destroy one electron in the single - particle state @xmath130 . we thank dg for using the superscript @xmath186 in their notation @xmath183 and @xmath184 , but it is superfluous . the creation and annihilation operators @xmath187 and @xmath188 used in i [ the same as those entering the present eqs . ( [ eq - fourier ] ) ] are already unequivocally defined in textbooks , and it is in this sense that we used them throughout . in view of the aforementioned , the need to translate in the comment obvious results expressed in the second quantization in i into the first quantization language [ e. g. , the attempt to `` demonstrate '' that the matrix element @xmath189 of eq . ( 4 ) in the comment is real ] is understandable . but then it is not surprising that understanding what are the actual constraints of i or the generality of the demonstration of i is not obvious . remarkably , although dg reproduce our expression of @xmath78 in the comment , they completely overlook the content of this expression : namely , the fact the bcs are general , and so are the pertaining orbitals , which are not necessarily real . not coincidental is also the fact that dg indicate as a way - out solution just a particular case of i , wherein the incorrectness of the dg variational ansatz can be immediately understood ( cf . [ sec - complex ] ) . above , we preferred to respond to and rebut in detail all the issued raised by the comment . in fact , this represents the main objective of our reply . still , we note that this analysis of all the concrete aspects is actually not necessary . in sec . vii of our recent work @xcite we pointed out more serious reasons why the dg variational approach fails . this criticism also applies to the case discussed above ; it comprises fundamental aspects not confined to a certain type of bcs , let they be in terms of wfs , fds or others . our criticism was presented in ref . in sufficient detail and will not be even summarized in this reply . a further fundamental reason why the dg approach is incorrect will be presented in sec . [ sec - isolated ] . the incorrect claims of dg made us aware of a limitation of the usefulness of the wf for transport , which we could not find in the literature . the wf ( @xmath190 ) is employed in many physical studies , including transport s , in spite of its physical limitations . the limitation known from textbooks @xcite traces back to the heisenberg s uncertainty principle . the wf can be negative and should be not interpreted as a probability distribution , but rather as `` one step in the calculation never the last step , since '' is not measurable but `` is used to calculate other quantities that can be measured the particle density and current '' and `` no problems are encountered as long as one avoids interpreting @xmath190 as a probability density '' ( quotations from ch . 3.8 , p. 203 of ref . ) . as noted above , in transport it is helpful to distinguish between incoming and outgoing electrons . to this aim , it is necessary to use a physical property enabling to indubitably assess that electrons are , say , left- or right - moving . the averages of the particle current operator @xmath191 used above or the _ physical _ momentum do represent such properties . to see whether the momentum `` variable '' @xmath0 of the wf @xmath192 justifies to speak of left- or right - moving electrons depending on the sign of @xmath0 , let us consider @xmath112 noninteracting electrons confined within a one - dimensional square well of width @xmath193 and infinite height . ( this could be the isolated electrode considered by dg . ) electrons occupy energy levels @xmath194 , whose single - electron wave functions @xmath195 $ ] ( @xmath196 , @xmath197 ) are just those that dg @xcite incorrectly claim we would have used in ref . . in the ground state , the lowest @xmath112 levels are occupied up to the fermi `` momentum '' @xmath198 . computing the wigner function of this system is straightforward : @xmath199 ( see , e. g. , ref . , ch . 3.7 , pp . 202 - 203 ) . ( color online ) wigner function for two sizes @xmath193 computed at two points @xmath126 indicated in the legend . gold s fermi wave vector @xmath200nm@xmath201 is used.,scaledwidth=45.0% ] one might think that one could use the wf as if it were a distribution function in cases where its shape resembles a fermi distribution . let us inspect the curves for @xmath202 computed as indicated above and presented in fig . [ fig : wf ] . in fact , at smaller sizes ( close to the linear size of the dg s au@xmath203-clusters @xcite ) the wf does not bear much resemblance to a fermi function ( the lower curves of fig . [ fig : wf ] ) , at larger sizes ( much larger than those one could hope to tackle within ab initio calculations to correlated molecules , for which the dg approach @xcite was conceived ) the curves ( the upper part of fig . [ fig : wf ] ) become more similar to a step function , and one may think that this is encouraging . in reality , the contrary is true : as visible in fig . [ fig : wf ] , _ mathematically _ one can calculate the wf for positive and negative `` momentum '' variables @xmath0 _ separately_. however , this mathematical separation does not reflect a physical reality : for any single - particle eigenstate @xmath61 the electron momentum vanishes , @xmath204 ; left- and right - traveling waves are entangled with equal weight , and one can not speak of single - particle eigenstates representing left- or right - moving electrons only because wigner functions with positive or negative @xmath0-arguments can be computed . this represents a further limitation of the usefulness of the wf , not related to the heisenberg s principle , which is particularly relevant for transport . the current has a direction , and if one wants to unambiguously specify this direction , the wf @xmath202 is inappropriate ; a wf with negative ( positive ) `` momentum '' , @xmath205 ( @xmath206 ) , does not imply that left- and right - moving particles exist in the real physical world . so , using @xmath168 and @xmath169 as if they were true momentum distributions of incoming electrons , as dg did,@xcite is not justified in quantum mechanics . the above example demonstrates that , indeed , the textbook s warning mentioned in the beginning of this section is pertinent . in the course of our extensive critical investigations @xcite of the dg approach @xcite we became aware of a series of difficulties , which not only the dg s but also other approaches to nanotransport are faced with , which we want to bring to the reader s attention . to understand the problem , let us briefly consider the uncorrelated dot model of ref . linked to semi - infinite electrodes ( @xmath207 ) . exact transport calculations at arbitrary bias @xmath208 can be easily carried out within a multitude of approaches.@xcite besides the current @xmath32 and dot occupancy @xmath209,@xcite the occupancies of the sites in electrodes can also be computed.@xcite in fig . [ fig : ncluster ] , we present steady - state results for the electron number on the dot @xmath210 and that for nanoclusters @xmath211 centered on the dot and including several electrodes sites @xmath212 . as visible there , the changes in both @xmath208 ( source - drain voltage ) and @xmath155 ( gate potential ) yield variations in the dot occupancy @xmath210 . ( notice that there is no exchange of electrons with the gate . ) the total number @xmath211 of electrons in the nanoclusters also varies , it closely follows the change of @xmath210 ; the variations in the small difference @xmath213 could be hardly seen within the drawing accuracy of fig . [ fig : ncluster ] , and therefore are not shown there . the fact that @xmath213 is practically constant implies that the sites in electrodes remain practically unaffected by changes in @xmath208 and @xmath155 ; even the electrodes sites in the very close dot s vicinity are very little affected . the dependence @xmath214 is most important in the context of a transport approach . it demonstrates that , by varying @xmath208 and/or @xmath155 , a finite nanocluster exchanges electrons with the infinite electrodes linked to it , which act as reservoirs that can supply / withdraw electrons . ( color online ) dot charge @xmath209 and cluster excess charge @xmath215 for two gate potentials @xmath216 ( @xmath217 ) and @xmath218 ( @xmath219 ) . for both gate potentials , the employed cluster sizes @xmath212 given in parentheses increase upwards for , say , @xmath220 . notice that for larger sizes the excess charge rapidly saturates and closely follows the dot charge , and this gives a flavor on how fast the convergence to charge neutrality within more distant parts of electrodes is ( see the main text ) . energy unit is @xmath221 and @xmath222.,scaledwidth=45.0% ] in the light of the aforementioned , it becomes clear that it is _ strictly _ impossible to correctly describe the transport within a dg - like approach , which considers a small isolated cluster . within an approach like that initiated by dg @xcite and further scrutinized by us @xcite one attempts to describe the transport using such a cluster characterized by a many - electron wave function @xmath223 to be determined by a constrained minimization of cluster s energy for @xmath49 . this wave function merely describes a nanocluster with a given number of electrons @xmath112 , which can not be varied by changing source - drain ( @xmath208 ) or gate ( @xmath155 ) voltages : * _ whatever _ the property chosen to be constrained ( wf , electron momentum distribution or else ) , * _ whatever _ the boundary conditions ( obc , pbcs , moebius , general twisted or else ) which might be chosen for the finite `` electrodes '' included in the cluster ( the au@xmath203-clusters of ref . ) , * let the many - electron wave function @xmath223 be determined within _ exact _ full ci expansions ( i. e , using the whole hilbert space ) as done by us @xcite and not only approximately , by using a reduced number of configurations selected by monte - carlo sampling , as dg did.@xcite even the most `` carefully '' chosen constraints can at most , e. g. , acceptably account for the change the number of electrons in the device ( dot ) . but this change will inevitably modify the charge of the neighboring sites in electrodes . it is easy to imagine how profound will be the impact on the transport , e. g , in correlated or switchable devices . it is hard to conceive that the transport could be reasonably described within this framework , even if the cluster would be very large ( what is obviously hard within the presently available ab initio quantum - chemical calculations for correlated systems ) : because it is by no way obvious / necessary that the local charge density be incorrectly described only at the remote ends of the finite `` electrodes '' , which would allow to suppose / hope that the electric field within the device will be little affected . a further difficulty is , of course , the fact that the number of electrons @xmath224 of the coupled nanocluster is generally noninteger ( cf . fig . [ fig : ncluster ] ) , but we do not discuss this issue here.@xcite one may further ask whether a `` careful '' choice of a certain charge state of the dot ( molecule ) instead of the neutral species could help . no , it does not . for the above model , one can also compute the exact time - dependent dot population @xmath225 by suddenly coupling at @xmath226 a dot with population @xmath227 to infinite electrodes.@xcite the result for @xmath49 within the wide band limit ( @xmath228 , where @xmath229 , @xmath230 , and @xmath231 ) is @xcite @xmath232 \nonumber \\ & & - \frac{2\gamma}{\pi } e^{-\gamma t } \left ( \int_{-\infty}^{-\varepsilon_g - ev/2 } + \int_{-\infty}^{-\varepsilon_g + ev/2 } \right ) d\,\varepsilon \ \frac{\cos(\varepsilon t)}{\varepsilon^2 + \gamma^2 } . \nonumber\end{aligned}\ ] ] ( color online ) curves for the time - dependent dot charge @xmath233 showing that the asymptotic values do not keep track of the initial occupancy @xmath227 . the values of the source - drain bias @xmath208 , energy offset @xmath155 , and hybridisation @xmath234 are given in the legend.,scaledwidth=45.0% ] as seen in fig . [ fig : nd ] , which visualizes the result expressed by eq . ( [ eq - nd ] ) , the asymptotic value of the dot population @xmath235 , which corresponds to the steady state , does not `` remember '' the initial value @xmath236 ; it is determined by the applied voltages . the foregoing analysis makes it also clear that not only a variational ansatz like the dg s is incorrect . the above list ( a ) ( c ) can be enlarged by stating that , _ whatever _ the ansatz ( let it be a variational ansatz better than the dg s or based on e. g. , liouville @xcite or schrdinger equations ) used to compute the wave function for a fixed @xmath112 , it will fail . the aforementioned flaw is serious and can be unambiguously traced back to the use of a finite ( small ) isolated cluster . electric transport can only occur in an open system , and allowing electron exchange with environment ( infinite electrodes ) is indispensable . most commonly , this is done via the embedding self - energies within the keldysh - negf - approach,@xcite which enable the nanocluster to change the number of electrons it contains . most important for the present reply , we have demonstrated above that the dg s claims of the comment are incorrect and our critique of i is untouched . the unphysical current ( @xmath2 ) obtained in i is the _ result _ of the dg defective ansatz and by no means emerges from pretended unsuitable constraints of real orbitals , as dg incorrectly claim . using complex orbitals does not in the least change the unphysical prediction of the dg variational ansatz ; furthermore , it confirms more directly its incorrectness . in fact , the complex orbitals given by dg represents nothing but a _ particular _ case of the general case considered in i , for which our demonstration holds . for completeness , we have worked out this case in detail to explicitly specify the labels of the incoming charge carriers ( not necessarily electrons ) , to demonstrate that they can be constrained without concomitantly constraining the outgoing carriers , and to show that the distribution of the outgoing carrieres represents the _ outcome _ of transport calculations . of course , fully consistent with the unphysical prediction @xmath2 , the distribution of the outgoing carriers _ deduced _ within the defective dg ansatz is found equal to that of the incoming carriers . to conclude , dg concede that our idea of constraining populations `` is an interesting alternative to the wigner function '' , but argue that it is essential for this to choose complex states . we have shown above that the usage of the complex orbitals _ just _ of the form indicated by them as key point of a way - out solution does not mender the dg variational approach . these are not only complex , they are _ just _ of the form @xmath237 , which dg give ( and _ incorrectly _ attempt to suggest that they used them in ref . ) . consequently , they must now accept that their variational ansatz for transport is incorrect . the analysis done in conjunction with the issues raised by dg has led us to reveal two aspects of more general relevance for the transport theory . first , we have presented an example illustrating a limitation of the wigner function @xmath192 important for transport , namely that the sign of @xmath0 is not necessarily related to the direction of motion in the real world . second , we have presented an important physical reason why transport approaches , which , like the dg s , use information pertaining to a finite isolated cluster are inappropriate . this enlarges the basis of our critique recently formulated in sec . vii of ref . , where two other fundamental reasons were exposed . to summarize our detailed investigations on the dg approach,@xcite we can state that this approach lamentably fails because virtually all its ingredients are incorrect : * the dg approach imposes boundary conditions as if the wf were a true particle distribution , which is not justified quantum mechanically . replacing the wf - constraints by the boundary conditions most justified physically ( namely , the fermi distributions ) attempted in i does not remedy this approach . * even if the wf were a true momentum distribution , by uncritically constraining @xmath168 and @xmath169 it is very possible to erroneously constrain the _ outgoing _ majority charge carriers and not _ incoming _ ones . ( this is _ just _ the case in refs . . ) * the dg approach attempts to describe the transport by using a finite cluster within calculations based on a variational principle ( entropy maximization ) , which turned out to be problematic even if , unlike in the dg case , the limits of infinite time and infinite volume are taken in the correct order.@xcite * the dg approach aims at describing the transport by means of a wave function determined for a ( small ) isolated cluster , whose number of electrons is fixed , while a nanocluster in a real electric circuit does exchange electrons with the infinite electrodes to which it is connected . the lamentable failure of the dg approach was demonstrated by explicit calculations for the simplest uncorrelated and correlated , discrete and continuous models.@xcite they contradict well - established experimental and theoretical results , and it would make little sense to more amply document the incorrectness in _ many _ other cases.@xcite by contrast , dg could not present even a single example where it is valid . based on ingredients unfounded physically , it is not at all surprising that the currents predicted by the dg approach are completely unphysical and much poorly agreeing with experiment that more common approaches , contrary to the seemingly original success claimed in ref . . viii of ref . , we clearly showed that a standard negf - dft calculation yields currents slightly larger by a factor @xmath238 , while dg s currents @xcite represent @xmath239% of the experimental currents of the recent accurate experiment of ref . . the financial support provided by the deutsche forschungsgemeinschaft is gratefully acknowledged .

after iras @xcite showed that most galaxies in the nearby universe contain at least some dust , the main question moved from `` whether '' to `` when , where , and how much '' . `` when '' did galaxies started to be enriched in metals and dust and how did their dust content evolve as a function of redshift ? `` where '' is the dust located within a galaxy and how is it distributed among hubble types ? `` how much '' dust there is in galaxies , how much stellar light is absorbed by dust and how is it re - emitted ? more than one third of the bolometric luminosity of local galaxies is processed by dust into the infrared . the fraction of stellar radiation reprocessed into the iras 8120 @xmath0 m window is @xmath125%30% @xcite ; this implies an infrared energy fraction around 35%40% of the total bolometric energy within the local @xmath1100 mpc , once bolometric corrections from the iras window to the full infrared range are included @xcite . this average value comes with a large variation between galaxies , depending on luminosity , morphological type , and activity level ( see section 3 ) . the cosmic infrared background ( cib ) observed by cobe @xcite sets equally stringent limits on the total amount of stellar energy re - processed by dust into the infrared at _ all _ redshifts . about half , and possibly more , of the background radiation between 0.1 @xmath0 m and 1000 @xmath0 m is emerging at wavelengths @xmath240 @xmath0 m @xcite . this implies that at some epoch(s ) in the past galaxies have been infrared luminous at least as much as , and possibly more than , today to produce the observed balance between uv / optical and infrared energy in the cosmic background . bright sources are also found to be about two orders of magnitude more numerous in the past than today ( section 5.2 and * ? ? ? the infrared luminosity of a galaxy is related to both its dust content and nuclear / star - formation activity . the lower metal content of distant galaxies does not necessarily make them more transparent systems than galaxies today . the dust content of galaxies is , in fact , proportional to both metallicity and gas content ; as stellar populations evolve in time and more gas is locked into stars , the metallicity of a galaxy increases while its gas content decreases . the two trends are opposite to each other , and the dust content reaches a peak sometimes during the evolution of the galaxy . models predict that galaxies at z@xmath112 have about twice the dust opacity of local galaxies @xcite . high - redshift , star - forming galaxies are indirectly observed to have large gas column densities ; the average _ sfr per unit galactic area _ increases by more than an order of magnitude between redshift 0 and 3 @xcite , which is indication of a similar increase in the average gas column density via the global schmidt law @xcite . the presence of dust in galaxies limits our ability to interpret the local and distant universe , because dust extinction dims and redden the galaxy light in the traditional uv to near - ir windows , where the vast majority of the data have been obtained . dust effects hamper the interpretation of galaxies seds in terms of their fundamental parameters , such as age , stellar population mix , star formation rates ( sfrs ) , and stellar initial mass function ( imf ) . since young stellar populations are on average more deeply embedded in dust clouds than older stars , dust opacity variations within a galaxy impact the derivation of a galaxy s star formation history , by altering the census of the populations in the different age bins . heavy dust extinction in nearby starburst galaxies still prevents establishing whether the stellar imf in these systems is different from that of more quiescent environments @xcite . the increase of dust content with luminosity in spirals may be responsible for the non - linear behavior of the tully - fisher relation , and this translates into biased distance estimates @xcite . determinations of luminosity functions , mass - to - light ratios , and number counts , especially at short wavelengths , suffer from both the presence of dust and the variations of dust content from galaxy to galaxy . light dimming by dust biases measurements of galaxies sfrs , both locally @xcite and at high redshift @xcite , and is the cause of uncertain cross - calibrations between different indicators . the evolution with redshift of the sfr of galaxies per unit comoving volume is still debated : unclear is whether it has a peak around z@xmath113 @xcite or is mostly flat between z@xmath11 and z@xmath14 @xcite . answering this question , which requires understanding the dust opacity characteristics and the nature of high redshift galaxies , can discriminate between different flavors of galaxy formation scenarios @xcite . finally , measuring the fraction of metals locked into dust at different epochs is a crucial ingredient for tracing the chemical evolution of galaxies . the quantification of the dust opacity of galaxies of all types and at all redshifts as a function of wavelength has required the collective effort of many investigators over the last @xmath11015 years and is still , in many aspects , an on - going process . this review attempts to capture the main results obtained so far ; it concentrates on star - forming galaxies , because of their key role for intepreting the star formation history and evolution of galaxies . the topic has acquired with time large breadth and numerous publications have appeared in the field ; despite the effort to include as many relevant results as possible , some important contributions may have been inadvertenly overlooked . after iras , the infrared space observatory ( iso ) continued to shed light on the dust emission properties of galaxies , and comprehensive reviews are given in @xcite and in the proceedings of the 1998 paris conference ` the universe as seen by iso ' @xcite . the status of the topic before iso is reported in the proceedings of the 1994 cardiff ( u.k . ) conference ` the opacity of spiral disks ' @xcite and of the 1996 johannesburg ( south africa ) conference ` new extragalactic perspectives in the new south africa ' @xcite . throughout the following , the adopted cosmology is a flat universe with @xmath3=0 and hubble constant h@xmath4=65 km s@xmath5 mpc@xmath5 . ` local universe ' refers to a sphere of 100200 mpc radius centered on the milky way . the transfer of radiation through dust is governed by the well - known equation ( e.g. , * ? ? ? * ; * ? ? ? * ) : @xmath6 where i@xmath7(s,@xmath8 ) is the radiation intensity , ds is the element of linear distance along the direction of propagation , @xmath9 is the element of solid angle as seen by the dust grain , @xmath10 is the opacity ( absorption plus scattering ) of the dust per unit separation s , @xmath11 is the emissivity of the sources along the direction of propagation , @xmath12 is the albedo , i.e. the fraction of the opacity due to scattering , and @xmath13 is the phase function of the dust grains expressed as a function of @xmath14 , the angle between the directions determined by the incident radiation ( @xmath15 ) and the scattered radiation ( @xmath16 ) . one commonly used expression of the scattering phase function , the henyey - greenstein function @xcite , is characterized by a single parameter , the asymmetry parameter @xmath17=@xmath18cos(@xmath14)@xmath19 , which is the average of the cosine of the scattering angle . determining the dust opacity and recovering the intrinsic physical properties of complex systems like galaxies typically requires the full treatment of the transfer of radiation through dust ( equation 1 ) . the efforts to solve the integro - differential equation 1 in the case of galactic environments have generated increasingly sophisticated radiative transfer models and codes to deal with the complexity of both the dust physical properties and the dust / emitter geometry . a representative , although non - exhaustive , list of models and their salient characteristics is given in table [ tbl-1 ] . a detailed description and comparison between these models and codes is beyond the scope of the present review and the interested reader is referred to the papers listed in the table . although the simplifying assumptions implemented in many cases often appear to yield as good a solution as more complex treatments ( see , for instance , * ? ? ? * ) , the large number of free parameters is a fundamental shortcoming of radiative transfer models . workarounds to this shortcoming go from assembling large , coherent sets of independent observational data on just one or a small number of objects to making over - simplistic assumption on the age and metallicity mix of the galaxy stellar populations , and often both . this clearly bears deeply on the generality and applicability of any result and may account to an extent for the wide variety of conclusions on the opacity of galaxies reached by different authors . a point - like source ( a star or an agn ) behind a screen of dust is the simplest source / dust geometry , and equation 1 has solution : @xmath20 where i@xmath21 and i@xmath22 are the attenuated ( observed ) and intrinsic radiation intensity , and @xmath23 is the optical depth of the dust screen and a@xmath7 that will be used later . ] . the total - to - selective extinction , i.e. the part of the optical depth which only depends on the physics of the dust grains and , therefore , only on the wavelength , is often expressed as k(@xmath24)=a@xmath25/e(b@xmath26v ) . the color excess e(b@xmath26v ) is defined as a@xmath27(b)@xmath26a@xmath27(v ) , hence k(b)@xmath26k(v)=1 . although many studies traditionally employ a ` mean galactic extinction curve ' to correct galactic and extragalactic sources for the effects of dust obscuration , the well - studied milky way interstellar extinction has been shown to be spatially variable and highly dependent on the dust grain environment along the line of sight ( see reviews of @xcite and @xcite ) . @xcite demonstrated that the variations can be effectively parametrized by the total - to - selective extinction at v , r(v)=a@xmath28/e(b@xmath26v ) . both the steepness of the far - uv ( @xmath290.3 @xmath0 m ) rise of the extinction curve and the strength of the 0.2175 @xmath0 m absorption feature decrease for increasing r(v ) , as schematically shown in figure [ fig1 ] @xcite . most sightlines have r(v)@xmath3026 , with the low values corresponding to the diffuse ism ( for which r(v)=3.1 is a reasonable average value , ( * ? ? ? * e.g. , ) ) and the high values pertaining to dense clouds . the large values of r(v ) in dense clouds , usually accompanied by a gray extinction curve , are consistent with dust grains having systematically larger sizes than in the diffuse medium @xcite . the only galaxies external to the milky way for which some measurements of the uv extinction curves exist are the magellanic clouds ( figure [ fig1 ] ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , although tentative measurements in m31 have been made @xcite . despite the large variation of extinction properties found within each of the magellanic clouds ( * ; * ? ? ? * ; * ? ? ? * e.g. , ) , the _ average _ curves of the clouds appear to have some intrinsic differences from the average milky way curve , including a smaller r(v ) ( r(v)@xmath312.42.9 , @xcite ) . this has been interpreted as an effect of the lower metallicity in the clouds , which causes the molecular clouds to be more diffuse @xcite and , possibly , to produce on average smaller dust grains @xcite . the weak or absent uv absorption ` bump ' at 0.2175 @xmath0 m in the extinction curves of the lmc 30 doradus region and of the smc bar , respectively , ( figure [ fig1 ] ) could be related to local processes in these regions , possibly to the ongoing star formation activity @xcite . although the bump carriers have not been positively identified yet ( e.g. , * ? ? ? * ) , mid and far infrared studies of star forming regions in the milky way have provided evidence that the environmental uv energy density level affects the dust properties , such as the grain size distribution , via destruction or coagulation , and/or the dust physical characteristics , via ionization state changes @xcite . thus , the environmental uv energy density appears to play as important a role as the metallicity in shaping the extinction curve . if confirmed by further observations , this hypothesis may explain the general absence of a 0.2175 @xmath0 m absorption ` bump ' in the extinction curve of starburst galaxies ( section 4.1 ) and may partially account for its weakness in qsos ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? in addition to absorbing stellar light , dust manifests itself through emission and absorption features and continuum emission at @xmath21 @xmath0 m . features in the near / mid - ir include , among others , the silicate absorption at 9.7 and 18 @xmath0 m , h@xmath32o and co@xmath32 ice absorption at 3.0 and 4.3 @xmath0 m ( see @xcite and @xcite for a list of other absorption features ) , and the emission bands at 3.3 , 6.2 , 7.7 , 8.6 , and 11.3 @xmath0 m @xcite . the continuum emission beyond a few @xmath0 m and up to @xmath14070 @xmath0 m is considered to be mainly due to discrete photon heating of very small dust grains , the smallest among which are transiently heated up to @xmath301,000 k ( e.g. , * ? * ; * ? ? ? * ) ; dust heated by massive stars to temperatures @xmath1970 k also contributes to the flux in the wavelength region below @xmath150 @xmath0 m , albeit with decreasing filling factor for increasing temperature @xcite . beyond 4070 @xmath0 m , the emission is mainly due to dust grains in nearly steady balance with the average heating by starlight ; the transition wavelength between predominance of discrete photon heating versus that of nearly - steady state depends on the average color temperature of the dust and , thus , on the mean uv energy density level of the galaxy @xcite . the original dust model of bare graphite and silicate grains @xcite has evolved over time by including more complex grain compositions and extended grain size distributions , to accomodate all the details of the absorption and emission , and the variability of the environment , while at the same time accounting for the constraints imposed by the heavy elements overall abundance ( * ? ? ? * e.g. , ) . polycyclic aromatic hydrocarbons @xcite , hydrogenated amorphous carbons @xcite , icy and organic refractory mantles @xcite , composite and/or porous grains @xcite , as well as grain size distributions that significantly depart from the original power law model @xcite , have all been suggested as possible alternatives / complements to the graphite / silicate model . in the late 80s , @xcite and @xcite revived the decade - long debate on whether galaxy disks are mainly transparent or opaque to dust . using multi - wavelength information and surface brightness - versus - inclination tests , respectively , the two groups of authors reached the similar conclusion that the disks of galaxies are fairly opaque , with total face - on extinctions in the b band a@xmath331 mag . ] . these results countered earlier findings , also based on the surface brigthness - versus - inclination tests , that galaxy disks are mostly transparent with face - on extinctions at b less than 1 mag @xcite . the reason why similar approaches can produce very different results can be found in the complexity of the absolute and relative distribution of stars and dust inside galaxies . when there is a significant departure from the simplest geometry of a point source ( a star ) behind a homogeneous screen of dust , the dust distribution , rather than the optical depth of the dust layer , becomes the dominant factor in determining the obscuration suffered by the stellar radiation @xcite . in particular , differential optical depth effects come into play . if , for instance , the stars are homogeneously mixed with the dust , the shorter the wavelength of the radiation the smaller the optical depth of the dust layer from which it can emerge ; thus , an observer will detect blue light only from the closest outer ` skin ' of the mixed distribution , while redder light will come from deeper layers . an extreme case is that of infinitely optically thick clumps in front of an extended light distribution : the stellar emission will come only from those lines of sight that do not intersect clumps , and the net result will be an unreddened , albeit dimmer than the original , spectral energy distribution ( sed ) . the scattering component of the extinction curve will also modify the general characteristics of the dust obscuration , as light will be scattered _ into the line of sight _ as well as out of it . the net result of all these effects is a general blueing of the emerging sed relative to the foreground homogeneous dust screen case , or , in other words , an often much greyer ` effective extinction ' , even in the presence of large dust optical depths @xcite . for illustrative purposes , figures [ fig2 ] and [ fig3 ] show how photometric and spectroscopic quantities are affected by the dust geometry for five ` toy models ' . the models represent plane - parallel geometries of dust and stars ( details are given in * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) for : ( 1 ) the simple case of a homogeneous , non - scattering dust screen foreground to the light source ; ( 2 ) a homogeneous , scattering dust screen foreground to the light source ( e.g. , a screen in close proximity of the source , so that scattering into the line of sight becomes an important effect ) ; ( 3 ) a poissonian distribution of clumps in front of the light source , with average number @xmath34=10 ; ( 4 ) a homogeneous mixture of dust and emitters ; ( 5 ) a homogeneous dust / emitter distribution averaged over all inclination angles , with the uv and ionized gas emission having the same scaleheight of the dust while the scaleheight of the optical light gradually increases to twice that of the dust at v and beyond . models 25 include scattering in the calculationsm feature is due to pure absorption ( no scattering ) @xcite . these @xmath35 and @xmath36 values are assumed here to be representative of dust properties , although the following caveats need to be kept in mind : some values , especially for the albedo , are still unsettled @xcite ; the generalization of the reflection nebulae results ( r(v)@xmath193.1 ) to the diffuse dust is not immediate @xcite ; the uv @xmath35 and @xmath36 values could be lower in the magellanic clouds than in the milky way @xcite . ] . in all cases , the light source is assumed to be extended and to be represented by a 300 myr old stellar population undergoing constant star formation , with solar metallicity and a salpeter imf in the mass range 0.1100 m@xmath37 @xcite . the ` blueing ' effect of geometries more complex than a foreground dust screen is evident from the comparison of the five models . at fixed optical depth @xmath38 ( fixed amount of dust ) , the reddening of the v@xmath26i color of a stellar population induced by a generic geometry of stars and dust will be less , and the effective extinction a@xmath39 lower , than that induced by a foreground screen ( figure [ fig2 ] , panels a and b ) . the same will be true for colors at other wavelengths ( figure [ fig2 ] ) , where , again , the foreground non - scattering screen will provide the maximum reddening . for mixtures of dust and stars ( models 4 and 5 ) , reddening of colors and effective extinction reach asymptotic values ; beyond certain values of the optical depth , an increase in @xmath38 corresponds to small or negligible changes in colors and a@xmath39 . this is due to the fact that the shortest wavelength emission is contributed only by the dust layers closest to the observer . one common consequence of the greyer effective extinction caused by complex dust / emitter geometries is that a limited number of diagnostics over a small wavelength range will lead to underestimates of the true galaxy opacity , especially if the actual dust geometry is not known and is ` a priori ' assumed to be foreground @xcite . to aid the comparison with data on local starbursts and high - redshift star - foming galaxies ( sections 4 and 5 ) , the models shown in figure [ fig3 ] are a function of the ratio of the attenuated - to - intrinsic balmer line ratio : @xmath40 and of the uv spectral slope . ] . the luminosity ratios of nebular hydrogen lines , like h@xmath41 , h@xmath42 , pa@xmath41 , br@xmath43 , etc . , are rather insensitive to the details of the underlying stellar population and of the imf , being affected at the @xmath15%10% level by variations of the gas physical conditions ( e.g. , * ? ? ? * ) ; they represent , therefore , accurate probes of dust reddening and geometry . table [ tbl-2 ] lists the commonly used hydrogen emission line ratios , and the differential extinction between each pair . when multiple line ratios widely separated in wavelength are available , they can constrain the dust geometry in the sampled wavelength range . the two drawbacks of using nebular lines are : ( 1 ) they are produced by ionizing photons , therefore an ionizing stellar population needs to be present in the galaxy ; and ( 2 ) they only probe the attenuation of the _ ionized gas _ , and such attenuation may or may not be directly related to that of the stellar population ( section 4 ) . the explicit expression of r@xmath44 depends on the dust geometry ; in the simple case of a foreground , non - scattering screen : @xmath45},\ ] ] where the values of k(h@xmath42 ) and k(h@xmath46 are given in table [ tbl-2 ] . degeneracies between the reddening induced by dust and that induced by variations in the age , metallicity , and imf of the stellar populations play an non - negligible role in opacity determinations of galaxies . these effects are especially important when only broad band colors or low resolution spectroscopy are available , and age - sensitive or imf - sensitive stellar features are not identifiable ( e.g. , discussion in * ? ? ? * ) . for the age - dust degeneracy , an example is given in figure [ fig4 ] ; here , the broad band colors and magnitudes of a dust - free , ageing stellar population produced by an instantaneous burst of star formation are compared with the analogous quantities of an increasingly extincted 6 myr old stellar population . a clumpy dust distribution with @xmath472 in front of the young population reddens and attenuates it enough that its broad - band uv , optical , and near - ir colors closely resemble those of a 20 myr old , dust - free population ( corresponding to the color v@xmath26i@xmath10.6 in figure [ fig4 ] ) . thus , the dusty , young population can be mistaken for a dust - free population a factor @xmath13 older . this example is just one out of a large range of possibilities in the multi - dimensional dust / age / imf / metallicity parameter space . in summary , the grey effective extinction produced by dust geometry and scattering , the age - imf - metallicity - dust degeneracy , together with uncertainties on the appropriate extinction curve to be applied to different galaxies ( section 2.1 ) and the absence of strong emission and absorption features below @xmath11 @xmath0 m ( apart from the @xmath10.22 absorption bump which is not universal ) , all contribute to make dust obscuration measurements very difficult in galaxies . efforts to determine the dust opacity of local galaxies have spawned the development of a number of independent techniques that have been applied to a large array of samples . the most widely used methods in the case of galaxy disks are tests of the dependence of the surface brightness on inclination , multi - wavelength comparisons , and statistical analysis of the color and number count variations induced by a foreground galaxy onto background sources . each method has its strengths and weaknesses , as briefly detailed below . as @xcite , @xcite , and @xcite pointed out , probing galaxy dust opacity by measuring the surface brightness dependence of galaxy disks on inclination can suffer from drastic selection biases . for increasing inclination , an apparent - magnitude - limited catalog will mimick the condition for transparent galaxies , with mildly increasing surface brightnesses , increasing diameters , and constant total luminosities . conversely , an apparent - diameter - limited catalog will mimick the conditions for opaque disks , with roughly constant surface brightnesses , increasing magnitudes and constant diameters as a function of increasing inclination @xcite . selection effects can , however , be controlled by combining different samples and by using galaxy redshifts / distances to discriminate among the different biases , and analysis of large samples of late - type ( mainly sbc and sc ) spirals @xcite has shown that the central regions of disk galaxies are opaque while the external regions beyond 23 scale lengths are almost completely transparent ; there is also a trend for luminous galaxies to be more opaque than less luminous galaxies @xcite . in bright , m@xmath4821 disk galaxies , the observed magnitude correction from edge - on to face - on , @xmath49m@xmath501 mag , corresponds to a face - on central optical depth is the optical depth of the dust at the geometrical center of the tri - dimensional dust distribution for a face - on disk galaxy . ] @xmath38(0)@xmath515 ; this implies effective face - on extinctions a@xmath520.40.5 mag , a@xmath530.2 mag , and a@xmath540.1 mag @xcite , for a double exponential distribution of dust and stars , and for practically transparent external galaxy regions @xcite . less luminous galaxies are progressively less opaque in their central regions , with a edge - on to face - on correction of @xmath49m@xmath500.5 mag at m@xmath5518 , corresponding to effective face - on extinctions a@xmath560.1 mag @xcite . qualitative results similar to those above have been obtained by @xcite and @xcite for a small number of edge - on late - type spirals using multi - wavelength kinematic information . the method , proposed by @xcite , consists of comparing rotation curves of an edge - on galaxy at widely different wavelengths . if the edge - on galaxy is optically thick at a given wavelength , say , h@xmath42 , the rotation curve measured at this wavelength shows apparent solid body rotation throughout the entire disk , because the h@xmath42 emission is coming only from the external regions of the galaxy . a comparison with a mostly unextincted rotation curve measured at a longer wavelength , such as hi , can then establish whether optical thickness at h@xmath42 exists . this method , though fairly direct in principle , may suffer from two limitations : first , dust in galaxies tends to be patchy and the h@xmath42 emission may be detected along the least extincted sight lines even in the presence of considerable opacity ; second , the method implicitly assumes that h@xmath42-emitting regions are uniformly distributed across the disk . the multi - wavelength comparison method provides an independent approach to the problem of determining the dust opacity of galaxies . in broad terms , the technique consists of using multiwavelength data to solve simultaneously for the intrinsic colors of the stellar populations and for the reddening / attenuation induced by the dust . generally , broad band images covering as large as possible wavelength baseline , often from b to k band , are used to determine the opacity of the disks , by exploiting the differential extinction property of dust @xcite . indeed , in face - on quiescent galaxies k - band emission appears negligibly affected by dust obscuration , with @xmath5710% of the light lost to dust absorption and scattering @xcite . the multiwavelength technique is the most widely adopted for measuring dust opacity in galaxies , because it employes easily accessible information on diverse galaxy types , and enables detailed modelling of optical depths as a function of galactic region . there are , however , three main weaknesses to the method : first , this technique needs as wide a wavelength coverage as possible , often difficult to obtain , to yield unambiguous results ; second , if the dust optical depth exceeds unity at some wavelength , different volumes of the galaxy will be sampled at different wavelengths ; third , the need for large datasets and detailed modelling makes obtaining results on large samples relatively cumbersome . results vary from author to author , due to the different sample selections , galaxy and dust distribution models , and measurement techniques applied . optical and near - ir image analysis gives typical central optical depths @xmath38(0 ) in the range 0.32.5 for types sab - sc @xcite , although higher values have been obtained by @xcite . for exponential stellar and dust profiles , these values of @xmath38(0 ) result in optical depths @xmath581 averaged over the inner 23 scale - lengths @xcite , and a@xmath520.30.4 mag and a@xmath590.15 mag over the same area . much lower values of the optical depth are found for late - type , low surface brightness galaxies , with face - on , central values @xmath38(0)@xmath10.15 @xcite . in galaxy disks , arm regions tend to be opaque , with @xmath471.45.5 ( a@xmath530.20.7 mag ) and peak values at @xmath471012 , while interarm regions are generally transparent , with @xmath470.31.1 ( a@xmath530.050.15 mag , @xcite , but see the higher interarm extinction values found by @xcite ) . energy balance estimates between dust absorption and emission can also be classified as a ` multiwavelength technique ' , but it is conceptually different from the one above . this technique employes direct measurements of the dust emission in the infrared to derive an estimate of the total stellar energy absorbed by dust , and then derive opacities at shorter wavelengths ; whenever possible , the wavelength baseline is extended to the uv to probe the direct light from massive stellar populations and constrain bolometric quantities @xcite . by combining uv and infrared ( iras ) data , @xcite , @xcite , and @xcite find typically lower @xmath60 values than those derived from optical / near - ir images . this is not incompatible with those results , since @xcite and @xcite s models assume plane - parallel dust and stars , instead of exponential profiles , and thus the opaque inner regions are averaged with the nearly transparent outer regions @xcite . still , the uv selection of the samples may bias the conclusions towards lower optical depths and the use of iras data leaves some uncertainty as to the bolometric correction of the dust emission . the use of iso data , that extend the infrared wavelength coverage out to 200 @xmath0 m , appear indeed to favor higher values of the effective extinction @xcite . @xcite find that opacity in disk galaxies is a function of luminosity ( figure [ fig5 ] ) , in agreeement with @xcite . these authors parametrize the opacity as : @xmath60=@xmath61 , with @xmath62=0.6 and l@xmath63 the visible luminosity of a l@xmath64 galaxy in the local universe . similarly , @xcite note that sb scd galaxies have @xmath650.8 and are on average 1.5 times more opaque than sa sab , and @xmath13 times more opaque than sd irr galaxies , with effective face - on extinctions a@xmath520.3 mag , a@xmath520.15 mag , and a@xmath660.1 mag for sb scd , sa sab , and sd - irr , respectively . all methods described so far use the stellar radiation of the galaxy itself as light source for the dust , with the complication that intrinsic colors , age and metallicity gradients of the stellar populations and the relative distributions of dust and stars need to be disentangled from the extinction proper . a technique that overcomes these problems is the one that uses background sources as ` light bulbs ' for the dust in foreground galaxies . in this case , the dust opacity is measured across the full thickness of the foreground galaxy . there are basically two approaches to the technique . in the first approach the foreground galaxy is partially projected onto a nearby background galaxy @xcite . in general , the background galaxy needs to present a smooth light profile , thus early types and preferably ellipticals are required ( see , however , * ? ? ? * ) ; the dimming and color changes induced by the foreground galaxy dust onto the isophotes of the background source provides a measurement of dust opacity . the drawbacks of this approach are twofold : ( 1 ) there are very few known pairs of galaxies that fullfil all the requirements and ( 2 ) the method is limited by the degree of symmetry of both galaxies @xcite . the second approach uses distant galaxies as background light sources . statistical variations in number counts and colors of the background galaxies relative to control fields provide a handle on the total dust optical depth of the foreground galaxy @xcite . this method overcomes most of the limitations of the first one , but it presents some of its own : ( 1 ) high angular resolution images ( e.g. , from hst ) are needed in order to identify background galaxies ; ( 2 ) even with such images it is often difficult to discriminate background sources from extended regions within the foreground galaxy ( e.g. , hii regions ) , especially in crowded areas . because of these limitations , the second method may also tend to avoid crowded and/or heavily extincted areas , thus biasing the results towards less dusty galactic regions . results from various authors indicate that the spiral arms of galaxies tend to be opaque with a@xmath670.52 mag and a@xmath680.31.5 mag at any radius , while the opacity of interarm regions decreases with increasing distance from the galaxy center , starting at a@xmath690.7 mag within 0.3 r@xmath70 and decreasing to practically zero at r@xmath70 @xcite . the dust masses inferred from these extinction measures agree within a factor @xmath12 with masses derived from the infrared / submm ( iso and scuba ) dust emission @xcite . one common trend generally found is the greyer - than - galactic reddening in the arm regions ( see , however , * ? ? ? the clumpiness of the dust within each resolution bin has been suggested to be responsible for the grey reddening @xcite . less agreement exists on the shape of the reddening curve in the interarm regions , due to the larger uncertainties resulting from the lower extinction values . the measured extinctions are front - to - back , i.e. represent the optical depth of the intervening dust between the observer and the background sources . to relate these numbers to those derived from the inclination and multi - wavelength tests , one has to make assumptions as to the distribution of stars relative to the dust in the foreground galaxy ; for a mixed geometry of stars and small dusty clumps , the measured optical depths correspond to effective i extinctions of a@xmath530.30.4 mag and a@xmath710.2 mag for the arms and the interarm regions , respectively . multi - wavelength optical and near - ir imaging of edge - on or nearly edge - on systems indicate that the scalelength of the dust is about 40% larger than the scalelength of the stars @xcite , while the dust / stars scaleheight ratio is in the range 0.250.75 , with a mean value of 0.5 , @xcite for late - type spirals ( sb and later ) and appears to have ratio @xmath11 in early - type spirals @xcite . iras 100 @xmath0 m @xcite and iso long wavelength maps @xcite confirm that cold dust emission extends beyond the limits of the optical disks along the radial direction , with scales that are @xmath140% larger than those of the b - band emitting stars , but still well within the hi disks ( e.g. , * ? ? ? using statistical measurements of color variations between background galaxies and control fields , @xcite give a @xmath512 @xmath72 detection of a@xmath690.04 in galaxy haloes , at a distance of about 60 kpc from the galaxy centers , along the optical major axis . spiral and irregular galaxies follow a global metallicity - luminosity ( and metallicity - mass ) relation , where the oxygen abundance increases by a factor @xmath1100 for an increase of @xmath110@xmath73 in absolute blue luminosity @xcite . irregulars occupy the low metallicity locus of the relation , with @xmath3010 times on average lower oxygen abundance than spirals . since the hi masses per unit l@xmath74 in irregulars are a factor @xmath303 larger than in spirals @xcite , the mass in interstellar metals per unit l@xmath74 is @xmath303 times smaller . adding the h@xmath32 gas into the balance does not change this conclusion , since the h@xmath32/hi mass ratio in irregulars appear to be no larger than that in spirals , even accounting for the uncertainties in the co - to - h@xmath32 conversion factor @xcite . measurements give the h@xmath32 mass as 25%50% of the hi mass in late spirals and as @xmath3020% of the hi mass in irregulars @xcite . smaller metal masses for the irregulars translate into smaller dust masses per unit l@xmath74 , for standard values of the metal - to - dust ratio @xcite , and into the expectation that irregular galaxies are at least 3 times more transparent at b than spiral galaxies , for similar dust / star geometries . the analysis of the blue , h@xmath42 , and iras - detected infrared emission confirms that irregular galaxies are relatively transparent systems @xcite . in a study of the magellanic clouds using background galaxies , @xcite find that the central regions have a@xmath750.25 in the lmc and a@xmath750.16 in the smc . in their uv selected samples , @xcite find that a 0.1 l@xmath64 galaxy has a mean optical depth about 1/3 of the optical depth of an l@xmath64 galaxy , and @xcite reach a similar conclusion by comparing sb - scd galaxies with sd - irr . thus , if the typical face - on effective extinction for an sc galaxy is a@xmath520.40.5 mag , the analogous quantities for an irregular are a@xmath520.15 and a@xmath530.050.10 . although early hubble type galaxies ( ellipticals and s0s ) do not generally fall under the category of ` star - forming ' galaxies , they occupy a niche in terms of dust opacity properties that is at the same time opposite and complementary to that of other galaxies . ellipticals and s0s are at the opposite end of the hubble classification relative to irregulars , but they appear to be no more opaque than those late systems . early type galaxies cover a metallicity range similar to that of late type spirals @xcite , but have total ( hi@xmath76h@xmath32 ) gas fractions per unit optical luminosity that are @xmath13 times or more smaller @xcite . thus , for standard interstellar metal - to - dust ratios , early type galaxies are expected to be more transparent than disk galaxies . the iras survey has shown that between 10% and 50% of all nearby early type systems ( 45% of ellipticals and 68% of s0s ) contain dust @xcite , although the involved masses are generally modest , typically in the range 10@xmath7710@xmath78 m@xmath37 or m@xmath79/l@xmath8010@xmath8110@xmath82 m@xmath37 l@xmath83 @xcite . the exact fraction of dusty systems appears to depend upon the adopted detection threshold @xcite , but it is likely a lower bound to the actual value , given the limits in the iras sensitivity . the 50% figure seems in agreement with the results of optical surveys : at least 40% of the local ellipticals and s0s show presence of disks , lanes and/or patchy distributions of dust in visual images @xcite . about 4050% , and possibly as many as 80% , of ellipticals also show evidence for circumnuclear dust on scales @xmath18100 pc , as seen in hst images ; the small - scale dust is mostly associated with nuclear radio activity @xcite . despite the abundance of evidence for dust , optical extinction determinations of dust features in ellipticals generally account for only @xmath11/10 to @xmath11/5 of the dust mass measured from the infrared ( iras ) emission ; the remaining 80%90% of the dust mass is probably associated with a component smoothly distributed across the galaxy , and thus difficult to detect with standard techniques @xcite . limited data from iso in the wavelength range 120200 @xmath0 m confirm these findings @xcite . using the elliptical galaxy model of @xcite , @xcite determine that the observed infrared luminosities are compatible with central optical depths of the diffuse component @xmath38(0)@xmath570.7 , with a typical value @xmath38(0)@xmath10.20.3 ( see , however , the somewhat larger values predicted by the models of @xcite ) . the corresponding effective extinction is a@xmath840.050.10 mag and a@xmath850.020.04 mag . the giant ellipticals found at the centers of cooling flow clusters are often surrounded by extended , @xmath3010100 kpc , filamentary , and dusty emission - line nebulae , with a higher frequency in those clusters with cooling times shorter than a hubble time @xcite . dust lanes in the central galaxies have , indeed , been directly observed @xcite . optical and uv emission line studies give extinction values in the range a@xmath860.32 mag for the dust associated with the nebula @xcite ; infrared and sub mm measurements give dust masses of the order of 10@xmath8710@xmath88 m@xmath37 , or m@xmath79/l@xmath8010@xmath8210@xmath89 m@xmath37 l@xmath83 , for these systems @xcite . the existing controversy on the origin of the dust associated with the central galaxies of cooling flow clusters will be briefly reviewed in section 6 . such galaxies are , however , not a statistically significant component of ellipticals in general , thus their contribution to the dust mass budget of early type galaxies is small . the galaxies discussed so far tend to be mostly ` quiescent ' , i.e. their bolometric output is _ not dominated _ by a single region of active star formation or by nuclear non - thermal activity , and have modest infrared outputs . their infrared luminosities in the 31000 @xmath0 m band are typically @xmath3010@xmath90 l@xmath37 and less than @xmath146@xmath9110@xmath92 l@xmath37 , with infrared - to - blue luminosity ratios l@xmath93/l@xmath940.40.5 @xcite , and @xmath301/2 or less of the stellar bolometric luminosity converted into infrared emission via dust reprocessing of the dust emission is derived from the infrared luminosity measured in the 8120 @xmath0 m or 40120 @xmath0 m iras window @xcite multiplied by @xmath11.4 or @xmath12.2 , respectively , to account for bolometric correction factors as modeled by @xcite from iso data . the blue luminosity is l@xmath95(0.44 @xmath0 m ) . ] ( see table [ tbl-3 ] ) . galaxies with l@xmath966@xmath9110@xmath92 l@xmath37 and l@xmath93/l@xmath971 are progressively associated with dustier and more active systems for increasing values of the infrared luminosity @xcite . the brightest infrared sources are not necessarily associated with the most massive galaxies , despite the existence of a relation between a galaxy luminosity ( mass ) and its dust content @xcite . infrared luminosity is a combination of dust content and activity : brighter sources have larger amounts of star formation and/or non - thermal activity embedded in dust . most of the luminosity from galaxies with l@xmath98l@xmath9910@xmath100 l@xmath10134 l@xmath64 appears associated with sources heavily obscured by dust @xcite . at the bright end of the infrared luminosity function @xcite are the ultraluminous infrared galaxies ( ulirgs ) with luminosities from l@xmath10210@xmath103 l@xmath37 up to @xmath18@xmath9110@xmath103 l@xmath37 @xcite , infrared - to - blue ratios around 30400 @xcite , and optical extinctions in excess of 1040 magnitudes @xcite . these sources show a continumm of properties as a function of increasing luminosity , with the fraction of starburst - dominated ulirgs going from @xmath180%85% at the low luminosity end to @xmath150% for l@xmath962@xmath9110@xmath103 l@xmath37 , and the remaining fraction being agn - dominated @xcite . comprehensive reviews of the properties of ulirgs are given in @xcite and @xcite . despite their large luminosities , bright infrared sources do not represent a major contributor to the energy budget of galaxies in the present - day universe . such objects are relatively rare within the local 100200 mpc : galaxies with l@xmath10410@xmath100 l@xmath37 and with l@xmath10410@xmath103 l@xmath37 represent @xmath16% and @xmath181% , respectively , of the total infrared emission in the local universe @xcite . however , the contribution of infrared bright sources to the galaxy energy budget increases with redshift ( see section 5.2 ) , as does their relevance in the framework of galaxy evolution . although the issue of dust opacity of galaxies is far from settled , an overall picture is apparent : galaxies in the local universe are only moderately opaque , and extreme values of the opacity are only found in the statistically non - dominant , more active systems . table [ tbl-3 ] summarizes the main conclusions of this section , by listing the effective extinction values for local galaxies according to morphological type . for disk and irregular galaxies , both the face - on and the inclination - averaged extinctions are given . for instance , in sb - scd galaxies , the typical inclination - averaged extinctions are 0.60.65 mag larger than face - on at 0.15 @xmath0 m , @xmath10.4 mag at b , @xmath10.25 mag at i , and 0.1 mag at k , for a stellar to dust scaleheight ratio that varies between 1 in the uv and 2 in the i and k bands ( model 5 of section 2.2 ) . among the non - active galaxies , luminous sb scd galaxies contain the largest amounts of dust , which absorb @xmath301/2 of their bolometric light . for comparison , dust emission represents between 35% and 40% of the bolometric luminosity of our milky way galaxy @xcite . the dust content generally decreases in less luminous galaxies and for earlier and later type galaxies ; dust absorbs less than @xmath115% of the bolometric light in e / s0 galaxies , and the fraction rises to @xmath3030% in irregulars ( table [ tbl-3 ] ) . interpolating from table [ tbl-3 ] , the inclination - averaged effective extinction of the stellar emission in late spiral galaxies at the restframe wavelength of h@xmath42 is a@xmath1050.60 mag . this value should be compared with the effective extinction measured for the h@xmath42 line emission , a@xmath1060.81.1 mag , from the ratio of the line to the reddening - free thermal radio emission @xcite . althought there are relatively large uncertainties associated with both numbers , the two extinction values suggest that the nebular emission is more reddened than the stellar continuum , with a@xmath1071.5 a@xmath108 . studies of controlled samples indicate a value of 2 as more typical for the ratio between the dust attenuation of the ionized gas and that of the stellar continuum , both in the case of spirals and irregulars @xcite and in the case of moderate - intensity starbursts ( section 4 ) . among local galaxies , starbursts occupy a special niche . the average uv and infrared luminosities ( both proxies for sfr ) per unit comoving volume increase with redshift by roughly an order of magnitude up to z@xmath112 @xcite . active star formation played a larger role in the past than today , and may have been a fundamental ingredient in the shaping of galaxy populations . nuclear and circumnuclear starbursts occur in galaxies spanning a wide range of properties , from almost dust - free , infrared - faint , low - mass irregulars ( e.g. , tol1924@xmath26416 and ngc1569 ) to dust - rich , massive spirals ( e.g , ngc5236 ) , to infrared - bright interacting galaxies ( e.g. , ngc7714 and ngc6090 ) , to mergers and ulirgs ( e.g. , ngc4194 and ngc1614 ; arp220 , * ? ? ? nuclear and circumnuclear bursts of star formation can be triggered by a variety of mechanisms , including secular evolution of bars @xcite and galaxy interactions and mergers @xcite . the latter produce the most luminous starbursts ( ulirgs ) , which rapidly deplete the molecular gas content of the galaxy as merging progresses @xcite . less extreme starbursts , hosted in galaxies with l@xmath1095@xmath9110@xmath100 l@xmath37 , obey a positive correlation between the luminosity of the starburst and the luminosity and mass of the host galaxy @xcite : @xmath110 where sfr@xmath111 is in m@xmath37 yr@xmath5 for a salpeter imf in the mass range 0.1100 m@xmath37 ( from the formulae of * ? ? ? equation 5 is probably a consequence of the fact that large sfrs are substained by the massive gas inflows that only a deep potential well can support @xcite . indeed , from causality arguments , the maximum luminosity of a starburst scales as @xmath112v@xmath113 , with @xmath72 the dispersion velocity of the gas and v the rotation velocity of the host galaxy @xcite . there are notable exceptions to this average trend : m82 is an example of a dwarf , m@xmath110@xmath92 m@xmath37 @xcite and l@xmath1145@xmath9110@xmath90 l@xmath37 , galaxy hosting a relatively powerful starburst , with sfr@xmath16 m@xmath37 yr@xmath5 and l@xmath1023@xmath9110@xmath92 l@xmath37 . a secondary effect of both the sfr@xmath111host galaxy mass correlation and the mass - metallicity correlation is that the more powerful starbursts are hosted in the more metal - rich and dustier host galaxies ( * ? ? ? * also figure [ fig7 ] , panel c ) : @xmath115 ( see , also , * ? ? ? although equation 6 is not applicable to ulirgs , still in the local universe there are no examples of ` naked ' arp220-like galaxies , emitting most of the luminosity produced by their sfr@xmath1100200 m@xmath37 yr@xmath5 directly in the uv . objects like arp220 tend to be heavily extincted , with a@xmath861545 mag for arp220 itself @xcite , and to be bright infrared sources ( section 3.5 ) . the rest of section 4 discusses the dust reddening and obscuration properties of the local starburst galaxies with l@xmath1095@xmath9110@xmath100 l@xmath37 . the specific sample contains starbursts that are uv luminous enough to have been detected by iue @xcite ; it will be termed ` uv - selected ' in what follows . the sample is not statistically complete or rigorously defined ; however , it is representative of the broad range of properties of local uv bright starbursts @xcite . these galaxies are not as extreme in dust extinction and activity level as ulirgs . they have sfrs per unit area in the range @xmath10.320 m@xmath37 yr@xmath5 kpc@xmath116 and cover a limited range in the parameter space of optical attenuation , a@xmath10602.5 mag . the attenuation refers to the integrated value across the starburst site ; individual regions or clumps can have effective extinctions as high as a@xmath10630 mag @xcite . the host galaxies cover a wide variety of late type morphologies , from grand design spirals to irregular and amorphous , to distuberd morphologies that result from interactions or merging processes . the starbursts mostly occupy the inner region of solid body rotation of the host galaxy , with sizes @xmath10.54 kpc @xcite . the uv selection @xcite aids the discrimination from their more opaque and infrared - brighter counterparts . although the uv selection results in some limitations to the applicability of the obscuration properties discussed below ( see end of section 4.2 ) , it does not entirely exclude dusty systems ; the sample contains objects in which up to 90%95% of the bolometric energy is re - processed by dust . table [ tbl-4 ] gives three representative examples each for both the high - luminosity , dust - rich end and the low - luminosity , dust - poor end of the uv - selected starbursts , together with the average of the sample . although the uv - selected starbursts form an heterogeneous set , their dust reddening and obcuration properties are remarkably uniform . higher dust optical depths produce redder uv to near ir seds and nebular emission line ratios ( figure [ fig6 ] ) , almost independent of the details of the dust distribution and of the intrinsic stellar population @xcite . the geometry of the dust that best describes the reddening of the ionized gas emission in these systems , in the wavelength range 0.48 @xmath0m2.2 @xmath0 m , is that of a foreground like distribution . figure [ fig6 ] , panel b , shows the comparison between observations and the models of section 2.2 for the attenuated - to - intrinsic hydrogen line ratios h@xmath42/h@xmath41 and h@xmath41/br@xmath43 . internal dust does not appear to be a major component in the starburst region(s ) , and the little present is likely to be in compact clumps @xcite . the main source of opacity appears to be given by dust that is external , or mostly external , to the starburst region ( although still internal to the host galaxy ) , similar to a ( clumpy ) dust shell surrounding a central starburst . this is , of course , verified only in the b to k wavelength range where the reddening of the nebular emission lines has been measured ; measurements at longer wavelengths could easily reveal more complex geometries . from a practical point of view , the data in figure [ fig6 ] , panel b , are well described by a simple foreground , non - scattering dust distribution , with mw extinction curve for the diffuse ism ( r(v)=3.1 ) ; the reddening of the ionized gas is thus parametrized by the color excess e(b@xmath26v)@xmath117 ( equation 4 ) . the stellar continuum colors show the trend to become redder for increasing reddening of the hydrogen emission line ratios ( figure [ fig6 ] , panels a the implications are twofold : ( 1 ) variations of the intrinsic sed from galaxy to galaxy are secondary relative to the trend induced by dust obscuration , and ( 2 ) within each starburst , the same foreground like dust geometry holds for both the ionized gas and the non - ionizing stellar continuum . the first implication is not surprising , at least at uv wavelengths . since we are observing the rayleigh - jeans part of the massive stars spectrum , the intrinsic uv spectral slope and @xmath118 refer to the intrinsic and observed , respectively , uv spectral slope measured between 0.125 @xmath0 m and 0.26@xmath0 m as defined in @xcite , see appendix b for more details . ] @xmath119 is relatively constant over a fairly large range of ages , with values @xmath262.7@xmath1202.2 for constant star formation over 10@xmath78 yr@xmath57age@xmath5710@xmath90 yr ( see * ? ? ? * and the table in appendix b ) . in the same conditions , the production of ionizing photons is also constant . in an instantaneous burst of star formation , the ionizing photons are more age - sensitive than the non - ionizing photons that produce the stellar continuum and disappear before appreciable changes in the uv spectral shape can be observed . the observed uv spectral slope @xmath118 of the uv - selected starbursts span the range @xmath262.5@xmath1210.4 , much larger than what expected from stellar population variations alone , and increases for increasing values of the attenuated - to - intrinsic hydrogen line ratios ( i.e. , color excess , figure [ fig6 ] ) . in addition to correlate with the reddening of the ionized gas , the reddening of the uv stellar continuum , as measured by @xmath118 , correlates with the the infrared - to - blue and the infrared - to - uv luminosity ratios , l@xmath93/l@xmath74 and l@xmath93/l@xmath122 ( ( figure [ fig7 ] and * ? ? ? * ; * ? ? ? these ratios are a measure of the total dust opacity in the starburst region , because the bulk of the starburst s energy is coming out in the uv b . in addition , dust _ absorption _ is the main process that removes light from the line of sight @xcite ; when large galactic regions are observed light scattered by dust out of the line of sight compensates , on average , that scattered into the line of sight . the opacity reddening correlation means that measurements of reddening can be used to infer the total dust absorption in uv - selected starbursts @xcite with an uncertainty , @xmath49a@xmath1230.6 mag for individual objects and @xmath120% when averaged over large samples @xcite , that is dominated by variations in the starburst populations and in the details of the dust geometry from galaxy to galaxy . such a correlation is expected for a foreground - like dust geometry ( figure [ fig3 ] , panel d ; see discussion in * ? ? ? the slopes of the correlations in figure [ fig6 ] identify the reddening suffered by the stellar continuum , a@xmath124a@xmath125=[k@xmath126(@xmath127)-k@xmath126(@xmath128)]e(b@xmath26v)@xmath117 , with k@xmath126(@xmath24 ) the obscuration curve , while the total obscuration measurements of figure [ fig7 ] provide the zeropoint for k@xmath126(@xmath24 ) ( see , also , * ? ? ? the intrinsic starburst flux density f@xmath129 is thus recovered from the observed flux density f@xmath22 via : @xmath130 where the obscuration curve for the stellar continuum , k@xmath126(@xmath24 ) , is given by @xcite : @xmath131 the above expression folds into a single functional form a variety of effects : extinction proper , scattering , and the geometrical distribution of the dust relative to the emitters . it is derived from the spatially integrated colors of the entire stellar population in the starburst , and represents the ` net ' obscuration of the population itself . as mentioned above , the obscuration curve has mainly an absorption component , as the effects of scattering are averaged out . thus , it should not be confused with the extinction curve k(@xmath24 ) defined in section 2.1 . equation 8 implies that 1 mag of reddening between b and v for the ionized gas , a@xmath132a@xmath133=1 mag , corresponds to a@xmath134a@xmath135=0.44 mag of reddening for the stellar continuum . this fact can be alternatively expressed as : @xmath136 in other words , the stellar continuum suffers roughly half of the reddening suffered by the ionized gas ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * see , also , figure [ fig3 ] , panels e and f ) . although e(b@xmath26v)@xmath137 parametrizes the amount of reddening of the stellar continuum , it is not a color excess proper as in the case of individual stars , because of the dust geometry effects folded in the expression of k@xmath126(@xmath24 ) . one obvious characteristic of the obscuration curve is the absence of the 0.2175 @xmath0 m bump , which is a prominent feature of the mw extinction curve ( figure [ fig1 ] ) . dust / emitter geometries that can dilute the bump include mixed distributions with large optical depths @xcite . however , mixed geometries do not account for the full range of observational data ( figure [ fig3 ] ) , because the sources that contribute to the emerging sed are located at progressively lower optical depths @xmath38 for decreasing wavelength . in particular , the reddest uv slope produced by mixed geometries is @xmath138 @xcite , to be compared with @xmath139 observed in starburst galaxies ( figure [ fig6 ] , panel a ) . thus , the lack of the feature is probably intrinsic to the _ extinction curve _ of the starbursts @xcite , and may be due to the high uv energy densities that characterize these regions ( sections 2.1 and 4.2 ) . this section describes a dust model that attempts to reconcile two apparently contradictory facts : ( 1 ) stars in starbursts are on average a factor @xmath302 less reddened than the ionized gas ( equation 9 ) , and ( 2 ) both stars and ionized gas are affected by the same foreground - like dust distribution ( figure [ fig6 ] ) . a schematic representation of a dust / star / gas configuration that can at the same time account for both effects is shown in figure [ fig8 ] . stars are born within optically thick molecular clouds , and the short - lived , most massive stars remain closely associated with their parental cloud and the gas they ionize for their entire lifetime ( e.g. , * ? ? ? these stars and their surrounding gas will be generally highly attenuated . as the starburst population evolves , the inside of the region of star formation becomes depleted of dust @xcite . with energy densities @xmath140100 times higher than in the local ism , starburst environments are rather inhospitable to dust . shocks from supernovae destroy dust grains , via grain - grain collisions and sputtering @xcite . the average lifetime of a refractory grain is @xmath18@xmath9110@xmath78 yr , for a supernova rate of 0.05 yr@xmath5 and an ism mass of 5@xmath9110@xmath88 m@xmath37 ( e.g. , * ? ? ? * ) , about one order of magnitude or more shorter than the typical lifetimes of starbursts @xcite . in addition , hot - star - winds- and supernova - driven outflows evacuate both interstellar gas and dust from the region @xcite . r136 , the central cluster in 30 doradus , is an example of a few million years old cluster surrounded by an evacuating region @xcite . ionized gas separated , in projection , from the hot stars has also been observed in nearby starbursts @xcite . the evacuated ( clumpy ) dust will act as a foreground - like distribution for both the gas ( also at the edges of the region ) and for the central stars ( figure [ fig8 ] , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ; however , the ionized gas will be more obscured than the stars , because of its spatial location and closer association with the dust @xcite . this gas is more likely to contribute to the nebular emission observed at uv optical wavelengths than the gas in molecular clouds , because of its milder obscuration . in the presence of multiple generations of starburst populations , the long - lived , non - ionizing stars have time to ` diffuse ' into regions of lower dust density @xcite , as their native clusters are disrupted by evaporation or by the host galaxy s gravitational field @xcite . cool stellar populations in the lmc have , for instance , been observed to be less embedded in dust than hotter stellar populations @xcite . a similar trend is observed in the young ( @xmath1820 myr ) open clusters of the milky way @xcite . massive , non - ionizing stars can still produce significant uv emission . hence , the integrated uv optical stellar continuum from the combination of the ageing , diffusing populations and the starburst - embedded young populations will be less obscured than the emission from the nebular gas @xcite . for the stellar continuum and the ionized gas to have correlated obscuration values ( equation 9 ) , multiple starburst generations or long - lasting star formation events ( rather than ` instantaneous ' events ) are needed to attain an equilibrium between the number of stars drifting out of molecular clouds / starburst regions and the number of newly - born , dusty stars . starbursts hosted in dwarf galaxies are observed to have on - going star formation over @xmath51200 myr @xcite , and durations could be as long in more massive starburst galaxies @xcite . although the obscuration curve can be widely applied to uv - selected starbursts , it is not necessarily applicable to systems with different stellar population or ism characteristics . in the case of non - starburst galaxies there is no obvious mechanism for creating a foreground - dust - like geometry . in extreme starbursts as the ulirgs , the non - applicability of the obscuration curve has been demonstrated observationally ( * ? ? ? * and figure [ fig7 ] , panel a ) . in these galaxies , the high central concentration of gas and dust is likely to induce more extreme geometries than in uv - selected galaxies , and to produce combinations of clumpy mixed and foreground distributions . galaxies whose infrared properties are intermediate between the ulirgs and the uv - selected galaxies will progressively deviate from the dust geometry of the latter @xcite . dust in subsonic hii regions is likely to be mixed with the stars , rather than having been pushed away or destroyed as in starbursts , and scattering of stellar light into the line of sight may be an important component in the attenuation budget @xcite . because of the differences in the physical environment between the two types of objects , the foreground - dust - like approximation no longer holds for subsonic hii regions and equation 8 does not apply . differences in the star formation history can also contribute to the difference between subsonic hii regions and starbursts , with instantaneous bursts being a better description for the former and extended star formation for the latter ; thus , differential attenuation between ionized gas and stars is not necessarily expected in the hii regions . supersonic hii regions could be at the boundary between the two extremes just discussed ; one example is 30 doradus , that has evolved to the point of resembling a mini - starburst @xcite . to conclude , the starburst obscuration curve is a purely empirical result , and its derivation is independent of any assumption about the starburst s population , the dust geometry , or the details of the extinction curve . it provides statistical estimates of the reddening and obscuration of local uv - selected starbursts , and deviations as large as @xmath10.6 mag from the mean obscuration at 0.16 @xmath0 m are to be expected on an case - by - case basis . the physics of the starburst environment provides an explanation of the ` foreground - dust - like ' behavior of the curve , of the correlation between the reddening of the ionized gas and that of the stellar continuum , and of the correlation between reddening and total obscuration . as a sample , local uv - selected starbursts are characterized by modest amounts of reddening and obscuration , that are correlated with the star formation activity ( equation 6 ) . the median color excess is e(b@xmath26v)@xmath1410.35 @xcite , corresponding to uv continuum attenuations of a@xmath1421.6,1.4,1.1 mag at 0.15,0.20,0.30 @xmath0 m , respectively ( table [ tbl-4 ] and * ? ? ? * ; * ? ? ? observables like the uv slope , the colors , and the luminosities and equivalent widths of the hydrogen lines can be readily related to the dust - induced color excess of the ionized gas and , through equation 9 , of the stellar continuum . the uv slope is correlated with e(b@xmath26v)@xmath117 ( figure [ fig6 ] panel a ) , and equation 8 gives : @xmath143 where a list of @xmath144 values is given in the table of appendix b. the slope of the correlation between @xmath118 and e(b - v)@xmath117 is somewhat steeper than that measured by @xcite directly from the data ( 1.88@xmath1450.28 ) , but the two are within 1 @xmath72 of each other . it should also be noted that equation 10 becomes a lower envelope to the data points in figure [ fig6 ] , panel a , for @xmath1462.25 , a value appropriate for dust - free , star - forming populations . this indicates that variations of the dust / stars / gas geometry from the schematic model of figure [ fig8 ] and/or of the stellar population from the theoretical extreme @xmath1472.25 has the effect of reddening the uv slope even for e(b - v)@xmath1410 . the vertical scatter in figure [ fig6 ] , panel a should thus be interpreted as due to such variations from galaxy to galaxy . reddening expressions for optical colors can be directly derived from equation 8 ( see , also , figure [ fig6 ] , panels c and d ) . a mixed uv optical color , 0.28@xmath26v , could be useful to infer the reddening of intermediate redshift ( @xmath18z@xmath1480.6 ) starburst galaxies observed in b and i , and the reddening of one relative to the other is given by : @xmath149 under the assumption that the restframe uv emission and the nebular lines are both due to massive stars , the reddening of the uv continuum relative to the nebular hydrogen emission is given by : @xmath150 figure [ fig6 ] , panel e shows the comparison between data of local galaxies and equation ( 13 ) @xcite . while the agreement between the data in the first four panels of figure [ fig6 ] and equation 8 is a direct consequence of the fact that those data were used to derive the starburst obscuration curve , the match between data and expected trend in the bottom two panels of the figure is a consistency check , since it was not imposed a priori . one immediate consequence of the differential reddening between gas and stars is that the equivalent widths of the nebular lines depend on the amount of gas obscuration ; for the h@xmath42 and h@xmath41 emission lines the relations are @xcite : @xmath151= -0.64 e(b - v)_{gas},\\ \log [ ew(h\alpha)_{a}/ew(h\alpha)_{o}]= -0.40 e(b - v)_{gas},\end{aligned}\ ] ] where the attenuated and intrinsic ews are directly compared . comparison of these trends with the data points is shown in panel f of figure [ fig6 ] ; the trend given by the obscuration curve marks an upper envelope to the ensemble of observed ews , because the presence of older stellar populations underlying the starbursts dilutes the contrast between the emission lines and the stellar continuum in the data . statistically speaking , the uv slope is a good estimator of the total obscuration affecting the starburst ( * ? ? ? * and figure [ fig7 ] ) ; in particular , the ratio of the dust infrared luminosity , l@xmath93 , to the stellar uv or blue luminosity , l@xmath122 or l@xmath74 , is related to @xmath118 via : @xmath152 \simeq { a_{0.16 , star}\over 2.5 } = 0.84 ( \beta_{26}-\beta_{26 , 0}),\\ \log \biggl[{1\over 3.85 } \biggl({l_{ir}\over l_b}\biggr)_{star } + 1\biggr]\simeq { a_{b , star}\over 2.5 } = 0.43 ( \beta_{26}-\beta_{26 , 0}).\end{aligned}\ ] ] the two constant values on the left - hand - side of equations 17 and 18 , 1.68 and 3.85 , represent the bolometric corrections of the stellar emission relative to the uv or b. the correction for the b - band is larger and subject to more uncertainties than that for the uv , hence the larger scatter in the data point of figure [ fig7 ] , panel b. equation 17 can be rewritten as : @xmath153 = 1.76 e(b - v)_{gas},\ ] ] where the equation above represents a lower envelope to the data ( figure ( 7 ) , panel d and * ? ? ? * ) , and has a scatter comparable to that of figure [ fig6 ] , panel a. the nature of the scatter in the @xmath118 vs. e(b - v)@xmath117 and in the l@xmath93/l@xmath122 vs. e(b - v)@xmath117 relations is the same : it is due to variations from galaxy to galaxy of the starburst populations and of the dust / star / gas geometry . as discussed in the introduction , the energy balance between uv optical and infrared cosmic background suggests that there have been infrared luminous galaxies at all epochs . the number density of quasars was @xmath19100 times higher at z@xmath12.5 than today ( e.g. , * ? ? ? * ) and star formation activity was @xmath13 to 10 times higher at z@xmath511 than at z@xmath10 @xcite , flattening or decreasing only beyond z@xmath112 @xcite . if the association between activity and dust opacity observed in the local universe ( equation 6 ) holds at high redshift as well , galaxies in the past may have been as dusty as or dustier than today ( section 5.2 ) . among the methods described in section 3 , the multiwavelength comparison is the only practical approach for measuring the dust opacity of galaxies beyond the local few hundred mpc . however , because of observational difficulties , the wavelength coverage of the typical sample is limited to a few windows , often with little overlap from sample to sample . measures of the content , nature , and opacity effects of the dust in distant galaxies are therefore still at a very early stage , and the picture is vastly incomplete and controversial . the very first evidence for presence of dust in high redshift systems came from studies of damped ly@xmath42 systems ( dlas , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? dlas are the largest hi column density absorption systems seen in the spectra of background quasars , with n(hi)@xmath15410@xmath155 cm@xmath116 . their metallicities cover the range @xmath11/1001/10 solar for z@xmath10.54 , with a mean value @xmath11/13 solar at z@xmath10.53 and little evidence for evolution with redshift @xcite . selection biases against the dustiest , and most metal - rich dlas and problems with dust depletion corrections have been advocated as possible reasons for the observed weak metallicity evolution ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * see , however , @xcite ) . the dust fraction in dlas is measured by comparing the abundance of dust - depleted metals such as cr ii , fe ii , and si ii against that of a metallicity tracer such as zn ii , all of them producing rest - frame uv absorption lines @xcite . the dust - to - metals ratio in dlas is in the range 50%60% of the milky - way value ; this , combined with the generally low metallicities , indicates that dlas are relatively transparent systems , with dust - to - gas ratios between 2% and 25% of the milky - way value @xcite . the absence of the 0.2175 @xmath0 m dust absorption associated with dlas led @xcite to conclude that the extinction curve in these systems is probably similar to the one observed in the magellanic clouds . more recently , though , @xcite reported a @xmath13 @xmath72 detection of the dust absorption feature in the composite spectra of mgii absorbers . although presence of modest quantities of dust is well established in dlas , the weak or absent evolution of the metal content and of the cosmological mass density with redshift @xcite questions their relationship to luminous galaxies , and favors the interpretation that dlas preferentially trace galaxies with extended gas distributions and low metal and dust contents @xcite and the unevolved outskirts of galaxies @xcite . in recent years , facilities / instruments like iso and the sub - millimeter common user bolometer array ( scuba ) at the james clerk maxwell telescope ( jcmt ) have achieved high enough sensitivities in the mid and far infrared to detect _ dust emission _ beyond a few @xmath0 m from statistically significant samples of galaxies at cosmological distances . studies on the iso and scuba surveys , especially the identification of the optical / near ir counterparts to the infrared sources and the redshift measurements , are still ongoing and the results described below should be considered in many instances preliminary . iso surveys have yielded @xmath301,500 sources in the redshift range 0@xmath57z@xmath571.5 , down to a sensitivity limit of @xmath300.1 mjy in the mid ir ( 6.75 @xmath0 m , 12 @xmath0 m , and 15 @xmath0 m ) and @xmath300.1 jy in the far ir ( 90 @xmath0 m and 175 @xmath0 m , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? full lists of all surveys are given in @xcite and @xcite . the number counts at 175 @xmath0 m account for @xmath110% of the cib @xcite , showing that the bulk of the background has not been resolved yet at these wavelengths . neverthless , the number counts at mid and far ir wavelengths are about 310 times above the no - evolution extrapolation of the local iras counts @xcite . the implication is a strong luminosity evolution , l@xmath157(1@xmath76z)@xmath158 , or a combination of luminosity and density evolution of the infrared sources out to z@xmath11 @xcite , that is comparable to or stronger than the luminosity evolution observed at restframe uv wavelengths out to the same redshift @xcite . the iso counts are thus dominated by luminous infrared galaxies at z@xmath571 , with l@xmath159l@xmath16010@xmath100 h@xmath161 l@xmath37 @xcite . although this conclusion is based on the highly uncertain conversion of the 15 @xmath0 m emission1 the flux density at 15 @xmath0 m is centered on the restframe emission of the aromatic features that dominate the region between 6 @xmath0 m and 12 @xmath0 m ; the relation between the intensity of the features and the total infrared emission is still unsettled @xcite . ] to a total infrared luminosity , it is qualitatively supported by the optical counterparts of the iso sources . close to 100 mid ir sources , with median redshift @xmath18z@xmath1480.70.8 have been so far identified at optical / near ir wavelengths @xcite [ @xmath18z@xmath1480.10.2 for the 12@xmath0 m sample @xcite ] . many of the restframe optical spectra show characteristics of dusty luminous starbursts , with current bursts of star formation superimposed on a prominent population of a stars , and with attenuation values a@xmath1620.51.8 mag @xcite . at the time of writing , scuba surveys at 850 @xmath0 m of ` blank ' fields and targeted lensed objects have detected a total of @xmath30100 sources with flux densities in the range @xmath10.5 mjy to @xmath5110 mjy and redshifts between z@xmath11 and 3 , although sources as far away as z@xmath305 may be present @xcite . the number counts down to 2 mjy and to 0.5 mjy account for @xmath120%30% and more than 90% , respectively , of the cib @xcite . however , the latter fraction is potentially affected by small number statistics : so far only a few sources have flux densities below 2 mjy , the scuba confusion limit in blank fields , all of them from the survey of lensed galaxies @xcite . observations at 850 @xmath0 m probe the raileigh - jeans tail of the thermal dust emission of galaxies at least to redshift @xmath3057 . a source at z@xmath1541 and with an 850 @xmath0 m flux density of 2 mjy corresponds to a luminosity l@xmath1022@xmath9110@xmath103 h@xmath161 l@xmath37 , for typical seds of local infrared selected galaxies @xcite ; this is roughly the luminosity of a ulirg like arp220 . the few optical / near ir counterparts which have been identified are faint , often with i@xmath1924 and k@xmath512021 @xcite ; in most cases the ratio l@xmath93/l@xmath1630.05 , where l@xmath164 is the luminosity in the observer frame r band , and the infrared emission accounts for more than 95%99% of the bolometric luminosity . it is indeed widely accepted , also on the basis of the few available multi - wavelength seds , that the scuba sources are ` cosmological ulirgs ' @xcite , although this generalization has been recently questioned @xcite . the median redshift of the surveys is @xmath18z@xmath1652.5 @xcite ; however , @xmath1810% of all the sources have reliable redshift determinations from optical / near ir counterparts , and a total of @xmath3050% of the sample have photometric redshifts , most often derived from the radio / sub - mm spectral - index relation of @xcite . the comoving number density of scuba sources in the redshift range z=13 brighter than 6 mjy is @xmath135@xmath9110@xmath81 h@xmath166 mpc@xmath89 in our cosmology @xcite , about two orders of magnitude larger than the corresponding number density of local ulirgs brighter than 10@xmath103 l@xmath37 @xcite . a strong luminosity evolution , possibly as strong as l@xmath157(1@xmath76z)@xmath167 to z@xmath12 , is also implied by the number counts ( e.g. , * ? ? ? although the statistical properties of the sources are often inferred using the star - forming arp220 as a template , a fraction of agns is present in the sample @xcite . current best estimates , based on the x - ray luminosity of the sources favor an agn fraction @xmath120% @xcite ; this value will generally depend on the source luminosity range , as it does in the local universe @xcite . multi - wavelength observations of optically - selected samples of intermediate and high redshift galaxies are generally limited to the rest - frame uv and optical windows , with few exceptions ( e.g. , * ? ? ? * ; * ? ? ? such surveys suffer the same limitations of local uv optical samples , with the added complication of the often incomplete wavelength coverage . by probing only the _ differential attenuation _ , the uv optical data provide , in principle , lower limits to the actual dust opacity of the galaxy ( section 2.2 , but see also section 4 ) . in order to differentiate an ageing stellar population from dust as the cause of the sed reddening , a variety of approaches have been adopted by different authors . common diagnostics include the balmer line decrement h@xmath42/h@xmath41 , line - to - continuum ratios , e.g. h@xmath42/uv and h@xmath41/uv , and multi - color sed fitting . limitations to each of these diagnostics are discussed in section 2.2 and in the sections below , as appropriate . for a uv - selected sample of star - forming galaxies at z@xmath300.2 , @xcite and @xcite derive an average attenuation a@xmath1060.8 mag and find tentative evidence that those systems follow an obscuration sfr correlation similar to that observed in more local galaxies @xcite . the authors use a combination of uv ( 0.20 @xmath0 m ) , optical ( h@xmath42 and h@xmath41 ) and radio ( 1.4 ghz ) data to analyze their sample , although their reddening estimates are mostly based on the subsample for which they can measure h@xmath42/h@xmath41 ratios . if the starburst obscuration curve ( section 4 ) applies to z@xmath300.2 star - forming galaxies , an attenuation a@xmath1060.8 mag for the nebular line corresponds to a@xmath1681.3 and a@xmath1231.4 for the uv stellar continuum . although these figures agree with local measurements of the extinction at uv and h@xmath42 in star - forming galaxies ( section 3.6 and table [ tbl-3 ] ) , the uv selection of the sample potentially biases the reddening measures towards low values by excluding more heavily extincted galaxies . the z@xmath11 balmer break sample of @xcite is also a restframe - uv / blue selected one . exploiting the presence of the balmer discontinuity at @xmath10.365 @xmath0 m in galaxy seds to select z@xmath11 candidates ( a derivative of the lyman - break technique , see next section ) , these authors have obtained spectroscopic confirmation for @xmath30700 galaxies . the sample contains a large variety of galaxies , from active starbursts to quiescent star - forming to post - starbursts . if the z@xmath11 galaxies with blue ( @xmath1690.3 ) seds are analogs of local starbursts and follow the same ir / uv - versus-@xmath41 relation , they also follow an obscuration sfr relation similar to the local one , with a comparable range of attenuations ( figure [ fig9 ] , panel a ) . however , for fixed opacity , the z@xmath11 galaxies tend to have on average brighter uv@xmath76ir luminosities ( larger sfrs ) than the z@xmath10 galaxies . thus , even taking selection effects into account , the distant galaxies are , on average , more actively star - forming for the same amount of obscuration than the local starbursts @xcite . @xcite , @xcite , and @xcite use restframe h@xmath42 observations and h@xmath42/uv comparisons to infer dust reddening in star - forming galaxies at redshift z@xmath112.2 . the i - band selected sample of @xcite corresponds to restframe @xmath1v for a z=1 galaxy : the longer wavelength selection of the samples mitigates , although does not remove completely , the bias against dust obscured galaxies introduced by uv selections [ for the canada - france redshift survey employed by @xcite , see @xcite , and for the sample used by @xcite , see @xcite ] . the presence of dust reddening in the z@xmath112.2 samples is derived from the discrepancy between the measured l@xmath170/l@xmath171 luminosity ratio and the ratio predicted by dust - free models / l@xmath172 ratio ; the small extrapolation to l@xmath170/l@xmath171 is done here assuming a constant star formation population . ] . figure [ fig10 ] compares the range of attenuated to intrinsic values ( l@xmath170/l@xmath171)@xmath173/(l@xmath170/l@xmath171)@xmath174 derived from data by the three groups of authors / l@xmath171)@xmath174 is an average of h@xmath42 and uv luminosities for a variety of imfs , star formation histories , and metallicities of stellar populations , see table 3 in @xcite . ] with predictions from the dust models of section 2.2 . geometries where dust and stars are homogeneously mixed , either in spheroidal or flattened distributions , appear less effective in reproducing the observed h@xmath42/uv ratios than foreground geometries ( figure [ fig10 ] ) . if the foreground and starburst distributions bracket the possible range of dust / emitter morphologies for the z@xmath112 h@xmath42 galaxies , the front - to - back optical depths are @xmath470.64.9 mag ( figure [ fig10 ] , bottom panel ) , and the effective attenuations are a@xmath1060.64.3 mag , a@xmath1751.55 mag , a@xmath12337 mag , for smc extinction . mw - like and smc - like extinction curves give results that are less than 0.1 mag apart at @xmath1760.28 @xmath0 m , but @xmath11.2 mag different at @xmath24=0.16 @xmath0 m , with larger obscurations given by the smc extinction curve . uncertainties in the metallicity , imf , and star formation history of the galaxy stellar populations introduce a factor @xmath512 additional scatter on the inferred opacities ( table 3 in * ? ? ? the major limitation is , however , that dust opacities measured from the current h@xmath42 surveys of galaxies are probably not representative of the z@xmath112 population as a whole . by number , @xcite s and @xcite s samples are small ( 13 galaxies and 5 galaxies , respectively ) ; @xcite s sample , albeit 2.5 times larger than @xcite s , does not measure the h@xmath42 emission on the same galaxies for which uv is available . for such small samples sensitivity limits induce a strong selection effect : the observed galaxies will be among the brightest in h@xmath42 and , in addition , contamination from agn / seyfert nuclei is not excluded @xcite . therefore , such surveys sample the high end of the activity and dust obscuration ranges for optically - detectable systems at z@xmath112 ( * ? ? ? * ; * ? ? ? * and equation 6 ) . in summary , the limited information available on galaxies at z@xmath301 suggests that dust opacities cover a range similar to that of local star - forming and starburst galaxies ; they may follow a similar relation a@xmath177(sfr)@xmath178 ( see equation 6 ) , with the proportionality constant revised downward relative to the local sample . to date , the surveys that have most efficiently secured large populations of galaxies beyond z@xmath12 are those that exploit the lyman discontinuity at restframe 0.0912 @xmath0 m to identify high - redshift candidates ; these are the so called lyman break galaxies ( lbgs , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? so far , there are @xmath11,000 spectroscopically confirmed lbgs at z@xmath13 , and @xmath170 at z@xmath14 . the lbg population is a substantial component of the high redshift `` zoo '' : their comoving number density at z=3 is n(z=3)@xmath13@xmath9110@xmath116 h@xmath166 mpc@xmath89 @xcite , comparable to the density of local galaxies @xcite . by selection , i.e. due to the need of presenting a relatively blue continuum longward of ly@xmath42 , this high redshift population tends to be biased against systems with predominantly old populations , or systems with large amounts of dust extinction . indeed , most lbgs show evidence for ongoing or recent star formation activity @xcite . from a purely phenomenological point of view , lbgs resemble local uv - selected starburst galaxies ( section 4 ) . the obscuration corrected sfrs per unit galactic area are @xmath10.225 m@xmath37 yr@xmath5 kpc@xmath116 , for a 0.1100 m@xmath37 salpeter imf @xcite , corresponding to @xmath10.5%50% the maximum sfr per unit galactic area measured in the local universe @xcite ; this is similar to the observed range in nearby uv - selected starbursts . restframe uv spectra show a wealth of absorption features , and often p - cygni profiles in the civ(1550 ) line @xcite , which indicates the predominance of young , massive stars . nebular emission lines observed in the brightest lbgs @xcite have intensities comparable to those of local starbursts @xcite . the observed blueshifts in the uv interstellar absorption lines relative to the systemic redshifts of lbgs have been interpreted as bulk gas outflows with velocities v@xmath1200500 km s@xmath5 @xcite , again very similar to what is observed in nearby starbursts @xcite . finally , more luminous lbgs tend to be dustier ( figure [ fig9 ] , panel b and equation 6 ; * ? ? ? * ; * ? ? ? however , dissimilarities also exist between the two classes of galaxies . star formation extends over a large fraction of the galactic body in lbgs , typically 530 h@xmath161 kpc@xmath113 @xcite , while it is confined within the inner kpc@xmath113 in local starbursts . as a result , lbgs are , on average , more rapidly star - forming than either z@xmath10 or z@xmath11 galaxies ( figure [ fig9 ] ; * ? ? ? * ; * ? ? ? the bulk outflows in lbgs have velocities that are similar to those of local infrared bright starbursts like ngc6240 and a few times larger than those of hii galaxies @xcite . the implications of such differences are not fully understood yet . establishing how much dust lbgs contain is the subject of ongoing debate . metals are definitely present in these systems . although the exact value of the metallicity has not been constrained , it most likely lies between 0.1 z@xmath37 and 0.7 z@xmath37 @xcite . the uv spectral slopes of lbgs have median value @xmath1791.5 @xcite , significantly redder than the value @xmath1802.2@xmath262.7 expected for a dust - free star - forming population ( appendix b ) . if interpreted in the same fashion as the uv slope of nearby uv - selected starbursts , lbgs suffer from dust obscuration at the level of a@xmath12312 mag @xcite , and , possibly , more @xcite . however , a number of uncertainties and systematics limit the usefulness of uv data alone to derive obscuration estimates . first and foremost , the high redshift galaxy for which reddening corrections are sought needs to possess active , on - going star - formation ( i.e. , not to be a post - starburst ) ; this appears to be true for the lbgs as a population , but may not be true on a individual basis @xcite . second , for most lbgs the uv slope is derived from photometry alone ; the conversion of g@xmath181 colors filters have central wavelengths 0.483 @xmath0 m and 0.693 @xmath0 m , respectively @xcite . ] to @xmath118 suffers from uncertainties due to the presence of both intrinsic ( stellar lines and lyman break ) and intervening ( lyman@xmath42 forest ) absorption in the photometric filters @xcite . finally , because of the age dust degeneracy , the intrinsic uv slope is a free parameter that can lead to @xmath191 mag difference in the derived obscuration value at 0.16 @xmath0 m ( compare * ? ? ? * ; * ? ? ? supplementing the uv with restframe optical data provides further leverage for determining the level of dust reddening in lbgs . the negligible correlation between l@xmath182/l@xmath122 and the uv slope found for a sample of @xmath5715 luminous lbgs(0.5007 @xmath0 m ) emission line @xcite are not discussed because of the strong dependence of the nebular line intensity , up to one order of magnitude variation , on the chemical and physical conditions of the gas @xcite . ] supports the notion that lbgs are subject to modest reddening @xcite . although this result is still dominated by the large scatter induced by both observational and model uncertainties @xcite , it is consistent with what is observed in local uv - selected starbursts1.5 corresponds to obscurations a@xmath1231.2 mag and a@xmath1830.50 mag , and a differential obscuration between the two of @xmath10.7 mag ; the latter is probably too large to be reconciled with observational data . highly inclined disks are not considered in the obscuration budget , as they would probably be too dim in the uv to be selected as lbgs . ] , for which an obscuration of 12 mag at 0.16 @xmath0 m corresponds to the modest reddening a@xmath184a@xmath1830.150.3 ( section 4.3 ) . although the age dust degeneracy can not be easily broken even with the addition of long wavelength data , extending the wavelength coverage of lbgs to the restframe optical has helped delimiting the locus of allowed combinations in the age dust plane @xcite . using complementary datasets of lbgs in terms of luminosity range , @xcite and @xcite have established that there is an inverse correlation between the age of the stellar populations in a galaxy and the amount of dust reddening it suffers : the older the galaxy , the lower its reddening . in agreement with previous results @xcite , lbgs whose seds are compatible with young ( @xmath5710@xmath88 yr ) stellar populations are also dusty . whatever dust is present , it is not preferentially obscuring selected regions of the galaxies , because of the similarity between the restframe uv and optical morphologies @xcite . a typical lbg obscuration of a@xmath12312 mag , with a median of @xmath11.5 mag and a high - end tail around 4.55 mag @xcite , is currently the most broadly accepted value . despite all the uncertainties in estimating dust obscuration in lbgs , the inferred opacity values are modest enough that the high redshift galaxies are expected to be faint sub - mm emitters , with 850 @xmath0 m flux densities at or below the scuba detection limit of @xmath112 mjy @xcite . out of @xmath5110 lbgs targeted with scuba , only a couple have been detected @xcite . table [ tbl-5 ] gives three examples ( two lbgs and one scuba source ) of highly reddened z@xmath13 galaxies , plus the average values of a typical lbg . the three examples are meant to show some of the most attenuated cases that are also detected at restframe uv . table [ tbl-5 ] demonstrates that the exact prediction of the flux density at restframe @xmath30200 @xmath0 m is heavily dependent on the adopted dust temperature(s ) and emissivity , because long infrared wavelengths only probe the raileigh - jeans tail of the thermal dust emission in galaxies ( figure [ fig11 ] ) . local galaxies , although extensively used as analogs of high redshift galaxies , are of limited help in this respect , given the wide range of properties they show . if the flux density of a local galaxy beyond 6070 @xmath0 m is parametrized by a modified blackbody with dust emissivity @xmath185 : @xmath186 ulirgs and bright ir - selected galaxies tend to be well described by either a single - temperature component with t@xmath140 k and @xmath1871.5 @xcite or by a warm / cool two - temperature model with @xmath188=2 @xcite ; however , there is tentative indication that uv bright , metal - poor starbursts have hotter dust seds , with t@xmath150 k and @xmath188=2 ( tol1924416 , * ? ? ? equation 20 , which is used for purely illustrative purposes , is clearly an over - simplification of the dust sed in local galaxies , which is characterized by a range of temperatures and , at restframe @xmath294070 @xmath0 m , by the emission from non - equilibrium dust grains ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * see , also , section 2.1 ) . in starbursts , the latter component represents @xmath125% of the total infrared emission . given the uncertainty of the sub - mm measures and the current poor understanding of the detailed physics of dust emission in both local and high - redshift galaxies , the opacity of lbgs is still relatively unconstrained by sub - mm observations ( * ? ? ? * ; * ? ? ? * and table [ tbl-5 ] ) . further complications ( and uncertainties ) arise when the comparison between predictions and observations of infrared flux densities is made for individual objects @xcite ; reddening / obscuration corrections at uv - optical wavelengths are accurate only in a statistical sense and , generally , do not apply on a one - to - one basis to individual objects . for instance , if the dust geometry in a particular system is such that the scattering out of the line of sight is larger than that into the line of sight , the infrared energy , which is due to the absorption part of the extinction only , will be overestimated when derived from uv data . there is so far limited overlap between infrared and optical surveys at intermediate and high redshift ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? consequently multi - wavelength coverage from the restframe @xmath1uv optical to the infrared is available only for a small number of galaxies . to make matters worse , observational limitations lead to vastly different selection criteria for the optical surveys in different redshift bins @xcite . thus , conclusions about the dust opacity of galaxy populations as a whole and as a function of redshift are at the moment based on ` reasonable ' extrapolations of existing data , rather than hard evidence . available infrared and optical data support the scenario that z@xmath301 star forming galaxies cover at least the same range of dust opacities as local galaxies . the luminosity ( or combined luminosity and density ) evolution implied by the iso counts suggests that luminous infrared galaxies , with l@xmath159l@xmath16010@xmath100 h@xmath161 l@xmath37 , were common at z@xmath571 . the range of dust attenuation values at uv wavelengths inferred for optically - selected z@xmath11 galaxies is comparable to those of local disk galaxies and uv selected starbursts ( figure [ fig9 ] ) , although the distant galaxies tend to be more luminous ( more active ) on average . larger samples with more extended wavelength coverage are needed before a more complete picture can be drawn in this redshift range . the ism of z@xmath13 galaxies has been observed to be metal - polluted to at least @xmath300.1 z@xmath37 , implying that non - negligible amounts of dust must be present in the galaxy population at large by z=3 and , in many cases , earlier @xcite . the observed properties of the z@xmath191 scuba sources and of the lbgs suggest that both objects are the high redshift analogs of local actively star - forming galaxies , where dust content and level of activity determines a continuum of characteristics . the scuba sources are at the high end of such continuum , with powerful activity , both thermal and non - thermal , and large dust contents that make them similar to local ulirgs ; more than 95% of their bolometric light is absorbed by dust . lbgs appear to span a larger range of the continuum of dust content and activity level , although they tend to have , on average , lower values than the scuba sources . typical uv attenuations are a@xmath12312 mag , corresponding to 35%95% of the bolometric light being absorbed by dust . assuming the local infrared luminosity function can be applied to high redshift sources as well , the comoving number density of z@xmath303 scuba galaxies with 850 @xmath0 m flux densities brighter than 2 mjy and 0.5 mjy ( corresponding to @xmath12@xmath9110@xmath103 l@xmath37 and @xmath15@xmath9110@xmath100 l@xmath37 for an arp220-like sed ) would be @xmath304@xmath9110@xmath82 h@xmath166 mpc@xmath89 and @xmath3010@xmath116 h@xmath166 mpc@xmath89 , respectively . high - redshift infrared galaxies as bright as arp220 are almost 100 times less numerous than lbgs ; for comparison , in the local universe ulirgs are more than 4 orders of magnitude less numerous than optically selected galaxies . at lower 850 @xmath0 m luminosities , z@xmath13 scuba sources are about as numerous as lbgs at the same redshift , but there could be large overlap between the two samples at the mjy level . lbgs are , in fact , predicted to occupy the faint end of the scuba 850 @xmath0 m source counts , with flux densities of a few mjy or less @xcite . thus , although scuba sources down to 0.5 mjy at all redshifts already account for @xmath1990% of the cib , a non - negligible fraction of this , at least 25% , could be due to lbgs @xcite . intracluster and intergalactic dust are briefly discussed in view of their potential importance in the interpretation of the intermediate and high redshift universe . the amount , nature , and sources of the dust in the intracluster ( icm ) and intergalactic ( igm ) media are still a controversial matter , despite the large number of independent researches tackling the issue . the metallicity of the present - day icm is fairly high , about 1/31/2 solar , implying that a non - negligible fraction of the gas in the icm has been ejected from the cluster s galaxies @xcite . the high metallicity is not automatically accompanied , however , by a high dust content of the icm . the icm hot gas effectively destroys dust through sputtering . the typical timescale for sputtering in a @xmath5110@xmath87 k hot ambient gas is t@xmath18910@xmath78a / n , where t@xmath190 is measured in yrs , a is the size of the dust grain in @xmath0 m , and n is the density of the icm in cm@xmath89 @xcite . in the high - density cores of clusters ( n@xmath5110@xmath89 cm@xmath89 ) , a 0.1 @xmath0 m dust grain survives a few times 10@xmath88 yr , with smaller grains being destroyed on shorter timescales than the large grains @xcite . small grains can survive only outside the cores of clusters , i.e. , beyond the central @xmath30mpc . dust needs to be constantly replenished in the icm , and the likely sources are enriched winds from the cluster s galaxies , intracluster stars , and , in dynamically young clusters , winds from infalling spiral galaxies @xcite . the timescales for the dust injection in the icm are of order 10@xmath90 yr or longer , implying that the grain size distribution , and therefore the shape of the extinction curve , depends mainly on the sputtering efficiency at the selected cluster location @xcite . various direct and indirect methods have been employed to detect dust in the icm . deficits in background counts of galaxies , quasars , and clusters behind galaxy clusters of various richness have been interpreted as extinction in the range a@xmath670.20.5 , with patches that can exceed a@xmath1921 mag @xcite . conversely , reddening measurements of cluster members @xcite or background galaxies @xcite give upper limits of a@xmath670.2 mag . the two methods thus give contradictory results and neither is completely bias - free : deficits in background counts can be due not only to dust , but also to confusion ; low reddening values can be attributed to a gray extinction curve , greyer than a milky - way - like curve , because of the selective destruction of small grains in the icm . tentative infrared detections of icm dust emission , using iras and iso data @xcite , are either at low significance level , or have been argued to be due to emission from cluster members @xcite . presence of dust is well established in the centers of cooling flow clusters ( see discussion in section 3.4 ) , but the issue of its origin is still controversial . in particular , it is not clear whether this is an extended component , such as condensates from the cooled cluster gas or a by - product of star formation triggered by the cooling flow , @xcite , or is directly associated with the central galaxy , possibly from a merging event that the galaxy underwent @xcite . the milky - way - like form of the extinction curve in the central galaxies has been extensively used as an argument to support the galactic origin of the dust in cooling flow clusters ; as remarked above , icm - processed dust would tend to be depleted in small grains and therefore produce a gray extinction @xcite . however , if soft x - ray absorption is present in the centers of the clusters , dust absorption in the cooling flow would appear to be the direct physical explanation @xcite . the tentative detection of the neutral oxygen edge , and thus of dust , in the perseus cluster @xcite and the widespread presence of co emission in low - radio - emitting cooling flow clusters @xcite are countered by xmm data that do not support the presence of large amounts of cool gas in the center of some of the clusters and , therefore , do not require a large amount of dust to explain the soft x - ray spectral shape @xcite . at present times , the presence and nature of dust in the igm is even more uncertain than that in the icm , though it is widely accepted that the igm is metal - enriched . hydrodynamical cdm simulations identify the ly@xmath42 forest as the igm , i.e. , modest baryon density fluctuations in a ionizing background @xcite . the ly@xmath42 forest accounts for a significant fraction , @xmath1950% , of the baryons at z@xmath303 @xcite ; its metallicity at z@xmath12.53 is @xmath10.21% z@xmath37 down to column densities n(hi)@xmath110@xmath193 cm@xmath116 @xcite , while metallicity values at lower column densities remain uncertain @xcite . these values could be lower limits to the igm metallicity , if there is a ` missing metals ' problem as pointed out by @xcite and @xcite on the basis of the metals produced by star formation by z@xmath12.5 . the metallicity of the ly@xmath42 forest increases for decreasing redshift , being @xmath110 times more metal rich at z@xmath10.5 than at z@xmath12.53 @xcite . in the local universe , the ly@xmath42 forest accounts for only @xmath3020% of the baryons @xcite and matter associated with galaxies and clusters for another @xmath3025%30% @xcite ; thus , @xmath150% of the baryons at z=0 , and possibly of the metals , are contained in a yet undetected phase of the igm , proposed to be hot gas @xcite . various mechanisms have been considered for the metal , and possibly dust , pollution of the igm ; they include supernova ejecta from galaxies and/or subgalactic structures , dynamical encounters between galaxies , ram - pressure stripping , and population iii stars @xcite . the different mechanisms are non - exclusive , although each one could be dominant relative to the other at specific cosmological epochs . early attempts to detect diffuse dust in the igm investigated the evolution of the mean quasar spectral index with redshift @xcite , with the basic result that presence of igm dust is negligible for milky - way - like extinction , up to at least z@xmath302 @xcite . milky - way - like igm dust should also leave an imprint on the cosmic microwave background , as it would re - emit in the sub - mm / mm regime the absorbed background radiation @xcite . the tight cobe upper limit to deviations from blackbody emission @xcite limits the igm dust to a@xmath1940.3 mag to z=5 , with negligible value to z=3 @xcite . however , if sputtering is efficient in the hot igm , dust has a greyer extinction than milky - way - like ( * ? ? ? * for a challenge to the sputtering argument see @xcite ) . gray igm dust can be present in larger quantities than milky - way - like dust without violating the cobe limits , but its emission would represent @xmath5175% of the cib at 850 @xmath0 m @xcite . source counts with scuba account already for @xmath1990% of the cib ( section 5.2 ) , which challenges the case for large amounts of gray dust . so far , the strongest limits have come from observations of cosmological supernovae . for gray dust with minimum grain size @xmath10.1 @xmath0 m , @xcite s model gives a@xmath74-a@xmath1940.01 and a@xmath74-a@xmath500.2 for a@xmath860.20.3 to z=0.5 ; at z=2 , a@xmath860.60.7 mag . using observations of type 1a supernovae at z@xmath10.5 and z@xmath11.7 , @xcite and @xcite have placed a limit of a@xmath860.10.15 to the amount of diffuse , gray igm dust to z=0.5 . in summary , although the actual numbers are still not settled , igm dust is likely to produce small attenuations at most redshifts . current observational data indicate that the population of local galaxies is characterized by modest values of the dust opacity . typically , 50% or less of the total stellar energy is absorbed by dust . the same evidence points out that the definition of ` modest opacity ' is wavelength - dependent ; for example , in the uv , at 0.15 @xmath0 m , a non active galaxy can lose up to 80% of its energy to dust . luminous , intermediate to late type spirals are dustier than ellipticals , by @xmath11.01.4 mag in the uv , and than irregulars , by 0.50.8 mag . luminous galaxies are on average more metal rich and dust rich than their faint counterparts . dust opacity also increases with thermal and non - thermal activity , with the high - end of the distribution given by the ulirgs . although the wealth of data and wavelength coverage for local galaxies has started to uncover the dust opacity of these systems , we are still missing a few crucial pieces of the puzzle . the largest among these is the one - to - one relation between absorbed stellar seds and dust emission seds . at first order , hotter stars are expected to heat the dust to higher temperatures than cooler stars @xcite , but we do nt have yet the direct link between individual dust emission and heating components ; we do nt know the dependence of the dust sed on parameters like metallicity , radiation field , gas density , galactic environment , etc . ; and questions still surround the relation between radiation intensity and pah emission , despite the progresses made with iso @xcite . the main reason for these limitations to our comprehension of the dust absorption emission relationship is the large mismatch in angular resolution between uv - to - near ir observations and current mid / far infrared data . while observations probing the direct stellar light can easily reach sub - arcsecond resolution , existing data on dust emission are at the arcminute resolution level , especially at wavelengths above @xmath140 @xmath0 m where the bulk of the energy from dust emerges . upcoming infrared missions like sirtf ( the space infrared telescope facility ) , sofia ( the stratospheric observatory for infrared astronomy ) , and first ( now the herschel space observatory ) promise to bridge this gap . they will also push the wavelength coverage all the way to 600700 @xmath0 m , thus probing the cold dust emission in local galaxies , which is the repository of most of the dust mass . uv and optical selected , intermediate and high redshift , galaxy surveys carry the inherent limitation that physical parameters directly derived from the observations are affected by dust in a non - easily quantifiable way . this is further complicated by the fact that instrumental sensitivity limits favor detection of the luminous representatives of any galaxy class , that may tend to have relatively high dust content . estimates of the sfrs in intermediate and high redshift galaxies using balmer lines @xcite and infrared emission @xcite are generally a few times higher than the analogous estimates using the restframe uv emission @xcite . as discussed in section 5 , presence of dust in galaxies at cosmological distances is reasonably established , but the quantification of its effects as a function of redshift is still an ongoing process . linked to this is the puzzle of the nature of the uv selected and ir selected sources , and of their relationship : whether they represent overlapping populations , or the two ends of a continuum of properties possibly parametrized by activity and dust content is still an open question . the absence of full characterization of the dust emission and absorption in distant galaxies is mostly due to the scant overlap between optically selected and infrared selected surveys . this is due to sensitivity limitations at both ends of the spectrum , to angular resolution limitations at long wavelengths , to the small number of objects in some of the samples , and to the sparse wavelength coverage of the dust emission sed by current sub - mm / radio instrumentation . sirtf and first initially , and then alma ( the atacama large millimeter array ) and ngst ( the next generation space telescope ) will be able to fill the gaps . in particular the last two instruments , with their sub - arcsecond resolution and high sensitivities , should mark a break - through in our understanding of the high redshift universe and of its stellar , gas , and dust content up to z@xmath120 . the author is grateful to mark dickinson , megan donahue , paul goudfrooij , claus leitherer , and mark voit for insightful comments on specific sections of this review . she thanks kurt adelberger for data on the lyman break and balmer break galaxies , and max pettini and chuck steidel for the uv spectrum of ms1512-cb58 . she is also indebted with veronique buat , bruce draine , tim heckman , nino panagia , marcia rieke , and max pettini for a critical reading of the manuscript , which has greatly improved its content and presentation . the dimming effects of light are generally termed ` extinction ' only in the simple geometrical case of a point source behind a dust screen ( section 2.1 ) , where the attenuation suffered by the source can be directly related to the optical depth of the dust screen . in this review , the extinction at wavelength @xmath24 is indicated as : @xmath195 ( equation 2 ) . the dimming of background galaxies due to the dust in a foreground galaxy can generally be considered extinction , because the foreground galaxy can be modeled in first approximation as a homogeneous screen in front of point sources ( section 3.1 ) . for more general distributions of dust and emitters , the dimming and reddening of the light is no longer directly related to the extinction or the optical depth of the dust layer , and is called indiscriminately ` effective extinction ' , ` obscuration ' , or ` attenuation ' . the effective extinction a@xmath7 is defined as : @xmath196 where m@xmath197 is the magnitude of the dust free source and m@xmath198 is the measured ( attenuated ) magnitude ; a@xmath7 measures the magnitudes of attenuation suffered by an extended source at the specified wavelength / band . by definition , a@xmath199a@xmath25 for a foreground , homogeneous , non - scattering dust screen . for dust / emitter distributions that do not have spherical symmetry , such as spiral galaxies , the face - on effective extinction is termed a@xmath200 , while a@xmath7 indicates the effective extinction averaged over all inclinations . throughout this review , two different definitions of the uv spectral slope , @xmath41 and @xmath118 have been used . in this context , the spectral slope is derived from the fit f(@xmath24)@xmath201 within a specified wavelength range , with f(@xmath24 ) the flux density of the source . the various definitions reflect historical reasons driven mostly by observational constraints . the default @xmath41 adopted in this review ( section 2.2 ) is measured between 0.125 @xmath0 m and 0.30@xmath0 m , using a set of wavelength windows that avoid the strong stellar and interstellar absorption features . an example of such windows is given in @xcite , although the ones used to derive @xmath41 here are better tuned for use with model spectra rather than with observed spectra , and extend all the way to 0.3 @xmath0 m . this definition of @xmath41 has been adopted to ensure an appropriate comparison between different extinction curves and dust models , as in figures [ fig2 ] and [ fig3 ] , because the fitting windows are selected to avoid the strong and wide dust feature at 0.2175 @xmath0 m . the uv slope @xmath118 used in section 4 is a fit through the 10 windows of @xcite , and corresponds to deriving the slope in the wavelength range 0.1250.26 @xmath0 m , using spectral regions mostly devoid of strong absorption lines . this definition of the uv slope is observation - driven , as it was used to obtain measurements from iue spectra of a local uv - selected starburst sample . the wavelength range is bluer than that of @xmath41 above , and was chosen by @xcite to avoid including in the fits the older stellar populations underlying the starbursts . these older populations can contribute to the spectra starting around 0.270.28 @xmath0 m , when the fluxes of f - stars become non - negligible . for reference , a list of @xmath118 values for dust - free stellar populations , both for instantanous burst and for constant star formation , is given in table [ tbl-6 ] . an convenient definition of the uv - slope for studies of high - redshift galaxies is @xmath202 , measured roughly between @xmath10.12 @xmath0 m and @xmath10.18@xmath0 m . at z@xmath13 , the optical window give information shortward of @xmath10.2 @xmath0 m , thus forcing the uv slope fitting to a much shorter range than possible with local galaxies . the details of the actual wavelength range vary with redshift and with the observing technique ( whether spectroscopy or imaging , and depending on the filters used ) . the relation between the three quantities , as given by models of stellar populations reddened by the starburst obscuration curve of section 4 , is : @xmath203 @xmath204 the error bar following each coefficient refers to the maximum variance observed by varying a number of parameters in the models and fits : ( a ) stellar populations from both @xcite and @xcite models ; ( b ) stellar populations with constant star formation in the range 51000 myr , and instantaneous bursts in the age range 512 myr ; ( c ) both 0.2 z@xmath37 and z@xmath37 metallicities ; ( d ) fitting windows that progressively include or exclude weak absorption lines in the spectra . the latter turns out to be the main driver of the variance , while changes in the characteristics of the stellar populations have only a small impact . the relations between the three slopes would , however , change from equations ( b1)(b2 ) if more extreme cases of stellar population mix , such as the addition of an old ( age@xmath191 gyr ) component , were considered ; in this case , @xmath41 and @xmath118 would be strongly affected , because of the long wavelength baseline . of interest for high - redshift studies is the consideration that while @xmath118 and @xmath41 follow each other closely , @xmath202 is always redder ( more positive ) than the other two , and the discrepancy increases for increasing value of the slopes . the reason for this counter - intuitive behavior is the presence of the well - known ` iron curtain ' at long uv wavelengths : this is a wide dip in the 0.230.28 @xmath0 m range of the uv spectra due to the presence of a large number of closely spaced fe absorption lines @xcite . the uv slopes fitted with the longer wavelength baseline have windows in this continuum - depressed region , hence their more negative values . increasing dust obscuration affects the short - wavelength slope more than the long - wavelength ones ; hence the trend for @xmath202 to become progressively redder than @xmath118 as their values increase ( i.e. , as the uv spectra become more dust reddened ) . @xcite derive an empirical relation between @xmath118 and a short - wavelength uv slope @xmath205 ( close , but not identical , to @xmath202 ) from iue spectra of the uv - selected starbursts . their findings , that @xmath118=@xmath2060.16 , are consistent with the long wavelength slope being bluer than the short wavelength one . however , they do not measure an increasingly redder @xmath205 for increasing values of both slopes . a possible reason is that noise in the iue data may have prevented the detection of such effect . in this review , it is assumed that equations ( b1)(b2 ) are valid for uv - selected starbursts . adelberger , k.l . , & steidel , c.c . , 2000 , , 544 , 218 adelberger , k.l . , steidel , c.c . , giavalisco , m. , dickinson , m.e . , pettini , m. , & kellogg , m. 1998 , , 505 , 18 aguirre , a. 1999 , , 525 , 583 aguirre , a. , & haman , z. 2000 , , 532 , 28 aguirre , a. , hernquist , l. , schaye , j. , katz , n. , weinberg , d.h . , & gardner , j. 2001 , , accepted ( astroph/0105065 ) allen , s.w . 1995 , , 276 , 947 aloisi , a. , tosi , m. , & greggio , l. 1999 , , 118 , 302 alonso - herrero , a. , engelbracht , c.w . , rieke , m.j . , rieke , g.h . , & quillen , a.c . 2001 , , 546 , 952 altieri , b , metcalfe , l. , kneib , j.p . , mcbreen , b. , aussel , h. et al . 1999 , , 343 , l65 alton , p.b . , trewhella , m. , davies , j.i . , evans , r. , bianchi , s. , gear , w. , thronson , h.a . , valentijn , e. , & witt , a.n . , 1998 , , 335 , 807 arimoto , n. , sofue , y. , & tsojimoto , t. 1996 , pasj , 48 , 275 arnaud , k.a . , & mushotzky , r.f . 1998 , , 501 , 119 aussel , h. , cesarsky , c.j . , elbaz , d. , & starck , j.l . 1999 , , 342 , 313 baker , a.j . , lutz , d. , genzel , r. , tacconi , l.j . , & lehnert , m.d . 2001 , , in press ( astroph/0104345 ) barger , a.j . , cowie , l.l . , mushotzky , r.f . , & richards , e.a . 2001 , , 121 , 662 barger , a.j . , cowie , l.l . , & richards , e.a . , 2000 , , 119 , 2092 barger , a.j . , cowie , l.l . , & sanders , d.b . 1999a , , 518 , l5 barger , a.j . , cowie , l.l . , sanders , d.b . , fulton , e. , taniguchi , y. , sato , y. , kawara , k. , & okuda , h. 1998 , nature , 294 , 248 barger , a.j . , cowie , l.l . , smail , i. , ivison , r.j . , blain , a.w . , & kneib , j .- 1999b , , 117 , 2656 barlow , t.a . , & tytler , d. 1998 , , 115 , 1725 baugh , c.m . , cole , s. , frenk , c.s . , & lacey , c.g . 1998 , , 498 , 504 beck , s.c . , turner , j.l . , ho , p.t . , lacy , j.h . , & kelly , d.m . 1996 , , 457 , 610 beckman , j.e . , peletier , r.f . , knapen , j.h . , corradi , r.l.m . , & gentet , l.j . 1996 , , 467 , 175 bell , e.f . 2001 , private communication bell , e.f . , gordon , k.d . , kennicutt , r.c . , & zaritsky d. 2001 , in prep . bell , e.f . , & kennicutt , r.c . 2001 , , in press ( astroph/0010340 ) berlind , a.a . , quillen , a.c . , pogge , r.w . , & sellgren , k. 1997 , , 114 , 107 bianchi , l. , clayton , g.c . , bohlin , r.c . , hutchings , j.b . , & massey , p. 1996 , , 471 , 203 bianchi , s. , ferrara , a. , davies , j.i . , & alton , p.b . 2000 , , 311 , 601 blain , a.w . , ivison , r.j . , kneib , j .- & smail , i. 1999a , in the hy - redshift universe , ed . bunker & w.j.m . van breugel , asp conf . 193 ( asp : san francisco ) , 246 blain , a.w . , jameson , a. , smail , i. , longair , m.s . , kneib , j .- p . , & ivison , r.j . 1999b , , 309 , 715 blain , a.w . , kneib , j .- p . , ivison , r.j . , & smail , i. 1999c , , 512 , l87 blain , a.w . , smail , i. , ivison , r.j . , & kneib , j .- 1999d , , 302 , 632 block , d.l . , & greenberg , j.m . 1996 , new extragalactic perspectives in the new south africa ( dordrecht : kap ) block , d.l . , witt , a.n . , grosbol , p. , stockton , a. , & moneti , a. 1994 , , 288 , 383 bogart , r. , & wagoner , r. 1973 , , 181 , 609 boiss , p. , le brun , v. , bergeron , j. , & deharveng , j .- m . 1998 , , 333 , 841 bond , j.r . , carr , b.j . , & hogan , c.j . 1991 , , 367 , 420 boselli , a. , lequeux , j. , sauvage , m. , boulade , o. , boulanger , f. , et al . 1998 , , 335 , 53 bosma , a. , byun , y. , freeman , k.c . , & athanassoula , e. 1992 , , 400 , l21 bouchet , p. , lequeux , j. , maurice , e. , prevot , l. , & prevot - burnichon , m.l . 1985 , , 149 , 330 boulanger , f. , beichman , c. , dsert , f.x . , helou , g. , prault , m. , & ryter , c. 1988 , , 332 , 328 boyle , b.j . , fong , r. , & shanks , t. 1988 , , 231 , 897 bregman , j.n . , snider , b.a . , grego , l. , & cox , c.v . 1998 , , 499 , 670 brodie , j.p . , & huchra , j.p.1991 , , 379 , 157 bruzual , g. , & charlot , s. 2001 , in preparation bruzual , g.a . , magris , g.c . , & calvet , n. 1988 , , 333 , 673 buat , v. , & burgarella , d. 1998 , , 334 , 772 buat , v. , & xu , c. 1996 , , 306 , 61 bunker , a.j . 1999 , in photometric redshifts and high redshift galaxies , conf . ser . 191 , eds . weymann , l.j . storrie - lombardi , m. sawicki , & r.j . brunner ( san francisco : asp ) , 317 burgh , e.b . 2001 , phd thesis ( the johns hopkins university , u.s.a . ) burstein , d. , haynes , m.p . , & faber , s.m . 1991 , nature , 353 , 515 byun , y.i . 1993 , , 105 , 993 byun , y.i . , freeman , k.c . , & kylafis , n.d . 1994 , , 432 , 114 calzetti , d. , 1997a , , 113 , 162 calzetti , d. 1997b , in the ultaviolet universe at low and high redshift , eds . waller , m.n . fanelli , j.e . hollis , & a.c . danks , aip conf . 408 ( aip : woodbury ) , 403 calzetti , d. , armus , l. , bohlin , r.c . , kinney , a.l . , koornneef , j. & storchi - bergmann , t. 2000 , , 533 , 682 calzetti , d. , bohlin , r.c . , gordon , k.d . , witt , a.n . , & bianchi , l. 1995a , , 446 , l97 calzetti , d. , bohlin , r.c . , kinney , a.l . , storchi - bergmann , t. , & heckman , t.m . 1995b , , 443 , 136 calzetti , d. , conselice , c.j . , gallagher , j.s . , & kinney , a.l . 1999 , , 118 , 797 calzetti , d. , & heckman , t.m . 1999 , , 519 , 27 calzetti , d. , kinney , a.l . , & storchi - bergmann , t. 1994 , apj , 429 , 582 calzetti , d. , kinney , a.l . , & storchi - bergmann , t. 1996 , , 458 , 132 calzetti , d. , meurer , g.r . , bohlin , r.c . , garnett , d.r . , kinney , a.l . , leitherer , c. , & storchi - bergmann , t. 1997 , , 114 , 1834 caplan , j. , & deharveng , l. 1986 , , 155 , 297 cardelli , j.a . , clayton , g.c . , & mathis , j.s . 1989 , , 345 , 245 carilli , c.l . , & yun , m.s . 1999 , , 513 , l13 carollo , c.m . , franx , m. , illingworth , g.d . , & forbes , d.a . 1997 , , 481 , 710 cen , r. , & bryan , g.l . 2001 , , 546 , l81 cen , r. , miralda - escud , j. , ostriker , j.p . , & rauch , m. 1994 , , 437 , l9 cen , r. , & ostriker , j.p . 1999 , , 514 , 1 cesarsky , d. , lequeux , j. , abergel , a. , perault , m. , palazzi , e. , & tran , d. 1996 , , 315 , l305 chandrasekhar , s. 1960 , radiative trnasfer , new york : dover chapman , s.c . , richards , e.a . , lewis , g.f . , wilson , g. , & barger , a.j . 2001 , , 548 , l147 chapman , s.c . , scott , d. , steidel , c.c . , borys , c. , halpern , m. , et al . 2000 , , 319 , 318 charlot , s. , & fall , s.m . 2000 , , 539 , 718 charlot , s. , & longhetti , m. 2001 , , in press ( astroph/0101097 ) chary , r. , & elbaz , d. 2001 , , in press ( astroph/0103067 ) cheng , f.h . , gaskell , c.m . , & koratkar , a.p . 1991 , , 370 , 487 chini , r. , krgel , e. , & kreysa , e. 1986 , , 167 , 315 choloniewski , j. 1991 , , 250 , 486 clements , d.l . , dsert , f .- x . , & franceschini , a. 2001 , , in press ( astroph/0103242 ) clements , d.l . , dsert , f .- x . , franceschini , a. , reach , w.t . , baker , a.c . , davies , j.k . , & cesarsky , c. 1999 , , 346 , 383 clements , d.l . , sutherland , w.j . , saunders , w. , efstathiou , g.p . , mcmahon , r.g . 1996 , , 279 , 459 corradi , r.l.m . , beckman , j.e . , & simonneau , e. 1996 , , 282 , 1005 cowie , l.l , songaila , a. , & barger , a.j . 1999 , , 118 , 603 cox , c.v . , bregman , j.n . , & schombert , j.m . 1995 , , 99 , 405 cox , p. , & kessler , m.f . 1999 , the universe as seen by iso ( noordwijk : esa publications division ) cox , p. , krgel , e. , & mezger , p.g . 1986 , , 155 , 380 crawford , c.s . , allen , s.w . , ebeling , h. , edge , a.c . , & fabian , a.c . 1999 , , 306 , 857 dale , d.a . , helou , g. , contursi , a. , silbermann , n.a . , & kolhatkar , s. 2001 , , 549 , 215 davies , j.i . , alton , p.b . , trewhella , m. , evans , r. , & bianchi , s. 1999 , , 304 , 495 davies , j.l . , & burstein , d. 1995 , the opacity of spiral disks ( dordrecht : kap ) davies , j.i . , phillips , s. , booyce , p.j . , & disney , m.j . 1993 , , 260 , 491 dsert , f .- x . , boulanger , f. , & puget , j.l . 1990 , , 237 , 215 de jong , t. , clegg , p.e . , soifer , b.t . , rowan - robinson , m. , habing , h.j . , houck , j.r . , aumann , h.h . , & raimond , e. 1984 , , 278 , l67 de vaucouleurs , g. 1959a , , 64 , 397 de vaucouleurs , g. 1959b , handb der physik , 53 , 311 devereux , n.a . , & young , j.s . 1991 , , 371 , 515 di bartolomeo , a. , barbaro , g. , & perinotto , m. 1995 , , 277 , 1279 de young , d.s . , & heckman , t.m . 1994 , , 431 , 598 dickinson , m.e . 1998 , in the hubble deep field , stsci may symposium , eds . m. livio , s.m . fall & p. madau ( cambridge : cup ) , 219 dickinson , m.e . 2000 , phil . royal soc . lond . a , 358 , 2001 disney , m. , davies , j. , & phillips , s. 1989 , , 239 , 939 dole , h. , gispert , r. , lagache , g. , puget , j .- l . , bouchet , f.r . , et al . 2001 , , in press ( astroph/0103434 ) domingue , d.l . , keel , w.c . , ryder , s.d . , white , r.e . 1999 , , 118 , 1542 donahue , m. , mack , j. , voit , g.m . , sparks , w. , elston , r. , & maloney , p.r . 2000 , , 545 , 670 draine , b.t . , & anderson , n. 1985 , , 292 , 494 draine , b.t . , & lee , h.m . 1984 , , 285 , 89 draine , b.t . , & salpeter , e.e . 1979 , , 231 , 77 duley , w.w . , jones , a.p . , & williams , d.a . 1989 , , 236 , 709 dunne , l. , eales , s. , edmunds , m. , ivison , r. , alexander , p. , & clements , d.l . , 2000 , , 315 , 115 dunne , l. , & eales , s.a . 2001 , , in press ( astroph/0106362 ) dutra , c.m . , bica , e. , clari , j.j . , piatti , a.e . , & ahumada , a.v . 2001 , , in press ( astroph/0103518 ) dwek , e. , rephaeli , y. , & mather , j.c . 1990 , , 350 , 104 eales , s.a . , lilly , s. , gear , w. , dunne , l. , bond , j.r . , hammer , f. , le fevre , o. , & crampton , d. 1999 , , 515 , 518 eales , s.a . , lilly , s. , webb , t. , dunne , l. , gear , w. , clements , d. , & yun , m. 2000 , , 120 , 2244 ebneter , k. , & balick , b. 1985 , , 90 , 183 ebneter , k. , djorgovski , s. , & davis , m. 1988 , , 95 , 422 edge , a.c . , ivison , r.j . , smail , i. , blain , a.w . , & kneib , j .- 1999 , , 306 , 599 edge , a.c . , 2001 , , in press ( astroph/0106225 ) edmunds , m.g . , & phillips , s. , 1997 , , 292 , 733 efstathiou , a. , oliver , s. , rowan - robinson , m. , surace , c. , summer , t. , et al . 2000 , , 319 , 1169 efstathiou , a. , rowan - robinson , m. , & siebenmorgen , r. 2000 , , 313 , 734 elbaz , d. 1999 , in building galaxies : from the primordial universe to the present , eds . f. hammer , thuan , t.x . , cayatte , v. , guiderdoni , b. , & j. tran thanh van ( singapore : world scientific publishing ) , 195 elbaz , d. , cesarsky , c.j . , fadda , d. , aussel , h. , dsert , f.x . , et al . 1999 , , 351 , l37 ellingson , e. , yee , h.k.c . , bechtold , j. , & elston , r. 1996 , , 466 , l71 ellison , s.l . , lewis , g.f . , pettini , m. , chaffee , f.h . , & irwin , m.j . 1999 , , 520 , 456 ellison , s.l . , songaila , a. , schaye , j. , & pettini , m. 2000 , , 120 , 1175 ellison , s.l . , yan , l. , hook , i.m . , pettini , m. , wall , j.v . , & schaver , p. 2001 , , submitted elmegreen , d.m . 1980 , , 43 , 37 emsellem , e. 1995 , , 303 , 673 fabian , a.c . , johnstone , r.m . , & daines , s.j . 1994 , , 271 , 737 fabian , a.c . , smail , i. , iwasaka , k. , allen , s.w . , blain , a.w . , crawford , c.s . , ettori , s. , ivison , r.j . , johnstone , r.m . , kneib , j .- p . , & wilman , r.j . 2000 , , 315 , l8 fanelli , m.n . , oconnell , r.w . , & thuan , t.x . 1988 , , 334 , 665 fall , s.m . , & pei , y.c . 1993 , , 402 , 479 fall , s.m . , pei , y.c . , & mcmahon , r.g . 1989 , , 341 , l5 ferguson , h.c . 1993 , , 263 , 343 ferguson , a.m.n . , gallagher , j.s . , & wyse , r.f.g . 1998 , , 116 , 673 ferrara , a. , bianchi , s. , cimatti , a. , & giovanardi , c. 1999a , , 123 , 437 ferrara , a. , nath , b. , sethi , s.k . , shchekinov , y. 1999b , , 303 , 301 ferrara , a. , pettini , m. , & shchekinov , y. 2000 , , 319 , 539 ferrara , a. , & tolstoy , e. 2000 , , 313 , 291 fitzpatrick , e.l . 1986 , , 92 , 1068 fitzpatrick , e.l . 1999 , , 111 , 63 fixsen , d.j . , cheng , e.s . , gales , j.m . , mather , j.c . , shafer , r.a . , & wright , e.l . 1996 , , 473 , 576 fixsen , d.j . , dwek , e. , mather , j.c . , bennett , c.l . , & shafer , r.a . 1998 , , 508 , 123 flores , h. , hammer , f. , dsert , f.x . , cesarsky , c.j . , thuan , t. , et al . 1999a , , 343 , 389 flores , h. , hammer , f. , thuan , t.x . , cesarsky , c.j . , dsert , f.x . , et al . 1999b , , 517 , 148 forbes , d.a . , franx , m. , & illingworth , g.d . 1995 , , 109 , 1988 frster schreiber , n.m . , genzel , r. , lutz , d. , kunze , d. , & sternberg , a. 2001 , , 552 , 544 fox , m.j . , efstathiou , a. , rowan - robinson , m. , dunlop , j.s . , scott , s. et al . 2001 , , submitted ( astroph/0107585 ) franceschini , a. , mazzei , p. , dezotti , g. , & danese , l. 1994 , , 427 , 140 friedli , d. , & benz , w. 1995 , , 301 , 649 fukugita , m. , hogan , c.j . , & peebles , p.j.e . 1998 , , 503 , 518 furton , d.g . , laiho , j.w . , & witt , a.n . 1999 , , 526 , 752 gasda , s. , & leitherer , c. 1996 , in from stars to galaxies : the impact of stellar physics on galaxy evolution , eds . c. leitherer , u. fritze - von alvensleben & j. huchra , asp conf . 98 ( asp : san francisco ) , 451 ge , j. , bechtold , j. , & kulkarni , v.p . 2001 , , 547 , l1 genzel , r. , & cesarsky , c.j . 2000 , , in press genzel , r. , lutz , d. , sturm , e. , egami , e. , kunze , d. , moorwood , a.f.m . , rigopoulou , d. , spoon , h.w.w . , sternberg , a. , tacconi - garman , l.e . , tacconi , l. , & thatte , n. 1998 , , 498 , 579 giavalisco , m. , livio , m. , bohlin , r.c . , macchetto , f.d . , & stecher , t.p . 1996a , , 112 , 369 giavalisco , m. , steidel , c.c . , adelberger , k.l . , dickinson , m.e . , pettini , m. , & kellogg , m. 1998 , , 503 , 543 giavalisco , m. , steidel , c.c . , & macchetto , f.d . , 1996b , , 470 , 189 giovanelli , r. , haynes , m.p . , salzer , j.j . , wegner , g. , da costa , l.n . , & freudling , w. 1994 , , 107 , 2036 giovanelli , r. , haynes , m.p . , salzer , j.j . , wegner , g. , da costa , l.n . , & freudling , w. 1995 , , 110 , 1059 glazebrook , k. , blake , c. , economou , f. , lilly , s. , & colless , m. 1999 , , 306 , 843 gnedin , n.y . 1998 , , 294 , 407 goad , j.w . , & roberts , m.s . 1981 , , 250 , 79 goldader , j.d . , meurer , g.r . , heckman , t.m . , seibert , m. , sanders , d. , calzetti , d. , & steidel , c.c . 2001 , in prep . gonzlez , r.a . , allen , r.j . , dirsch , b. , ferguson , h.c . , calzetti , d. , & panagia , n. 1998 , , 506 , 152 gozalez delgado , r.m . , leitherer , c. , heckman , t.m . , lowenthal , j.d . , ferguson , h.c . , & robert , c. 1998 , , 495 , 698 gordon , k.d . , calzetti , d. , & witt , a.n . 1997 , , 487 , 625 gordon , k.d . , & clayton , g.c . 1998 , , 500 , 816 gordon , k.d . , clayton , g.c . , witt , a.n . , & misselt , k.a . 2000 , , 533 , 236 gordon , k.d . , misselt , k.a . , witt , a.n . , & clayton , g.c . 2001 , , in press ( astroph/0011575 ) goudfrooij , p. , hansen , l. , jorgensen , h.e . , & norgaard - nielsen , h.u . 1994 , , 105 , 341 goudfrooij , p. , & de jong , t. 1995 , , 298,784 granato , g.l . , lacey , c.g . , silva , l. , bressan , a. , baugh , c.m . , cole , s. , & frenk , c.s . 2000 , , 542 , 710 greenberg , j.m . 1984 , sci . , 250 , no . 6 , 124 greenberg , j.m . 1989 , in interstellar dust , iau symp . 135 , eds . allamandola & a.g.g.m . tielens ( dordrecht : kap ) , 239 greenberg , j.m . , gillette , j.s . , munz caro , g.m . , mahajan , t.b . , zare , r.n . , 2000 , , 531 , l71 greggio , l. , tosi , m. , clampin , m. , de marchi , g. , leitherer , c. , nota , a. , & sirianni , m. 1998 , , 504 , 725 guhathakurta , p. , tyson , j.a . , & majewski , s.r . 1990 , , 357 , l9 guiderdoni , b. , hivon , e. , bouchet , f.r . , & maffei , b. 1998 , , 295 , 877 haas , m. 1998 , , 337 , l1 hansen , l. , jrgensen , h.e . , & nrgaard - nielsen , h.u . 1995 , , 297 , 13 hauser , m.g . , arendt , r.g . , kelsall , t. , dwek , e. , odegard , n. , et al . 1998 , , 508 , 25 hauser , m.g . , & dwek , e. 2001 , , in press . heckman , t.m . 1994 , in mass - transfer - induced activity in galaxies , ed . i. shloshman ( cambridge : cup ) , 234 heckman , t.m . , in origins , eds . woodward , j.m . shull , & h.a . thronson ( san francisco : asp conf . series ) 148 , 127 heckman , t.m . , armus , l. , & miley , g.k . 1990 , , 74 , 833 heckman , t.m . , baum , s.a . , van breugel , w.j.m . , & mccarthy , p. 1989 , , 338 , 48 heckman , t.m . , & leitherer , c. 1997 , , 114 , 69 heckman , t.m . , lehnert , m.d . , strickland , d.k . , & armus , l. 2000 , , 129 , 493 heckman , t.m . , robert , c. , leitherer , c. , garnett , d.r . , & van der rydt , f. 1998 , , 503 , 646 helou , g. 1986 , , 311 , l33 helou , g. , khan , i.r . , malek , l. , & boehmer , l. 1988 , , 68 , 151 henyey , l.g . , & greenstein , j.l . 1941 , , 93 , 70 hernquist , l. , katz , n. , weinberg , d.h . , & miralda - escud , j. 1996 , , 457 , l51 hibbard , j.e . 1997 , in star formation near and far , the 7th annual astrophysics conference in maryland , eds . holt & l.g . mundy ( woodbury : aip ) 393 , 259 holmberg , e. 1958 , medd . lunds astr . ii , no . 136 hornschemeier , a.e . , brandt , w.n . , garmire , g.p . , schneider , d.p . , broos , p.s . , townsley , l.k . , bautz , m.w . , burrows , d.n . , chartas , g. , feigelson , e.d . , griffith , r. , lumb , d. , nousek , j.a . , & sargent , w.l.w . 2000 , , 541 , 49 hopkins , a.m. , connolly , a.j . , haarsma , d.b . , & cram , l.e . 2001 , , in press ( astroph/0103253 ) hu , e.m . 1992 , , 391 , 608 hu , e.m . , cowie , l.l . , kaaret , p. , jenkins , e.b . , york , d.g . , & roesler , f.l . 1983 , , 275 , l27 hughes , d.h . , serjeant , s. , dunlop , j. , rowan - robinson , m. , blain , a. , ivison , r. , et al . 1998 , nature , 394 , 241 huizinga , e. , & van albada , t.s . 1992 , , 254 , 677 hunter , d.a . 1997 , , 109 , 937 hunter , d.a . , & gallagher , j.s . 1997 , , 475 , 65 hunter , d.a . , gallagher , j.s . , rice , w.l . , & gillett , f.c . 1989 , , 336 , 152 hunter , d.a . , gillett , f.c . , gallagher , j.s . , rice , w.l . , & low , f.j . , 1984 , , 303 , 171 hunter , d.a . , kaufman , m. , hollenbach , d.j . , rubin , r.h , malhotra , s. , dale , d.a . , brahuer , j.r . , silbermann , n.a . , helou , g. , contursi , a. & lord , s.d . 2001 , , in press ( astroph/0012354 ) hunter , d.a . , & sage , l. 1993 , , 105 , 374 hunter , d.a . , & thronson , h.a . 1996 , , 461 , 202 ivison , r.j . , smail , i. , barger , a.j . , kneib , j .- blain , a.w . , owen , f.n . , kerr , t.h . , & cowie , l.l . 2000 , , 315 , 209 ivison , r.j . , smail , i. , le borgne , j .- f . , blain , a.w . , kneib , j .- p . , bezecourt , j. , kerr , t.h . , & davies , j.k . 1998 , , 298 , 583 jansen , r.a . 2000 , phd thesis jenniskens , p. 1993 , a&a , 274 , 653 jones , h. , davies , j.i . , & trewhella , m. 1996 , , 283 , 316 jones , a.p . , tielens , a.g.g.m . , hollenbach , d.j . , & mckee , c.f . 1994 , , 433 , 797 jura , m. 1977 , nature , 266 , 702 jura , m. , kim , d.w . , knapp , g.r . , & guhathakurta , p. 1987 , , 312 , l11 kaastra , j.s . , ferrigno , c. , tamura , t. , paerels , f.b.s . , peterson , j.r . , & mittaz , j.p.d . 2001 , , 365 , l99 karachentsev , i.d . , & lipovetskii , v.a . 1969 , sov . , 12 , 909 kawara , k. , sato , y. , matsuhara , h. taniguchi , y. , okuda , h. et al . 1998 , , 336 , l9 keel , w.c . 1983 , , 88 , 1579 keel , w.c . , kennicutt , r.c . , hummel , e. , & van der hulst , j.m . 1985 , , 90 , 708 keel , w.c . , & white , r.e . 2001a , , in press ( astroph/0011370 ) keel , w.c . , & white , r.e . 2001b , in press ( astroph/0106056 ) kennicutt jr . , r.c . 1983 , , 272 , 54 kennicutt jr . 1998 , , 36 , 189 kim , s .- h . , martin , p.g . , & hendry , p.d . 1994 , , 422 , 164 kim , s.s . , morris , m. , & lee , h.m . 1999 , , 525 , 228 kim , d.c . , & sanders , d.b . 1998 , , 119 , 41 kinney , a.l . , bohlin , r.c . , calzetti , d. , panagia , n. , & wyse , r.f.g . , , 86 , 5 knapp , g.r . , guhathakurta , p. , kim , d.w . , & jura , m. , 70 , 329 koornneef , j. , & code , a.d . 1981 , , 247 , 860 kuchinski , l.e . , terndrup , d.m . , gordon , k.d . , & witt , a.n . 1998 , , 115 , 1438 kunth , d. , mas - hesse , j.m . , terlevich , e. , terlevich , r. , lequeux , j. , & fall , s.m . 1998 , , 334 , 11 kylafis , n.d . , & bahcall , j.n . 1987 , , 317 , 637 lanzetta , k.m . , chen , h .- w . , fernndez - soto , a. , pascarelle , s. , yahata , n. , & yahil , a. 1999 , in the hy - redshift universe : galaxy formation and evolution at high redshift , eds . bunker & w.j.m . van breugel ( asp : san francisco ) , 544 larson , r.b . , & tinsley , b.m . 1978 , , 219 , 46 lauer , t.r . , ajhar , e.a . , byun , y .- i . , dressler , a. , faber , s.m . , grillmair , c.,kormendy , j. , richstone , d. , & tremaine , s. 1995 , , 110 , 2622 le brun , v. , bergeron , j. , boiss , p. , & deharveng , j .- 1997 , , 321 , 733 lger , a. , & puget , j.l . 1984 , , 137 , l5 lehnert , m.d . , & heckman , t.m . 1996 , , 472 , 546 leisawitz , d. , & hauser , m.g . 1988 , , 332 , 954 leitherer , c. , & heckman , t.m . 1995 , , 96 , 9 leitherer , c. , leao , j. , heckman , t.m . , lennon , d. , pettini , m. , & robert , c. 2001 , , 550 , in press leitherer , c. , schaerer , d. , goldader , g.d . , gonzalez delgado , r.m . , robert , c. , kune , d.f . , de mello , d.f . , devost , d. , & heckman , t.m . 1999 , , 123 , 3 lequeux , j. , maurice , e. , prevot - burnichon , m.l . , prevot , l. , & rocca - volmerange , b. 1982 , , 113 , l15 lequeux , j. , maurice , e. , prevot , l. , prevot - burnichon , m .- l . , & rocca - volmerange , b. 1984 , structure and evolution of the magellanic clouds , i.a.u . symposium no . 108 , s. van den bergh & k.s . de boer eds . ( dordrecht : reidel ) , 405 li , a. , & draine , b.t . 2001 , , 554 , 778 lilly , s.j . , eales , s.a . , gear , w.k.p . , hammer , f. , le fevre , o. , crampton , d. , bond , j.r . , & dunne , l. 1999 , , 518 , 641 lilly , s.j . , lefevre , o. , crampton , d. , hammer , f. , & tresse , l. 1995 , , 455 , 50 lilly , s.j . , lefevre , o. , hammer , f. , & crampton , d. 1996 , , 460 , l1 lisenfeld , u. , isaak , k.g . , & hills , r. 2000 , , 312 , 433 loeb , a. , & haiman , z. 1997 , , 490 , 571 londale - persson , c.j . , & helou , g. 1987 , , 314 , 513 lowenthal , j.d . , koo , d.c . , guzman , r. , gallego , j. , phillips , a.c . , et al . 1997 , , 481 , 673 lu , l. , sargent , w.l.w . , barlow , t.a . , churchill , c.w . , & vogt , s.s . 1996 , , 107 , 475 lutz , d. , feuchtgruber , h. , genzel , r. , kunze , d. , rigopoulou , d. , et al . 1996 , , 315 , l269 lutz , d. , spoon , h.w.w . , rigopoulou , d. , moorwood , a.f.m . , & genzel , r. 1998 , , 505 , l103 mac low , m .- m . , & ferrara , a. 1999 , , 513 , 142 madau , p. ferguson , h.c . , dickinson , m.e . , giavalisco , m. , steidel , c.c . , & fruchter , a. 1996 , , 283 , 1388 madau , p. , ferrar , a. , & rees , m.j . 2001 , , in press maz - apellniz , j. , & walborn , n.r . 2000 , poster presented at the iau xxivth general assembly , manchester malhotra , s. 1997 , , 488 , l101 maoz , d. 1995 , , 455 , l115 martin , m.c . 1998 , , 131 , 77 marzke , r.o . , huchra , j.p . , & geller , m.j . 1994 , , 428 , 43 mas - hesse , j.m . , arnault , p. , & kunth , d. 1989 , ap&ss , 157 , 131 mathis , j.s . 1990 , araa , 28 , 37 mathis , j.s . 1996 , , 472 , 643 mathis , j.s . , mezger , p.g . , & panagia , n. 1983 , , 128 , 212 mathis , j.s . , rumpl , w. , & nordsieck , k.h . 1977 , , 217 , 425 mathis , j.s . , & whiffen , g. 1989 , , 341 , 808 matthews , l.d . , & wood , k. 2001 , , 548 , 150 mccarthy , p.j . , yan , l. , freudling , w. , teplitz , h.i . , malumuth , e.m . , et al 1999 , , 520 , 548 mckee , c.f . 1989 , in interstellar dust , iau symposium 135 , eds . allamandola & a.g.g.m . tielens ( dordracht : kap ) , 431 mcquade , k. , calzetti , d. , & kinney , a.l . 1995 , , 97 , 331 meier , d.s . , turner , j.l . , crosthwaite , l.p . , & beck , s.c . 2001 , , in press ( astroph/0011582 ) merluzzi , p. 1998 , , 338 , 807 meurer , g.r . , heckman , t.m . , & calzetti , d. 1999 , , 521 , 64 meurer , g.r . , heckman , t.m . , lehnert , m.d . , leitherer , c. , & lowenthal , j. 1997 , , 114 , 54 meyer , d.m . , & york , d.g . 1987 , , 319 , l45 meyer , d.m . , & roth , k.c . 1990 , , 363 , 57 mezger , p.g . , mathis , j.s . , & panagia , n. 1982 , , 105 , 372 mihos , j.c . , & hernquist , l. 1996 , , 464 , 641 misiriotis , a. , kylafis , n.d . , papamastorakis , j. , xilouris , e.m . 2000 , , 353 , 117 misselt , k.a . , clayton , g.c . , & gordon , k.d . 1999 , , 515 , 128 misselt , k.a . , gordon , k.d . , clayton , g.c . , & wolff , m.j . 2001 , , in press ( astroph/0011576 ) moorwood , a.f.m . , van der werf , p.p . , cuby , j.c . , & oliva , e. 2000 , , 362 , 9 moriondo , g. , giovanelli , r. , & haynes , m.p . 1998 , , 338 , 795 murphy , t.w . , soifer , b.t . , matthews , k. , armus , l. , & kiger , j.r . 2001 , , 121 , 97 mushotzky , r.f . , loewenstein , m. , arnaud , k.a . , tamura , t. , fukazawa , y. , matsushita , k. , kikuchi , k. , & hatsukade , i. 1996 , , 466 , 686 mushotzky , r.f . , & loewenstein , m. , 1997 , , 481 , l63 nandy , k. , morgan , d.h . , willis , a.j . , wilson , r. , gondhalekar , p.m. 1981 , , 196 , 955 nath , b.b . , & trentham , n. 1997 , , 291 , 505 natta , a. , & panagia , n. 1976 , , 50 , 191 natta , a. , & panagia , n. 1984 , , 287 , 228 nelson , a.e . , zaritsky , d. , & cutri , r.m . 1998 , , 115 , 2273 niklas , s. , klein , u. , & wielebinski , r. 1997 , , 322 , 19 norman , c.a , sellwood , j.a . , & hasan , h. 1996 , , 462 , 114 oliver , s.j . , goldschmidt , p. , franceschini , a. , serjeant , s.b.g . , efstathiou , a. , et al . 1997 , , 289 , 471 ortolani , s. , renzini , a. , gilmozzi , r. , marconi , g. , babuy , b. , bica , e. , & rich , m.r . 1995 , nature , 377 , 701 osterbrock , d.e . 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( mill valley : university science books ) ostriker , j.p . , & gnedin , n.y . 1996 , , 472 , l63 ouchi , m. , yamada , t. , kawai , h. , & ohta , k. 1999 , , 517 , l19 pak , s. , jaffe , d.t . , van dishoeck , e.f . , johansson , l.e.b . , & booth , r.s . , 1998 , , 498 , 735 papovich , c. , dickinson , m. , & ferguson , h.c . 2001 , , in press peacock , j.a . , rowan - robinson , m. , blain , a.w . , dunlop , j.s . , efstathiou , a. , et al . 2000 , , 318 , 535 pei , y.c . , fall , s.m . , & bechtold , j. 1991 , , 378 , 6 pei , y.c , fall , s.m . , & hauser , m.g . 1999 , , 522 , 604 peletier , r.f . , valentijn , e.a . , moorwood , a.f.m . , freudling , w. , knapen , j.h . , & beckman , j.e . 1995 , , 300 , l1 peletier , r.f . , & willner , s.p . 1992 , , 103 , 1761 penton , s.v . , shull , j.m . , & stocke , j.t . 2000 , , 544 , 150 peterson , j.r . , paerels , f.b.s . , kaastra , j.s . , arnaud , m. , reiprich , t.h . , fabian , a.c . , mushotzky , r.f . , jernigan , j.g . , & sakelliou , i. 2001 , , 365 , l104 petitjean , p. , muecket , j.p . , & kates , r.e . 1995 , , 295 , l9 pettini , m. 1999 , in chemical evolution from zero to high redshift , ed . j. walsh & m. rosa ( berlin : springer ) , 233 pettini , m. , boksenberg , a. , & hunstead , r.w . , 1990 , , 348 , 48 pettini , m. , ellison , s.l . , steidel , c.c . , shapley , a.e . , & bowen , d.v . 2000a , , 532 , 65 pettini , m. , ellison , s.l . , steidel , c.c . , & bowen , d.v . 1999 , , 510 , 576 pettini , m. , kellogg , m. , steidel , c.c . , dickinson , m.e . , adelberger , k.l . , & giavalisco , m. 1998 , , 508 , 539 pettini , m. , king , d.l . , smith , l.j . , & hunstead , r.w . 1997a , , 478 , 536 pettini , m. , shapley , a.e . , steidel , c.c . , cuby , j .- g . , dickinson , m. , moorwood , a.f.m . , adelberger , k.l . , & giavalisco , m. 2001 , , 554 , 981 pettini , m. , smith , l.j . , hunstead , r.w . , & king , d.l . 1994 , , 426 , 79 pettini , m. , smith , l.j . , king , d.l . , & hunstead , r.w . , & king , d.l . 1997b , , 486 , 665 pettini , m. , steidel , c.c . , adelberger , k.l . , dickinson , m.e . , & giavalisco , m. 2000b , , 528 , 96 pinkney , j. , holtzman , j. , garasi , c. , et al . 1996 , , 468 , l13 pizagno , j. , & rix , h .- w . 1998 , , 116 , 2191 poggianti , b.m . , & wu , h. 2000 , , 529 , 157 popescu , c.c . , tuffs , r.j . , fischera , j. , & vlk , h. 2000 , , 354 , 480 prochaska , j.x . , & wolfe , a.m. 2000 , , 533 , l5 puget , j .- l . , abergel , a. , bernard , j .- p . , boulanger , f. , burton , w.b . , desert , f .- x . , & hartmann , d. 1996 , , 308 , l5 puget , j.l . , lagache , g. , clements , d.l , reach , w.t . , aussel , h. , et al . 1999 , , 345 , 29 quillen , a.c . , rieke , g.h . , rieke , m.j . , caldwell , n. , & engelbracht , c. 1999 , , 518 , 632 radovich , m. , kahanp , j. , & lemke , d. 2001 , , in press ( astroph/0108075 ) rao , s.m . , & turnshek , d.a . 2000 , , 130 , 1 rauch , m. , miralda - escud , j. , sargent , w.l.w . , barlow , t.a . , weinberg , d.h . , hernquist , l. , katz , n. , cen , r. , & ostriker , j.p . 1997 , , 489 , 7 reach , w.t . , boulanger , f. , contursi , a. , & lequeux , j. , 2000 , , 361 , 895 renzini , a. 1999 , in chemical evolution from zero to high redshift , ed . j. walsh & m. rosa ( berlin : springer ) , 185 rieke , g.h . , & lebofsky , m.j . 1985 , , 288 , 618 rieke , g.h . , & lebofsky , m.j . 1986 , , 304 , 326 rieke , g.h . , lebofsky , m.j . , thompson , r. i. , low , f.j . , & tokunaga , a.t . 1980 , , 238 , 24 rieke , g.h . , loken , k. , rieke , m.j . , & tamblyn , p. 1993 , , 412 , 99 riess , a. , filippenko , a.v . , liu , m.c . , et al . 2000 , , 536 , 62 riess , a. , nugent , p. , smith , b. , tonry , j. , gilliland , r. , et al . 2001 , , in press rigopoulou , d. , franceschini , a. , aussel , h. , genzel , r. , van der werf , p. , et al . 2000 , , 537 , l85 rix , h .- w . , & rieke , m.j . 1993 , , 418 , 123 roberts , m.s . , & haynes , m.p . 1994 , , 32 , 115 roberts , m.s . , hogg , d.e . , bregman , j.n . , forman , w.r . , & jones , c. 1991 , , 75 , 751 romani , r.w . , & maoz , d. 1992 , , 386 , 36 rowan - robinson , m. , & crawford , j. 1989 , , 238 , 523 rowan - robinson , m. , & efstathiou , a. 1993 , , 263 , 675 rowan - robinson , m. , mann , r.g . , oliver , s.g . , efstathiou , a. , eaton , n. et al . 1997 , , 289 , 490 rowan - robinson , m. , oliver , s. , efstathiou , a. , gruppioni , c. , serjeant , s. , et al . 1999 , in the universe as seen by iso , eds . p. cox & m.f . kessler ( noordwijk : esa publications division ) 1011 rowan - robinson , m. , negroponte , j. , & silk , j. 1979 , nature , 281 , 635 sadler , e.m . , & gerhard , o.e . 1985 , , 214 , 177 sanders , d.b . , & mirabel , i.f . 1996 , , 34 , 749 sanders , d.b . , soifer , b.t . , elias , j.h , madore , b.f . , matthews , k. , neugebauer , g. , & scoville , n.z . 1988 , , 325 , 74 savaglio , s. , panagia , n. , & stiavelli , m. 2000 , in cosmic evolution and galaxy formation : structure , interactions , and feedback , asp conf . 215 , eds . j. franco , e. terlevich , o. lpez cruz , & i. aretxaga ( asp : chelsea ) , 65 sawicki , m. 2001 , , in press ( astroph/0102208 ) sawicki , m. , & yee , h.k.c . 1998 , , 115 , 1329 schutte , w.a . , van der hucht , k.a . , whittet , d.c.b . , boogert , a.c.a . , tielens , a.g.g.m . , 1998 , , 337 , 261 scott , d. , lagache , g. , borys , c. , chapman , s.c . , halpern , m. , et al . 2000 , , 357 , l5 scott , s.e . , fox , m.j . , dunlop , j.s . , serjeant , s. , peacock , j.a . , et al . 2001 , , submitted ( astroph/0107446 ) scoville , n.z . , evans , a.s . , dinshaw , n. , thompson , r. , rieke , m.,schneider , g. , low , f.j . , hines , d. , stobie , e. , becklin , e. , & epps , h. 1998 , , 492 , l107 scowen , p.a . , hester , j.j . , sankrit , r. , gallagher , j.s . , ballester , g.e . , et al . 1998 , , 116 , 163 sellgren , k. 1984 , , 277 , 623 sellgren , k. 1994 , the first symposium on the infrared cirrus and diffuse insterstellar clouds , cutri & latter , san francisco : asp , 58 , 243 sellgren , k. 2000 , spectrochimica acta a , feb . 2001 , in press ( astroph/0010138 ) serjeant , s. , oliver , s. , rowan - robinson , m. , crockett , h. , missoulis , v. , et al . 2000 , , 316 , 768 severgnini , p. , maiolino , r. , salvati , m. , axon , d. , cimatti , a. , fiore , f. , gilli , r. , la franca , f. , marconi , a. , matt , g. , risaliti , g. , & vignali , c. 2000 , , 360 , 457 shapley , a.e . , steidel , c.c . , adelberger , k.l . , dickinson , m. , giavalisco , m. , & pettini , m. 2001 , in prep . shaver , p.a . , hook , i.m . , jackson , c.a . , wall , j.v . , & kellermann , k.i . 1999 , in highly redshifted radio lines , asp conf . carilli , s.j.e . radford , k.m . menten & g.i . langston ( asp : san francisco ) , 163 shioya , y. , trentham , n. , & taniguchi , y. 2001 , , in press ( atroph/0011467 ) silva , l. , granato , g.l . , bressan , a. , & danese , l. 1998 , , 509 , 103 simonsen , j.t . , & hannestad , s. 1999 , , 351 , 1 skillman , e.d . , kennicutt , r .. , & hodge , p.w . 1989 , , 347 , 875 smail , i. , ivison , r.j . , & blain , a.w . 1997 , , 490 , l5 smail , i. , ivison , r.j . , blain , a.w . , & kneib , j .- p . 1998 , , 507 , l21 smail , i. , ivison , r.j . , kneib , j .- blain , a.w . , barger , a.j . , owen , f.n . , & morrison , g. 1999 , , 308 , 1061 smail , i. , ivison , r.j . , owen , f.n . , blain , a.w . , & kneib , j .- 2000 , , 528 , 612 sobolev , v.v . 1963 , a treatise on radiative transfer , princeton : van nostrand sodroski , t.j . , bennett , c. , boggess , n. , dwek , e. , franz , b.a . , hauser , m.g . , kelsall , t. , moseley , s.h . , odegard , n. , silverberg , r.f . , & weiland , j.l . 1994 , , 428 , 638 sofue , y. , reuter , h .- p . , krause , m. , wielebinski , r. , & nakai , n. 1992 , , 395 , 126 soifer , b.t . , houck , j.r . , & neugebauer , g. 1987 , , 25 , 187 soifer , b.t . , & neugebauer , g. 1991 , , 101 , 354 songaila , a. & cowie , l.l . 1996 , , 112 , 335 sparks , w.b . , ford , h.c . , & kinney , a.l . 1993 , , 413 , 531 sparks , w.b . , macchetto , f. , & golombek , d. 1989 , , 345 , 153 sprayberry , d. , & foltz , c.b . 1992 , , 390 , 39 steidel , c.c . , adelberger , k.l . , giavalisco , m. , dickinson , m.e . , & pettini , m. 1999 , , 519 , 1 steidel , c.c . , giavalisco , m. , dickinson , m.e . , & adelberger , k.l . 1996a , , 112 , 352 steidel , c.c . , giavalisco , m. , pettini , m. , dickinson , m. , & adelberger , k. 1996b , , 462 , l17 steidel , c.c . , & hamilton , d. 1992 , , 104 , 941 stickel , m. , lemke , d. , mattila , k. , haikala , l.k . , & haas , m. 1998 , , 329 , 55 storchi - bergmann , t. , kinney , a.l . , & challis , p. 1995 , , 98 , 103 sullivan , m. , treyer , m.a . , ellis , r.s . , bridges , t.j . , milliard , b. , & donas , j. 2000 , , 312 , 442 sullivan , m. , mobasher , b. , chan , b. , cram , l. , ellis , r. , treyer , m. , & hopkins , a. 2001 , , in press ( astroph/0104425 ) szalay , a.s . , hollsi , j. , & tth , g. 1989 , , 339 , l5 tamura , t. , kaastra , j.s . , peterson , j.r . , paerels , f.b.s . , mittaz , j.p.d . , trudolyubov , s.p . , stewart , g. , fabian , a.c . , mushotzky , r.f . , lumb , d.h . , & ikebe , y. 2001 , , 365 , l87 taniguchi , y. , cowie , l.l . , sato , y. , sanders , d.b . , kawara , k. et al . 1998 , , 328 , l9 teplitz , h.i . , malkan , m.a . , steidel , c.c . , mclean , i.s . , becklin , e.e . , et al . 2000a , , 542 , 18 teplitz , h.i . , mclean , i.s . , becklin , e.e . , figer , d.f . , gilbert , a.m. , et al . 2000b , , 533 , l65 theuns , t. , leonard , a. , efstathiou , g. , pearce , f.r . , & thomas , p.a . 1998 , , 301 , 478 tielens , a.g.g.m . 1989 , in interstellar dust , iau symp . 135 , eds . allamandola & a.g.g.m . tielens ( dordrecht : kap ) , 239 tran , h. , tsvetanov , z. , ford , h.c . , davies , j. , jaffe , w. , van den bosch , f.c . , & rest , a. 2001 , , accepted ( astroph/0102292 ) tremonti , c.a . , calzetti , d. , leitherer , c. , & heckman , t.m . 2001 , , in press ( astroph/0103432 ) trentham , n. , kormendy , j.k . , & sanders , d.b . 1999 , , 117 , 2152 trewhella , m. 1998 , , 297 , 807 trewhella , m. , madore , b. , & kuchinski , l. 1999 , observational cosmology : the development of galaxy systems , giuricin , mezzetti & salucci , san francisco : asp , 176 , 454 trewhella , m. , davies , j.i . , alton , p.b . , bianchi , s. , & madore , b.f . 2000 , , 543 , 153 trewhella , m. , davies , j.i . , disney , m.j . , & jones , h.g.w . 1997 , , 288 , 397 turnshek , d.a . , rao , s.m . , & nestor , d.b . 2001 , in extragalactic gas at low redshift , asp conf . , j. mulchaey and j. stocke eds . ( asp : ... ) , in press ( astroph/0108142 ) tytler , d. , fan , x.m . , burles , s. , cottrell , l. , davis , c. , kirkman , d. , & zuo , l. 1995 , in qso absorption lines , ed . g. meylan ( garching : eso ) , 289 valentijn , e.a . 1990 , nature , 346 , 153 van den bergh , s. , 1974 , , 193 , 63 van der werf , p.p . , kraiberg knudsen , k. , labb , i. , & franx , marijn 2001 , in the far - infrared and submillimeterspectral energy distributions of active and starburst galaxies , eds . p. barthel , b. wilkes , & i. van bemmel ( elsevier : the netherlands ) , ... van dokkum , p.g . & franx , m. 1995 , , 110 , 2027 vrosi , f. , & dwek , e. 1999 , , 523 , 265 veilleux , s. , kim , d .- , & sanders , d.b . 1999 , , 522 , 113 verter , f. , & hodge , p. 1995 , , 446 , 616 vladilo , g. 1998 , , 493 , 583 voit , g.m . , & donahue , m. 1992 , , 452 , 164 voit , g.m . , & donahue , m. 1997 , , 486 , 242 voshchinnikov , n.v . & mathis , j.s . 1999 , , 526 , 257 wainscoat , r.j , hyland , a.r . , & freeman , k.c . 1990 , , 348 , 85 walborn , n.r . 1991 , in massive stars in starbursts , eds . c. leitherer , n.r . walborn , t.m . heckman , & c.a . norman ( cambridge : cup ) , 145 walborn , n.r . , barb , r.h . , brandner , w. , rubio , m. , grebel , e.k . , & probst , r.g . 1999 , , 117 , 225 wang , b. 1991 , , 383 , l37 wang , b. , & heckman , t.m . 1996 , , 457 , 645 wang , j. , heckman , t.m . , & lehnert , m.d . 1998 , , 515 , 97 weingartner , j.c . , & draine , b.t . 2001 , , 548 , 296 white , s.d.m . , & frenk , c.s . 1991 , , 379 , 52 white , r.e . , & keel , w.c . 1992 , nature , 359 , 129 white , r.e . , keel , w.c . , & conselice , c.j . 2000 , , 542 , 761 whittle , m. 1992 , , 79 , 49 wilson , c.d . , walker , c.e . , & thornely , m.d . 1997 , , 483 , 210 wise , m.w . , oconnell , r.w . , bregman , j.n . , & roberts , m.s . 1993 , , 405 , 94 wise , m.w . , & silva , d.r . 1996 , , 461 , 155 witt , a.n . , friedmann , b.c . , & sasseen , t.p . 1997 , , 481 , 809 witt , a.n , & gordon , k.d . 1996 , , 463 , 681 witt , a.n . , & gordon , k.d . 2000 , , 528 , 799 witt , a.n . , thronson , h.a . , & capuano , j.m . 1992 , , 393 , 611 wolff , m.j . , clayton , g.c . , & gibson , s.j . 1998 , , 503 , 815 wright , e.l . 1981 , , 250 , 1 xilouris , e.m . , kylafis , n.d . , papamastorakis , j. , paleologou , e.v . , & haerendel , g. 1997 , , 325 , 135 xilouris , e.m . , byun , y.i . , kylafis , n.d . , paleologou , e.v . , & papamastorakis , j. , 1999 , , 344 , 868 xu , c. 2000 , , 541 , 134 xu , c. , & buat , v. 1995 , , 293 , l65 xu , c. , & helou , g. 1996 , , 456 , 163 yadat , r.k.s . , & sagar , r. 2001 , , in press yamamoto , t.m . , & vansevicius , v. 1999 , pasj , 51 , 405 yan , l. , mccarthy , p.j . , freudling , w. , teplitz , h.j . , malumuth , e.m . , weymann , r.j . , & malkan , m.a . 1999 , , 519 , l47 young , j.s . , & scoville , n.z . 1991 , , 29 , 581 yu , g. & solomon , p.m. 1999 , , 512 , l99 yun , m.s . , reddy , n.a . , & condon , j.j . 2001 , , accepted ( astroph/0102154 ) zaritsky , d. 1994 , , 108 , 1619 zaritsky , d. 1999 , , 118 , 2824 zaritsky , d. , kennicutt , r.c . , & huchra , j.p . 1994 , , 420 , 87 zhang , y. , anninos , p. , & norman , m.l . 1995 , , 453 , l57 zuo , l. , beaver , e.a . , burbidge , e.m . , cohen , r.d . , junkkarinen , v.t . , & lyons , r.w . 1997 , , 477 , 568 zwicky , f. , 1957 , in morphological astronomy ( berlin : springer - verlag ) , 101 llllll spherical & homog . & central/ & & & + & & spherical & @xcite&mw & multiple density profiles , + & & & & & multiple scale - lengths + & & & @xcite & & starburst / ir modelling , + & & & & & central source only + & & & @xcite & & multiple scale - lengths + & & & @xcite & & also external source + & & & @xcite&mw , smc & multiple dust / star density + & & & & & profiles + & & & @xcite & & starburst / ir modelling , + & & & & & multiple spheres , central + & & & & & source only + & & & @xcite&mw , smc&updated albedo / phase + & & & & & function values + & & & & & + spherical & clumpy & central/ & & & + & & spherical & @xcite & mw & central source only + & & & @xcite & & also external source + & & & @xcite&mw , smc & updated albedo / phase + & & & & & function values + & & & & & + plane - paral.&homog.&plane - paral.&@xcite&mw , lmc & multiple scale - heights + & & & @xcite & & no scattering + & & & @xcite&mw & multiple scale - heights + & & & @xcite & & multiple scale - heights + & & & & & + expon . & homog . & expon . & @xcite&mw & edge - on only , isothermal + & & & & & vertical distribution + & & & @xcite & & no scattering + & & & @xcite&mw & stellar bulge , multiple + & & & & & scale lengths / heights + & & & @xcite&mw & face - on only + & & & @xcite&mw & stellar bulge + & & & @xcite&mw , smc & stellar bulge , multiple + & & & & & scale lengths / heights + & & & & & + expon . & clumpy & expon . & @xcite&mw & stellar bulge + & & & @xcite&mw & stellar bulge , ir + & & & & & modelling + & & & @xcite&mw & clumping of stars as well + & & & & & + arbitrary&homog./ & arbitrary & & & + & clumpy & & @xcite & & multi - phase clumpy medium + & & & & & and emitters + & & & @xcite&all&multi - phase clumpy medium + & & & & & and emitters , polarization , + & & & & & ir modelling + lrrr line & @xmath127,@xmath128 & l@xmath207/l@xmath208 & k(@xmath128)@xmath26k(@xmath127 ) + ratio & ( @xmath0m,@xmath0 m ) & & + h@xmath42 , h@xmath41 & 0.6563,0.4861 & 2.87 & 1.163 + h@xmath42 , h@xmath43 & 0.6563,0.4340 & 6.16 & 1.628 + br@xmath43 , h@xmath41 & 2.166,0.4861 & 0.0302 & 3.352 + br@xmath43 , pa@xmath41 & 2.166,1.282 & 0.175 & 0.457 + pa@xmath42 , h@xmath42 & 1.876,0.6563 & 0.123 & 2.104 + lccccccccc e / s0&0.050.15 & 0.100.20 & & 0.050.10 & & 0.020.04 & & @xmath180.02 & + sa sab&&0.400.65 & 0.901.25&0.200.40&0.500.75 & 0.100.15 & 0.300.40&@xmath570.05 & @xmath570.15 + sb scd&0.450.65 & 0.600.80 & 1.201.45 & 0.300.50&0.650.95 & 0.150.20&0.400.45 & 0.050.10 & 0.150.20 + irr & 0.250.40 & 0.300.45 & 0.600.75 & 0.100.15 & 0.300.40 & @xmath10.05 & @xmath10.15 & @xmath570.02 & @xmath570.06 + lcrrrrrrrr name & d & morphology & l@xmath209 & sfr & l@xmath79/l@xmath209 & a@xmath210 & a@xmath74 & a@xmath211 & a@xmath170 + & ( mpc ) & & ( l@xmath37 ) & ( m@xmath37 yr@xmath5 ) & & ( mag ) & ( mag ) & ( mag ) & ( mag ) + + ngc1614 & 74 . & sbc , merger . & 4.8e+11 & 54.9 & 0.91 & 5.67 & 2.78 & 1.20 & 3.08 + ngc4194 & 40 . & ibm , merger & 1.1e+11 & 9.9 & 0.86 & 3.20 & 1.57 & 0.68 & 1.74 + ngc6090 & 135 . & sd , merger & 4.1e+11 & 52.4 & 0.88 & 3.41 & 1.67 & 0.72 & 1.85 + + ngc1140 & 23 . & ibm & 9.1e+09 & 1.1 & 0.59 & 0.83 & 0.40 & 0.17 & 0.45 + ngc1510 & 12 . & sa0 & 8.1e+08 & 0.06 & 0.42 & 0.52 & 0.25 & 0.11 & 0.28 + tol1924@xmath26416 & 44 . & irr & 1.9e+10 & 2.8 & 0.42 & 0.52 & 0.25 & 0.11 & 0.28 + + & & & & & & 1.60 & 0.78 & 0.34 & 0.86 + rlrrrr & parameter & smmj14011@xmath760252 & ms 1512@xmath26cb58 & west mmd11 & l@xmath64 lbg + 1 & z & 2.565 & 2.729 & 2.982 & @xmath13 + 2 & ab@xmath212 & 21.25 & 20.41 & 24.05 & @xmath124.5 + 3 & g@xmath181 & 0.69 & 0.72 & 0.80 & 0.6 + 4 & @xmath118 & @xmath260.74 & @xmath261.34 & @xmath260.90 & @xmath261.5 + 5 & a@xmath213 & 3.1 & 1.8 & 2.7 & 1.5 + 6 & e(b@xmath26v)@xmath137(uv ) & 0.31 & 0.18 & 0.27 & 0.15 + 7 & e(b@xmath26v)@xmath137(h@xmath41 ) & & 0.1@xmath1450.1 & 0@xmath1450.2 & + 8 & f@xmath214(ir ) & 3.7e@xmath2613 & 2.3e@xmath2613 & 1.9e@xmath2614 & 3.2e@xmath2615 + 9 & f@xmath215(450 ) & 42@xmath1457 & & 22@xmath14523 & @xmath1810 + 10 & f@xmath214(450 ) & 25.9 , 71.6 & 19.5 , 51.9 & 2.1 , 5.3 & 0.36 , 0.89 + 11 & f@xmath215(850 ) & 15@xmath1452 & @xmath184 & 5.5@xmath1451.4 & @xmath181 + 12 & f@xmath214(850 ) & 4.0 , 19.0 & 3.2 , 14.6 & 0.4 , 1.6 & 0.06 , 0.27 + lrr age ( myr ) & @xmath216 & @xmath217 + 2 & @xmath262.70 & @xmath262.68 + 5 & @xmath262.50 & @xmath262.61 + 10 & @xmath262.45 & @xmath262.57 + 15 & @xmath262.30 & @xmath262.55 + 30 & @xmath262.09 & @xmath262.50 + 50 & @xmath261.93 & @xmath262.45 + 100 & @xmath261.61&@xmath262.38 + 300 & @xmath260.20&@xmath262.28 + 500 & & @xmath262.26 + 900 & & @xmath262.23 +

derived equivalences between finite dimensional algebras are known to be fundamental in representation theory and applications . unfortunately , still very few positive results are known about the structure of derived equivalences and about homological invariants . for instance , it is not known when ( a ) a derived equivalence between algebras @xmath0 and @xmath3 induces derived equivalences between certain centraliser subalgebras @xmath6 and @xmath7 , or in case of group algebras between subgroup algebras . it is also not known when ( b ) the existence of a derived equivalence implies that @xmath0 and @xmath3 share homological invariants such as global or dominant dimension . for known classes of derived equivalences , both questions are known to have dauntingly negative answers . the aim of this article is to identify large classes of algebras where both problems do have positive solutions . a starting point , and some hope , may be provided by the class of self - injective algebras , which have both global and dominant dimension infinite , except in the semisimple case . under some mild assumption , derived equivalences are known to preserve the property of being self - injective , hence also the values of these two homological dimensions . in full generality , derived equivalences between self - injective algebras @xmath0 and @xmath3 induce stable equivalences of morita type and thus there are equalities between global and also between dominant dimensions of @xmath0 and @xmath3 . this motivates considering algebras that are closely related to self - injective algebras and considering derived equivalences inducing stable equivalences of morita type , leading to the following more precise versions of ( a ) and ( b ) : the key concept to address both questions and to identify suitable and interesting classes of algebras is @xmath1-dominant dimension ( to be defined in [ defn : strong dominant dimension ] ) , where @xmath1 is the nakayama functor sending projective to injective modules over an algebra . concerning ( a ) , an assumption is needed to identify unique and non - zero associated self - injective centraliser subalgebras ( first used by martinez - villa @xcite ) of @xmath0 and @xmath3 , respectively , which then can be investigated for derived equivalence . in general , it may happen that an algebra @xmath0 having zero as associated self - injective algebra is derived equivalent to an algebra @xmath3 having a non - zero associated self - injective algebra ; see ( * ? ? ? * section 5 ) for an example . the condition that both algebras have faithful strongly projective - injective modules allows to identify non - trivial associated self - injective centraliser subalgebras ( lemma [ lem : two dominant dimensions coincide ] and definition [ defn : associatedselfinjectivealgebra ] ) , and it is strong enough to solve problem ( a ) : : _ let @xmath0 and @xmath3 be finite dimensional algebras of @xmath1-dominant dimension at least one , and let @xmath8 and @xmath9 be their associated self - injective centraliser subalgebras . if @xmath0 and @xmath3 are derived equivalent , then also @xmath2 and @xmath4 are derived equivalent . _ the proof is based on a stronger result ( theorem [ thm : restriction theorem ] ) , which shows that the given derived equivalence between @xmath0 and @xmath3 restricts to certain subcategories that are shown to determine the derived categories of @xmath2 and @xmath4 . the class of algebras of @xmath1-dominant dimension at least one contains all self - injective algebras , but also the morita algebras introduced in @xcite , which are characterised in ( theorem [ prop : characterise morita algebras ] ) as the algebras having @xmath1-dominant dimension at least two ; their @xmath1-dominant dimension coincides with the classical dominant dimension . morita algebras in turn contain gendo - symmetric algebras and hence several classes of algebras of interest in algebraic lie theory such as classical or quantised schur algebras and blocks of the bgg - category @xmath10 of semisimple complex lie algebras ; these algebras usually have finite global dimension , but are related to self - injective algebras by schur - weyl dualities . special cases of the derived restriction theorem state for instance : ( 1 ) two classical or quantised schur algebras @xmath11 ( with @xmath5 ) are derived equivalent only if the corresponding group algebras of symmetric groups or hecke algebras are so ( for the latter a derived equivalence classification is known by chuang and rouquier @xcite ) . + ( 2 ) auslander algebras of self - injective algebras of finite representation type are derived equivalent if and only if the self - injective algebras are so ( the latter derived equivalences are known by work of asashiba @xcite ) , and in this case the auslander algebras moreover are stably equivalent of morita type ( corollary [ cor : restriction theorem for auslander algebras ] ) . problem ( b ) does not , in general , have a positive answer for algebras of @xmath1-dominant dimension at least one ; only upper bounds for the differences in dimensions can be given ( which are valid in general , proposition [ prop : bound global dimension ] and theorem [ thm : bound the difference ] ) . to identify subclasses of algebras where problem ( b ) has a positive answer , two approaches are developed here : one approach identifies special derived equivalences , which do preserve both global and dominant dimensions : these are the ( iterated ) almost @xmath1-stable derived equivalences ( introduced in @xcite ) . these equivalences are known to induce stable equivalences of morita type . standard equivalences between self - injective algebras are of this form . the same is true for a larger class of algebras introduced here , the almost self - injective algebras , which include all self - injective algebras and also some algebras of finite global dimension , for instance schur algebras of finite representation type : the second approach concentrates on directly identifying global and dominant dimension inside some derived category ; in the case of dominant dimension , the associated self - injective centraliser subalgebra occurring in the derived restriction theorem is used . this approach works ( under the assumption of having dominant dimension at least one ) for all split algebras ( e.g. , algebras over an algebraically closed field ) having an anti - automorphism ( for instance , a duality ) preserving simples : dualities , i.e. involutory anti - automorphisms fixing simples , exist for instance for all cellular algebras . the second invariance theorem covers in particular classical and quantised schur algebras and even their blocks . in fact , the invariance property is strong enough ( theorems [ thm : global dimension of blocks of schur algebras ] and [ thm : dominant dimension of blocks of schur algebras ] ) to determine these dimensions by explicit combinatorial formulae in terms of weights and ( quantum ) characteristics , for all blocks of such schur algebras , using the derived equivalences constructed by chuang and rouquier . the main results of this article are motivated by various results in the literature , which are extended and applied here : the idea to use schur - weyl dualities and schur functors to compare homological data of self - injective algebras such as group algebras of symmetric groups and of their schur algebras ( quasi - hereditary covers ) , of finite global dimension , has been developed in @xcite . there it has been demonstrated that dominant dimension is not only crucial for existence of schur - weyl duality , @xcite , but also for the quality of the schur functor in preserving homological data , although typically on one side of schur - weyl duality there is an algebra of infinite global and dominant dimension and on the other side there is an algebra with finite such dimensions . these are derived equivalent only in degenerate ( semisimple ) cases . the same approach has been demonstrated to work for hochschild cohomology in @xcite , where also a first instance of the derived restriction theorem has appeared . the concept of ( iterated ) almost @xmath1-stable derived equivalence and its useful properties have been developed in @xcite , where the focus has been on the resulting stable equivalences of morita type ( which imply invariance of global and dominant dimension ) . chuang and rouquier s derived equivalence classification of blocks of symmetric groups and of some related algebras provides an important supply of examples . it turns out , however , that already these equivalences , more precisely those between quantised schur algebras , are not always iterated @xmath1-stable , which forces us to use the second approach to derived invariants , using anti - automorphisms fixing simples , in this case . after recalling two major homological invariants of algebras , global dimension and dominant dimension , a new variation of dominant dimension , @xmath1-dominant dimension , is introduced that will turn out to provide a crucial assumption in the main results . in the second subsection , the two main classes of algebras considered here will be discussed and related to @xmath1-dominant dimension ; these are the morita algebras , which will get characterised in terms of @xmath1-dominant dimension , and the new class of almost self - injective algebras . throughout , @xmath12 is an arbitrary field of any characteristic . algebras are _ finite dimensional _ @xmath12-vector spaces and , unless stated otherwise , modules are finitely generated left modules . when @xmath0 is an algebra , @xmath13 denotes the opposite algebra of @xmath0 , and @xmath14 is the enveloping algebra @xmath15 . let @xmath16 ( respectively @xmath17 ) be the category of all ( respectively all finitely generated ) left @xmath0-modules , and @xmath18 ( respectively @xmath19 ) the full subcategory of @xmath17 whose objects are the projective ( respectively injective ) left @xmath0-modules . let @xmath20 be the usual @xmath12-duality functor @xmath21 and @xmath22 the nakayama functor . we follow the conventions from @xcite . let @xmath23 be an additive category . an object @xmath24 in @xmath23 is called _ strongly indecomposable _ if @xmath25 is a local ring . an object @xmath26 in @xmath23 is called _ basic _ if @xmath26 is a direct sum of strongly indecomposable objects of multiplicity one each . for an object @xmath27 in @xmath23 , we write @xmath28 for the full subcategory of @xmath23 consisting of all direct summands of finite direct sum of copies of @xmath27 . by @xmath29 or @xmath30 we denote the composition of morphisms @xmath31 and @xmath32 in @xmath23 . a morphism @xmath33 is said to be _ radical _ in @xmath23 , if for morphisms @xmath34 and @xmath35 , the composition @xmath36 never is an isomorphism . in contrast to the composition rule for morphisms , we write @xmath37 for the composition of two functors @xmath38 and @xmath39 between additive categories . given a finite dimensional @xmath12-algebra @xmath0 , there are many homological invariants around to measure the complexity of @xmath0 from different points of view , and global dimension is the most widely used one . by definition , the global dimension of @xmath0 , denoted by @xmath40 , is the smallest number @xmath41 or @xmath42 such that @xmath43 for any @xmath44 and all @xmath45 . the following ( well - known ) characterisation can be found for instance in ( * ? ? ? * corollary 3.8 ) . [ lem : characterise global dimension ] let @xmath0 be a @xmath12-algebra . if @xmath46 , then @xmath40 is the largest number @xmath41 such that @xmath47 . dominant dimension was introduced by nakayama , and developed later mainly by morita and tachikawa , see @xcite for more information , and @xcite for a recent development partly motivating our aims and results . [ defn : dominant dimension ] let @xmath0 be a @xmath12-algebra . the _ dominant dimension _ of @xmath0 , denoted by @xmath48 , is defined to be the largest number @xmath49 ( or @xmath42 ) such that in a minimal injective resolution @xmath50 of the left regular @xmath0-module , @xmath51 is projective for all @xmath52 ( or @xmath42 ) . thus , @xmath53 not being projective , that is @xmath54 , is equivalent to @xmath0 not having a faithful projective - injective module . the module @xmath55 is projective and injective if and only if so is @xmath56 in @xmath57 . it follows that @xmath58 and thus @xmath48 can be defined alternatively via right @xmath0-modules . if @xmath59 , then there exists a unique ( up to isomorphism ) minimal faithful right @xmath0-module ( and also a unique up to isomorphism minimal faithful left @xmath0-module ) . it must be projective and injective , hence of the form @xmath60 for some idempotent @xmath61 in @xmath0 . if @xmath62 , then @xmath60 is a faithful balanced bimodule , i.e. , there is a double centralizer property , namely @xmath63 canonically . algebras of infinite dominant dimension are conjectured to be self - injective ( this is the celebrated _ nakayama conjecture _ ) , see @xcite . the following characterisation of dominant dimension is due to mller . [ prop : muller characterisation ] let @xmath0 be a @xmath12-algebra of dominant dimension at least @xmath64 . let @xmath60 be a minimal faithful right @xmath0-module and @xmath65 be an integer . then @xmath66 if and only if @xmath67 for @xmath68 . of particular interest later on will be certain derived equivalences , ( iterated ) almost @xmath1-stable derived equivalences , defined in @xcite . here , a certain subclass of both projective and injective ( projective - injective for short ) modules is crucial . this motivates the following variation of dominant dimension , which is crucial for our main results . [ defn : strong projective injective modules ] let @xmath0 be a @xmath12-algebra and @xmath69 be the nakayama functor . a projective @xmath0-module @xmath70 is said to be _ strongly projective - injective _ if @xmath71 is projective for all @xmath72 . by @xmath73 we denote the full subcategory of @xmath18 consisting of strongly projective - injective @xmath0-modules . in @xcite , strongly projective - injective modules are called @xmath1-stably projective ; since this may be misunderstood as implying @xmath1-stable , we use a different terminology here . strongly projective - injective modules are projective and injective , which justifies their name . an easy proof goes as follows , see also ( * ? ? ? * lemma 2.3 ) . first note that @xmath70 is strongly projective - injective if and only if so is each of its direct summands . thus we may assume that @xmath70 is indecomposable . since the nakayama functor @xmath74 sends indecomposable projective modules to indecomposable injective modules , it follows that @xmath75 are indecomposable projective - injective for all @xmath72 . but there are only finitely many indecomposable objects in @xmath18 , so there must exist @xmath76 such that @xmath77 . using again that @xmath78 is an equivalence , we deduce that @xmath79 . in particular , @xmath70 is both projective and injective . [ defn : strong dominant dimension ] let @xmath0 be a @xmath12-algebra . @xmath1-dominant dimension _ of @xmath0 , denoted by @xmath80 , is defined to be the largest number @xmath49 ( or @xmath42 ) such that in a minimal injective resolution @xmath50 of the left regular @xmath0-module , @xmath51 is strongly projective - injective for all @xmath52 ( or @xmath42 ) . by definition , @xmath81 , but in general there is no equality . here is an example illustrating the difference between these two dimensions . let @xmath0 be the path algebra @xmath82 of the quiver @xmath83 . then @xmath84 is projective - injective , but not strongly projective - injective , since @xmath85 and @xmath86 is not projective . as a result , @xmath87 , while @xmath88 . in our context , @xmath1-dominant dimension is important , since it allows to identify particular self - injective centraliser subalgebras : [ lem : two dominant dimensions coincide ] let @xmath0 be a @xmath12-algebra . if @xmath89 , then all projective - injective @xmath0-modules are strongly projective - injective , and thus @xmath90 . in this case , endomorphism rings of minimal faithful left @xmath0-modules are self - injective . suppose @xmath89 . then , a minimal faithful left @xmath0-module is of the form @xmath91 , and strongly projective - injective . we will use its endomorphism ring @xmath6 as ` the largest self - injective centraliser subalgebra ' . since @xmath89 , the injective envelope @xmath92 of @xmath93 is strongly projective - injective . let @xmath70 be an indecomposable projective - injective @xmath0-module . the composition is a split monomorphism . thus @xmath70 is a direct summand of @xmath92 , and in particular strongly projective - injective . consequently , all projective - injective @xmath0-modules are strongly projective - injective . hence the two dominant dimensions coincide . let @xmath91 be a minimal faithful left @xmath0-module . by assumption , it is strongly projective - injective . hence @xmath94 belongs to @xmath95 , and in particular @xmath96 , that is , @xmath6 is self - injective . the endomorphism ring of a strongly projective - injective @xmath0-module in general may not be self - injective , even when assuming @xmath89 . for instance , let @xmath0 be the self - injective nakayama algebra with cyclic quiver , three simple modules and @xmath97 . then each indecomposable projective module is injective as well , and even strongly projective - injective , but the endomorphism ring of a sum of two non - isomorphic indecomposable projective modules never is self - injective . the proof of lemma [ lem : two dominant dimensions coincide ] works not only for @xmath91 , but also for any direct sum of copies of @xmath91 . [ defn : associatedselfinjectivealgebra ] let @xmath0 be a @xmath12-algebra with @xmath89 and let @xmath91 be a minimal faithful left @xmath0-module . then the centraliser algebra @xmath6 is called the _ associated self - injective algebra of @xmath0_. these associated self - injective subalgebras have been introduced and strongly used before in work of martinez - villa @xcite reducing validity of the auslander - reiten conjecture on stable equivalences preserving the number of non - projective simple modules to the case of self - injective algebras . there , the setup is more general and the associated self - injective algebras have been allowed to be zero , which does not make sense in our situation as we need a strong connection between the given algebra and its associated self - injective subalgebra . the term ` associated self - injective algebra ' first occurred in @xcite . the term ` morita algebras ' ( not related to morita rings occurring in morita contexts ) was coined by kerner and yamagata in @xcite , when they investigated algebras first studied by morita @xcite as endomorphism rings of generators over self - injective algebras . the subclass of morita algebras consisting of endomorphism rings of generators over symmetric algebras , called gendo - symmetric algebras , was introduced and studied independently in @xcite . [ def : morita algebras ] a @xmath12-algebra @xmath0 is called a _ morita algebra _ if @xmath0 is isomorphic to for some self - injective algebra @xmath98 and some module @xmath99 . @xmath0 is called _ gendo - symmetric _ if in addition @xmath98 is symmetric . gendo - symmetric algebras form a large class of algebras , cutting across traditional boundaries such as finite or infinite global dimension . examples of finite global dimension include classical and quantised schur algebras @xmath11 ( with @xmath5 ) , blocks of the bernstein - gelfand - gelfand category @xmath10 and many other algebras occurring in algebraic lie theory and elsewhere . examples of infinite global dimension include symmetric algebras , for instance group algebras and hecke algebras . morita algebras include in addition self - injective , and in particular frobenius algebras , and their auslander algebras . morita algebras have been characterised in several ways , see @xcite . for our purposes , a new characterisation is needed in terms of @xmath1-dominant dimension ( section [ subsec : global dimension and dominant dimension ] ) : [ prop : characterise morita algebras ] let @xmath0 be a @xmath12-algebra . then @xmath100 if and only if @xmath0 is a morita algebra . suppose @xmath100 . then by lemma [ lem : two dominant dimensions coincide ] , the minimal faithful @xmath0-module @xmath91 is ( strongly ) projective - injective and its endomorphism ring @xmath6 is self - injective . moreover , @xmath0 has dominant dimension at least two , which implies a double centraliser property on @xmath91 , between @xmath0 and @xmath6 . therefore , @xmath0 is a morita algebra . conversely , if @xmath0 is a morita algebra , then @xmath0 is isomorphic to @xmath101 for some self - injective algebra @xmath98 and a generator @xmath27 in @xmath102 . let @xmath103 be the direct sum of all pairwise non - isomorphic indecomposable projective @xmath98-modules . by definition , @xmath103 is a direct summand of @xmath27 and @xmath104 . moreover , as left @xmath98-modules and there are isomorphisms of @xmath0-modules @xmath105 here , the isomorphism @xmath106 follows from @xmath107 . the isomorphism @xmath108 uses and the isomorphism @xmath109 follows from tensor - hom adjointness . @xmath110 uses @xmath111 . this proves the claim . now we construct an injective presentation of the left regular @xmath0-module @xmath112 ( or @xmath101 ) as follows : take an injective presentation @xmath113 of @xmath27 and apply @xmath114 to obtain the exact sequence @xmath115 of left note that both @xmath84 and @xmath116 are projective @xmath98-modules and thus belong to @xmath117 . therefore , @xmath118 for @xmath119 , and so @xmath100 . let @xmath0 be a morita algebra . then @xmath120 . this follows immediately from lemma [ lem : two dominant dimensions coincide ] and proposition [ prop : characterise morita algebras ] . gendo - symmetric algebras appeared first in @xcite , see @xcite for further information . in our context , there is the following characterisation : [ prop : properties on gendo - symmetric algebras ] let @xmath0 be a @xmath12-algebra . then @xmath0 is gendo - symmetric if and only if @xmath121 as @xmath0-bimodules . if @xmath0 is gendo - symmetric , then @xmath122 . the first claim follows from ( * ? ? ? * theorem 3.2 ) and the second one from ( * ? ? ? * proposition 3.3 ) . alternatively , the characterisation of @xmath48 also follows from proposition [ prop : muller characterisation ] combined with the following claim : since @xmath0 is a gendo - symmetric algebra , dualising both sides of the isomorphism @xmath123 yields an isomorphism @xmath124 as @xmath0-bimodules . thus @xmath125 is isomorphic to @xmath91 , and in particular it is projective - injective . now , for a minimal injective resolution @xmath126 of the left regular @xmath0-module , the first @xmath48-terms are projective - injective , hence belong to @xmath127 . applying @xmath128 to the sequence @xmath129 , and comparing the cohomologies of @xmath130 and @xmath131 , proves the claim . the second class of algebras we are going to study generalises self - injective algebras . an algebra @xmath0 is called an _ almost self - injective algebra _ , if @xmath89 and there is at most one indecomposable projective @xmath0-module that is not injective . among the examples are schur algebras of finite representation type , which have finite global dimension and which are morita algebras . schur algebras , and thus morita algebras , in general are not almost self - injective . schur algebras of finite representation type are examples of gendo - brauer tree algebras described and classified in @xcite . these algebras are representation - finite gendo - symmetric and in addition biserial ; the corresponding symmetric algebras are brauer tree algebras . conversely , almost self - injective algebras need not be morita algebras as the following example illustrates . let @xmath0 be the @xmath12-algebra given by the quiver @xmath132^\delta & 1 \ar@/^/[l]^\alpha \ar@/^/[r]^\beta & 3 \ar@/^/[l]^\theta}\ ] ] and relations @xmath133 . the loewy series of the indecomposable projective left @xmath0-modules are @xmath134 then both @xmath116 and @xmath135 are strongly projective - injective , and @xmath136 . thus @xmath0 is an almost self - injective algebra , but not a morita algebra . after recalling fundamental facts of derived morita theory , basic properties of standard equivalences will be shown and then almost @xmath1-stable derived equivalences will be explained , thus providing crucial tools for proofs later on . let @xmath23 be an additive category . a complex @xmath137 over @xmath138 is a sequence of morphisms @xmath139 in @xmath23 of the form @xmath140 with @xmath141 for all @xmath142 . we call @xmath137 a _ radical _ complex if all @xmath139 are radical morphisms . we denote by @xmath143 ( respectively @xmath144 ) the category of complexes ( respectively bounded complexes ) over @xmath138 , and by @xmath145 ( respectively @xmath146 ) the corresponding homotopy category . @xmath147 ( respectively @xmath148 ) is the derived category of complexes ( respectively bounded complexes ) over @xmath138 when @xmath138 is abelian . homotopy categories and derived categories are prominent examples of _ triangulated categories_. for an algebra @xmath0 , we write @xmath149 , @xmath150 and @xmath151 for @xmath152 , @xmath153 and @xmath154 respectively , where @xmath155 stands for @xmath156 or @xmath157 . [ lem : morphisms in derived categories ] let @xmath0 be an algebra . then : @xmath106 every complex of @xmath0-modules is isomorphic to a radical complex in the homotopy category @xmath158 . @xmath108 two radical complexes are isomorphic in the homotopy category @xmath158 if and only if so they are in @xmath159 . @xmath109 for two complexes @xmath137 and @xmath160 , if there exists an integer @xmath161 such that @xmath137 has no cohomology in degrees larger than @xmath161 ( i.e. , @xmath162 for @xmath163 ) , and @xmath160 has no cohomology in degrees smaller than @xmath161 ( i.e. , @xmath164 for @xmath165 ) , then @xmath166 + in particular , for any complex @xmath167 of @xmath0-modules , @xmath168)$ ] is isomorphic to its @xmath169-th cohomology @xmath170 for all @xmath169 . the first two statements are taken from @xcite ; the remaining statements can be shown by using truncations and cartan - eilenberg resolutions of complexes . the tensor product of two complexes @xmath137 and @xmath160 in @xmath171 is defined to be the total complex of the double complex with its @xmath172-term @xmath173 , and their tensor product in @xmath174 is the tensor product of their projective resolutions in @xmath171 @xcite . let @xmath137 be a complex over @xmath138 of the form @xmath175 . then the _ brutal truncations _ of @xmath137 are defined by cutting off the left or right hand part of the complex : @xmath176 and @xmath177 . there is an exact sequence of complexes @xmath178 , which also defines a triangle . _ derived equivalences _ are by definition equivalences of derived categories that preserve the triangulated structures , that is shift and triangles . two algebras _ @xmath0 and @xmath3 are derived equivalent _ if there is a derived equivalence between their derived categories . despite their importance , derived equivalences are still rather unknown and even basic questions are still open . the equivalence relation between algebras defined by derived equivalence does , however , admit a very satisfactory theory , known as _ morita theory for derived categories _ , due to rickard and ( more generally for dg algebras ) to keller . [ thm : morita theory for derived category ] let @xmath0 and @xmath3 be two @xmath12-algebras . the following statements are equivalent . * @xmath179 and @xmath180 are equivalent as triangulated categories . * @xmath181 and @xmath182 are equivalent as triangulated categories . * @xmath183 and @xmath184 are equivalent as triangulated categories . * @xmath185 and @xmath186 are equivalent as triangulated categories . * there exists a complex @xmath187 such that @xmath188 and * * @xmath189)=0 $ ] unless @xmath190 ; * * @xmath191 generates @xmath185 as triangulated category . the complex @xmath192 in ( 5 ) is called a _ tilting complex_. for any derived equivalence from @xmath179 to @xmath180 , the image of @xmath0 in @xmath180 is a tilting complex , and so is the preimage of @xmath3 in @xmath179 . it is not known whether the equivalences in ( 1)-(4 ) determine each other uniquely ( see ( * ? ? ? * section 7 ) ) . to fix the ambiguity , rickard @xcite associated to each derived equivalence a _ standard derived equivalence_. a complex @xmath193 is called a _ two - sided tilting complex _ if and only if @xmath194 for some complex @xmath195 . the complex @xmath196 is called an _ inverse _ of @xmath197 . the functor @xmath198 ( respectively , @xmath199 ) is a triangle equivalence with @xmath200 ( respectively , @xmath201 ) as a quasi - inverse . such a derived equivalence is called a _ standard derived equivalence_. it has been proved in @xcite that each derived equivalence @xmath202 induces a derived equivalence @xmath203 from @xmath204 to @xmath205 , and the image @xmath197 of @xmath206 under @xmath203 is a two - sided tilting complex such that @xmath207 for all @xmath208 . it is not known whether @xmath209 and @xmath210 agree on morphisms . two - sided tilting complexes can be used to provide further derived equivalences . let @xmath197 be a two - sided tilting complex in @xmath211 with an inverse @xmath196 . then , for each algebra @xmath212 , the functor @xmath213 is a triangle equivalence with @xmath200 as a quasi - inverse , and the functor @xmath214 defines a derived equivalence with @xmath215 as a quasi - inverse , see @xcite for more details . here are some properties of the standard derived equivalence @xmath216 , to be used later on . [ lem : standard derived equivalence ] let @xmath0 and @xmath3 be derived equivalent @xmath12-algebras . let @xmath216 be a standard derived equivalence from @xmath174 to @xmath217 defined by a two - sided tilting complex @xmath197 in @xmath218 as defined above in ( [ eqn : two sided standard derived equivalence ] ) . then for any complexes @xmath137 and @xmath160 in @xmath174 , there are isomorphisms in @xmath217 * @xmath219 , @xmath220 ; * @xmath221 ; * @xmath222 ; * @xmath223 . \(1 ) follows from the isomorphism @xmath224 and ( 2 ) follows from ( * ? ? ? * proposition 5.2 ) . to show ( 3 ) and ( 4 ) , note that the derived functor @xmath225 from @xmath226 to @xmath227 is right adjoint to the derived functor @xmath210 ( see @xcite ) . thus it is naturally isomorphic to the derived functor @xmath200 . similarly , the two derived functors @xmath228 and @xmath210 are naturally isomorphic . ( 3 ) then follows by a series of isomorphisms in @xmath229 : @xmath230 here the isomorphisms marked by @xmath231 follow by tensor - hom adjointness , and the isomorphism marked by @xmath232 follows from @xmath233 . to prove ( 4 ) , observe that @xmath234 in @xmath235 . thus by ( 1 ) and ( 3 ) @xmath236 in @xmath217 . let @xmath237 be a triangulated category , and let @xmath238 , i=1 , 2 $ ] be triangles in @xmath237 . if one of the following conditions is satisfied @xmath106 @xmath239)$ ] for @xmath240 ; @xmath108 @xmath241 , z)$ ] for @xmath240 , then @xmath242 and @xmath243 are isomorphic . [ lemma - isomorphic - triangles ] assume that condition ( 1 ) is satisfied . applying @xmath244 to @xmath245 ( @xmath240 ) yields isomorphisms @xmath246 . this means that each morphism factorises uniquely through @xmath247 . in particular , @xmath248 and @xmath249 for some . it follows that @xmath250 and @xmath251 , and thus by uniqueness , @xmath252 . hence @xmath253 is an isomorphism , and the commutative diagram @xmath254^{f_1}\ar[d]_{h } & z\ar@{=}[d]\\ y\ar[r]^{f_2 } & z } \ ] ] extends to an isomorphism between @xmath242 and @xmath243 . the proof is similar when assuming ( 2 ) . derived equivalences , by definition , preserve all triangles , especially the following type . let @xmath255 be an idempotent in the @xmath12-algebra @xmath0 . there are canonical triangles associated to @xmath61 in @xmath227 : @xmath256 \\ \mathbf{(ct2 ) } \ , \ , \ , \eta_e^a & : \quad a { \stackrel{\rho_e}{{\longrightarrow } } } { { \rm\bf r}{\rm hom}}_{eae}(ea , ea ) { \longrightarrow}v_a(e ) { \longrightarrow}a[1 ] \end{aligned}\ ] ] where @xmath257 is induced by the multiplication map @xmath258 , and @xmath259 is induced by the canonical morphism @xmath260 . the triangle @xmath261 plays a crucial role in the analysis of recollements of derived categories @xcite . the following result implies that the property of being a canonical triangle is preserved under certain derived equivalences . [ lem : derived - equivalence - preserves - canonical - triangle ] let @xmath0 and @xmath3 be derived equivalent @xmath12-algebras , and @xmath262 a two - sided tilting complex with inverse @xmath263 . let @xmath61 and @xmath264 be idempotents in @xmath0 and @xmath3 respectively . assume that the standard derived equivalence @xmath265 restricts to a triangle equivalence @xmath266 . then @xmath267 and @xmath268 as triangles in @xmath229 . since @xmath269 is an equivalence of additive categories with a quasi - inverse @xmath270 , it follows that @xmath271 is an equivalence of triangulated categories with a quasi - inverse @xmath272 . similarly , the derived functor @xmath273 defines an equivalence from @xmath274 to @xmath275 with a quasi - inverse @xmath276 . note that @xmath277 and ( see @xcite ) . @xmath278 _ proof of claim . _ by assumption , @xmath279 belongs to @xmath280 . similarly , hence ( a ) . there are the following isomorphisms @xmath281 it follows that @xmath282 . by symmetry , also @xmath283 , hence ( b ) . to prove the first statement in ( c ) , consider the isomorphisms @xmath284 which use @xmath271 being an equivalence with quasi - inverse @xmath272 and @xmath285 . the second statement in ( c ) follows similarly . now , the isomorphisms ( the second one using ( a ) and ( c ) ) @xmath286 combined with lemma [ lem : standard derived equivalence ] ( 1 ) yield that @xmath287 is a triangle of the following form @xmath288\ ] ] in @xmath217 , where @xmath289 . so , the triangles @xmath290 and @xmath291 have at least two terms in common . to identify them as triangles , note that @xmath292 and since @xmath293 belongs to @xmath294 by ( b ) . then @xmath295)=0= { { \rm hom } } _ { { \mathscr{d}({{b^{\mathrm{e}}{\sf \mbox{-}mod}}})}}(bf{\otimes^{\rm\bf l}}_{fbf}fb , u_b(f)'[i ] ) , \forall i\in \mathbb{z}.\ ] ] here , the vanishing follows by adjointness . thus @xmath290 and @xmath291 are isomorphic triangles in @xmath217 by lemma [ lemma - isomorphic - triangles ] . so , @xmath290 is as claimed . it remains to check the claim about @xmath296 . the canonical triangle @xmath297 is isomorphic to @xmath298 . _ proof of claim . _ applying @xmath299 to @xmath261 results in a triangle @xmath300,\ ] ] and @xmath301 by adjointness . let @xmath302 . by definition , @xmath303 . using adjointness again , this implies @xmath304 , that is , @xmath305 . similarly , @xmath306 . it follows that @xmath307 , { { \rm\bf r}{\rm hom}}_{eae}(ea , ea))=0={{\rm hom } } _ { { \mathscr{d}({{a^{\rm e}{\sf \mbox{-}mod}}})}}(v_a(e)[i ] , { { \rm\bf r}{\rm hom}}_{eae}(ea , ea))\ ] ] for all @xmath142 . by lemma [ lemma - isomorphic - triangles ] , the triangles @xmath308 and @xmath297 are isomorphic , which proves the claim . by the first part of the proof , @xmath267 . applying lemma [ lem : standard derived equivalence ] ( 3 ) shows that @xmath309 . derived equivalences in general fail to preserve homological invariants such as global or dominant dimension . in this respect , stable equivalences of morita type behave much better . unfortunately , derived equivalences between algebras that are not self - injective , in general do not induce stable equivalences . the problem of finding derived equivalences , which do induce stable equivalences of morita type , has been addressed in @xcite by introducing a new class of derived equivalences , called _ almost @xmath1-stable derived equivalences _ , and relating them with stable equivalences . as a crucial feature , these derived equivalences preserve many homological invariants . let @xmath310 be a derived equivalence between two @xmath12-algebras @xmath0 and @xmath3 . we call @xmath209 an _ almost @xmath1-stable derived equivalence _ if the following conditions are satisfied . @xmath106 the radical tilting complex @xmath311 in @xmath186 has nonzero terms only in positive degrees , that is , @xmath312 for all @xmath313 ; the radical tilting complex @xmath314 in @xmath185 has nonzero terms only in negative degrees , that is , @xmath315 for all @xmath72 . @xmath108 @xmath316 and @xmath317 . note that we may assume without loss of generality that the tilting complex @xmath318 is radical by lemma [ lem : morphisms in derived categories ] . the two conditions for @xmath192 are equivalent to those for @xmath319 @xcite . to generalise this type of derived equivalence , but to keep many interesting properties , the following _ iterated almost @xmath1-stable derived equivalences _ have been introduced . let @xmath310 be a derived equivalence between two @xmath12-algebras @xmath0 and @xmath3 . we call @xmath209 an _ iterated almost @xmath1-stable derived equivalence _ , if there exists a sequence of derived equivalences @xmath320 of @xmath12-algebras @xmath321 @xmath322 for some @xmath323 with @xmath324 and @xmath325 such that each @xmath326 or @xmath327 is an almost @xmath1-stable derived equivalence and @xmath328 . in section [ sec : derived equivalences for morita algebras ] we will see that all derived equivalences between certain classes of algebras are iterated almost @xmath1-stable derived equivalences . this will make full use of the characterisations developed in @xcite . the crucial property in our context is : [ prop : almost nu - stable preserve dimensions ] let @xmath310 be an iterated almost @xmath1-stable derived equivalences between @xmath12-algebras . then @xmath329 and @xmath330 . the derived restriction theorem , to be proved in this section , states that derived equivalences between two algebras restrict to derived equivalences between their associated self - injective centraliser subalgebras , provided the two given algebras have @xmath1-dominant dimension at least one . a subcategory of the bounded homotopy category of projective modules will be defined and shown to be invariant under derived equivalence , under the assumption on @xmath1-dominant dimension . this will be the key ingredient in the proof of the derived restriction theorem . recall that for an algebra @xmath0 with @xmath89 and minimal faithful left module @xmath91 , the algebra @xmath331 is called the _ associated self - injective algebra of @xmath0_. then the category @xmath73 is additively generated by @xmath91 . that @xmath98 is self - injective has been shown in the proof of lemma [ lem : two dominant dimensions coincide ] . suppose now that @xmath0 and @xmath3 are derived equivalent @xmath12-algebras . like dominant dimension , @xmath1-dominant dimension is not a derived invariant . in particular , @xmath89 and @xmath332 may happen , and there may be no reasonable way to define a non - zero associated self - injective algebra for @xmath3 in this case . an example of such a situation has been given , for different reasons , in ( * ? ? ? * section 5 ) . therefore , it seems difficult to deduce any connection between the associated self - injective algebras of @xmath0 and @xmath3 in general without assuming both algebras have @xmath1-dominant dimension at least one . if we , however , assume that both @xmath0 and @xmath3 have @xmath1-dominant dimension at least @xmath333 , then theorem [ thm : restriction theorem ] below shows that any derived equivalence between @xmath0 and @xmath3 restricts to a derived equivalence between their associated self - injective algebras . the main tool for showing is the following subcategory of the homotopy category . [ defn : nu - stable complex ] let @xmath0 be a @xmath12-algebra and @xmath74 be the nakayama functor . define @xmath334 note that for @xmath335 , the complex @xmath336 is defined componentwise . when @xmath0 has arbitrary @xmath1-dominant dimension , a complex in @xmath337 need not be isomorphic in @xmath338 to a complex in @xmath339 ; this is illustrated by an example below . however : [ prop : nu - stable complexes ] let @xmath0 be a @xmath12-algebra with @xmath89 . then @xmath339 is the smallest triangulated full subcategory of @xmath185 that contains @xmath337 and is closed under taking direct summands . in particular , every complex in @xmath337 is isomorphic in @xmath185 to a complex in the category @xmath339 . let @xmath340 be the smallest triangulated full subcategory of @xmath185 which contains @xmath337 and is closed under taking direct summands . let @xmath103 be a basic additive generator of @xmath73 . then @xmath341 is isomorphic to @xmath103 in @xmath342 , because @xmath341 is again basic and strongly projective - injective and has the same number of indecomposable direct summands as @xmath103 . therefore @xmath103 belongs to @xmath337 . since @xmath339 is the smallest triangulated full subcategory of @xmath185 which contains @xmath103 and is closed under taking direct summands , it follows that @xmath343 . to finish the proof , we need to show @xmath344 , that is , every radical complex in @xmath337 is isomorphic in @xmath185 to a complex in @xmath339 . without loss of generality , we assume that @xmath345 . let @xmath346 . so the complex @xmath347 is of the form @xmath348 we will prove by induction on @xmath161 that @xmath347 is isomorphic in @xmath185 to a complex in @xmath339 . if @xmath190 , then @xmath349 in @xmath183 implies that @xmath350 in @xmath342 , and so @xmath351 . in general , we first prove : @xmath352 is strongly projective - injective . since @xmath347 is a radical complex in @xmath185 and @xmath78 is an equivalence , it follows that @xmath336 is a radical complex in @xmath353 . let @xmath354 be a quasi - isomorphism . then @xmath336 is an injective resolution of @xmath347 . by the construction of cartan - eilenberg injective resolutions @xmath347 admits an injective resolution @xmath355 with the properties : @xmath355 is quasi - isomorphic to @xmath347 , and @xmath356 for @xmath313 and @xmath53 is the injective envelope of @xmath352 . by the uniqueness of injective resolutions up to homotopy , the radical complex @xmath336 and the complex @xmath355 are isomorphic in @xmath357 , and therefore @xmath358 is a direct summand of @xmath53 by lemma [ lem : morphisms in derived categories ] . since @xmath89 , the injective envelope @xmath53 of @xmath352 is strongly projective - injective . it follows that @xmath358 and hence @xmath352 are strongly projective - injective as well . let @xmath359 be the mapping cone of @xmath360 , a complex of the form @xmath361}{{\longrightarrow } } } p^1\oplus \nu_{_a}(p^0){\longrightarrow}\cdots { \longrightarrow}\nu_{_a}(p^n){\longrightarrow}0\ ] ] where @xmath362 is placed in degree @xmath363 . since @xmath360 is a quasi - isomorphism , it follows that @xmath359 is an acyclic complex , and thus the morphism @xmath364 : p^0\to p^1\oplus \nu_{_a}(p^0)$ ] splits in @xmath342 because @xmath352 is injective . let @xmath365 and @xmath366 be morphisms in @xmath342 such that @xmath367 . then @xmath368 is invertible in @xmath369 by the assumption that @xmath370 is a radical morphism . as a result , @xmath371 is a split monomorphism , and even an isomorphism as @xmath352 and @xmath358 have the same number of indecomposable direct summands . now , let @xmath372 be the brutal truncation of @xmath347 and let @xmath373 be the corresponding truncation of the chain morphism @xmath360 . starting from the triangle , the following commutative diagram in @xmath374 gives a morphism of triangles in @xmath374 @xmath375\ar[r]\ar[d]^{f^0[-1 ] } & { p^{\bullet}}_{{\geqslant}1}\ar[r]\ar[d]^{{f^{\bullet}}_{{\geqslant}1 } } & { p^{\bullet}}\ar[r]\ar[d]^{{f^{\bullet } } } & p^0\ar[d]^{f^0}\\ \nu_{_a}(p^0)[-1]\ar[r ] & \nu_{_a}({p^{\bullet}}_{{\geqslant}1 } ) \ar[r ] & \nu_{_a}({p^{\bullet}})\ar[r ] & \nu_{_a}(p^0 ) . } \ ] ] since @xmath360 is a quasi - isomorphism ( by assumption ) and so is @xmath371 ( by the arguments above ) , it follows that @xmath376 is a quasi - isomorphism . therefore @xmath377 , and by induction @xmath378 $ ] is isomorphic in @xmath158 to a complex in @xmath339 . using @xmath379 implies that @xmath347 is isomorphic in @xmath158 to a complex in @xmath339 . _ example . _ without the assumption on @xmath1-dominant dimension , proposition [ prop : nu - stable complexes ] may fail in general : let @xmath0 be the @xmath12-algebra given by the quiver @xmath380^{\alpha } & 2\ar@/^/[l]^{\beta}\ar@/^/[r]^{\gamma } & 3\ar@/^/[l]^{\delta } } \ ] ] and relations @xmath381 . the indecomposable projective left @xmath0-modules are @xmath382 the indecomposable injective left @xmath0-modules are @xmath383 let @xmath347 be the complex @xmath384 , where @xmath385 is the unique ( up to scalar ) non - zero map , and @xmath84 is placed in degree zero . then @xmath336 is a complex of the form @xmath386 . the obvious surjective maps @xmath387 and @xmath388 define a chain map from @xmath347 to @xmath389 and is a quasi - isomorphism . as a result , @xmath390 , but @xmath347 does not belong to @xmath339 , since @xmath391 . [ thm : restriction theorem ] let @xmath0 and @xmath3 be derived equivalent @xmath12-algebras , both of @xmath1-dominant dimension at least @xmath333 . then any derived equivalence @xmath392 restricts to an equivalence of triangulated subcategories . without loss of generality , we may assume that @xmath209 is a standard derived equivalence . then @xmath209 induces @xmath393 as triangulated subcategories , and @xmath394 in @xmath184 , for any @xmath395 . thus @xmath396 and so by proposition [ prop : nu - stable complexes ] . let @xmath397 be a quasi - inverse of @xmath209 . the same arguments applied to @xmath397 imply @xmath398 . therefore , @xmath209 induces an equivalence @xmath399 . _ theorem [ thm : restriction theorem ] has two predecessors : a special case , stated only for gendo - symmetric algebras , was proved in @xcite , where it was used to relate the hochschild cohomology of @xmath0 and that of its associated self - injective algebra . a result in the same spirit as theorem [ thm : restriction theorem ] was obtained in @xcite without any restriction on algebras , but assuming @xmath209 to be an ( iterated ) almost @xmath1-stable derived equivalence . [ cor : restriction theorem for morita algebras ] let @xmath0 and @xmath3 be derived equivalent @xmath12-algebras , both of @xmath1-dominant dimension at least @xmath333 . then the associated self - injective algebras of @xmath0 and @xmath3 are derived equivalent . in particular , every derived equivalence of morita algebras induces a derived equivalence of their associated self - injective algebras . let @xmath98 be an associated self - injective algebra of @xmath0 . then by definition @xmath400 as additive categories , and thus @xmath401 as triangulated categories . the statement then follows directly from theorem [ thm : restriction theorem ] . an application of corollary [ cor : restriction theorem for morita algebras ] is to auslander algebras : a ( finite dimensional ) @xmath12-algebra @xmath0 is said to be of _ finite representation type _ , if there are only finitely many indecomposable @xmath0-modules ( up to isomorphism ) . the _ auslander algebra _ @xmath402 of @xmath0 is defined to be the endomorphism ring of the direct sum of all pairwise non - isomorphic indecomposable left @xmath0-modules . [ cor : restriction theorem for auslander algebras ] let @xmath0 and @xmath3 be self - injective @xmath12-algebras , both of finite representation type . let @xmath402 and @xmath403 be the auslander algebras of @xmath0 and @xmath3 respectively . then \(1 ) @xmath402 and @xmath403 are derived equivalent if and only if @xmath0 and @xmath3 are derived equivalent . \(2 ) if @xmath402 and @xmath403 are derived equivalent , then they are stably equivalent of morita type . \(1 ) if @xmath0 and @xmath3 are derived equivalent , then by ( * ? ? ? * corollary 3.13 ) , their auslander algebras @xmath402 and @xmath403 are derived equivalent . conversely , if @xmath402 and @xmath403 are derived equivalent , then by corollary [ cor : restriction theorem for morita algebras ] @xmath0 and @xmath3 are derived equivalent , since @xmath0 and @xmath3 are the associated self - injective algebras of @xmath402 and @xmath403 respectively . \(2 ) if @xmath402 and @xmath403 are derived equivalent , then @xmath0 and @xmath3 are derived equivalent by ( 1 ) , and thus stably equivalent of morita type by ( * ? ? ? * corollary 5.5 ) . now ( * ? ? ? * theorem 1.1 ) implies that @xmath402 and @xmath403 are stably equivalent of morita type . in this section , the two invariance theorems will be proven . for almost self - injective algebras the approach is to show that standard equivalences have a special form ; they are iterated almost @xmath1-stable and therefore preserve both global and dominant dimension . for algebras with a duality , a rather different approach will be taken , identifying the two homological dimensions inside the derived category . before addressing invariance , in the first subsection the general question is addressed how much homological dimensions can vary under derived equivalences . as it is well - known , the difference of global dimensions of two derived equivalent algebras is bounded by the length of a tilting complex inducing a derived equivalence , see for example ( * ? ? ? * section 12.5(b ) ) . more precisely , let @xmath0 be a @xmath12-algebra , and define the _ length _ of a radical complex @xmath137 in @xmath404 to be @xmath405 the length of an arbitrary complex @xmath160 in @xmath404 is defined to be the length of the unique radical complex that is isomorphic to @xmath160 in @xmath404 ( lemma [ lem : morphisms in derived categories ] ) . [ prop : bound global dimension ] let @xmath406 be a derived equivalence between @xmath12-algebras . then @xmath407 . naively , one may expect that dominant dimension behaves similarly under derived equivalences . here is a counterexample : let @xmath65 be an integer , and let @xmath0 be the @xmath12-algebra given by the quiver @xmath408_{\alpha_1 } & 3\ar[l]_{\alpha_2 } & \cdots \ar[l]_{\alpha_3 } & 2n\ar[l]_(.45){\alpha_{2n-1 } } & 2n+1\ar[l]_{\alpha_{2n } } } \ ] ] and relations @xmath409 . let @xmath410 denote the simple left @xmath0-module corresponding to the vertex @xmath169 and @xmath411 be the projective cover of @xmath410 . the modules @xmath411 are projective - injective for @xmath412 . the projective dimensions of the simple modules are @xmath413 , @xmath414 for @xmath415 , and the minimal injective resolution of the left regular @xmath0-module is of the form @xmath416 consequently , @xmath417 . let @xmath418 be the apr - tilting module ( see , for instance , ( * ? ? ? * vi.2.8 ) ) associated with the projective simple @xmath0-module @xmath419 , and let @xmath420 . then @xmath421 and therefore the derived equivalence between @xmath0 and @xmath3 induced by @xmath422 is given by a two - term tilting complex . by direct computation , @xmath3 is seen to be isomorphic to the @xmath12-algebra given by the same quiver as @xmath0 but with different relations @xmath423 for @xmath424 . as a result , @xmath425 and the difference between dominant dimensions of @xmath0 and @xmath3 is @xmath426 , although the derived equivalence is induced by a tilting module . although this example smashes any hope to bound the difference of dominant dimensions of derived equivalent algebras in terms of lengths of tilting complexes , there are still some cases where both global dimension and dominant dimension behave nicely , see @xcite and @xcite . note that both algebras in the example above are of @xmath1-dominant dimension @xmath427 . this suggests to restrict attention to algebras of @xmath1-dominant dimension at least @xmath333 - a restriction that has been needed for theorem [ thm : restriction theorem ] and that will be further justified by the invariance results later on . [ thm : bound the difference ] let @xmath0 and @xmath3 be @xmath12-algebras , both of @xmath1-dominant dimension at least @xmath333 and derived equivalent by @xmath310 . then @xmath428 . let @xmath429 and @xmath430 . since @xmath431 by ( * ? ? ? * lemma 2.1 ) , it is enough to show that @xmath432 if @xmath433 , there is nothing left to prove . assume that @xmath209 is a standard derived equivalence , @xmath434 and @xmath435 is a radical tilting complex in @xmath186 of the following form ( up to degree shift ) @xmath436 where @xmath352 is nonzero and placed in degree @xmath427 . take a cartan - eilenberg injective resolution @xmath355 of @xmath347 that is the total complex of the double complex obtained by taking minimal injective resolutions of @xmath437 for all @xmath169 . since @xmath438 and @xmath439 , the modules @xmath51 are strongly projective - injective for @xmath440 by lemma [ lem : two dominant dimensions coincide ] . let @xmath441 and @xmath442 be the brutal truncations of @xmath355 . by definition of brutal truncation , there is a triangle in @xmath443 @xmath444{\longrightarrow}{i^{\bullet}}_{{\geqslant}m}{\longrightarrow}{i^{\bullet}}{\longrightarrow}{i^{\bullet}}_{<m}.\ ] ] the standard derived equivalence @xmath209 lifts to a derived equivalence ( denoted by @xmath209 again ) from @xmath374 to @xmath443 . applying @xmath445 to the triangle @xmath231 , we obtain the triangle in @xmath374 @xmath446{\longrightarrow}\mathcal{f}^{-1}({i^{\bullet}}_{{\geqslant}m}){\longrightarrow}\mathcal{f}^{-1}({i^{\bullet}}){\longrightarrow}\mathcal{f}^{-1}({i^{\bullet}}_{<m } ) \end{aligned}\ ] ] and therefore the long exact sequence @xmath447 note that @xmath448 in @xmath374 , and @xmath449 belongs to @xmath339 by the construction of @xmath441 and theorem [ thm : restriction theorem ] . moreover , lemma [ lem : morphisms in derived categories ] implies for @xmath450 , @xmath451 ) \cong { { \rm hom } } _ { { \mathscr{d}(b)}}({p^{\bullet } } , { i^{\bullet}}_{{\geqslant}m}[i ] ) = 0 \end{aligned}\ ] ] since @xmath452)^p = 0 $ ] for @xmath453 and @xmath450 . therefore , from the long exact sequence ( [ eqn : long exact sequence ] ) , @xmath454 for @xmath313 , @xmath455 , and for @xmath456 . hence @xmath449 is isomorphic to a radical complex in @xmath339 of the form @xmath457 such that @xmath458 is exact . consequently the dominant dimension of @xmath0 is at least @xmath459 , as desired . a special case of theorem [ thm : bound the difference ] is : [ cor : bound difference for morita algebras ] let @xmath0 and @xmath3 be morita algebras . if there is a derived equivalence , then @xmath428 . interactions between derived equivalences and stable equivalences frequently seem to be of particular interest , see @xcite and the references therein . * corollary 5.5 ) and corollary [ cor : restriction theorem for auslander algebras ] state that , for self - injective algebras and auslander algebras of finite representation type self - injective algebras , derived equivalences imply stable equivalences of morita type . the following theorem implies that the same holds for almost self - injective algebras , by characterising all derived equivalences among them as iterated almost @xmath1-stable derived equivalences . [ thm : derived equivalences for almost self - injective ] let @xmath0 and @xmath3 be derived equivalent almost self - injective algebras . then any standard derived equivalence @xmath392 is an iterated almost @xmath1-stable derived equivalence ( up to shifts ) . in particular , derived equivalent almost self - injective algebras are stably equivalent of morita type . to prove this theorem , we need some preparations . first , we recall some basics on @xmath460-split sequences introduced in @xcite , see also @xcite . let @xmath461 be an additive category and let @xmath462 be a full subcategory of @xmath461 . a morphism @xmath463 in @xmath461 is a _ right @xmath462-approximation _ , if @xmath464 and for any @xmath465 , the canonical morphism @xmath466 is an epimorphism . we call @xmath264 _ right minimal _ if an equality @xmath467 implies that @xmath468 is an isomorphism , for left @xmath462-approximations and left minimal morphisms are defined similarly . let @xmath138 be a triangulated category and let @xmath469 be a full ( not necessarily triangulated ) additive subcategory of @xmath138 . a triangle in @xmath138 @xmath470\ ] ] is called a _ @xmath469-split triangle _ if @xmath264 is a left @xmath469-approximation and @xmath41 is a right @xmath469-approximation . a full subcategory @xmath471 of @xmath138 is called a _ tilting subcategory _ if @xmath472)=0 $ ] for all @xmath473 and @xmath23 itself is the only triangulated subcategory of @xmath23 that contains @xmath474 and is closed under taking direct summands . an object @xmath422 in @xmath138 is a _ tilting object _ if @xmath475 is a tilting subcategory of @xmath138 . for example , all tilting complexes over an algebra @xmath0 are tilting objects in @xmath185 . [ lem : d - split sequences ] let @xmath138 be a triangulated category , and @xmath469 an additive full subcategory of @xmath138 . let @xmath476 $ ] be a @xmath469-split triangle . then : \(1 ) suppose that @xmath264 is left minimal and @xmath41 is right minimal , and that @xmath477 and @xmath478 are decompositions of @xmath24 and @xmath26 into strongly indecomposable direct summands . then @xmath479 . in particular , if indecomposable objects in @xmath138 are strongly indecomposable , then @xmath24 is indecomposable if and only if so is @xmath26 . \(2 ) if @xmath480 is a tilting subcategory of @xmath23 and @xmath481 is injective , then @xmath482 is also a tilting subcategory of @xmath23 . \(3 ) if @xmath482 is a tilting subcategory of @xmath23 and @xmath483 is injective , then @xmath484 is also a tilting subcategory of @xmath23 . \(2 ) and ( 3 ) follow from ( * ? ? ? * theorem 2.32 ) . it remains to prove ( 1 ) . if @xmath485 , then and we are done . now we assume that @xmath486 . then @xmath487 and @xmath488 , since otherwise @xmath489 or @xmath490 , which contradicts the minimality of @xmath264 and @xmath41 . to proceed , let @xmath491 and @xmath492 . for any @xmath493 and @xmath494 , there exist @xmath495 with @xmath496 , and @xmath497 with @xmath498 since @xmath264 is a left @xmath460-approximation . therefore , @xmath499 , which implies @xmath500 . similarly @xmath501 and so @xmath502 is a two - sided ideal of @xmath25 . to see @xmath503 , it suffices to show , by ( * ? ? ? * theorem 15.3 , p. 166 ) , that @xmath504 is invertible in @xmath25 for each @xmath493 . let @xmath496 for some @xmath495 . in the diagram in @xmath23 @xmath505\ar[r]^{-h[-1 ] } \ar@{=}[d]^{\mathrm{id } } & x \ar[r]^f \ar[d]^{1-\alpha } & d \ar[r]^g \ar@{-->}[d]^\beta & y \ar@{=}[d]^{\mathrm{id}}\\ y[-1]\ar[r]^{-h[-1 ] } & x \ar[r]^f & d \ar[r]^g & y}\ ] ] the first square commutes because @xmath506\cdot \alpha = h[-1]\cdot f\cdot u=0 $ ] , and @xmath507 exists so that the other squares commute by axioms of triangulated categories . since @xmath41 is a right minimal morphism , @xmath507 is an isomorphism , and thus again by axioms of triangulated categories , @xmath504 is an isomorphism . by similar arguments , @xmath508 is a two - sided ideal of @xmath509 and it is contained in the jacobson radical @xmath510 of @xmath509 . we first construct a ring homomorphism @xmath511 as follows . for each @xmath512 , there exists a commutative diagram in @xmath23 @xmath513^f \ar[d]^\alpha & d \ar[r]^g \ar@{.>}[d]^\beta & y \ar[r]^h \ar@{-->}[d]^\gamma & x[1]\ar[d]^{\alpha[1]}\\ x \ar[r]^f & d \ar[r]^g & y \ar[r]^h & x[1]}\ ] ] where @xmath507 exists since @xmath264 is a left @xmath460-approximation , and @xmath514 exists by axioms of triangulated categories . if @xmath515 and @xmath516 are different choices such that @xmath517 and @xmath518 , then @xmath519 which implies that @xmath520 for some @xmath521 . in other words , the image of @xmath514 in the quotient ring @xmath522 is well - defined , and thus @xmath523 is well - defined . then @xmath524 is a ring homomorphism . it is surjective since @xmath41 is a right @xmath460-approximation . the kernel of @xmath524 equals @xmath502 . for any @xmath493 , there exists @xmath525 with @xmath526 . so @xmath527 @xmath528\cdot v[1]=0 $ ] which implies that @xmath523 factors through @xmath41 and thus @xmath529 . on the other hand , for any , there exists @xmath521 with @xmath530 , and so @xmath506\cdot \alpha = \phi(\alpha)[-1]\cdot h[-1 ] = ( u\cdot g\cdot h)[-1]=0 $ ] which implies that @xmath468 factors through @xmath264 and thus . altogether , @xmath524 induces an isomorphism @xmath531 , and hence further an isomorphism @xmath532 since both @xmath24 and @xmath26 are decomposed into strongly indecomposable direct summands , it follows that both rings @xmath25 and @xmath509 are semi - perfect rings by , and therefore the isomorphism above implies that @xmath25 and @xmath509 have the same number of simple left modules , or equivalently the same number of indecomposable projective left modules . consequently , @xmath24 and @xmath26 have the same number of indecomposable direct summands , that is @xmath479 . [ lem : tilting complex over almost self - injective algebras ] let @xmath0 be an almost self - injective algebra , but not self - injective . let @xmath103 be an additive generator of @xmath73 and let @xmath70 be the unique indecomposable projective left @xmath0-module such that @xmath533 . let @xmath534 be an indecomposable radical complex in @xmath185 . if @xmath535 is a tilting complex over @xmath0 , then at least one of the following two assertions holds true : \(1 ) there exists @xmath536 such that @xmath537 , @xmath315 for all @xmath72 and for all @xmath538 , and @xmath539 in @xmath73 for all @xmath540 ; \(2 ) there exists @xmath541 such that @xmath542 , @xmath315 for all @xmath313 and for all @xmath543 , and @xmath539 in @xmath73 for all @xmath544 . _ special case . _ the complex @xmath192 has only one nonzero term , that is , @xmath545 $ ] for some @xmath546 and @xmath547 . then @xmath548 is indecomposable since @xmath192 is indecomposable . if , then @xmath549 which contradicts the assumption that @xmath535 , as a tilting complex , generates @xmath185 . since each indecomposable projective left @xmath0-module is either isomorphic to @xmath70 or strongly projective - injective , by the definition of almost self - injective algebras @xmath548 must be isomorphic to @xmath70 . we still have to show that @xmath550 . assume @xmath551 . then by the self - orthogonality of tilting complexes , , which implies @xmath552 , a contradiction to @xmath89 . so , @xmath553 , which satisfies both conditions ( 1 ) and ( 2 ) . in the _ general case _ , @xmath554 is an indecomposable radical complex of the following form @xmath555 where @xmath556 are nonzero projective left @xmath0-modules and @xmath557 . using the self - orthogonality of the tilting complex @xmath535 , we are going to check : to see ( a ) , assume on the contrary that @xmath558 is a nonzero common direct summand of both @xmath559 and @xmath560 . let @xmath561 be a split epimorphism from @xmath559 to @xmath558 and @xmath562 be a split monomorphism from @xmath558 to @xmath560 . then the composition @xmath563 from @xmath559 to @xmath560 defines a nonzero morphism in @xmath564)$ ] , because @xmath192 is a radical complex . but this contradicts the assumption that @xmath192 is self - orthogonal and @xmath565 , which forces any morphism from @xmath192 to @xmath566 $ ] in @xmath185 to be zero . similarly , ( b ) and ( c ) follow since for any @xmath551 , @xmath567)=0= { { \rm hom } } _ { { \mathscr{k}^{\rm b}({a\mbox{{\sf -proj}}})}}(e , { t^{\bullet}}[m]).\ ] ] as a consequence of the claim , @xmath568 or @xmath569 . indeed , if @xmath570 and @xmath571 , then by ( b ) and ( c ) , @xmath559 and @xmath560 have a common direct summand @xmath70 , which contradicts ( a ) . we will finish the proof by analysing these two cases . _ case @xmath572 . _ by ( b ) , @xmath573 and then @xmath574 by ( a ) . let @xmath575 be the brutal truncation of @xmath192 , which by definition provides the following triangle in @xmath185 : @xmath576{\stackrel{f}{{\longrightarrow } } } t^0 { \stackrel{g}{{\longrightarrow } } } { t^{\bullet}}{\longrightarrow}{t^{\bullet}}_{<0 } \end{aligned}\ ] ] where @xmath264 is the chain morphism induced by @xmath577 , and @xmath41 is the chain morphism induced by @xmath578 . applying @xmath579 to this triangle gives the short exact sequence @xmath580,e){\rightarrow}0 \end{aligned}\ ] ] since @xmath581 trivially and @xmath582,e)=0 $ ] by @xmath535 being tilting . in particular , @xmath583 is injective , and @xmath264 is a left @xmath584-approximation . the morphism @xmath264 is left minimal . for any @xmath585 with @xmath586 , we have @xmath587 . thus @xmath561 defines a chain morphism @xmath588 in @xmath185 with @xmath589 and @xmath590 for all @xmath473 . now @xmath192 being an indecomposable complex in @xmath185 implies that @xmath591 is a local @xmath12-algebra , and therefore @xmath592 is either nilpotent or invertible . if @xmath592 is nilpotent in @xmath591 , say @xmath593 for some @xmath547 , then there exists a homotopy morphism such that @xmath594 . by definition of @xmath192 , the map @xmath595 is zero . thus , and hence @xmath596 is a split monomorphism . but this contradicts @xmath192 being a radical complex . it follows that @xmath264 is a left minimal morphism . applying @xmath597 to the triangle ( [ eqn : key triangle ] ) implies that @xmath41 is a right @xmath584-approximation . since @xmath41 induces an isomorphism @xmath598 , @xmath41 is a right minimal morphism . consequently , the triangle ( [ eqn : key triangle ] ) is a @xmath584-split sequence , with @xmath583 being an injective morphism . thus , by lemma [ lem : d - split sequences ] ( 1 ) and ( 3 ) , @xmath575 is an indecomposable radical complex , and @xmath599\oplus e$ ] is a tilting complex over @xmath0 . replacing @xmath192 by @xmath599 $ ] and repeating the same arguments @xmath600 times , we finally get that @xmath559 is indecomposable and @xmath601 is a tilting complex over @xmath0 , whereby @xmath559 is isomorphic to @xmath70 and the assertion ( 1 ) holds . _ case @xmath602 . _ by similar arguments as in the previous case , the assertion ( 2 ) in the statement can be verified for @xmath192 . since the case @xmath603 has been shown already , all cases have been settled . with these preparations , we are now able to prove theorem [ thm : derived equivalences for almost self - injective ] . first we assume that one of @xmath0 or @xmath3 is self - injective . when the ground field is algebraically closed , being self - injective is invariant under derived equivalence , by @xcite . under our assumption @xmath604 , invariance of being self - injective can be shown as follows , without any restriction on the ground field . by assumption , both @xmath0 and @xmath3 have @xmath604 . then @xmath0 is self - injective if and only if @xmath605 , and similarly for @xmath3 . by theorem [ thm : restriction theorem ] , the given derived equivalence @xmath209 induces an equivalence @xmath606 of triangulated subcategories . hence , both @xmath0 and @xmath3 are self - injective . now , ( * ? ? ? * proposition 3.8 ) implies that @xmath209 is an almost @xmath1-stable derived equivalence up to degree shift . next , we assume that neither @xmath0 nor @xmath3 is self - injective . let @xmath607 and @xmath608 be basic additive generators of @xmath73 and @xmath609 respectively . then the number of indecomposable direct summands of @xmath608 is exactly one less than the number of the simple left @xmath3-modules . since @xmath89 and @xmath438 , the derived equivalence @xmath209 induces an equivalence @xmath610 of triangulated subcategories , again by theorem [ thm : restriction theorem ] . let @xmath611 be the image of @xmath607 in @xmath186 under the equivalence . then @xmath612 generates @xmath613 , and @xmath614)=0 $ ] unless @xmath615 . without loss of generality , we assume that @xmath611 is , up to degree shift , a radical complex of the form @xmath616 where @xmath617 for all @xmath169 . by ( * ? ? ? * proposition 4.1 ) , there exists a complex @xmath137 in @xmath186 such that @xmath618 is a basic tilting complex over @xmath3 , inducing an almost @xmath1-stable derived equivalence @xmath619 where @xmath620 . let @xmath192 be a radical complex such that @xmath621 . then @xmath622 . in particular , @xmath623 is a tilting complex over @xmath0 and @xmath192 is indecomposable . hence by lemma [ lem : tilting complex over almost self - injective algebras ] and ( * ? ? ? * theorem 1.3(5 ) ) , @xmath37 is an iterated almost @xmath1-stable derived equivalence up to degree shift , and thus @xmath624 is an iterated almost @xmath1-stable derived equivalence . let @xmath0 and @xmath3 be self - injective @xmath12-algebras and let @xmath24 ( respectively @xmath26 ) be an indecomposable left @xmath0-module ( respectively left @xmath3-module ) . if @xmath625 and @xmath626 are derived equivalent , then every standard derived equivalence between them is an iterated almost @xmath1-stable derived equivalence ( up to degree shift ) , and the two endomorphism algebras are stably equivalent of morita type . by definition [ def : morita algebras ] , both endomorphism algebras are almost self - injective algebras . the statements then follow directly from theorem [ thm : derived equivalences for almost self - injective ] . _ example . _ theorem [ thm : derived equivalences for almost self - injective ] may fail if the algebras @xmath0 and @xmath3 are not assumed to be almost self - injective . here is an example . let @xmath627/(x^2,y^2)$ ] , and let @xmath628 be the unique simple @xmath629-module . then @xmath629 is a self - injective @xmath12-algebra and the auslander - reiten sequence @xmath630 is an @xmath631-split sequence in the sense of @xcite . by ( * theorem 1.1 ) , the endomorphism algebras @xmath632 and @xmath633 are derived equivalent . by direct checking , both @xmath0 and @xmath3 are seen to have @xmath1-dominant dimension at least two , but none of them is an almost self - injective algebra . since @xmath634 and @xmath635 , the algebras @xmath0 and @xmath3 can not be stably equivalent of morita type . combining proposition [ prop : almost nu - stable preserve dimensions ] and theorem [ thm : derived equivalences for almost self - injective ] yields the _ first invariance theorem _ : [ cor : class almost self - injective algebras ] derived equivalences between almost self - injective algebras preserve both global dimension and dominant dimension . many algebras have anti - automorphisms that preserve simple modules . prominent examples are involutory such anti - automorphisms , often called dualities , which are fundamental ingredients of the definition of cellular algebras . among the examples are group algebras of symmetric groups in any characteristic , many hecke algebras , brauer algebras , temperley - lieb algebras and many classes of schur algebras . in particular , quasi - hereditary algebras with duality occur frequently in algebraic lie theory . they have been studied in @xcite from the point of view of dominant dimension , motivating the concept of gendo - symmetric algebras and our present investigation of invariance properties . the anti - automorphisms defining dualities often occur as shadows of cartan involutions . a much larger class of algebras ( with anti - automorphisms not required to be involutions ) will be shown now to satisfy homological invariance properties . after establishing various elementary properties of the class of algebras with anti - automorphism preserving simples , both global and dominant dimension will get identified explicitly in the derived category , in terms of complexes occurring in canonical triangles . now we state precisely what we mean by an anti - automorphism preserving simples , and then we collect basic properties of algebras with such anti - automorphisms . see also @xcite and the references therein for further discussion of the special case of dualities . let @xmath0 be a @xmath12-algebra and let @xmath636 be an algebra anti - automorphism . for each ( left ) @xmath0-module @xmath27 , the @xmath514-twist of @xmath27 , denoted by @xmath637 , is defined to be the right @xmath0-module that equals @xmath27 as a @xmath12-vector space , and is equipped with the right @xmath0-action : @xmath638 the notation @xmath639 is reserved for this action . we define the twist of a right @xmath0-module or an @xmath0-bimodule similarly . for an @xmath0-bimodule @xmath27 , there are two different right @xmath0-module structures on @xmath640 : the usual one from the right @xmath0-action on @xmath27 and the twisted one from the left @xmath0-action on @xmath27 . to emphasize the difference , we write @xmath641 for the former and @xmath642 for the latter . if for every simple @xmath0-module @xmath628 , the @xmath12-dual of @xmath643 is isomorphic to @xmath628 as @xmath0-modules , then we say that _ @xmath514 fixes all simple modules_. a @xmath12-algebra will be called @xmath12-split , if the endomorphism rings of simple @xmath0-modules are just @xmath12 . then simple @xmath0-bimodules are tensor products of simple left @xmath0-modules with simple right @xmath0-modules . [ lem : twist bimodule ] let @xmath0 be a split @xmath12-algebra with an anti - automorphism @xmath644 fixing all simple @xmath0-modules . then : @xmath106 for each primitive idempotent @xmath61 in @xmath0 , the idempotent @xmath645 is conjugate to @xmath61 in @xmath0 . @xmath108 for each projective - injective @xmath0-module @xmath70 , the @xmath0-module @xmath646 is projective - injective . @xmath109 @xmath647 . in particular , @xmath0 is a morita algebra if and only if @xmath62 . @xmath110 @xmath648 , @xmath649 , and @xmath650 in @xmath651 , for every simple @xmath0-bimodule @xmath652 . @xmath653 let @xmath61 be a basic idempotent such that @xmath654 . then in @xmath655 , @xmath656 @xmath657 if @xmath27 is a non - zero @xmath0-bimodule with @xmath658 in @xmath651 , then @xmath659 . @xmath660 if @xmath27 and @xmath661 are @xmath0-bimodules with @xmath658 and @xmath662 in @xmath663 , then in @xmath651 @xmath664 @xmath665 if @xmath666 is a complex in @xmath667 and @xmath668 is an integer such that : @xmath669 , and @xmath670 for all @xmath671 , then @xmath672 , ( \operatorname{\mathrm{d}}{m^{\bullet}})[-d])\neq 0\}$ ] . \(1 ) let @xmath673 be the @xmath12-linear map defined by @xmath674 , for any @xmath675 . then @xmath676 for any @xmath677 . ( note that here @xmath91 gets twisted by @xmath678 . ) so @xmath679 is a right @xmath0-module isomorphism . now for each simple left @xmath0-module @xmath628 , @xmath680 by the assumption on @xmath644 . therefore , @xmath61 and @xmath645 are conjugate idempotents in @xmath0 . \(2 ) we may assume that @xmath70 is an indecomposable projective - injective @xmath0-module , thus of the form @xmath91 for some idempotent @xmath61 . then @xmath681 and @xmath682 by ( 1 ) . therefore , @xmath60 is projective - injective and @xmath683 is so , too . \(3 ) let @xmath91 be a basic projective - injective @xmath0-module such that every projective - injective @xmath0-module @xmath70 belongs to @xmath95 . then @xmath684 as @xmath0-modules by ( 2 ) . in other words , every projective - injective @xmath0-module is a strongly projective - injective @xmath0-module . so by definition , @xmath685 . the claim about morita algebras then follows from proposition [ prop : characterise morita algebras ] . \(4 ) let @xmath686 be the @xmath12-linear morphism defined by : @xmath687 for @xmath688 . then for any @xmath689 , there are equalities @xmath690 in @xmath691 . thus @xmath692 is an @xmath0-bimodule isomorphism and so is the @xmath12-dual of @xmath692 , that is , @xmath693 as @xmath0-bimodules . since @xmath0 is a split @xmath12-algebra , it follows that every simple @xmath0-bimodule @xmath652 is of the form @xmath694 , where @xmath628 and @xmath695 are simple left @xmath0-modules . thus , there is a series of isomorphisms of @xmath0-bimodules @xmath696 as the anti - automorphism @xmath644 fixes simple @xmath0-modules by assumption . \(5 ) since @xmath697 , the minimal faithful @xmath0-module @xmath91 is basic and projective - injective . by ( 2 ) , @xmath698 is also basic and projective - injective . as a result , @xmath699 as @xmath0-modules , and therefore @xmath700 as @xmath701-bimodules for some automorphism @xmath702 of @xmath6 . so for all @xmath703 @xmath704 \(6 ) since @xmath705 , there exists a simple @xmath0-bimodule @xmath652 such that @xmath706 . therefore @xmath707 thus the composition of an epimorphism from @xmath27 to @xmath652 followed by a monomorphism from @xmath652 to @xmath708 in @xmath651 defines a nonzero morphism from @xmath27 to @xmath708 . in particular , @xmath709 . \(7 ) since @xmath658 and @xmath710 in @xmath663 as @xmath0-bimodules , it follows that @xmath711 as @xmath0-modules for all @xmath712 . @xmath665 by definition , the complex @xmath713 $ ] has vanishing cohomology in degrees larger than @xmath427 , and @xmath714 $ ] has vanishing cohomology in degrees smaller than @xmath427 . this implies that @xmath715 , ( \operatorname{\mathrm{d}}{m^{\bullet}})[-d])=0 $ ] for all @xmath716 . moreover , by lemma [ lem : morphisms in derived categories](3 ) , @xmath717 , ( \operatorname{\mathrm{d}}{m^{\bullet}})[-m])$ ] is isomorphic to @xmath718 , which is nonzero by ( 6 ) . [ thm : class with anti - automorphisms ] let @xmath0 and @xmath3 be derived equivalent split @xmath12-algebras . \(a ) if both @xmath0 and @xmath3 have anti - automorphisms fixing all simple modules , then . \(b ) if furthermore both @xmath0 and @xmath3 have dominant dimension at least @xmath333 , then . for the proof of the theorem , we need the following lemma . let @xmath0 be a @xmath12-split algebra with an anti - automorphism @xmath644 fixing all simple @xmath0-modules , and let @xmath61 be an idempotent such that @xmath91 is basic and @xmath654 . then , in the canonical triangle @xmath719,\ ] ] for each integer @xmath169 , we have @xmath720 as @xmath0-bimodules . [ lem : canonical - complex - va ] first of all , it follows from lemma [ lem : twist bimodule](1 ) that @xmath721 and @xmath91 are isomorphic left @xmath0-modules , and consequently @xmath645 and @xmath61 are conjugate in @xmath0 , namely , there is an invertible element @xmath722 such that @xmath723 . we fix this notation throughout the proof . let @xmath724 be the canonical map sending each element @xmath688 to the map @xmath725 . from the long exact sequence of cohomologies induced by the canonical triangle * ct2 * , it follows that @xmath726 for each @xmath712 , we have the following isomorphisms of @xmath0-bimodules @xmath727 where the last isomorphism follows from lemma [ lem : twist bimodule ] ( 5 ) . thus , for each integer @xmath72 , there is an isomorphism @xmath720 of @xmath0-bimodules . it remains to consider the cohomologies of @xmath728 in degrees @xmath427 and @xmath363 . for this purpose , we need four @xmath0-bimodules isomorphisms . we write down these isomorphisms explicitly , and it is straightforward to check that they are really @xmath0-bimodule homomorphisms . the first is @xmath729 note that @xmath723 and @xmath730 . let @xmath731 be the map sending @xmath732 to @xmath733 . the second isomorphism reads as follows . @xmath734 by assumption @xmath91 is basic and @xmath654 . then @xmath735 is again in @xmath95 , basic and has the same number of indecomposable direct summands as @xmath91 . hence @xmath736 as left @xmath0-modules . it follows that @xmath737 as left @xmath0-modules . equivalently , @xmath738 as right @xmath0-modules , and there is some automorphism @xmath739 of @xmath6 such that @xmath740 as @xmath6-@xmath0-bimodules . let @xmath741 be such an isomorphism . the third isomorphism is the canonical map @xmath742 and the fourth isomorphism is a composition @xmath743 from the proof of lemma [ lem : twist bimodule ] ( 4 ) , the map @xmath744 is an @xmath0-bimodule isomorphism . then it is straightforward to check that the diagram @xmath745^(.4){\eta}\ar[dddd]^{\omega } & { { \rm hom } } _ { eae}(ea , ea)\ar[d]^{c_{\omega}}\\ & { } ^{\omega}{{\rm hom } } _ { \omega(e)a\omega(e)}(a\omega(e ) , a\omega(e))^{\omega}\ar[d]^{c_u}\\ & { } ^{\omega}{{\rm hom } } _ { eae}(ae , ae)^{\omega}\ar[d]^{\operatorname{\mathrm{d}}}\\ & { } ^{\omega}{{\rm hom } } _ { eae}(\operatorname{\mathrm{d}}(ae ) , \operatorname{\mathrm{d}}(ae))^{\omega}\ar[d]^{c_{\tau}}\\ { } ^{\omega}\!\!a^{\omega}\ar[r]^(.4){\eta } & { } ^{\omega}{{\rm hom } } _ { eae}(ea , ea)^{\omega } } \quad\quad ( \star)\ ] ] is commutative . actually , for each @xmath688 , the image of @xmath746 under @xmath747 is @xmath748 , which sends each @xmath749 to @xmath750 . the image of @xmath746 under @xmath751 is @xmath752 . then one can check that there is a commutative diagram @xmath753^{\tau^{-1}}\ar[r]^{r_{\omega(a ) } } & ea\ar[d]^{\tau^{-1}}\\ \operatorname{\mathrm{d}}(ae)\ar[r]^{\operatorname{\mathrm{d}}(r_{u}^{-1}\omega^{-1}r_a\omega r_u ) } & \operatorname{\mathrm{d}}(ae ) } \ ] ] of right @xmath0-modules . this means precisely that the image of @xmath746 under @xmath754 is the same as its image under @xmath751 . thus , the diagram @xmath755 is commutative , and induces another commutative diagram @xmath756 & \operatorname{{ker}}\eta\ar[r]\ar@{-->}[d]^{\cong } & a\ar[d]^{\omega}\ar[r]^(.3){\eta } & { { \rm hom } } _ { eae}(ea , ea)\ar[d]^{c_{\omega}c_u\operatorname{\mathrm{d}}c_{\tau}}\ar[r ] & \operatorname{{coker}}\eta\ar[r]\ar@{-->}[d]^{\cong } & 0\\ 0\ar[r ] & { } ^{\omega}(\operatorname{{ker}}\eta)^{\omega}\ar[r ] & { } ^{\omega}\!\!a^{\omega}\ar[r]^(.3){\eta } & { } ^{\omega}{{\rm hom } } _ { eae}(ea , ea)^{\omega}\ar[r ] & { } ^{\omega}(\operatorname{{coker}}\eta)^{\omega}\ar[r ] & 0 } \ ] ] where the rows are exact and vertical maps are @xmath0-bimodule isomorphisms . this finishes the proof . \(a ) proposition [ prop : bound global dimension ] implies that if @xmath0 or @xmath3 has infinite global dimension , then so does the other and we are done . now assume @xmath0 and @xmath3 to have finite global dimension and @xmath757 to be a standard derived equivalence . let @xmath41 and @xmath758 be the global dimensions of @xmath0 and @xmath3 respectively , and let @xmath759 denote the bounded complex @xmath760 in @xmath761 , and @xmath762 the bounded complex @xmath763 in @xmath764 . then lemma [ lem : characterise global dimension ] can be reformulated as @xmath765 by lemma [ lem : twist bimodule ] ( 4 ) and ( 7 ) and by @xmath12-duality , @xmath766 as @xmath0-bimodules . note that @xmath767 . hence , by lemma [ lem : twist bimodule](8 ) , we get @xmath768 , ( \operatorname{\mathrm{d}}{r_a^{\bullet}})[-i])\neq 0\}.\ ] ] similarly @xmath769 , ( \operatorname{\mathrm{d}}{r_b^{\bullet}})[-i])\neq 0\}$ ] . let @xmath770 be the derived equivalence induced from the given standard derived equivalence between @xmath0 and @xmath3 . then @xmath771 and @xmath772 in @xmath773 by lemma [ lem : standard derived equivalence ] ( 1 ) , ( 3 ) and ( 4 ) . hence , @xmath774 , that is , @xmath0 and @xmath3 have the same global dimension . \(b ) let @xmath385 and @xmath775 be the @xmath1-dominant dimensions of @xmath0 and @xmath3 respectively . since both @xmath385 and @xmath775 are at least one by assumption , lemma [ lem : twist bimodule ] ( 3 ) ( or lemma [ lem : two dominant dimensions coincide ] ) implies that there is no need to distinguish between dominant dimension and @xmath1-dominant dimension . let @xmath61 and @xmath264 be basic idempotents in @xmath0 and @xmath3 , respectively , such that @xmath654 and @xmath776 . let @xmath777 $ ] and be the canonical triangles ( denoted by ( ct2 ) in [ subsectionstandarddereq ] ) associated to the idempotents @xmath61 and @xmath264 respectively . let @xmath778 . we claim that @xmath779 . as @xmath780 , the module @xmath60 is a faithful right @xmath0-module and @xmath781 is injective . in this case @xmath782 , and @xmath783 vanishes if and only if @xmath781 is also surjective , if and only if @xmath784 ( see @xcite ) . this implies that @xmath550 when @xmath785 . if @xmath784 , then it follows from mller s characterisation ( proposition [ prop : muller characterisation ] ) that @xmath779 . without loss of generality , the derived equivalence @xmath786 can be assumed to be standard and to be given by a two - sided tilting complex @xmath197 in @xmath218 . let @xmath787 be the induced standard derived equivalence and @xmath788 . then by theorem [ thm : restriction theorem ] and lemma [ lem : derived - equivalence - preserves - canonical - triangle ] , @xmath789 as triangles in @xmath790 , and in particular , @xmath791 in @xmath790 . thus , it follows from lemma [ lem : canonical - complex - va ] and lemma [ lem : twist bimodule](8 ) that @xmath792 , v_a(e)[-i])\neq 0\ } , \mbox { and } \ ] ] @xmath793 , v_b(f)[-i])\neq 0\}.\ ] ] since @xmath794 and @xmath795 , one has @xmath796 , and @xmath797 . the characterisations of global and dominant dimension , respectively , inside the derived category of the enveloping algebras , provided by the proof may be of independent interest . let @xmath0 be an algebra with an anti - automorphism preserving simples . @xmath798 let @xmath759 denote the bounded complex @xmath760 in @xmath761 . then the global dimension of @xmath0 , if finite , is determined in the category @xmath761 as @xmath799 , ( \operatorname{\mathrm{d}}{r_a^{\bullet}})[-i])\neq 0\}.\ ] ] @xmath800 suppose @xmath61 is an idempotent in @xmath0 such that @xmath91 is basic and @xmath654 . then the dominant dimension of @xmath0 is determined in the category @xmath761 by the canonical triangle @xmath719 , \mathrm { as}\ ] ] @xmath801 , v_a(e)[-i ] ) \neq 0\}.\ ] ] the proof of theorem [ thm : class with anti - automorphisms ] implies that ( a ) holds , and that ( b ) holds in case that @xmath59 . suppose that @xmath802 . then @xmath60 can not be a faithful @xmath0-module , and thus the canonical map @xmath803 is not injective . it follows that @xmath804 , which is @xmath805 , is nonzero . then , in this case , ( b ) follows from lemma [ lem : twist bimodule](8 ) . in the current context , the complexes @xmath759 used to identify global dimension are preserved at least by standard derived equivalences . the derived restriction theorem ( corollary [ cor : restriction theorem for morita algebras ] ) and theorem [ thm : restriction theorem ] upon which it is based , are a crucial ingredient of the above proof of theorem [ thm : class with anti - automorphisms ] . together with lemma [ lem : derived - equivalence - preserves - canonical - triangle ] , they guarantee that the canonical triangle * ( ct2 ) * is preserved under standard derived equivalences between algebras with positive @xmath1-dominant dimensions . let @xmath0 and @xmath3 be algebras with anti - automorphisms preserving simples . then a standard derived equivalence from @xmath0 to @xmath3 induces a standard equivalence from @xmath761 to @xmath773 that sends @xmath759 to @xmath806 . when an algebra @xmath0 with an anti - automorphism preserving simples is derived equivalent to an algebra @xmath3 , possibly without such an anti - automorphism , the proof of theorem [ thm : class with anti - automorphisms ] provides an inequality : [ cor : inequality ] let @xmath0 be an algebra with an anti - automorphism preserving simples and suppose @xmath0 is derived equivalent to an algebra @xmath3 . then there is an inequality . the characterisation of @xmath807 given in the proof of theorem [ thm : class with anti - automorphisms ] remains valid . then the non - vanishing over @xmath0 of @xmath808 , ( \operatorname{\mathrm{d}}{r_b^{\bullet}})[-g])\neq 0\}$ ] , implies a contradiction to @xmath809 . the following example shows that the assumption in the second invariance theorem [ thm : class with anti - automorphisms ] , requiring both algebras admitting an anti - automorphism preserving simples , can not be relaxed ; the inequality in corollary [ cor : inequality ] is in general not an equality . in this example , the algebra @xmath0 has an anti - automorphism , while @xmath3 does not . the algebras @xmath0 and @xmath3 are derived equivalent . the global dimension of @xmath3 is strictly bigger than that of @xmath0 . let the algebra @xmath0 be given by the quiver @xmath810 with relations @xmath811 . this is a _ dual extension _ ( defined by c.c.xi @xcite ) of the path algebra of @xmath812 . it is a quasi - hereditary cellular algebra , and has global dimension @xmath64 . let @xmath411 be the indecomposable projective @xmath0-modules corresponding to the vertex @xmath169 , and let @xmath813 be the direct sum of three indecomposable complexes @xmath814 , @xmath815 and @xmath816 , where @xmath817 $ ] and , for each @xmath818 , the complex @xmath819 is the complex @xmath820 with @xmath116 in degree @xmath363 . then @xmath813 is a tilting complex , inducing a derived equivalence between @xmath0 and the endomorphism algebra @xmath3 of @xmath813 . the algebra @xmath3 has the following quiver @xmath821 with relations @xmath822 . the algebra @xmath0 has an anti - automorphism fixing simple modules , while @xmath3 can not have such an anti - automorphism . ( if an algebra @xmath629 has such an anti - automorphism , then @xmath823 for all simple @xmath629-modules @xmath824 . this implies that , in the quiver of the algebra @xmath629 , for any two vertices @xmath746 and @xmath825 , the number of arrows from @xmath746 to @xmath825 coincides with the number of arrows in the opposite direction , from @xmath825 to @xmath746 ) . a direct calculation shows that the algebra @xmath3 has global dimension @xmath826 , which is strictly larger than @xmath827 . cellular algebras by definition have anti - automorphisms preserving simples ; hence this large and widely studied class of algebras is covered by the second invariance theorem [ thm : class with anti - automorphisms ] . [ cor : cellular ] suppose that @xmath0 and @xmath3 are derived equivalent cellular algebras over a field . then @xmath0 and @xmath3 have the same global dimension . if both @xmath0 and @xmath3 have positive dominant dimension , then their dominant dimensions are equal . in the next section , it will be shown how to use this result to compute homological dimensions of blocks of quantised schur algebras . global dimensions of classical and quantised schur algebras @xmath11 ( with @xmath5 ) have been determined by totaro @xcite and donkin @xcite . dominant dimensions of these algebras have been obtained more recently in @xcite . the aim of this section is to determine the global and dominant dimensions of all blocks of these algebras , that is , of the indecomposable algebra direct summands . to recall some basics on @xmath828-schur algebras and their blocks ( see @xcite and the references therein ) , let @xmath12 be a field of characteristic @xmath427 or @xmath829 , and @xmath828 a non - zero element in @xmath12 . let @xmath830 be the smallest integer such that @xmath831 , and set @xmath832 if no such integer exists . for a natural number @xmath833 , let @xmath834 be the symmetric group on @xmath833 letters and let @xmath835 be the associated hecke algebra that is defined by generators @xmath836 and relations @xmath837 for each composition @xmath838 of @xmath833 , that is , a sequence of @xmath161 non - negative integers summing up to @xmath833 , let @xmath839 be the associated parabolic hecke algebra that is isomorphic to @xmath840 as @xmath12-algebras . then @xmath839 naturally can be seen as a @xmath12-subalgebra of @xmath835 . the algebra @xmath839 has a trivial module @xmath12 with all @xmath841 s acting as scalar @xmath828 . let @xmath842 be the permutation module over @xmath835 that is defined as the induced module @xmath843 . the @xmath828-schur algebra @xmath844 is then defined to be the endomorphism ring @xmath845 where @xmath846 ranges over all compositions @xmath847 of @xmath833 into @xmath161 parts , where @xmath161 is any natural number . for each composition @xmath846 of @xmath833 , there is a unique associated composition @xmath848 ( called _ partition _ ) obtained by rearranging the entries of @xmath846 weakly decreasing ; the permutation module @xmath849 is isomorphic to @xmath842 and is a direct sum of the ( indecomposable ) young modules @xmath850 with multiplicities @xmath851 ( @xmath829-kostka number ) where @xmath852 ranges over all partitions of @xmath833 . these multiplicities are known to satisfy @xmath853 and @xmath854 unless @xmath855 in the dominance ordering on partitions . when @xmath828 is not a root of unity , then the @xmath828-schur algebra @xmath844 is semisimple , and thus the global dimensions of @xmath844 and its blocks are @xmath427 , and the dominant dimensions of @xmath844 and its blocks are @xmath42 . in the following , we always assume that @xmath828 is a root a unity , and therefore @xmath856 . if @xmath857 and @xmath858 is the @xmath830-adic expansion of @xmath833 , then we set @xmath859 ; if @xmath860 , @xmath861 and @xmath862 are the @xmath830-adic expansion of @xmath833 and the @xmath829-adic expansion of @xmath863 respectively , then we set @xmath864 . the global dimension of @xmath844 for @xmath865 in this case has been given by totaro @xcite for @xmath866 , that is for the classical schur algebra , and by donkin @xcite in general . a lower bound for the dominant dimension of @xmath844 has been obtained ( implicitly ) by kleshchev and nakano @xcite for @xmath866 and by donkin ( * ? ? ? * proposition 10.5 ) in general . it was shown later in @xcite that this lower bound is an upper bound , too . to determine the global and dominant dimensions of blocks of the @xmath828-schur algebras , we will use the setup of algebras with a duality . each @xmath828-schur algebra has an anti - automorphism that fixes all simple modules , and there is a block decomposition that is invariant under the involution @xmath867 where @xmath868 and @xmath702 ranges over all @xmath830-core partitions of @xmath869 . moreover , for @xmath870 , the two blocks @xmath871 of @xmath844 and @xmath872 of @xmath873 are derived equivalent if @xmath874 @xcite . chuang and rouquier have shown in @xcite , that @xmath871 and @xmath877 are derived equivalent . as all block algebras of @xmath828-schur algebras have anti - automorphisms fixing all simple modules , theorem [ thm : class with anti - automorphisms ] can be applied and we get @xmath878 . now we are going to compute the global dimension of @xmath875 by induction on @xmath885 ; we have to show that it equals @xmath886 . if @xmath887 , then the block algebra @xmath875 is semisimple ( @xcite ) , and thus @xmath888 . now we assume that @xmath889 . note that the @xmath828-schur algebra @xmath890 is of global dimension @xmath886 by theorem [ thm : global dimension of schur algebras ] , and has a block subalgebra @xmath891 . it follows that @xmath892 since all other block subalgebras of @xmath890 are @xmath893 with @xmath894 , and thus @xmath895 by induction . in terms of the cover theory introduced by rouquier @xcite , the @xmath828-schur algebra @xmath844 is a quasi - hereditary cover of the hecke algebra @xmath835 of covering degree @xmath896 by theorem [ thm : dominant dimension of schur algebras ] and @xcite , that is , @xmath844 is a @xmath896-cover , but not an @xmath830-cover of @xmath835 . the following result implies a particular property of the cover ; each block of @xmath844 is a quasi - hereditary cover of the corresponding block of @xmath835 , of the same dominant dimension . this property may be formulated as saying that the covering is uniform of covering degree @xmath897 . if @xmath887 , then @xmath871 is semisimple and has dominant dimension infinity . now we assume that @xmath889 . since @xmath828 is a root of unity , it follows that @xmath899 , and thus by theorem [ thm : dominant dimension of schur algebras ] @xmath900 note that all block algebras have anti - automorphisms fixing all simple modules . by @xcite , the block subalgebra @xmath875 of @xmath844 and the principal block subalgebra @xmath891 of @xmath890 are derived equivalent , and thus have the same dominant dimension by theorem [ thm : class with anti - automorphisms ] . therefore we only need to show @xmath901 . let @xmath61 be an idempotent in @xmath891 such that @xmath902 is a minimal faithful @xmath891-module . then @xmath903 is a block subalgebra of @xmath904 and the @xmath905-module @xmath906 is isomorphic to a direct sum of those young modules @xmath850 that belong to the block @xmath905 ( see @xcite ) . by proposition [ prop : muller characterisation ] and the inequality ( [ eqn : lower bound for dominant dimension ] ) , to finish the proof , it suffices to show : set @xmath907 and @xmath908 , which are two partitions of @xmath909 . then the young module @xmath850 belongs to the block @xmath905 and by definition @xmath910 . as a result , by mackey s decomposition theorem @xmath911 here , the first isomorphism uses the definition of permutation modules and the second one uses adjointness ; non - vanishing of the third extension space follows by identifying @xmath912 with @xmath913 and then using the known cohomology of @xmath914 . so there is a direct summand @xmath915 of @xmath916 such that @xmath917 . in this case , the young module @xmath915 must belong to the block @xmath905 , too . this proves the claim and the theorem . _ remark . _ ( 1 ) when @xmath12 has characteristic zero or bigger than the weight @xmath885 , the dominant dimension of the blocks @xmath871 have been determined in @xcite by using chuang and tan s complete description of the corresponding rouquier blocks @xcite . \(2 ) all blocks @xmath918 of the @xmath828-schur algebra @xmath844 are also stably equivalent of morita type , and hence have the same global and dominant dimensions , as well as the same representation dimension . indeed , by carefully examining the tilting complexes constructed by chuang and rouquier @xcite , we see that the induced derived equivalences are almost @xmath1-stable . however , the derived equivalences between general blocks constructed by chuang and rouquier @xcite are not almost @xmath1-stable . for instance , when @xmath919 , the group algebras of @xmath920 and of @xmath921 , respectively , each have a unique block of @xmath829-weight @xmath826 . these two blocks correspond to each other under the reflection @xmath922 of the weyl group of the affine kac - moody algebra @xmath923 . the derived equivalences constructed in @xcite , relating these two blocks , are not almost @xmath1-stable . i. assem , d. simson and a. skowronski , _ elements of the representation theory of associative algebras . _ 1 . techniques of representation theory . london mathematical society student texts , * 65*. cambridge university press , cambridge , 2006 . s. donkin , tilting modules for algebraic groups and finite dimensional algebras , in _ handbook of tilting theory _ , london mathematical society lecture note series * 332 * , cambridge university press , cambridge , 2007 .

electrons are characterized by their quantized spin and charge . in conventional electronic devices , only the charge degree of freedom has been employed for the control of the electron transport . a new field of spintronics @xcite was born from both experimental and theoretical efforts to harness the electron s spin degree of freedom in order to control the electric current in the devices . one of typical spintronic devices is the spin valve which is a hybrid structure of ferromagnetic ( fm ) metal / nonmagnetic ( nm ) material / fm metal . the current passing through the spin valve depends on the magnetization configuration of two fm metals . in the collinear case , usually more current flows through the spin valve in the parallel configuration than in the antiparallel configuration . difference in resistance between the two is called magnetoresistance . in the noncollinear case or when two magnetization orientations are neither parallel nor antiparallel , the spin polarized current from one fm electrode exerts the spin torque @xcite on the other fm electrode , and induces the magnetization dynamics . examples of a spin valve are giant magnetoresistance ( gmr ) devices,@xcite magnetic tunnel junctions,@xcite nanopillars,@xcite etc . in contrast with vertical spin valves , lateral spin valves are characterized by their multi terminal functionalities and so are more favorable for integration into semiconductor electronics . due to increased spacing between terminals , efficient spin injection and detection have been a very hot issue . the spin injection and detection experiments in the two - terminal geometry are obscured by other effects like anisotropic magnetoresistance , hall effect , etc . this defect was overcome by adopting the nonlocal spin valve geometry @xcite similar to the schematic device structure in fig.[3fmlead ] . the original spin valve devices contain two fm electrodes ( vs. three fm electrodes in fig . [ 3fmlead ] ) contacting the nonmagnetic base electrode . in this lateral spin valve system , the spin transport was clearly observed with al wires @xcite by spatially separating the spin current path from the charge current path and thereby removing other undesirable effects . the spin polarized current flows from the left of n ( base electrode ) into f1 . that is , spin polarized electrons are injected from f1 into base electrode n and is drained to the left of n. due to asymmetry of two spin states in fm , the number of injected spin - up and spin - down electrons is different . flowing from the left end of n into f1 . the electrode f1 is called the spin - injecting probe , while f@xmath0 with @xmath1 will be called the nonlocal ( voltage ) probe in this paper . ( b ) the one - dimensional model geometry of the spin valve system in ( a ) . [ 3fmlead],width=264 ] in addition to charge current in the left of n ( @xmath2 ) , diffusion of injected spins generates spin current flowing to left and right of n symmetrically . pure spin current to the right of n was detected @xcite with another fm electrode by measuring the spin - dependent nonlocal voltage drop . the nonlocal spin injection and detection technique was also used to observe @xcite the ( inverse ) spin hall effect in diffusive nonmagnetic metallic strips . in these experiments , the spatial separation of charge and spin currents as well as the efficient spin injection are essential to observing the charge hall voltage induced by the spin current . recently experimental groups @xcite studied the spin transport in the lateral spin valves with the three fm electrodes as shown in fig . [ 3fmlead ] . f1 is the spin - injection electrode ( f@xmath3 ) as usual , while f2 and f3 are the nonlocal voltage probes located outside the charge current path . while the spin polarized electrons are injected from f1 into n and are drained into the left of n , the transresistance is measured between f1 and f2 or between f1 and f3 . in the former case , f2 is the spin - detecting electrode f@xmath4 and f3 is the additional electrode f@xmath5 . in the latter case , the roles of f2 and f3 are switched . the important issue is how sensitively the transresistance is affected by f@xmath5 . one experimental group @xcite found that f@xmath5 can drain the spin current and thereby significantly suppress the transresistance . they concluded that such additional fm electrode is relevant to the spin injection and accumulation in the multi terminal lateral spin valves . on the other hand , another group @xcite found that the transresistance is weakly affected by f@xmath5 even when the contact resistance between f@xmath5 and n is ohmic . moreover the transresistance was observed @xcite to be independent of the magnetization orientation of f@xmath5 ( parallel or antiparallel to that of f@xmath3 ) . they concluded that such additional fm electrodes are irrelevant to the spin injection and accumulation in the multi terminal lateral spin valves . the existing experimental results seem to be contradictory to each other . motivated by this experimental situation , we study theoretically the spin transport in the lateral spin valve with three ferromagnetic electrodes as schematically shown in fig . [ 3fmlead ] . we adopt the semiclassical spin drift - diffusion ( sdd ) equation @xcite for the one - dimensional device structure and study the mutual effect of fm electrodes on their nonlocal spin signals such as nonlocal voltage drop and leakage spin current . we find that the efficiency of the spin current leakage into fm electrodes depends on the relative magnitude of junction resistance and the bulk resistance ( defined over spin diffusion length ) in fm and n electrodes . the voltage drop in @xmath6 is proportional to its leakage spin current with the proportionality constant given by the _ effective _ spin resistance which depends on the magnetization orientation of f@xmath4 . nonlocal spin signals are sensitive to the position of f@xmath5 relative to the positions of f@xmath3 and f@xmath4 . when f@xmath5 is located in between f@xmath3 and f@xmath4 , the transresistance can be either significantly or weakly affected by f@xmath5 depending on the relative magnitude of junction resistance and spin resistance . the effect of f@xmath5 is weak when f@xmath5 is located outside the region between f@xmath3 and f@xmath4 . even though the magnitude of the spin current and nonlocal spin signals may be modified by f@xmath5 , the flow direction of the spin current in the whole device is set by the magnetization orientation of f@xmath3 and so the nonlocal spin signals are independent of the magnetization orientation of f@xmath5 . this surprising result is already observed @xcite in experiments and is the direct consequence of no charge current in f@xmath5 . based on decoupling of charge and spin modes in the sdd equation and the kirchhoff rules at the junction , we also show that the relationship between nonlocal spin signals and magnetization holds true even in realistic three - dimensional samples . these interesting properties in fact originate from zero charge current in nonlocal voltage electrodes . irrelevance of magnetization orientation of additional fm electrode to nonlocal spin signals implies that even additional _ nonmagnetic _ electrode can modify nonlocal spin signals in the spin - detecting probe . our theoretical study may be useful for clarifying the conflicting roles @xcite of an additional fm electrode in the lateral spin valve devices . in addition , our study is relevant to device applications , because the multi terminal functionality is essential for device applications of lateral spin valves . the rest of this paper is organized as follows . in sec . ii , the spin drift - diffusion equation is briefly introduced and the detailed algebras for the lateral spin valve with three fm electrodes are included . the results of our work for spin valves relevant to experiments are presented in sec . iii . in sec . iv , our work is summarized and its relevance to experiments is discussed . some algebraic details and interesting results are included in three appendices . from now on we are going to confine our discussion to the collinear magnetizations of three fm electrodes in spin valves and so we consider the spin polarized transport in a steady state . the noncollinear magnetizations go beyond the scope of our paper since the current flow in the noncollinear magnetizations generates the spin transfer torque and induces the magnetization dynamics . in the collinear and diffusive transport , the spin drift - diffusion equations @xcite have been very useful for understanding phenomenologically the spin polarized transport in the spin valve systems . later the sdd equations were derived @xcite from the semiclassical boltzmann equation under the assumption that the spin diffusion length ( sdl ) is larger than the mean free path ( mfp ) . using the numerical solution of the spin dependent boltzmann equation , the validity of the sdd equations was further extended @xcite to the case when the sdl is comparable to the mfp . the sdd equations have been widely used for analyzing the spin injection experiments in various device geometries . the sdd formalism was also applied to the study of spin transfer torque @xcite in the case of noncollinear magnetizations . the sdd equations in the collinear magnetizations are written down for the spin - dependent electrochemical potential @xmath7 and electric current density @xmath8 . here @xmath9 represents the spin - up ( @xmath10 ) and spin - down ( @xmath11 ) states , respectively . the presence of the spin flip scattering in bulk mixes two spin states and the sdd equations can be written down in a matrix form . @xmath12 here @xmath13 is the diffusion constant for spin direction @xmath9 and @xmath14 is the average spin - flip time for an electron from the spin direction @xmath10 to @xmath11 . @xmath15 is the conductivity for electrons with spin @xmath16 and @xmath17 is the absolute value of electron charge . the matrix differential equation for the electrochemical potential can be solved @xcite by analyzing the eigenvalues and eigenvectors of the matrix in the sdd equation . one eigenvalue is @xmath18 and the corresponding eigenvector is @xmath19 . the other eigenvalue defines the spin diffusion length @xmath20 and is given by the expression @xmath21 and its eigenvector is @xmath22 . to find this form of the eigenvector , the einstein relation as well as the detailed balance relation are already invoked . obviously the first eigenvector ( charge mode ) does not discriminate between two spin states , while the second one ( spin mode ) does . in this section we analyze the spin polarized transport in the spin valve system based on the one - dimensional sdd equations . the device structure is displayed in fig . [ 3fmlead](a ) , where the base electrode is contacted to three ferromagnetic ( fm ) electrodes . our primary goal is to understand the mutual influence of the ferromagnetic electrodes on the nonlocal spin signals such as the voltage drops and the leakage spin currents . fm leads are labeled as f@xmath0 with @xmath23 from left to right . the one - dimensional geometry , corresponding to the device structure , is displayed in fig . [ 3fmlead](b ) , where the junctions between the base electrode and the fm leads are labeled as @xmath24 ( @xmath23 ) . in experiments , the base electrode is nonmagnetic , but we are going to consider the case of magnetic base electrode with its nonzero bulk spin polarization @xmath25 . nonmagnetic case is recovered by a simple replacement @xmath26 . the spin polarization ( sp ) in each fm lead is denoted as @xmath27 which is defined by the spin asymmetry in the spin dependent conductivity @xmath28 . @xmath29 with the total conductivity @xmath30 , the spin - up and spin - down conductivities can be written as @xmath31 for the base electrode , the spin polarization ( @xmath25 ) in conductivity and the spin - dependent conductivities ( @xmath32 ) are defined in a similar manner . when the spin polarized electrons are injected from f1 into the base electrode n and is drained to left , the electrochemical potential in the fm leads can be written as @xmath33 \begin{pmatrix } 1 \cr 1 \end{pmatrix } \nonumber\\ & & - i_i { \cal r}_i e^{-y/\lambda_i } \begin{pmatrix } ( 1+\beta_i)^{-1 } \cr - ( 1-\beta_i)^{-1 } \end{pmatrix}.\end{aligned}\ ] ] here @xmath34 is the kronecker delta function . the spin dependent current is determined by the equation @xmath35 the charge current is given by the expression @xmath36 and flows only in f1 but not in f@xmath0 with @xmath1 . the @xmath0-th fm lead is contacted to the base electrode at @xmath24 . @xmath37 , @xmath38 , @xmath39 , and @xmath27 are the cross sectional area , the spin diffusion length , conductivity , and bulk spin polarization in conductivity of the @xmath0-th fm lead , respectively . @xmath40 , defined by the relation @xmath41 is the resistance of the fm electrode over the spin diffusion length . due to an exponential decay of spin current , this definition of resistance makes sense physically when discussing the spin current . @xmath42 is the voltage drop at each ferromagnetic electrode far away from the junction with the base electrode and is induced by the nonequilibrium spin injection and diffusion . note that the common fermi energy is dropped in writing the electrochemical potentials in this paper , because the overall constant energy shift does not change physics . the spin current in f@xmath0 @xmath43 is given by the expression @xmath44 the first term is the spin - polarized driving current , while the second comes from the spin accumulation and diffusion . though no charge current flows in the region @xmath45 , the spin current is induced in the base electrode due to the spin injection , accumulation and diffusion . the spin current decays exponentially over the spin diffusion length and in turn leaks into the other fm electrodes . @xmath46 measures the magnitude of this leakage spin current at the interface between the base electrode and f@xmath0 . the leakage spin current also decays exponentially over the sdl in the fm electrodes . the set of six unknown parameters @xmath47 are to be determined by the kirchhoff rules at the junctions . note that @xmath42 and @xmath46 are null for @xmath1 , when the injected current is not spin polarized . @xmath48 can be nonzero for the tunneling barrier even when the injected current is not spin polarized . hence we may call @xmath47 for @xmath1 as the _ nonlocal spin signals_. in the common base electrode , we have the electrochemical potential for spin - up(@xmath10 ) and spin - down(@xmath11 ) electrons @xmath49 here @xmath50 is the step function and the spin dependent current is computed from @xmath51 @xmath52 , @xmath20 , @xmath53 , and @xmath25 are the cross sectional area , the spin diffusion length , conductivity , and bulk spin polarization in conductivity of the base electrode , respectively . @xmath54 , defined by the relation @xmath55 is the resistance over the spin diffusion length in the base electrode . the charge current @xmath56 flows only at the section @xmath57 and the spin current @xmath58 is given by the equation @xmath59 the additional three unknown parameters @xmath60 are introduced for the base electrode . @xmath61 represents the redistribution of spin current in the base electrode due to its leakage into the voltage probes ( fm electrodes with @xmath62 ) . the electrochemical potentials are constructed such that the charge current is conserved at any junction in the device ( charge continuity equation ) . no net charge current flows to the right of @xmath63 and the currents for spin - up and spin - down electrons are equal in their magnitude and opposite in their sign or flowing direction . this symmetry in spin current is strictly obeyed in the nonlocal spin valve geometry even in the presence of the multiple ferromagnetic electrodes to the right of the spin - injecting fm electrode f1 . obviously the spin currents @xmath46 and @xmath61 are modified by the presence of other fm electrodes , which is our main research interest in this work . there are nine unknown parameters @xmath64 ( @xmath23 ) , which should be determined by the boundary conditions or kirchhoff rules at the junctions . as already mentioned in the above , the electrochemical potentials are constructed such that the charge current is conserved . in our work , the spin flip scattering is neglected at the interface , but is taken into account in bulks . in this case , the spin current is conserved at each junction and the following relations are obtained @xmath65 we consider the case of dirty interface between the ferromagnetic electrodes and the base electrode . due to a finite value of resistance at the interface , the electrochemical potential across the junction is not continuous and should be determined by the ohm s law . @xcite @xmath66 here @xmath67 is the difference of the electrochemical potentials at @xmath24 between the base electrode and the f@xmath0 electrode . @xmath68 is the spin dependent junction resistance between the base electrode and f@xmath0 , and is defined in terms of the spin polarization @xmath69 of junction resistance . @xmath71 is the total junction resistance or @xmath72 . the clean or transparent contact can be recovered by a simple replacement @xmath73 . for the sign of @xmath25 s ( spin polarization ) , we are going to adopt the following convention . when the spin - up(-down ) electrons belong to the majority(minority ) channel at the fermi level , the sign of @xmath25 s is positive . on the other hand , the sign of @xmath25 s is negative when the spin - up(-down ) electrons belong to the minority(majority ) channel . according to our convention , the sign of @xmath25 s is reversed under the magnetization reversal . the same convention applies to the sign of @xmath74 s which are introduced to define the spin polarization in the resistance of the interface . after some algebra as detailed in appendix [ dalgebra ] , we find the expressions for @xmath75 and @xmath42 , which contain all the information about the spin polarized transport in the one - dimensional spin valve . @xmath76 , \\ \frac{j_i}{i } & = & - \frac{1}{2 } g_{i1 } [ ( \beta_1 -\beta ) r_1 + ( \gamma_1 - \beta ) r_{t1 } ] , \\ \label{drop_v } \frac{v_i}{i } & = & - [ ( \beta - \beta_1)^2 r_1 + ( \beta^2 - 2\beta \gamma_1 + 1 ) r_{t1 } ] \delta_{i,1 } \nonumber\\ & & + [ ( \beta_i - \beta ) r_i + ( \gamma_i-\beta ) r_{ti } ] g_{i1 } \nonumber\\ & & \times [ ( \beta_1 - \beta ) r_1 + ( \gamma_1 - \beta ) r_{t1}].\end{aligned}\ ] ] @xmath77 is the element of the matrix @xmath78 defined in eq . ( [ cond_matrix ] ) of appendix [ dalgebra ] and has the dimension of conductance . the other set of material parameters , @xmath79 and @xmath80 , are introduced in appendix [ dalgebra ] and their definitions are repeated here for readers . @xmath81 these new material parameters need our special attention . they have the dimension of resistance and deserve their own terminology . they are already called the _ spin resistance _ in the literature . first of all , the spin resistance is introduced to simplify the algebra as shown in appendix [ dalgebra ] . as the above equations show , this spin resistance determines the nonlocal spin signals such as the voltage drops and the leakage spin currents in the voltage probes . more physical insights on the spin resistance are elaborated on in appendix [ s_resist ] . since we are interested in the nonlocal transport measurements , we focus on the leakage spin currents and voltage drops in the voltage probes ( f@xmath0 with @xmath1 ) . the leakage spin current in the nonlocal voltage probes is given by the expression [ eq . ( [ spin_i ] ) ] @xmath82 i.\end{aligned}\ ] ] this relation for the nonlocal spin current @xmath46 suggests that the conductance matrix @xmath78 contains all the information about the mutual effect of nonlocal voltage probes . since the conductance @xmath83 does not depend on the magnetization configuration of the fm electrodes , the leaking spin current does not depend on the magnetization orientation of voltage probes , but instead depends on the magnetization orientation of the spin - injecting fm electrode ( f1 ) and the base electrode ( if ferromagnetic , @xmath84 ) . this important observation can be understood as follows . the flow direction of spin current ( the sign of @xmath46 ) , in the base electrode as well as in the fm electrodes , is obviously determined by the magnetization configuration in the spin - injecting electrode . this means the flow direction of spin current can not be altered by the change of magnetic configurations in nonlocal voltage probes . this is due to the fact that the nonequilibrium spin current is generated by the spin - injecting electrode , but not by nonlocal voltage probes . another important observation is that the magnitude of spin current or @xmath46 is not modified under the magnetization reversal of nonlocal voltage probes , which in fact derives from the symmetry in the sdd equations . this property of @xmath46 derives from decoupling of spin and charge modes in sdd equations as well as the zero charge current in nonlocal voltage probes . detailed analysis can be found in appendix [ 3d - case ] . the relation @xmath85 simply reflects the conservation of spin current at the interface between the voltage probe and the base electrode . @xmath86 defined below is ubiquitous in the expressions of @xmath46 and @xmath42 . @xmath87 i.\end{aligned}\ ] ] we may call @xmath86 as the _ spin potential _ which is the source from the spin - injecting electrode and drives the spin current in the spin valve . the leakage spin current in nonlocal probes can be written as @xmath88 the spin current in the spin valve device can be expressed in terms of the spin potential @xmath86 and the conductance matrix @xmath78 as @xmath89 e^{-y/\lambda_1 } , \\ i_i^{s } & = & g_{i1 } v_1^{s } e^{-y/\lambda_i } , i \neq 1.\end{aligned}\ ] ] the voltage drop in the nonlocal voltage probes is given by the expression [ eq . ( [ drop_v ] ) ] @xmath90 g_{i1 } \nonumber\\ & & \times [ ( \beta_1 - \beta ) r_1 + ( \gamma_1 - \beta ) r_{t1 } ] i.\end{aligned}\ ] ] note that the voltage drop can be written as the product of spin current and some sort of spin resistance as @xmath91 i_i.\end{aligned}\ ] ] @xmath42 is the effective measure of weighted averaging the spin - up and spin - down electrochemical potentials in f@xmath0 and so depends on the magnetization configuration of f@xmath0 . in addition , @xmath42 can be understood as a shift in the fermi level in order to satisfy the condition of zero charge current in the nonlocal voltage probes . look at appendix [ s_resist ] for details . though the leakage spin current @xmath46 in f@xmath0 is independent of the magnetization orientations of all nonlocal voltage probes ( parallel or antiparallel to that of spin - injecting probe ) , the voltage drop @xmath42 depends on the magnetization orientation of f@xmath0 , the spin - injecting probe and the base electrode , but not on that of other voltage probes . in this section our discussion is confined to the spin valve system with three ferromagnetic electrodes and _ nonmagnetic _ ( @xmath92 ) base electrode . the current @xmath93 is injected from the left of nonmagnetic base electrode and is drained to f1 ( see fig . [ 3fmlead ] ) . though the charge current is null to the right of the contact between f1 and the base electrode , the finite spin current is induced everywhere by the spin injection , accumulation and diffusion . with two nonlocal fm electrodes ( labeled as f2 and f3 ) contacted with the nonmagnetic base electrode to the right of f1 , we want to study the mutual influence of two nonlocal fm electrodes on their voltage drops and leakage spin currents or non - local spin signals . with three fm electrodes , the dimension of conductance matrix @xmath78 is @xmath94 . @xmath95 here @xmath96 for @xmath23 and @xmath97 , @xmath98 , and @xmath99 , where @xmath100 is the distance between the contacts of the @xmath0-th and @xmath101-th fm electrodes with the base electrode . after inserting the explicit expressions of @xmath83 into eq . ( [ nonlocal_i ] ) with @xmath26 or @xmath102 we find the leakage spin currents @xmath103 and @xmath104 in the nonlocal electrodes to be given by the expressions @xmath105 , \\ \label{3lead_i3 } i_3 & = & - \frac{ir}{2d } e^{-l_{13}/\lambda } ~ ( \beta_1 r_1 + \gamma_1 r_{t1 } ) ( r_2 + r_{t2 } ) .\end{aligned}\ ] ] here @xmath106 is the determinant of the matrix @xmath107 and is given by the expression @xmath108 obviously the exponentially decaying factor can be extracted out as @xmath109 ( @xmath62 ) , where @xmath110 is negative . as mentioned in the previous section , the leaking spin current ( @xmath103 and @xmath104 ) does not depend on the magnetization configuration of the nonlocal voltage probes ( f2 and f3 ) , but only on the magnetization configuration of the spin - injecting electrode ( f1 ) . this is clearly explained by the fact that the spin current is generated by the spin - injecting electrode . the nonlocal voltage drops @xmath111 and @xmath112 are related to their leakage spin currents as @xmath113 this relation suggests that the nonlocal voltage drop @xmath42 is proportional to the corresponding nonlocal spin current @xmath46 and the proportionality constant is the effective spin resistance which depends on the magnetization orientation of f@xmath0 . this resistance is the intrinsic material properties of the relevant fm electrode so that the effect of the other fm electrode is completely embedded into the nonlocal spin current . this means that we can discuss the effect of multiple fm electrodes on transresistance in terms of the leakage spin current . from eqs . ( [ 3lead_i2 ] ) , ( [ 3lead_i3 ] ) and ( [ 3lead_v ] ) , transresistance ( @xmath114 and @xmath115 ) in f2 and f3 electrodes are reduced to the following forms . @xmath116 , \\ \label{3lead_tr3 } r_{s3 } & = & -\frac{r}{2d } e^{-l_{13}/\lambda } ~ ( \beta_1 r_1 + \gamma_1 r_{t1 } ) ( r_2 + r_{t2 } ) \nonumber\\ & & \times ( \beta_3 r_3 + \gamma_3 r_{t3 } ) .\end{aligned}\ ] ] we denote the transresistance in the absence of an additional fm electrode using the superscript as @xmath117 and @xmath118 . the same notations with superscript will be used for nonlocal voltage drop and spin current . the effect of other nonlocal fm electrode on the transresistance can be quantified by computing the ratio : @xmath119 with @xmath62 . it follows from eq . ( [ 3lead_i ] ) and ( [ 3lead_v ] ) that @xmath120 the effect of an additional fm electrode on the transresistance can be measured by how much the nonlocal spin current is reduced or by the change of the conductance matrix @xmath78 under the other fm electrode . two important facts can be read off from eqs . ( [ 3lead_tr2 ] ) and ( [ 3lead_tr3 ] ) . ( i ) the transresistance of one fm electrode ( say , f3 ) does not depend on magnetization orientation of the other electrode ( say , f2 ) . that is , the transresistance @xmath121 in f3 does not depend on the spin polarization , @xmath122 and @xmath123 , of f2 . in ref . , the transresistance was observed to be independent of the magnetization orientation ( parallel or antiparallel to f3 ) of the intervening fm electrode f2 , which is supported by our theoretical results . however this fact does not necessarily mean @xcite that the observed transresistance is not influenced by the additional fm electrode . the transresistance can be either significantly changed or weakly influenced by the presence of the additional intervening fm electrode , depending on sample and material parameters as we shall show below . ( ii ) the transresistance can be modified @xcite even when the additional contacting electrode is nonmagnetic . the relative magnitude of the interface and bulk resistance ( defined over the sdl ) plays an important role in determining the transresistance . irrespective of the magnetic or nonmagnetic nature of the intervening electrode , the transresistance will be influenced only by spin resistance and the interface quality . according to eq . ( [ ratio0 ] ) , the effect of an additional fm electrode on the transresistance is equivalent to its effect on the nonlocal leakage spin current . after the spin current is injected from f1 electrode , it will flow into both directions in n and will leak into nonlocal probes . from this perspective we can expect that the effect will be much stronger when an additional fm electrode lies in between two ( spin - injecting and spin - detecting ) fm electrodes than when it lies outside two electrodes . in the former case , the nonlocal spin current in spin - detecting probe will be reduced proportionally by the amount of spin current drained into an intervening electrode . in the latter case , the injected spin current leaks into the spin - detecting probe first and then into an additional fm electrode , so that the effect will be weaker . mathematically this difference between two cases comes from the asymmetry between eq . ( [ 3lead_i2 ] ) and eq . ( [ 3lead_i3 ] ) . under the index exchange @xmath124 , @xmath103 and @xmath104 are inequivalent due to the additional term @xmath125 in @xmath103 . to be more quantitative , let us consider @xmath103 when @xmath126 s in eq . ( [ d ] ) are all much less than a unity . this is a good approximation in all - metallic lateral spin valve systems since the spacing between the electrodes is comparable to the sdl which is of the order of few hundred nanometers . under this approximation , we can readily show that @xmath127 therefore , @xmath128 is very weakly influenced by the fm electrode f3 and the transresistance is almost the same as that in the absence of the electrode f3 . that is , the transresistance is not much changed by the additional electrode ( f3 ) when it is contacted to the outside of f1 ( spin current injected ) and f2 ( spin current detected ) . however , the effect of an additional fm electrode f3 can not be neglected if f2 and f3 are closer to each other than the sdl . so much for this case . we now focus on the case when an additional fm electrode lies in between the spin - injecting and spin - detecting electrodes . that is , we study the effect of f2 on the nonlocal spin signals for f3 . in the absence of the intervening fm electrode f2 , @xmath118 is @xcite @xmath129 note also that @xmath118 can be obtained from @xmath121 by taking the limit @xmath130 or when the second intervening f2 is effectively decoupled from the nonmagnetic base electrode . the effect of the second intervening electrode f2 on the transresistance @xmath121 can be quantified by computing the ratio @xmath131 , @xmath132 below this general relation will be reduced to the simple forms case by case . in order to provide some physical insights , let us consider the case when @xmath133 . we find the simple form of @xmath131 @xmath134 the reduction of @xmath121 stems from the leakage of spin current into the intervening electrode f2 . the efficiency of spin leakage into f2 is quantified by the relative magnitude of the serial resistance @xmath135 in f2 and the resistance @xmath136 in base electrode over the spin diffusion length . we can understand qualitatively the results of eq . ( [ ratio2 ] ) as follows . spins are injected from f1 into the base electrode n and in turn diffuse into left and right of n. that is , the spin current flows in n and leaks into nonlocal probes . just like charge transport , the spin current at the junction with f2 will continue to flow in n and also leak into f2 . if the effective spin resistance @xmath135 of f2 is much larger than the spin resistance @xmath136 of n , the leakage into f2 will be negligible and the spin current will mostly continue to flow in n. the leakage spin current @xmath103 is larger ( smaller ) if the effective spin resistance @xmath135 of f2 is smaller ( larger ) , compared to the spin resistance @xmath136 of n. obviously the leakage into f2 reduces the spin current in n and in turn reduces the leakage spin current @xmath104 . the larger ( smaller ) @xmath135 is , the larger ( smaller ) is @xmath104 . the spin diffusion length ( sdl ) is of the order of a few hundred nanometers ( nm ) in nonmagnetic metals and the sdl in fm metals is of the order of a few nm to a few tens of nm . the resistivity depends on the sample quality such as the impurities , defects , etc . though sdl is two orders of magnitude different between fm and nm , the relative magnitude of resistance ( @xmath137 : ferromagnetic metal , @xmath138 : nonmagnetic metal ) defined over the sdl can be varied from device to device . roughly @xmath139 in the spin valve devices . usually the interface between the fm electrodes and the nonmagnetic base electrode is ohmic ( @xmath140 ) , but not in the tunneling regime . in real materials , we have the following order in resistance : @xmath141 . for our theoretical study , we will consider both cases of ohmic and tunneling interfaces as well as other parameter regimes . we consider the clean interface between the base electrode and the fm electrodes : @xmath142 . to get the simple expression of @xmath121 , we take the limit @xmath143 . suppose that the fm electrodes are the same material with roughly the same @xmath144 for @xmath23 . if the exponentially decaying factors ( @xmath126 ) are negligible , we find the simple form of the transresistance ratio @xmath145 when @xmath146 , the transresistance will be strongly suppressed by the additional intervening fm electrode . on the other hand , the transresistance will be fractionally reduced when @xmath137 is comparable to @xmath136 . in the other extreme case of @xmath147 , the transresistance wo nt be affected by the intervening fm electrode . as noted in the above , the transresistance can be affected by the nonmagnetic electrode , @xmath148 and @xmath149 . let us study this case in detail . in the clean limit of interface , @xmath150 since @xmath151 with @xmath148 , we obtain the similar result as in the previous paragraph depending on the relative magnitude of @xmath152 . if the contacts between the base electrode and f1 , f3 are clean , but the contact with the intervening electrode f2 is in the tunneling regime , the effect of an additional electrode on the transresistance is negligible . when the junction resistance is dominant compared to the resistance over the spin - diffusion length in the fm lead and the base electrode , or when @xmath153 , the expressions of the voltage drop , eqs . ( [ 3lead_tr2 ] ) and ( [ 3lead_tr3 ] ) , are simplified as @xmath154 the voltage drop at each junction is not influenced by the presence of the other fm leads , when the junctions lie in the tunneling regime . in general , the expression of @xmath112 is not affected by the presence of the second fm lead ( additional fm lead ) as far as the contact is in the tunneling regime . when @xmath155 , @xmath156 so that the second fm lead is effectively disconnected from the base electrode . when the accumulated spin is diffused efficiently into the second intervening fm lead , its effect may not be negligible . we still assume that the contacts with f1 and f3 lie in the tunneling regime . let us see the extreme case of a transparent contact of f2 electrode to the base electrode . in this case , we may set @xmath157 and the desired voltage drop is given by the expression @xmath158 that is , the transresistance can be changed by the second intervening electrode f2 if f2 is in clean contact with the base electrode or if spin leakage into f2 is efficient . using the one - dimensional spin drift - diffusion equations , we studied theoretically the mutual effect of ferromagnetic electrodes on non - local spin signals ( the leakage spin currents and the voltage drops ) in the lateral spin valve with three ferromagnetic electrodes . we found the generic expression of the leakage spin current , eq . ( [ nonlocal_i ] ) , and also a very simple relation , eq . ( [ nonlocal_v2 ] ) , between the nonlocal voltage drop and the leakage spin current . ( [ ratio0 ] ) tells us that the effect of an additional electrode on the transresistance can be discussed in terms of the leakage spin current and in turn in terms of the conductance matrix . the measured non - local spin signals depend on the position of an additional fm electrode relative to the spin - injecting and spin - detecting electrodes . when the additional electrode lies outside of the two fm electrodes , non - local spin signals are found to be weakly influenced due to the exponentially decaying spin coherence over the sdl . on the other hand , when it is located in between the two fm electrodes , the non - local spin signals can be strongly modified provided the junction resistance is lower than or comparable to the spin resistance defined over the spin diffusion length in the fm electrodes and the nonmagnetic base electrode . if the junction resistance is high , the non - local spin signals are weakly modified even when the additional fm electrode is located in between the two fm electrodes the most general expression for the transresistance ratio is given by eq . ( [ ratio ] ) . in general , the non - local spin signal is not much modified when the additional electrode is in tunneling contact with the base electrode , but is fractionally reduced when the contact is ohmic . we also found that the non - local spin signals are independent of the magnetization orientation of the additional fm electrode , which agrees with the experimental observation.@xcite this result suggests that even the intervening _ nonmagnetic _ electrode can change non - local spin signals , which was already observed @xcite experimentally . since our theoretical study is based on the one - dimensional device structure , some care is needed when we try to apply our theoretical results to interpretation of experimental data . strictly speaking , the experimental spin valve structure is not one - dimensional in terms of the current distribution . et al . _ numerically showed @xcite that the nonlocal voltage drop depends strongly on the spatial distribution of the spin - polarized current . the one - dimensional approximation is valid when the current is uniformly distributed through the contact . when the contact is clean between the fm electrode and the base nonmagnetic electrode , the current flow may well not be uniform through the interface @xcite and may be short circuited . in this case , the nonlocal spin signals may deviate from its theoretical estimate based on one - dimensional sdd equations . keeping these restrictions in mind , let us apply our theoretical results to two experimental works . @xcite for numerical estimation ( @xmath159 in this section ) , we take examples of co / cu / co and py / cu / py lateral spin valves and use the following sample size and material parameters . the thickness and width of the nonmagnetic base electrode are taken as 80 nm and 300 nm , respectively . the width of all the ferromagnetic layers is assumed to be the same as 100 nm . the separation between nearest ferromagnetic layers is taken as 200 nm , which gives 300 nm of center - to - center distance . we use material parameters measured at low temperatures . the parameters for cu are @xmath160m@xcite and @xmath161m.@xcite for co , we use @xmath162,@xcite @xmath163,@xcite @xmath164m,@xcite @xmath165 nm,@xcite and @xmath166m@xmath167.@xcite for py , we take @xmath168,@xcite @xmath169,@xcite @xmath170m,@xcite @xmath171 nm,@xcite and @xmath172m@xmath167.@xcite for the co / cu / co spin valve , @xmath173 m@xmath174 , @xmath175 m@xmath174 , and @xmath176 m@xmath174 are obtained . the estimated spin signal is reduced to the value @xmath177 by the intervening f2 electrode . @xmath178 is also reduced by the factor 0.87 due to f3 electrode . for the py / cu / py spin valve , we have @xmath173 m@xmath174 , @xmath179 m@xmath174 , and @xmath180 m@xmath174 . the reduced spin signal @xmath181 by the f2 electrode is @xmath182 . @xmath178 is reduced by the factor 0.87 due to f3 electrode . since the sdl of co is rather long , @xmath183 is comparable to @xmath184 in the co / cu / co case and @xmath185 is large . since , in the py / cu , @xmath184 is bigger than @xmath183 and @xmath186 , @xmath185 is small . the rather significant reduction of estimated @xmath178 in both cases stems from our choice of the long sdl of cu at low temperatures . the long sdl means that the chemical potential splitting between opposite spin directions , though exponentially decaying , remains significant up to the position of the f3 electrode . the significant leakage of spin currents into f3 results in reduction of the spin signal . at room temperature , the sdl of the base electrode ( nonmagnetic metal ) is a few hundreds nanometers such that the reduction of @xmath178 by the f3 electrode is only a few percents . for the experimental conditions in refs . and , @xmath184 is comparable to @xmath135 and we can estimate theoretical value of @xmath185 , eq . ( [ ratio2 ] ) : @xmath187 although the observed @xmath185 is smaller for ref . and is close to a unity for ref . . as pointed out in ref . , the contact between the permalloy electrode and the base ag wire is very clean and the point injection and detection of current is suggested . in this case , the current distribution in the devices may well be nonuniform so that our one - dimensional theory can not be straightforwardly applied . we believe that the nonuniform current distribution is the main reason why some of our theoretical estimates are in poor agreement with the results of ref . . we may discuss the relevance of the device dimensionality based on the effective spin resistance . according to our theoretical analysis , nonlocal spin signals are determined by the relative magnitude of junction resistance and spin resistance in fm and nm electrodes . this relevant resistance is defined under the assumption that the current distribution is uniform in the device . when the current distribution is not uniform as in real devices , we may still be able to define the spin resistance using the effective cross sectional area which is smaller than the geometrical cross section of the sample . nonuniform current distribution tends to increase junction resistance as well as spin resistance , and will modify the magnitude of nonlocal spin signals . this may be one of the reasons for the discrepancy between two experimental results.@xcite nonlocal spin signals ( the leakage spin current and the voltage drop ) in one nonlocal fm electrode are shown not to depend on the magnetization orientation ( parallel or antiparallel ) of the other nonlocal fm electrode . we believe this symmetry of nonlocal spin signals are robust against the sample dimensionality , though their magnitude is sensitive to samples . the spin current is generated by the spin - injecting fm electrode and so its flow direction can not be changed by the magnetization orientation of nonlocal fm electrodes . in addition , the magnitude of spin current does not depend on the magnetization orientation of nonlocal fm electrodes . this property derives from both decoupling of spin and charge modes in the sdd equations and zero charge current in nonlocal voltage probes . hence our conclusion about the relationship between nonlocal spin signals and magnetization in nonlocal voltage probes wo nt depend on the sample dimensionality and qualities . this point is demonstrated more explicitly in appendix [ 3d - case ] . finally , we would like to discuss the properties of transresistance ( tr ) and ( longitudinal ) magnetoresistance ( mr ) in spin valves under magnetization reversal . obviously , both tr and mr are modified under magnetization reversal of two probing fm electrodes . under magnetization reversal , tr changes its sign while mr changes its value . note that mr , in general , consists of the two contributions : one part ( background ) remains the same but the other changes its sign under magnetization reversal . let us consider the effect of an additional fm electrode ( f@xmath5 ) on tr and mr . for the vertical spin valves , it wo nt be easy to implement f@xmath5 . so we consider tr and mr in the lateral spin valves with f@xmath5 . usually mr is obscured by other effects in the lateral spin valves as mentioned before . however , with increased sdl , mr was successfully measured @xcite in the carbon nanotube and graphenes . to measure mr in the lateral spin valves of fig . [ 3fmlead ] , f1 and f3 are both current and voltage probes . based on the results of appendix [ 3d - case ] , we can argue that mr should be independent of the magnetization orientation of f2 , @xcite because there is no charge current in f2 ( an additional electrode ) . explicit calculation , @xcite using the sdd equations , confirms this claim . that is , both tr and mr are independent of the magnetization orientation of f@xmath5 . on the other hand , tr and mr depend on the magnetization orientation of ferromagnetic electrodes through which the charge current flows . this work was supported by the korea science and engineering foundation ( kosef ) grant funded by the korea government ( most ) ( no . r01 - 2005 - 000 - 10303 - 0 ) , by the src / erc program of most / kosef ( r11 - 2000 - 071 ) and by postech core research program . in this appendix we show that the algebraic manipulation can be highly simplified by a proper definition of material parameters and by the vector and matrix notations . kirchhoff rules lead to the constraints given by eqs . ( [ rule_i ] ) and ( [ rule_v ] ) . ( [ rule_v ] ) can be written down explicitly leading to the following six relations ( @xmath23 ) @xmath188 \nonumber\\ & & = \left [ ( 1 \pm \beta_1 ) i ~ \delta_{i,1 } \pm i_i \right ] \times \frac{{\cal r}_{ti}}{1 \pm \gamma_i}.\end{aligned}\ ] ] for algebraic convenience , we introduce new material parameters @xmath189 and @xmath190 . these new material parameters highly simplify the complicated algebra as well as determine the spin currents in the nonlocal voltage probes . then the voltage drop @xmath42 can be written as @xmath191.\end{aligned}\ ] ] addition and difference of two @xmath42 s lead to @xmath192 and @xmath193 it is much more convenient to introduce the matrix notation for the algebraic manipulation . @xmath194 , @xmath195 , @xmath196 , and @xmath197 . here the superscript @xmath198 represents the transpose of row vectors so that its effect is to change them into column vectors . with these notations , kirchhoff rules can be written in compact forms as @xmath199 ~ |i \rangle \nonumber\\ & & - \beta r { \bf a } |j \rangle , \\ 0 & = & ( { \bf r } + { \bf r}_t ) |i \rangle - r { \bf a } |j \rangle \nonumber\\ & & + ( \beta_1 - \gamma_1 ) r_{t1 } i ~ |1 \rangle .\end{aligned}\ ] ] here @xmath200 and @xmath201 are diagonal matrices with diagonal elements @xmath79 and @xmath80 , respectively . similarly , @xmath202 and @xmath203 are diagonal matrices with diagonal elements representing the spin polarization of each fm electrode and the junction resistance , respectively . @xmath204 is the matrix with its elements given by @xmath205 . formally , the unknown parameters can be written in a more compact matrix form as @xmath206 i { \bf g } |1 \rangle , \\ & & + [ ( \beta_1 - \beta ) r_1 + ( \gamma_1 - \beta ) r_{t1 } ] \nonumber\\ & & \times [ ( \hat{\beta } - \beta ) { \bf r } + ( \hat{\gamma } - \beta ) { \bf r}_t ] i ~ { \bf g } |1 \rangle .\end{aligned}\ ] ] here the matrix @xmath78 is defined by the expression @xmath207^{-1}.\end{aligned}\ ] ] the matrix @xmath78 , with the dimension of conductance , is independent of magnetization configurations ( parallel or antiparallel to the spin - injecting electrode f1 ) of the fm electrodes . the set of parameters , @xmath75 and @xmath42 , contains all the information about the spin polarized transport in nonlocal spin valves . in components , we have the spin currents @xmath208 , \\ \frac{j_i}{i } & = & -\frac{1}{2 } g_{i1 } [ ( \beta_1 - \beta ) r_1 + ( \gamma_1 - \beta ) r_{t1 } ] , \end{aligned}\ ] ] and the voltage drops in the fm electrodes @xmath209 \delta_{i,1 } \nonumber\\ & & + [ ( \beta_i - \beta ) r_i + ( \gamma_i - \beta ) r_{ti } ] \nonumber\\ & & \times g_{i1 } [ ( \beta_1 - \beta ) r_1 + ( \gamma_1 - \beta ) r_{t1}].\end{aligned}\ ] ] note that the final results are written down in a very compact form , using new material parameters as well as the conductance matrix . spin resistance was defined in order to simplify the algebra . in this section we are going to infuse some physical meaning into spin resistance . let us recast eq . ( [ spin_r ] ) for @xmath1 ( nonlocal voltage probes ) into a more illuminating form as @xmath210 where @xmath211 acts as the effective electric potential of the base electrode at the junction with f@xmath0 and is defined by @xmath212 for electrons with negative charge , @xmath42 is the electric potential for both spin directions far into f@xmath0 , and @xmath213 is the electric potential for spin - up and spin - down electrons , respectively , of the base electrode at the junction with f@xmath0 . refer to eqs . ( [ ecp_fm ] ) and ( [ ecp_n ] ) . we can deduce that the current @xmath214 at the interface flows into ( out of ) f@xmath0 for spin - up ( spin - down ) electrons . the left hand side ( lhs ) of eq . ( [ spin_r2 ] ) represents the electric potential difference for both spin directions between the base electrode and f@xmath0 at the deep inside . the right hand side ( rhs ) is the product of the current @xmath214 and the effective spin - dependent resistance . the sign in front represents correctly the flowing direction of the spin - up and spin - down current , respectively . the first term in the parenthesis is the spin - dependent tunnel resistance as defined in eq . ( [ r_ts ] ) . the second term is none other than the spin resistance , which was introduced in the main text . with this spin resistance , eq . ( [ spin_r2 ] ) is the effective ohm s law for the leakage spin - up and spin - down currents . when @xmath211 is eliminated from eq . ( [ spin_r2 ] ) , the relation between the nonlocal voltage drop @xmath42 and the leakage spin current @xmath46 or eq . ( [ nonlocal_v2 ] ) is obtained . from eq . ( [ spin_r2 ] ) , we can deduce the physical meaning of @xmath42 . no charge current flows in the nonlocal fm electrodes . @xmath42 represents the shift of the electrochemical potential in f@xmath0 to satisfy the constraint of no charge current flow . if we eliminate @xmath42 from eq . ( [ spin_r2 ] ) , we find the following relation @xmath215 the material parameters are defined in appendix [ dalgebra ] . how do we interpret this relation ? this relation can be considered as the ohm s law for the leakage spin current . the lhs is the product of the spin current @xmath46 and the effective resistance . from the standpoint of f@xmath0 , the spin current @xmath46 flows from both sides of the base electrode ( @xmath136 ) through the junction ( @xmath80 ) and into f@xmath0 ( @xmath79 ) . hence the effective resistance is @xmath216 as in the above equation . the rhs is the effective spin potential which combines the source term from f1 and the sink terms from other nonlocal fm electrodes . one of key results of our paper is the independence of the spin accumulation , the spin current , and the nonlocal voltage on the magnetization directions of electrodes . this appendix is aimed to provide an insight into the origin of this independence in three - dimensional situations . we again consider the geometry in fig . 1 . similar notations will be used . the three - dimensional sdd equation is given by eq . ( [ sdde_3d ] ) and the associated charge and spin current densities @xmath217 , @xmath218 are given by @xmath219 similar relations hold for @xmath220 , @xmath221 , and @xmath222 in the f@xmath0 electrode . the system is subject to the following boundary conditions . from the condition of no leakage current to air or insulating substrate , @xmath223 should hold at the sample boundaries facing air or insulating substrate . here @xmath224 and @xmath225 denote normal vectors perpendicular to the boundaries . from the constraint of the current continuity applied to the interface between the base electrode and the electrode f@xmath0 , one finds that the following relation should hold at the interface , @xmath226 the ohm s law provides another boundary condition for the interface , @xmath227 when combined with these boundary conditions , the sdd equation completely fixes the spin - dependent electrochemical potentials . here we remark that the current density @xmath228 , instead of the current @xmath229 , appears in eq . ( [ 3d - ohms - law ] ) and thus @xmath230 in eq . ( [ 3d - ohms - law ] ) amounts to the spin - dependent junction resistance _ per unit area _ , instead of the junction resistance . we also remark that the tunneling barrier at a junction may not be uniform in realistic experimental situations and such non - uniformity can be taken into account by simply regarding @xmath231 as a position - dependent quantity since eq . ( [ 3d - ohms - law ] ) remains still valid even for the nonuniform barrier as long as the tunnelling current remains perpendicular to the interface . in order to examine the dependence of the spin accumulation on the magnetization directions , we reexpress the involved equations in terms of the spin accumulation @xmath232 and the charge potential @xmath233 . the sdd equation ( [ sdde_3d ] ) is again decomposed into the following two decoupled equations ( spin mode and charge mode ) , @xcite @xmath234 the boundary conditions for the spin accumulation can be derived from eq . ( [ 3d - no - leakage - current ] ) and one obtains @xmath235 from eq . ( [ 3d - current - continuity ] ) , one obtains @xmath236 and from eq . ( [ 3d - ohms - law ] ) , one obtains @xmath237 now we are ready to discuss the magnetization direction dependence of the spin accumulation , which is completely fixed from its sdd equation ( [ 3d - sdd - equation - spin - accumulation ] ) and boundary conditions ( [ 3d - no - leakage - current - spin - accumulation ] ) , ( [ 3d - current - continuity - spin - accumulation ] ) , and ( [ 3d - ohms - law - spin - accumulation ] ) . note that in these equations , all terms that depend on the magnetization directions are multiplied by the charge current density . thus the spin accumulation should be independent of the magnetization directions of electrodes in which the charge current density vanishes . next we discuss the magnetization direction dependence of the spin current . the spin current can be obtained from the spin accumulation as follows , @xmath238 then from the properties of the spin accumulation , it is evident that the spin current density should be independent of the magnetization directions of electrodes in which the charge current density vanishes . finally we discuss the charge potential , which is subject to the sdd equation ( [ 3d - sdd - equation - charge - potential ] ) . the boundary conditions for the charge potential can be derived from eqs . ( [ 3d - no - leakage - current ] ) , and one obtains @xmath239 from eq . ( [ 3d - current - continuity ] ) , one obtains @xmath240 and from eq . ( [ 3d - ohms - law ] ) , one obtains @xmath241 \nonumber \\ & & \,\,\,\,\,\ , \times \hat{\bf n}\cdot \nabla ( \mu_{i+}-\mu_{i- } ) \nonumber \\ & & + { { \cal r}_{ti } \over 8}\left [ ( 1+\beta_i)^2(1+\gamma_i)+(1-\beta_i)^2(1-\gamma_i)\right ] \hat{\bf n}\cdot { \bf j}^{\rm c}_i . \nonumber\end{aligned}\ ] ] note that the sdd equation ( [ 3d - sdd - equation - charge - potential ] ) and the boundary conditions ( [ 3d - no - leakage - current - charge - potential ] ) , ( [ 3d - current - continuity - charge - potential ] ) are not dependent on the magnetization directions of any electrodes . thus the magnetization direction dependence can arise only from the boundary condition ( [ 3d - ohms - law - charge - potential ] ) . from eq . ( [ 3d - ohms - law - charge - potential ] ) and from the properties of the spin accumulation , one then finds that the charge potential at the electrode f@xmath0 is independent of the magnetization direction of other non - current - carrying electrodes . this in turn implies that the nonlocal voltage measured between the electrode f@xmath0 and the base electrode ( @xmath242 ) should be independent of the magnetization directions of other non - current - carrying electrodes f@xmath101 @xmath243 . l. berger , phys . b * 54 * , 9353 ( 1996 ) . j. slonczewski , j. magn . magn . mater . * 159 * , l1 ( 1996 ) . for a recent review , see a. bratass , g. e. w. bauer , and p. j. kelly , phys . rep . * 427 * , 157 ( 2006 ) ; y. tserkovnyak , a. brataas , g. e. w. bauer , and b. i. halperin , rev . phys . * 77 * , 1375 ( 2005 ) . m. n. baibich , j. m. broto , a. fert , f. nguyen van dau , f. petroff , p. etienne , g. creuzet , a. friederich , and j. chazelas , phys . lett . * 61 * , 2472 ( 1988 ) . g. binasch , p. grnberg , f. saurenbach , and w. zinn , phys . b * 39 * , 4828 ( 1989 ) . f. j. jedema , h. b. heersche , a. t. filip , j. j. a. baselmans , and b. j. van wees , nature ( london ) * 416 * , 713 ( 2002 ) ; f. j. jedema , m. s. nijboer , a. t. filip , and b. j. van wees , phys . b * 67 * , 085319 ( 2003 ) .

optical spectra of active galactic nuclei ( agn ) exhibit an extremely wide variety of properties . the emission - line regions of agn have proven to be among the most complex of all astrophysical environments @xcite . agn have also been known to emit a substantial fraction of their luminosity as x - rays . multi - wavelength observations of well - defined source samples are effective in understanding the physical processes in agn . in the last decade , several correlations between optical emission line properties and x - ray properties were noted @xcite : for instance , the strong correlations between optical fe ii , [ o iii]@xmath1 line strengths , velocity width of h@xmath0 and the slope of soft x - ray continuum ( @xmath2 ) . these correlations received great interest previously and were expected to give important insight into the agn phenomenon . still , however , their interpretation remained unclear . it is therefore of great importance to test whether these correlations are fundamental properties or subtle selection effects , and search for new ones . clearly , a large and homogeneous x - ray selected sample would be advantageous in addressing these issues . we present such a sample here . x - ray agn samples are also of great interest for a number of other issues . agn of different types are the major contributors to the extragalactic soft and hard x - ray background ( xrb ) . the definitions of statistically significant agn samples using x - ray data are key to test agn number - flux relations , luminosity functions and cosmological evolution . in the course of the rosat all sky survey ( rass ) @xcite , many new agn have been identified by different teams , confirming that x - ray surveys are very efficient in finding new agn . @xcite performed an unbiased survey of all bright , @xmath3 , high galactic latitude sources in the rass without applying other selection criteria . * paper i ) identified a sample with high x - ray to optical flux ratio . @xcite and @xcite presented a catalog of sources in selected areas down to the rass limit . @xcite drew a high galactic latitude sample from the rass with soft pspc spectra ( @xmath4 ) , the corresponding sample of point - like x - ray sources with relatively hard x - ray spectra ( @xmath5 ) was presented by @xcite . @xcite published a catalog of northern agn in the rass based on their schmidt plate survey . these authors presented new agn , but they did not use their samples to address correlations issues . this will be done with the present sample . in paper i we described and presented the optical identification of a sample of 165 x - ray sources with high x - ray to optical flux ratio ( @xmath6 ) discovered in the rass . about 73% of the sources were identified as agn , including 115 emission line agn ( qsos and seyferts ) as well as 2 bl lac objects and 4 bl lac candidates . in this paper we investigate the optical emission line properties of the emission line agn that were identified as main optical counterparts of the sample and discuss their connection with x - ray properties . not included in the following discussion are the bl lac objects . the initial identification program excluded known x - ray sources that have counterparts in the simbad , ned and agn catalogs . motivated by the need of larger and better samples , we have cross - referenced the rass bright source catalog with the ned and agn catalogs . we find that 40 known emission line agn meet our selection criteria . , where cr and r represent x - ray count rate and r magnitude respectively ; declination @xmath7 ; galactic latitude @xmath8 ; optical counterparts within a circle with radius @xmath9 , where @xmath10 is the rass position error given by voges et al.(1996 ) ; optical counterparts with r magnitudes between 13.5 and 16.5 . ] in order to keep consistent emission line properties measurements , new optical spectra have been acquired for the 40 additional sources . these data , combined with the previous optical identifications , provide a large , bright and ( @xmath6 ) limited sample of emission line agn in the rosat x - ray band . the relatively complete sample is the largest x - ray selected agn sample with intense optical investigation so far . as described in paper i , our classifications and identifications were based on low resolution spectra obtained with a 2.16 m telescope at xinglong station , national astronomical observatories of china ( naoc ) . these spectra were exposed to reach a s / n suitable for a reliable classification . the quality of the majority spectra is sufficient to measure the properties of the emission lines , such as the line widths and line strengths . for spectra with inadequate s / n , new low resolution ( @xmath1110 fwhm ) optical spectroscopy was performed using the 2.16 m telescope at naoc . all objects in our emission line agn subsample are found to be seyfert 1 type agn . throughout this paper seyfert 1 galaxies and qsos will be discussed jointly , since these two classes are distinguished only by a certain luminosity level according to our classification criterion . we do not divide the seyfert 1 class into subclasses such as seyfert 1.5 - 1.9 or nls1 galaxies ( as defined by @xcite , @xcite ) , since the definitions of seyfert classifications depend on the resolution of the spectra used and the noise in the spectra @xcite . the paper is organized as follows . detailed optical measurements are given in sect . the statistical properties are investigated in sect . 3 . several diagnostically important correlations that exist between emission line and continuum properties are studied . 4 discusses the implications of the results , followed by a brief summary of the main conclusions of the paper . throughout this paper , a hubble constant of @xmath12 and a deceleration parameter of @xmath13 are assumed and all measured parameters are quoted for the rest frame of the source . the known sources as well as sources without high quality observations in the identification campaign were investigated during several observing runs performed from november 1998 through december 2000 . the low resolution spectra were taken with the naoc 2.16 m telescope and the omr spectrograph , using a tektronix @xmath14 ccd as detector . the grating of 300 g mm@xmath15 was employed in order to get large wavelength coverage . all observations were made through a 2.3@xmath16 slit that produced a resolution of 10 to 11 as measured from the comparison spectra . this set - up provided similar wavelength coverage ( 3800 8200 ) and resolution to the identification data . exposure time was generally between 1200s and 3600s depending on the brightness of the object . for a few objects , multiple exposures were performed in order to get a higher s / n spectrum . the exposures were combined prior to extraction where possible in order to help remove cosmic ray contamination . in cases when the target source moved across the ccd chip between exposures , the data were extracted separately from each frame and then combined . the raw data were reduced following standard procedures using iraf . the ccd reductions included bias subtraction , flatfield correction and cosmic - ray removal . wavelength calibration was carried out using helium - neon - argon lamps taken at the beginning and end of each exposure . the resulting wavelength accuracy is better than 1 . the flux calibration was derived with 2 to 3 observations of kpno standard stars @xcite per night . in many of the spectra there is a clear contribution from blends of fe ii line emission on both the blue and red sides of the h@xmath0-[o iii ] complex . these blends contaminate strong lines such as [ o iii]@xmath174959,5007 , and may alter the flux and width of h@xmath0+[o iii]@xmath174959,5007 . in order to reliably measure line parameters and to determine the strength of the fe ii emission , we have carefully removed the fe ii multiplets following the method described by @xcite , which relies on an fe ii template . the template used in the present work is the same as that of @xcite , namely the fe ii lines of i zw 1 , a bright nls1 widely known for the strong and narrow permitted fe ii emission @xcite . the observed spectra were wavelength - shifted to the rest frame according to the redshift.@xmath185007 to be the systemic redshift as most commonly used . the redshift was derived from fitting a gaussian to the upper half of the observed [ o iii]@xmath185007 line in each spectrum . ] the whole template was broadened to the fwhm of the broad h@xmath0 line by convolving with a gaussian and scaled to match the line intensities . the best match was then searched for in the two - dimensional parameter space of line width and line strength . from each object the best - fit fe ii spectrum was then subtracted . a successful fe ii subtraction showed a flat continuum between the h@xmath19 and h@xmath0 and between 51005400 ( which covers the fe ii multiplets 48 , 49 ) . the fe ii flux was measured between the rest wavelength 4434 and 4684 as in @xcite . the characteristic property of seyfert 1 type agn is the presence of broad emission lines . the line widths and their distributions provide important information on masses of the central black holes and mass distributions . the broad balmer lines in agn exhibit a wide variety of profile shapes and a large range in width @xcite and they are often strongly asymmetric @xcite . in many cases h@xmath0 is a mixture of broad and narrow components . differences in the relative strengths of these components account for much of the diversity of broad line profiles @xcite . the fe ii subtracted spectra were used to measure the non - fe ii line properties . in order to isolate the broad h@xmath0 component in all spectra , we have assumed that the emission line profiles can be represented by a single or a combination of gaussian profiles . @xcite claimed that the broad emission lines in nls1 galaxies were better fitted with lorentzian profiles than with gaussian profiles . however , as noted in @xcite , the choice of gaussian or lorentzian profiles as representatives of the observed emission lines may bear no physical meaning . we performed a multiple gaussian fit in the present work for its simplicity and a direct comparison with the measurements of previous studies ( e.g. @xcite ) . we adopted the method described by @xcite and used the iraf package specfit to measure blended lines . as a first step we tried to fit the emission lines with a single gaussian component . it worked well for most forbidden lines , e.g. , [ o iii]@xmath185007 . however , this simplest representation could not fit adequately the wings of h@xmath0 , although its core was nicely fitted in most of targets . in such cases , we then included one narrow component and one broad component to describe adequately the observed profile . ( note that when h@xmath0 measurements were taken from the literature , many of them refer to the whole line . ) for each assumed component , a first guess of the central wavelength , the total flux , and width of that component were specified . there were two parameters for each continuum component , i.e. , the flux and the slope . initially , we constructed a narrow component with a width determined from fitting the [ o iii ] lines . in some spectra such a guess failed to give a good fit and needed an additional component . we then left the width of the narrow component as a free parameter . it led to a nice fitting result by checking the residual , i.e. , the narrow component could be well isolated from the broad component . for h@xmath0 lines with asymmetric profile , the two gaussian components are not stationary in wavelength . the velocity shift between the broad and narrow components is defined as the shift of the broad component relative to the narrow component in units of km s@xmath15 ( see also zheng et al 2002 ) . as an illustration , we show the decomposition in figure 1 . the observed profile is represented by a solid line , the total fit by a dashed line , the individual gaussian components by a dot - dashed line and the differences between the data and the fit ( the residual ) by the lower dotted line . the broad component is referred as h@xmath20 , and the narrow component as h@xmath21 . the fwhm h@xmath20 is what we consider as representative of the broad line region ( blr ) . the instrumental resolution was determined from the fwhms of night sky spectra taken with each of the exposures and tested by comparison - lamp spectra . all quoted line widths were corrected for the instrumental resolution . the uncertainty of h@xmath0 velocity shift measurement is within 150 km s@xmath15 . to measure line fluxes and equivalent widths ( ew ) , power - law continua were fitted and subtracted from underneath the lines and the remaining line flux was integrated . for most objects , the uncertainty of flux measurement of emission lines is about 10% to 15% . the continuum level at the position of h@xmath0 @xmath184861 was measured from the fe ii subtracted spectra . all equivalent widths measurements in this paper refer to the continuum at that point to allow direct comparison with the measurements for other samples @xcite . the optical index @xmath22 was calculated using the continuum flux density at 4000 and 7000 in the rest frame . for the x - ray analysis , a single power law description was employed . this application has proven to be a fairly good approximation for agn in previous studies ( e.g. fiore et al . 1994 , brinkmann & siebert 1994 , schartel et al . 1996 , grupe et al . 1998 ) . because of the limited signal - to - noise ratio of the rass spectra we assumed that the absorption comes from our galaxy only . the assumption usually improves the reliability of the estimation of spectral index because of the reduced number of free parameters . the x - ray spectral slope @xmath23 was estimated using the rosat hardness ratios hr1 and hr2 , which are defined as @xcite : @xmath24 @xmath25 here , a , b , c and d denote the count rate in the energy range 0.10.4 kev , 0.52.0 kev , 0.50.9 kev and 0.92.0 kev , respectively . the full optical dataset was used to define the statistical properties of the sample and explore the correlations between various observed parameters . according to paper i , 115 sources ( about 73% of the total ) were identified as new seyfert 1s and qsos , together with 40 known emission line agn that meet our sample selection criteria , the agn subsample comprises 49 `` ultrasoft '' ( @xmath26 ) sources and 106 `` normal '' objects with mean @xmath27 . in the catalog in paper i we listed the absolute magnitude m@xmath28 and redshift @xmath29 . the distributions of the m@xmath28 values and redshift are plotted in figure 2 . as shown by these figures , the sample is dominated by moderate luminosity objects ( @xmath30 ) with an average redshift @xmath31 . table 1 lists the derived parameters . the columns list the following information : ( 2 ) redshift ; ( 3 ) fwhm of broad h@xmath0 ; ( 4 ) fwhm of narrow h@xmath0 ; ( 5 ) fwhm of [ o iii]@xmath185007 ; ( 6 ) , ( 7 ) and ( 8) the local equivalent widths of h@xmath0 broad component , fe ii and [ o iii]@xmath185007 ; ( 9 ) and ( 10 ) ratio of fluxes of [ o iii]@xmath185007 to broad and narrow h@xmath0 ; ( 11 ) ratio of fluxes of fe ii to broad h@xmath0 ; ( 12 ) velocity shift between h@xmath0 broad and narrow components , where a positive velocity corresponds to a redshifted broad component relative to narrow component ; ( 13 ) and ( 14 ) h@xmath0 broad and narrow components velocity shift with respect to the systemic frame , where a positive velocity indicates a redward shift with respect to systemic ; ( 15 ) and ( 16 ) spectral indices defined above ; ( 17 ) the monochromatic 250ev luminosity . the distribution of the broad h@xmath0 line width of the objects in our emission line agn subsample shows a significantly smaller fraction ( 3% ) of objects with fwhm @xmath32 2000 km s@xmath15 compared to another x - ray selected sample ( 18% ) @xcite . however , the fwhm measurements in @xcite refer to the full line profile and the widths of the broad components may have been underestimated . therefore , the real difference should be smaller . five out of 155 objects in our sample are nls1 galaxies without detectable broad balmer line components of fwhm @xmath33 2000 km s@xmath15 but with strong fe ii emission , which agrees within the error limits with that of @xcite ( about 2% ) . the distribution of the fwhm of the broad component of h@xmath0 in the sample is displayed in figure 3 . the mean value of 4350 km s@xmath15 is similar to the hard sample of @xcite . the velocity shift distribution of h@xmath34 with respect to [ o iii]@xmath1 is shown in figure 4a . positive velocity corresponds to a redward h@xmath0 with respect to systemic , while a negative velocity represents a blueshift . the peak of this distribution is near zero , but has a mean redshift of 78 km s@xmath15 with standard deviation of 305 km s@xmath15 . this shift is in good agreement with the previous studies @xcite , i.e. , balmer emission lines gives the same redshift to within 100 km s@xmath15 of narrow forbidden emission ( _ e.g. _ [ o iii]@xmath185007 ) in low redshift ( @xmath35 ) qsos and seyfert galaxies . in contrast , a systematic mean redward shift of 520 km s@xmath15 was found for a high redshift ( @xmath36 ) qsos sample @xcite . @xcite proposed this observed trend of increased balmer redshift with increased systemic redshift represents a luminosity dependency . for comparison , we also present the h@xmath21 [ o iii]@xmath1 shift in figure 4b . the mean of the distribution is 8 km s@xmath15 with a standard deviation of 176 km s@xmath15 . we can see from the figures that the dispersion of the h@xmath21 [ o iii]@xmath1 shift is significantly narrower than the h@xmath34 [ o iii]@xmath1 shift , which is expected by the different emitting regions between h@xmath0 broad component and narrow component , as well as [ o iii]@xmath1 . in a similar manner to the h@xmath34 [ o iii]@xmath1 shift , figure 4c illustrates the velocity differences between the broad and narrow components of h@xmath0 . the redshift ( or blueshift ) is defined as the shift of the broad component relative to the narrow component in units of km s@xmath15 . positive velocities refer to redshifts with respect to the narrow component velocities , whereas negative velocities indicate blueshifts . we emphasis that both the h@xmath0 redshift ( or blueshift ) in the present work and the asymmetry parameter in @xcite are measures for line profile , but characterized differently . the latter is a measure of the shift between the centroids at @xmath37 and @xmath38 maximum . the mean redshift of the distribution is found to be 70 km s@xmath15 with a standard deviation of 357 km s@xmath15 . correlation analyses involving the line profiles of h@xmath0 are investigated in section 3.3 . in this subsection , we explore whether the various emission - line and continuum properties correlate with one another . for this purpose , we calculated the spearman rank - order correlation matrix @xcite , along with its significance matrix for measured properties . the complete correlation coefficient matrix is shown in table 2 . the number of parameter pairs included in the trials ranges from 86 to 149 . spectra with s / n @xmath39 were excluded from the analyses including optical properties . the probability of the null correlation , ps , for a sample with corresponding correlation coefficient rs is also given in table 2 for entries with @xmath40 . a set of 12 different properties results in @xmath41 correlation coefficients ; therefore , at this level of significance , we would expect @xmath42 spurious events . among 66 trials , 31 correlations ( both positive and negative ) are found with two - sided probabilities @xmath40 . four are due to the dependent parameters , e.g. , relative strength of fe ii to h@xmath0 and fe ii ew ; 12 are degenerate correlations , i.e. , different measures of the same property ( e.g. , [ o iii ] ew , [ o iii]/h@xmath34 and [ o iii]/h@xmath21 ) all correlate with another property ( e.g. , fe ii ew ) ; and therefore 15 are independent correlations at the @xmath43 confidence level . the well - known anti - correlation between the x - ray spectral slope @xmath44 and the fwhm of h@xmath0 @xcite is not prominent in our sample ( @xmath45 , @xmath46 ) as is the case in the @xcite sample ( @xmath47 , @xmath48 ) . @xcite suggested that the relation between h@xmath0 width and @xmath44 is not a linear correlation , but probably resulted from a `` zone of avoidance '' @xcite , i.e. , broad line objects always tend to show flat x - ray spectra , whereas there is a very large scatter in the x - ray spectral steepness of nls1 galaxies , with several as flat as normal seyfert 1s ( xu et al . the correlation is more pronounced among the agn with higher luminosity as in the samples of @xcite and @xcite . for objects with @xmath49 the significance becomes @xmath50 . the correlation between @xmath44 and h@xmath34 is illustrated in figure 5 . the anti - correlation between the width of h@xmath0 and fe ii / h@xmath0 @xcite is strong ( @xmath51 , @xmath52 ) in the present data whereas it is only weak ( @xmath47 , @xmath53 ) if fwhm h@xmath0 is compared directly with ew fe ii . instead , there is a strong correlation ( @xmath54 , @xmath52 ) between fwhm h@xmath0 and ew h@xmath0 . the lack of correlation between the line width and fe ii strength ( when measured independently with h@xmath0 ) in @xcite , @xcite and the present work has confirmed that the anti - correlation arises because h@xmath0 gets weaker as the lines get narrower @xcite . in addition to our confirmation of well documented correlations between different agn properties , we report on three newly discovered correlations involving the h@xmath0 redshift ( or blueshift ) , flux ratios of fe ii to h@xmath0 broad component ( fe ii / h@xmath34 ) and [ o iii ] to h@xmath0 narrow component ( [ o iii]/h@xmath21 ) . the correlations covering [ o iii]/h@xmath21 have never been investigated by other authors so far . we find the relative strength of fe ii to h@xmath34 ( fe ii / h@xmath34 ) significantly correlated to both the h@xmath0 redshift blueshift and flux ratio of fe ii to h@xmath0 whole line was detected in @xcite . we emphasize that our ratio refers to the broad component in h@xmath0 only . note the sign difference between this paper and @xcite . the positive correlation quoted by @xcite becomes negative in the notation used herein . ] ( @xmath55 , @xmath52 ) and the relative strength of [ o iii ] to h@xmath21 ( [ o iii]/h@xmath21 , @xmath56 , @xmath52 ) . a correlation test was also applied between h@xmath0 redshift and [ o iii]/h@xmath21 . the correlation is prominent in our sample ( @xmath57 , @xmath58 ) . figure 6 shows the correlations . there is a trend that strong fe ii / h@xmath34weak [ o iii]/h@xmath21 objects tend to have blueshifts in h@xmath0 while strong [ o iii]/h@xmath21 weak fe ii / h@xmath34 objects tend to have redshifts in h@xmath0 . these strong correlations between fe ii / h@xmath34 , h@xmath0 redshift and [ o iii]/h@xmath21 must reflect some physical connection between broad and narrow line emitting regions . one key to understand the central engines of agn lies in examining their local environment . the standard paradigm proposes that the surrounding gas is photoionized in physically distinct regions by radiation emerging from the central power source@xcite . however , the segregation of the ionized gas into broad - line region ( blr ) and narrow - line region ( nlr ) , as well as the details of the gas kinematics are still far from clear . optical spectroscopy of the ionized gas around galactic nuclei provides strong constraints on the excitation mechanisms . the following discussion addresses three key points : ( 1 ) how do the present results compare to previous work ; ( 2 ) which new correlations do we find and how can we understand them ; and ( 3 ) which future observations can we perform to distinguish between different suggested scenarios . a number of consistent correlations among observational parameters have been searched in order to understand the basic properties underlying the observed spectra . in particular , @xcite made a landmark study of the optical emission - line properties and continuum properties ( radio through x - ray ) of 87 low redshift pg quasars . a principal component analysis revealed `` eigenvector 1 '' ( ev1 ) links stronger fe ii emission , weaker [ o iii ] emission from the nlr , and narrower h@xmath0 ( blr ) with stronger line blue asymmetry . more recently , @xcite found these optical properties also go along with steeper soft x - ray spectra and claimed the soft x - ray slope as a related part of ev1 . we confirm these trends and present in addition three newly discovered correlations involving the h@xmath0 redshift , fe ii / h@xmath34 and [ o iii]/h@xmath21 . potentially , ev1 represents a fundamental physical driver that control the energy producing and radiation emitting processes . the two leading interpretations by far are that ev1 is driven by ( 1 ) the eddington ratio @xmath59 @xcite and ( 2 ) orientation effect . @xcite argued strongly that viewing angle is unlikely to drive ev1 , by assuming [ o iii ] emission is isotropic , originating from radii large enough to be free from orientation dependent obscuration . the isotropy of [ o iii ] emission has been questioned by the studies of radio - loud agn @xcite . however , a recent study of radio - quiet quasars @xcite showed a significant correlation between ev1 and orientation independent [ o ii ] emission , which implied that ev1 is not driven by orientation . moreover , the correlation between [ o ii ] and [ o iii ] emission indicated [ o iii ] emission is not dependent on orientation . the most promising interpretation at present is that ev1 is mostly governed by @xmath59 @xcite . this suggestion was based on the notion that the vertical thickness of the accretion disk , driven by the eddington ratio , controls the line strengths and continuum parameters . the striking correlations between the h@xmath0 redshift , fe ii / h@xmath34 and [ o iii]/h@xmath21 found in the present work give us linkages of the gas kinematics and ionization between blr and nlr . the question raised is what is the physical process behind the correlations ? can both the new and previous correlations be explained within one single scenario ? indeed , the @xmath59 interpretation provides plausible explanations for the observed trends . one of the two general scenarios in agn to explain asymmetric profile is in terms of an outflowing component in the blr with a change in obscuration by a central disk of clouds on the receding side . accreting systems in high @xmath59 agn will drive relatively thick outflows due to their larger photon luminosities per unit gravitational mass . as the velocity of the outflow increases , the h@xmath0 develops an increasing excess on the blue wing . the corresponding blue h@xmath34 velocity shift in comparison with [ o iii ] can be also well predicted because both the line asymmetry and the shift of the line centroids are effects of the same process . the link between blr and nlr in terms of density is less straight forward to predict . e.g. , due to the strong outflow , the nlr might be replaced by denser clouds . given a typical radial distribution of nlr clouds , [ o iii]/h@xmath21 peaks around densities of @xmath60 = 24 ( e.g. , komossa & schulz 1997 ) . high - density nlr clouds would therefore lead to a suppression of [ o iii ] strength . the same is true for the other limit , a low - density nlr . some authors favored a low - density blr of nls1 galaxies ( see , e.g. , the discussion in rodriguez - pascual et al . 1997 and komossa & mathur 2001 ) , which may indicate a low - density nlr as well , if both regions are coupled . the best way to measure directly the density of the nlr of nls1s in the future is to employ the density - sensitive line ratios [ o ii]@xmath183729/[o ii]@xmath183726 or [ s ii]@xmath186716/[s ii]@xmath186731 . these ratios are presently generally not reported in the literature . their measurements requires high s / n spectroscopy and/or resolution to detect and resolve these lines .- ratio measured by @xcite . this yields @xmath61 2.8 and does not deviate significantly from the values found for , e.g. , seyfert 1 and 2 galaxies . however , ngc 4051 is not an extreme nls1 in terms of x - ray steepness and fe ii emission . ] more indirectly , other line - ratios will also change in dependence of density and thus allow a density determination . for instance , a higher - density nlr would strongly boost the [ o i]@xmath186300 line @xcite . although it is more difficult to illustrate the direct connection between strong fe ii emission and large @xmath59 , some suggestions have been proposed . @xcite suggested the x - ray spectrum becomes steeper in high @xmath59 agn . if the euv x - ray spectrum plays an important role in the formation and confinement of the blr clouds ( e.g. , krolik , mckee & tarter 1981 ) , a steep x - ray spectrum coupled to a flatter euv spectrum will lead to a thicker blr ( e.g. , komossa & meerschweinchen 2000 ) , and thus stronger fe ii emission if the the low - excitation part of the blr is mechanically heated @xcite . increased metal abundances in more extreme nls1 galaxies @xcite would enhance these trends , i.e. lead to thicker blrs @xcite . supersolar iron abundance may additionally boost the strength of fe ii . in another model by @xcite , fe ii line emission is produced in an accretion disk . the agn with higher @xmath59 might simply have more mass in the accretion disk to produce stronger fe ii emission by postulating collisional excitation @xcite as the origin of the fe ii emission in general . the strong correlations of properties linking dynamics ( line widths ) , kinematics ( line shifts ) , ionization in blr and nlr ( fe ii and [ o iii ] strengths ) , and radiation process ( x - ray spectral shape ) are likely driven by an intrinsic properties , i.e. , the @xmath59 . in order to further test the favored scenario ( and get more constraints on , e.g. , the density of the nlr ) , larger and even more complete agn sample with higher resolution optical data is needed . this paper presents the statistical study of optical emission line properties of a bright agn sample selected from rass . the sample comprises 155 seyfert 1 type agn . our conclusions based on the correlation analysis can be succinctly summarized as follows : * the previously reported correlations between the strengths of fe ii , [ o iii ] emission and the x - ray slope are confirmed . strong optical fe ii blends go along with weak forbidden emission lines ( e.g. , [ o iii ] ) and steep x - ray slopes . * the anti - correlation between the x - ray spectral slope @xmath62 and the fwhm of h@xmath0 becomes prominent as far as the high luminosity agn are concerned . * striking correlations between the h@xmath0 redshift , fe ii / h@xmath34 and [ o iii]/h@xmath21 are found . there is a general trend that strong fe ii / h@xmath34weak [ o iii]/h@xmath21 objects tend to have blueshifts in h@xmath0 while strong [ o iii]/h@xmath21 weak fe ii / h@xmath34 objects tend to have redshifts in h@xmath0 . * we add new arguments that the observed trends are likely driven by @xmath59 and describe some future observations that could further test this suggestion . we thank drs . todd a. boroson and richard f. green for providing the fe ii template , drs . dirk grupe , simon vaughan , mira vron , philippe vron , tinggui wang , dieter engels for helpful discussions on the optical measurements , and the anonymous referee for many useful comments and suggestions . special thanks go also to the staff at xinglong station for their instrumental and observing help . d.x . acknowledges a mpg - cas exchange program . this research has made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , caltech , under contract with the national aeronautics and space administration . the rosat project is supported by the bundesministerium fr bildung und forschung ( bmbf ) and the max - planck - gesellschaft . this work was supported by nsfc under grant 19973014 . appenzeller , i. , zickgraf , f.j . , krautter , j. , & voges , w. 2000 , , 364 , 443 appenzeller , i. , thiering , i. , zickgraf , f .- j . , et al . 1998 , , 117 , 319 bade , n. , engels , d. , & voges , w. 1998 , , 127 , 145 bade , n. , fink , h. , engels , d. , et al . 1995 , , 110 , 469 baker , j.c . 1997 , , 286 , 23 boller , th . , brandt , w.n . , & fink , h. 1996 , , 305 , 53 boroson , t.a . , & green , r.f . 1992 , , 80 , 109 brinkmann , w. , & siebert , j. 1994 , , 285 , 812 brotherton , m.s . , wills , b.j . , francis , p.j . , & steidel , c.c . 1994 , , 430 , 495 corbin , m.r . 1995 , , 447 , 496 corbin , m. r. , & boroson , t. a. 1996 , , 107 , 69 corbin , m.r . 1997 , , 113 , 245 creshaw , d.m . 1986 , , 62 , 821 de robertis , m.m . 1985 , , 289 , 67 evans , i.n . 1988 , , 67 , 373 fiore , f. , elvis , m. , mcdowell j. c. , et al . 1994 , , 431 , 515 fischer , j .- u . , hasinger , g. , schwope , a.d . , et al . 1998 , astron . nachr . , 319 , 347 francis , p.j . , hewett , p.c . , foltz , c.b . , & chaffee , f.h . 1992 , , 398 , 476 gaskell , c. m. 1982 , , 263 , 79 gaskell , c.m . 1985 , , 291 , 112 gaskell , c.m . 2000 , new astronomy review , 44 , 563 goodrich , r.w . 1989a , , 340 , 190 goodrich , r.w . 1989b , , 342 , 224 grupe , d. , beuermann , k. , thomas , h .- c . , mannheim , k. , & fink h.h . 1998 , , 330 , 25 grupe , d. , beuermann , k. , mannheim , k. , & thomas , h .- c . 1999 , , 350 , 31 he s , r. , barthel , p.d . , & fosbury , r.a.e . 1993 , , 362 , 326 joly , m. 1987 , , 184 , 33 komossa , s. , grupe , d. , & janek , m. 2001 , in aip conf . 599 , x - ray astronomy 999 : stellar endpoints , agn and the diffuse background , ed . n. white , g. malaguti , & g. palumboi , 686 komossa , s. , & schulz , h. 1997 , , 323 , 31 komossa , s. , & mathur , s. 2001 , , 374 , 914 komossa , s. , & meerschweinchen , j. 2000 , , 354 , 411 krolik , j.h . , mckee , c.f . , tarter , c.b . 1981 , , 249 , 422 kuraszkiewicz , j. , wilkes , b.j . , brandt , w.n . , et al . 2000 , , 542 , 631 kwan , j. , cheng , f. , fang , l. , et al . 1995 , , 440 , 628 laor , a. , bahcall , j. n. , jannuzi , b.t . , schneider , d.p . , & green , r.f . 1995 , , 99 , 1 laor , a. , fiore , f. , martin , e. , et al . 1997 , , 477 , 93 lawrence , a. , elvis , m. , wilkes , b.j . , mchardy , i. & brandt , n. 1997 , , 285 , 879 massey , p. , & strobel , k. 1988 , , 328 , 315 mathur , s. 2000 , , 314 , l17 mcintosh , d.h . , rix , h .- w . , rieke , m.j . , & foltz , c.b . 1999 , , 517 , l73 miller , p. , rawlings , s. , saunders , r. , & eales , s. 1992 , , 254 , 93 netzer , h. , & wills , b.j . 1983 , , 275 , 445 oke , j.b . , & lauer , t.r . 1979 , , 230 , 360 osterbrock , d.e . 1977 , , 215 , 733 osterbrock , d.e . , & shuder , j.m . 1982 , , 49 , 149 osterbrock , d.e . , & pogge , r.w . 1985 , , 297 , 166 osterbrock , d.e . 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( university science books : mill valley , ca ) pounds , k.a . , done , c. , osborne , j.p . , 1995 , , 277 , l5 press , w.h . , teukolsky , s.a . , vetterling , w.t . , & flannery , b.p . 1992 , numerical recipes in fortran : the art of scientific computing ( 2nd ed . ; cambridge : university press ) phillips , m.m . 1978 , , 226 , 736 rodriguez - ardila , a. , pastoriza , m.g . , & donzelli , c.j . 2000a , , 126 , 63 rodriguez - ardila , a. , binette , l. , pastoriza , m.g . , & donzelli , c.j . 2000b , , 538 , 581 rodriguez - pascual p.m. , mas - hesse j.m . , santos - lleo m. , 1997 , , 327 , 72 schartel , n. , walter , r. , fink , h.h . , et al . 1996 , , 307 , 33 schwope , a.d . , hasinger , g. , lehmann , i. , et al . 2000 , astron . nachr . , 321 , 1 stripe , g.m . 1991 , , 247 , 3 thomas , h .- c . , beuermann , k. , reisch , k. , et al . 1998 , , 335 , 467 tytler , d. , & fan , x .- m . 1992 , , 79 , 1 vaughan , s. , edelson , r. , warwick , r.s . , et al . 2001 , , 327 , 673 veilleux , s. 1991 , , 369 , 331 vron - cetty , m.p . , vron , p. , & gonalves , a.c . 2001 , , 372 , 730 vron - cetty , m.p . , & vron , p. 2000 , , 10 , 81 voges , w. , boller , th . , dennerl , k. , et al . 1996 , in mpe report 263 , rontgenstrahlung from the universe , ed . zimmermann , j. trmper , & h. yorke , 637 voges , w. , et al . 1999 , , 349 , 389 wang , t. , brinkmann , w. , & bergeron , j. 1996 , , 309 , 81 wei , j.y . , xu , d.w . , dong , x.y . , & hu , j.y . 1999 , , 139 , 575 ( paper i ) wills , b.j . , brotherton , m.s . , fang , d. , steidel , c.c . , & sargent , w.l.w . 1993 , , 415 , 563 wills , b.j . , netzer , h. , & wills , d. 1985 , , 288 , 94 wills , b. 1982 , in iau symp . 97 , extragalactic radio sources , ed . heeschen & c.m . wade ( dordrecht : kluwer ) , 373 xu , d.w . , wei , j.y . , & hu , j.y . 1999 , , 517 , 622 zheng , x.z . , xia , x.y . , mao , s. , et al . 2002 , , 124 , 18 zickgraf , f .- j . , thiering , i. , krautter , j. , et al . 1997 , , 123 , 103 llrrrrrrrrrrrrrcc 1.0 mm 1rxs j000011.9 + 052318&0.040 & 3050 & 1050 & 600 & 37.3 & 26.7 & 22.5 & 0.6 & 1.8 & 0.7 & 40 & -50 & -90 & 1.2&2.08@xmath630.18 & 43.6 + 1rxs j000055.5 + 172346&0.215 & 6420 & 2290 & 830 & 75.5 & 64.9 & 8.5 & 0.1 & 0.9 & 0.9 & 640 & 150 & -490 & 0.6&1.73@xmath630.35 & 44.3 + 1rxs j000550.0 + 102249&0.095 & 4240 & - & 660 & 58.6 & 0.0 & 45.8 & 0.8 & - & 0.0 & 380 & 340 & -40 & 2.0&1.06@xmath630.47 & 43.4 + 1rxs j000915.0 + 243758&0.118 & 3880 & 1540 & 640 & 34.0 & 13.5 & 12.9 & 0.4 & 2.3 & 0.4 & -70 & -150 & -80 & 1.4&1.60@xmath630.43 & 44.0 + 1rxs j000953.0 + 371906&0.154 & 4240 & - & 1380 & 27.6 & 34.6 & 22.8 & 0.8 & - & 1.3 & - & 70 & - & - & 1.84@xmath630.46 & 44.3 + 1rxs j001121.2 + 373919&0.104 & 3770 & 1010 & 860 & 52.7 & 38.2 & 12.4 & 0.2 & 1.7 & 0.7 & 480 & 290 & -190 & - & 0.60@xmath630.49 & 43.5 + 1rxs j001452.7 + 270549&0.086 & 3350 & 870 & 760 & 28.7 & 25.5 & 24.4 & 0.8 & 2.3 & 0.9 & 230 & 200 & -30 & 1.1&1.69@xmath630.23 & 43.9 + 1rxs j002007.2 + 324423&0.082 & 3820 & 1380 & 950 & 38.1 & 78.7 & 3.6 & 0.1 & 0.3 & 2.1 & -230 & -260 & -30 & - & 1.62@xmath630.45 & 43.6 + 1rxs j002445.3 + 082047&0.067 & 1660 & 690 & 690 & 31.0 & 72.2 & 6.2 & 0.2 & 4.0 & 2.3 & -170 & -220 & -50 & - & 2.11@xmath630.50 & 43.3 + 1rxs j002811.6 + 310342&0.500 & - & - & 762 & - & - & 8.8 & - & - & - & - & - & - & 2.1&1.32@xmath630.25 & 45.5 + 1rxs j002902.7 + 195717&0.264 & 5720 & 740 & 720 & 71.0 & 24.7 & 14.4 & 0.2 & 6.2 & 0.3 & 720 & 260 & -460 & - & 0.66@xmath630.63 & 44.2 + 1rxs j003121.2 + 301558&0.200 & 3940 & 715 & 630 & 23.8 & 27.2 & 33.1 & 1.4 & 9.9 & 1.1 & -370 & -310 & 60 & - & 1.99@xmath630.52 & 44.4 + 1rxs j003221.1 + 242359&0.066 & 3460 & 1180 & 620 & 13.2 & 38.2 & 14.1 & 1.1 & 0.9 & 2.9 & -260 & -230 & 30 & - & 1.21@xmath630.33 & 43.1 + 1rxs j004459.4 + 192143&0.179 & 3900 & 1380 & 750 & 20.3 & 64.6 & 9.5 & 0.5 & 0.4 & 3.2 & -520 & -470 & 50 & 0.8&2.20@xmath630.24 & 44.6 + 1rxs j005034.2 + 044152&0.097 & 1780 & - & - & - & - & - & - & - & - & - & - & - & - & 1.88@xmath630.40 & 43.5 + 1rxs j005050.6 + 353645&0.058 & 5470 & 2300 & 1230 & 48.1 & 27.3 & 10.8 & 0.2 & 0.5 & 0.6 & 160 & 0 & -160 & 0.7&2.21@xmath630.15 & 44.0 + 1rxs j005607.6 + 254802&0.153 & 6600 & 1960 & 1230 & 73.3 & 61.9 & 13.8 & 0.2 & 0.9 & 0.8 & -760 & 90 & 850 & 1.1&1.71@xmath630.39 & 44.4 + 1rxs j011655.1 + 254937&0.099 & 4000 & 910 & 1070 & 56.4 & 40.1 & 43.1 & 0.8 & 3.0 & 0.7 & -50 & -110 & -60 & 0.0&- & - + 1rxs j011745.5 + 363718&0.107 & 3770 & 1720 & 700 & 68.3 & 51.8 & 17.2 & 0.3 & 0.6 & 0.8 & 40 & 230 & 190 & 0.7&1.03@xmath630.28 & 44.0 + 1rxs j011848.2 + 383626&0.216 & 5040 & 1590 & 1070 & 68.0 & 50.0 & 13.5 & 0.2 & 0.8 & 0.7 & -150 & 40 & 190 & - & 2.85@xmath630.43 & 44.8 + 1rxs j012556.6 + 351039&0.312 & 6790 & - & 570 & 78.3 & 21.0 & 12.6 & 0.2 & - & 0.3 & - & 0 & - & - & 0.86@xmath630.89 & 44.3 + 1rxs j012923.0 + 223100&0.231 & 4730 & 980 & 800 & 65.7 & 39.4 & 17.3 & 0.3 & 2.2 & 0.6 & 170 & 0 & -170 & - & 1.55@xmath630.71 & 44.4 + 1rxs j013738.5 + 852414&0.499 & 4270 & - & 710 & - & - & 8.0 & 0.2 & - & - & - & 180 & - & - & 0.80@xmath630.92 & 44.8 + 1rxs j014023.0 + 240255&0.072 & 4620 & 1300 & 1260 & 48.3 & 44.0 & 13.7 & 0.3 & 0.9 & 0.9 & -200 & -70 & 130 & 1.6&1.89@xmath630.37 & 44.0 + 1rxs j014025.2 + 382400&0.067 & 3110 & 800 & 620 & 72.2 & 38.2 & 22.6 & 0.3 & 4.0 & 0.5 & 380 & 290 & -90 & 1.7&2.02@xmath630.24 & 44.0 + 1rxs j015536.7 + 311525&0.135 & 3830 & 1750 & 910 & 24.8 & 42.6 & 9.3 & 0.4 & 0.6 & 1.7 & 60 & 270 & 200 & 0.5&2.34@xmath630.50 & 44.1 + 1rxs j015546.4 + 071902&0.069 & 3750 & 1090 & 800 & - & - & 26.7 & - & - & - & - & - & - & 0.0&1.22@xmath630.54 & 43.2 + 1rxs j015555.6 + 040620&0.136 & 6980 & 2030 & 750 & 51.9 & 63.1 & 9.9 & 0.2 & 1.1 & 1.2 & 0 & 0 & 0 & - & 1.15@xmath630.36 & 43.8 + 1rxs j015621.6 + 241838&0.155 & 3770 & 940 & 660 & 51.5 & 70.7 & 18.9 & 0.4 & 2.7 & 1.4 & -60 & -20 & 40 & - & 1.73@xmath630.85 & 44.3 + 1rxs j023601.9 + 082124&0.228 & 4390 & 1250 & 1230 & 82.1 & 74.2 & 21.2 & 0.3 & 1.7 & 0.9 & 250 & 80&-170 & - & 2.86@xmath630.62 & 45.5 + 1rxs j023739.9 + 181945&0.168 & 4560 & 950 & 850 & 87.4 & 58.1 & 15.8 & 0.2 & 5.1 & 0.7 & 160 & 30 & -130 & 0.8&- & - + 1rxs j024214.5 + 074442&0.451 & 4950 & 1410 & 700 & 59.2 & 0.0 & 17.2 & 0.3 & 0.9 & 0.0 & 60 & 0 & -60 & - & 1.71@xmath630.78 & 45.5 + 1rxs j033522.6 + 190739&0.189 & 6180 & - & 800 & 30.8 & 5.8 & 19.3 & 0.6 & - & 0.2 & - & 200 & - & 0.9&0.20@xmath630.45 & 44.2 + 1rxs j034308.0 + 185836&0.110 & 3420 & 1300 & 830 & 75.0 & 56.5 & 44.4 & 0.6 & 1.5 & 0.8 & 280 & 410 & 130 & 1.6&0.63@xmath630.62 & 43.8 + 1rxs j035147.6 + 103618&0.185 & 5100 & 1040 & 910 & 70.1 & 0.0 & 28.9 & 0.4 & 7.3 & 0.0 & 580 & 460 & -120 & 1.9&2.54@xmath630.66 & 45.2 + 1rxs j035345.9 + 195825&0.029 & 3800 & 1150 & 710 & 23.0 & 14.0 & 6.4 & 0.3 & 1.2 & 0.6 & 0 & 0 & 0 & - & 1.73@xmath630.72 & 43.2 + 1rxs j040513.2 + 173647&0.098 & 3850 & - & 920 & 67.9 & - & 9.0 & 0.1 & - & - & - & -310 & - & - & - & - + 1rxs j041147.1 + 132417&0.277 & 2960 & 1010 & 790 & 49.1 & 49.5 & 8.8 & 0.2 & 0.4 & 1.0 & -30 & 170 & 200 & - & 3.10@xmath631.07 & 45.8 + 1rxs j044428.6 + 122113&0.089 & 3140 & 1320 & 1280 & 33.8&106.3 & 30.5 & 0.9 & 1.4 & 3.1 & -490 & -250 & 240 & 1.2&1.33@xmath630.54 & 44.0 + 1rxs j062334.6 + 644542&0.086 & 4400 & 1550 & 990 & 53.0 & 0.0&117.3 & 2.2 & 3.6 & 0.0 & 680 & 490 & -190 & 2.0&0.87@xmath630.38 & 43.8 + 1rxs j065136.5 + 563732&0.286 & 4810 & 1320 & 670 & 52.0 & 55.5 & 13.8 & 0.3 & 2.0 & 1.1 & -630 & -680 & -50 & - & 1.53@xmath630.59 & 44.6 + 1rxs j065211.3 + 553436&0.064 & 2750 & 1010 & 620 & 25.8 & 40.3 & 39.4 & 1.5 & 3.1 & 1.6 & -510 & -400 & 110 & - & 0.47@xmath630.46 & 43.1 + 1rxs j071006.0 + 500243&0.154 & 4030 & 1600 & 620 & 43.7 & 60.4 & 7.3 & 0.2 & 0.2 & 1.4 & 40 & 250 & 210 & 0.0&0.91@xmath630.58 & 44.1 + 1rxs j071339.7 + 382043&0.123 & 4140 & 1200 & 700 & 58.0&102.5 & 17.0 & 0.3 & 0.7 & 1.8 & -490 & -340 & 150 & 0.5&1.33@xmath630.46 & 44.0 + 1rxs j071858.2 + 705920&0.066 & 4810 & 1310 & 850 & 20.3 & 57.7 & 38.2 & 1.9 & 1.6 & 2.8 & -280 & -250 & 30 & - & 1.48@xmath630.32 & 43.2 + 1rxs j071923.7 + 570840&0.175 & 6540 & 1150 & 960 & 102.7 & 0.0 & 81.4 & 0.8 & 3.2 & 0.0 & 670 & 560 & -110 & 0.2&1.01@xmath630.33 & 44.3 + 1rxs j072220.6 + 732041&0.350 & 3570 & 1320 & 820 & 59.7 & 76.4 & 10.6 & 0.2 & 0.7 & 1.3 & -150 & 300 & 450 & - & 2.71@xmath630.35 & 45.1 + 1rxs j072408.4 + 355309&0.138 & 7260 & 820 & 680 & 94.4 & 0.0 & 74.0 & 0.8&11.1 & 0.0 & -450 & -510 & -60 & 0.6&1.89@xmath630.18 & 44.7 + 1rxs j072826.6 + 303407&0.100 & 3980 & 1010 & 600 & 59.7 & 48.1 & 15.3 & 0.3 & 2.2 & 0.8 & 0 & 0 & 0 & 1.4&1.46@xmath630.53 & 43.7 + 1rxs j072912.9 + 410552&0.103 & 4460 & 1580 & 1040 & 47.6 & 41.6 & 29.4 & 0.6 & 3.7 & 0.9 & -230 & 190 & 420 & - & 1.28@xmath630.39 & 43.8 + 1rxs j073308.7 + 455511&0.142 & 8070 & 3240 & 930 & 58.1 & 48.9 & 16.7 & 0.3 & 0.3 & 0.8 & -440 & -190 & 250 & 0.1&1.28@xmath630.29 & 44.5 + 1rxs j073735.8 + 430856&0.119 & 4500 & 670 & 660 & 33.8 & - & 51.7 & 1.5&14.9 & - & 0 & 0 & 0 & - & - & - + 1rxs j074228.1 + 465648&0.170 & 4050 & - & 620 & 99.3 & 69.3 & 25.4 & 0.3 & - & 0.7 & - & 330 & - & 1.4&- & - + 1rxs j074311.3 + 742926&0.332 & 7380 & 1060 & 840 & 94.0 & 32.1 & 31.4 & 0.3 & 4.4 & 0.3 & -50 & -140 & -90 & - & 1.29@xmath630.26 & 44.9 + 1rxs j074759.2 + 205229&0.156 & 5540 & 820 & 690 & 71.7 & 48.7 & 12.4 & 0.2 & 3.5 & 0.7 & 130 & 70 & -60 & 1.2&0.97@xmath630.45 & 44.0 + 1rxs j074906.2 + 451039&0.192 & 4500 & 800 & 800 & 100.4 & 55.1 & 80.4 & 0.8 & 8.4 & 0.6 & 0 & 0 & 0 & 0.4&0.04@xmath630.84 & 43.6 + 1rxs j075026.6 + 270726&0.154 & - & - & - & - & - & - & - & - & - & - & - & - & - & 1.01@xmath630.38 & 43.7 + 1rxs j075039.5 + 624205&0.194 & 6300 & 1330 & 1500 & 58.5 & 18.0 & 22.3 & 0.4&11.0 & 0.3 & 20 & -90 & -110 & 0.9 & 1.69@xmath630.36 & 44.4 + 1rxs j075408.7 + 431611&0.350 & - & - & 570 & - & - & 31.5 & - & - & - & - & - & - & - & 1.38@xmath630.30 & 45.0 + 1rxs j075409.5 + 512436&0.331 & 3680 & - & 730 & 99.0&108.2 & 11.0 & 0.1 & - & 1.1 & - & 0 & - & - & -&- + 1rxs j075634.7 + 362232&0.884&-&- & - & - & - & - & - & - & - & - & - & - & - & 1.60@xmath630.42 & 45.8 + 1rxs j075910.0 + 115152&0.050 & 4470 & 1180 & 700 & 25.9 & 52.8 & 12.7 & 0.5 & 0.6 & 2.0 & -190 & -190 & 0 & 1.4 & 1.99@xmath630.32 & 43.3 + 1rxs j080132.3 + 473618&0.157 & 7840 & - & 900 & 119.0 & 69.7 & 13.1 & 0.1 & - & 0.6 & -530 & -190 & 340 & 0.4 & 1.29@xmath630.17 & 44.6 + 1rxs j080358.9 + 433248&0.451 & 5580 & 1340 & 670 & 75.6 & 0.0 & 38.4 & 0.5 & 1.3 & 0.0 & 390 & 570 & 180 & - & 2.31@xmath630.32 & 45.3 + 1rxs j080534.6 + 543132&0.406 & 5780 & 1530 & 730 & 51.5 & - & 30.1 & 0.6 & 1.5 & - & -370 & -250 & 120 & - & 2.01@xmath630.22 & 45.5 + 1rxs j080548.5 + 112617&0.144 & 4480 & 880 & 770 & 89.6 & 0.0 & 50.0 & 0.6 & 5.0 & 0.0 & 260 & 0 & -260 & 0.7 & 1.90@xmath630.22 & 44.2 + 1rxs j080639.9 + 724831&0.099 & 1760 & 740 & 820 & 13.7 & 34.4 & 61.9 & 4.5 & 6.2 & 2.5 & -190 & -220 & -30 & 0.7 & 1.25@xmath630.34 & 43.7 + 1rxs j080727.2 + 140534&0.074 & 2450 & - & - & 51.7 & 70.2&@xmath3220&@xmath320.4 & -&1.4 & - & 0 & - & - & 0.98@xmath630.45 & 43.3 + 1rxs j081237.9 + 435636&0.183 & 4110 & - & 700 & 51.5 & 21.6 & 11.8 & 0.2 & - & 0.4 & - & -90 & - & 0.4 & 1.14@xmath630.44 & 44.2 + 1rxs j081502.4 + 525304&0.126 & 4110 & 1080 & 770 & 75.8 & 41.2 & 14.7 & 0.2 & 1.7 & 0.5 & 360 & 230 & -130 & 0.2 & 1.62@xmath630.24 & 44.2 + 1rxs j081651.4 + 180252&0.158 & 3890 & 810 & 700 & 68.4 & 0.0 & 84.6 & 1.2 & 9.0 & 0.0 & 620 & 680 & 60 & 0.9 & 1.37@xmath630.20 & 44.5 + 1rxs j081718.6 + 520200&0.040&2170 & 470 & - & - & 25.2 & - & - & - & - & - & - & - & - & 1.09@xmath630.42 & 42.6 + 1rxs j081738.0 + 242333&0.283 & 5560 & 960 & 740 & 73.0 & 14.9&125.4 & 1.7 & 8.6 & 0.2 & 390 & 350 & -40 & - & 0.90@xmath630.74 & 44.3 + 1rxs j082633.1 + 074259&0.312 & 3460 & 1010 & 880 & 50.0 & 65.6 & 6.0 & 0.1 & 1.3 & 1.3 & -130 & 310 & 440 & - & - & - + 1rxs j082942.2 + 415442&0.126 & 7960 & 800 & 910 & 70.1 & 0.0 & 98.6 & 1.4&12.1 & 0.0 & 750 & 700 & -50 & 1.8 & 0.55@xmath630.39 & 43.5 + 1rxs j083737.3 + 254746&0.080 & - & - & 960 & 0.0 & - & 13.3 & - & - & - & - & - & - & - & - & - + 1rxs j083820.6 + 241025&0.043 & 4980 & 780 & 750 & 26.0 & 18.0 & 11.0 & 0.0 & 3.0 & 0.0 & -680 & -770 & -90 & - & 1.70@xmath630.29 & 43.2 + 1rxs j084026.2 + 033316&0.060 & 4940 & 1510 & 990 & 55.6 & 0.0 & 28.5 & 0.5 & 1.6 & 0.0 & 430 & 390 & -40 & 0.3 & 1.73@xmath630.21 & 43.6 + 1rxs j084302.0 + 030226&0.512 & 1920 & - & - & 53.1 & 43.3 & - & - & - & 0.8 & - & - & - & - & 2.25@xmath630.38 & 45.4 + 1rxs j084622.3 + 031318&0.107 & 4380 & 1000 & 960 & 71.6 & 49.7 & 24.2 & 0.3 & 4.1 & 0.7 & 0 & -150 & -150 & 0.9 & 1.81@xmath630.26 & 44.1 + 1rxs j084856.8 + 584150&0.134 & - & - & 890 & - & 0.0 & 23.3 & - & - & - & - & - & - & - & 1.29@xmath630.47 & 43.7 + 1rxs j084918.7 + 531806&0.112 & 2750 & 810 & 990 & 23.3 & 20.6 & 29.7 & 1.3 & 7.6 & 0.9 & -150 & -20 & 130 & 1.5 & 0.77@xmath630.48 & 43.2 + 1rxs j085233.3 + 422539&0.137 & 2670 & 750 & 700 & 14.8 & 43.3 & 22.8 & 1.5 & 2.7 & 2.9 & -250 & -220 & 30 & 1.1 & 1.83@xmath630.38 & 43.9 + 1rxs j085358.8 + 770054&0.106 & 5480 & 2300 & 620 & 23.1 & 40.8 & 11.7 & 0.5 & 0.8 & 1.8 & -640 & -540 & 100 & 1.6 & 1.34@xmath630.16 & 43.8 + 1rxs j085443.8 + 401912&0.152 & 5520 & 1100 & 810 & 60.6 & 37.1 & 19.7 & 0.3 & 1.9 & 0.6 & 460 & 440 & -20 & 1.8 & 0.80@xmath630.30 & 43.7 + 1rxs j085706.2 + 621050&0.160 & - & - & - & - & - & - & - & - & - & - & - & - & - & 0.08@xmath630.97 & 43.4 + 1rxs j085820.6 + 642100&0.115&3060 & 1260 & 970 & -&-&63.7 & -&- & - & - & -&- & - & 1.18@xmath630.40 & 43.6 + 1rxs j085902.0 + 484611&0.083 & 4420 & 1300 & 910 & 80.0 & 72.2 & 36.1 & 0.5 & 2.3 & 0.9 & -150 & -130 & 20 & 0.6 & 1.44@xmath630.11 & 44.1 + 1rxs j190910.3 + 665222&0.191 & 3830 & 600 & 670 & 97.0&111.7 & 7.0 & 0.1 & 1.1 & 1.2 & -290 & -40 & 250 & 0.9 & 2.55@xmath630.12 & 44.8 + 1rxs j193929.8 + 700752&0.116 & 4850 & 1690 & 1160 & 60.9 & 54.1 & 14.1 & 0.2 & 0.4 & 0.9 & -20 & 80 & 100 & 0.2 & 1.36@xmath630.15 & 44.4 + 1rxs j210550.5 + 041435&0.139 & - & - & - & - & - & - & - & - & - & - & - & - & - & 1.54@xmath630.54 & 43.9 + 1rxs j212654.3 + 112403&0.096 & 2730 & 1600 & 1120 & 66.7 & 33.3 & 14.1 & 0.2 & 0.8 & 0.5 & 780 & 350 & -430 & 0.7 & 1.60@xmath630.28 & 44.1 + 1rxs j214309.6 + 153658&0.385 & 7610 & - & 1230 & 72.7 & 0 & 10.9 & 0.1 & - & 0 & - & 120 & - & - & 1.00@xmath630.94 & 44.9 + 1rxs j214655.2 + 092102&0.984&- & - & - & - & - & - & - & - & - & - & - & - & - & 0.83@xmath630.61 & 45.6 + 1rxs j215912.7 + 095247&0.102 & 5380 & - & 1110 & 71.4 & 26.0 & 50.0 & 0.7 & - & 0.4 & - & 0 & - & 1.3 & 1.51@xmath630.25 & 44.2 + 1rxs j221537.8 + 290239&0.229 & 5980 & 600 & 640 & 70.0 & 0.0 & 54.5 & 0.8&11.8 & 0.0 & 150 & 220 & 70 & - & 1.02@xmath630.36 & 44.4 + 1rxs j221918.8 + 120757&0.082&2080 & - & - & - & - & - & - & - & - & - & -&- & - & 2.58@xmath630.22 & 44.4 + 1rxs j222537.5 + 225917&0.173 & 2150 & 1100 & 1100 & 52.5 & 39.4 & 17.8 & 0.3 & 4.7 & 0.7 & -40 & 0 & 40 & 0.6 & 1.50@xmath630.47 & 44.0 + 1rxs j222846.3 + 333500&0.090 & 3270 & 900 & 1260 & 44.4 & 24.1 & 46.1 & 1.0&11.0 & 0.5 & -70 & 0 & 70 & - & 1.65@xmath630.73 & 44.0 + 1rxs j222934.3 + 305720&0.319 & 7930 & 1390 & 1110 & 88.0 & 0.0 & 30.0 & 0.3 & 6.3 & 0.0 & 800 & 900 & 100 & - & 1.40@xmath630.43 & 44.8 + 1rxs j223019.5 + 163114&0.084 & 4270 & 1740 & 1310 & 55.8 & 34.9 & 17.2 & 0.3 & 1.2 & 0.6 & -20 & 80 & 100 & 0.8 & 1.39@xmath630.31 & 43.8 + 1rxs j224621.7 + 314206&0.145 & - & - & 660 & 0 & - & 25.3 & - & - & - & - & - & - & - & 1.48@xmath630.30 & 44.4 + 1rxs j224939.6 + 110016&0.084 & 4390 & 1440 & 1220 & 24.7 & 47.2 & 3.5 & 0.1 & 0.3 & 1.9 & -430 & -180 & 250 & 0.9 & 1.74@xmath630.25 & 44.0 + 1rxs j225148.5 + 341937&0.132 & 3050 & 1120 & 590 & 51.4 & 55.5 & 10.5 & 0.2 & 0.6 & 1.1 & 40 & 220 & 180 & 0.6 & 1.43@xmath630.22 & 44.5 + 1rxs j225209.6 + 264236&0.069 & 2160 & 700 & 700 & 16.7 & 39.4 & 15.5 & 0.9 & 1.3 & 2.4 & -200 & -200 & 0 & 1.3 & 1.75@xmath630.26 & 43.7 + 1rxs j225639.4 + 261846&0.176 & - & - & - & - & - & - & - & - & - & - & - & - & - & 0.87@xmath630.79 & 43.8 + 1rxs j230328.0 + 144341&0.080 & - & - & - & - & - & - & - & - & - & - & - & - & - & 0.20@xmath630.46 & 42.9 + 1rxs j231456.0 + 224325&0.171 & 3750 & 1230 & 1170 & 62.8&127.8 & 22.4 & 0.4 & 0.7 & 2.0 & -360 & -360 & 0 & 0.9 & 0.98@xmath630.35 & 44.1 + 1rxs j231517.5 + 182825&0.104 & 3310 & 1190 & 1200 & 25.0 & 18.3 & 10.7 & 0.4 & 6.4 & 0.7 & 0 & 0 & 0 & - & -&- + 1rxs j232554.6 + 215310&0.120 & 5260 & 950 & 770 & 125.2 & 0.0 & 62.5 & 0.5 & 5.5 & 0.0 & 390 & 330 & -60 & 1.1 & 1.14@xmath630.13 & 44.5 + 1rxs j232841.4 + 224853&0.129 & 5400 & - & 810 & 121.7 & 0.0 & 34.6 & 0.3 & - & 0.0 & - & 0 & - & 1.5 & 1.10@xmath630.54 & 43.7 + 1rxs j234728.8 + 242743&0.159 & 4270 & 2550 & 1350 & 105.1 & 0.0 & 75.2 & 0.7 & 2.2 & 0.0 & 980 & 770 & -210 & 1.7 & 0.64@xmath630.17 & 44.2 + 1rxs j234734.6 + 271910&0.646&- & - & - & - & - & - & - & - & - & - & - & - & - & -&- + 1rxs j235152.7 + 261938&0.040 & 3750 & 1850 & 1460 & 47.7 & 19.0 & 36.7 & 0.8 & 1.8 & 0.4 & -300 & -250 & 50 & 0.9 & 1.09@xmath630.21 & 43.3 + 1rxs j235754.3 + 132418&0.115 & 4290 & 2000 & 1480 & 102.3&101.9 & 35.5 & 0.3 & 0.9 & 1.0 & 40 & 380 & 340 & 1.5 & 1.67@xmath630.18 & 44.2 + iii zw 2 & 0.090 & 4990 & 700 & 630 & 72.1 & 33.3 & 47.1 & 0.7&10.0 & 0.5 & 140 & 0 & -140 & 0.5 & 0.68@xmath630.24 & 44.0 + 4c 25.01 & 0.284 & 4310 & 1050 & 840 & 74.6 & 79.1 & 38.8 & 0.5 & 8.2 & 1.1 & 140 & 140 & 0 & - & 1.51@xmath630.43 & 44.9 + pg 0026 + 12 & 0.142 & 4890 & 1120 & 690 & 36.1 & 19.2 & 14.8 & 0.4 & 1.5 & 0.5 & 0 & 360 & 360 & 0.9 & 0.98@xmath630.13 & 44.5 + s 10785 & 0.134 & 5970 & 1290 & 850 & 89.1 & 65.4 & 26.2 & 0.3 & 5.6 & 0.7 & 60 & 260 & 200 & 1.5 & 1.88@xmath630.21 & 44.6 + mark 1148 & 0.064 & 5560 & 840 & 800 & 135.3 & 22.8 & 55.0 & 0.4 & 8.4 & 0.2 & 190 & 290 & 100 & 1.2 & 0.52@xmath630.16 & 43.5 + pg 0052 + 251 & 0.155 & 5110 & 780 & 780 & 68.8 & 19.2 & 30.0 & 0.4 & 8.6 & 0.3 & 60 & 20 & -40 & 0.6 & 1.38@xmath630.18 & 44.9 + b2 0110 + 29 & 0.363 & 7360 & 680 & 750 & 77.6 & 0.0 & 63.1 & 0.8&15.0 & 0.0 & 490 & 520 & 30 & - & 1.51@xmath630.63 & 44.8 + 3c 48.0 & 0.367 & 4080 & 1840 & 1200 & 31.0 & 46.2 & 37.2 & 1.2 & 2.8 & 1.5 & -220 & 270 & 490 & 0.3 & 1.78@xmath630.11 & 45.9 + ms 02448 + 1928 & 0.176 & 4510 & 2000 & 780 & 70.0 & 59.5 & 23.9 & 0.3 & 1.1 & 0.9 & 780 & 720 & -60 & - & 1.93@xmath630.77 & 44.7 + ms 03574 + 1046 & 0.182 & 4160 & 1660 & 890 & 42.2 & 72.2 & 13.0 & 0.3 & 0.5 & 1.7 & -380 & -90 & 290 & 1.1 & 2.73@xmath631.10 & 45.3 + rx j04506 + 0642 & 0.118 & 3460 & 1150 & 940 & 22.5 & 45.2 & 11.0 & 0.5 & 0.6 & 2.0 & -460 & -360 & 100 & 0.7 & 1.36@xmath631.11 & 44.0 + ms 07007 + 6338 & 0.152 & 5070 & 2070 & 840 & 49.8 & 63.6 & 8.4 & 0.2 & 0.4 & 1.3 & 710 & 450 & -260 & 0.2 & 2.18@xmath630.33 & 44.4 + vii zw 118 & 0.079 & 4910 & 2300 & 860 & 58.1 & 71.4 & 14.3 & 0.2 & 0.5 & 1.2 & 490 & 680 & 190 & 0.5 & 1.95@xmath630.09 & 44.7 + f07144 + 4410 & 0.061 & 3820 & 1380 & 830 & 95.8 & 65.3&105.1 & 0.8 & 2.4 & 0.7 & -120 & -70 & 50 & 0.7 & 1.16@xmath630.21 & 44.1 + ms 07199 + 7100 & 0.125 & 4020 & 1330 & 680 & 54.8 & 50.9 & 26.3 & 0.5 & 7.2 & 0.9 & 800 & 230 & -570 & - & 1.72@xmath630.23 & 44.1 + rx j07491 + 2842 & 0.345 & 4710 & 820 & 620 & 84.3 & 57.7 & 12.3 & 0.1 & 4.8 & 0.7 & 250 & 140 & -100 & - & 2.10@xmath630.39 & 44.7 + rx j07527 + 2617 & 0.082 & 2940 & 1060 & 780 & 38.3 & 55.0 & 7.9 & 0.2 & 0.4 & 1.4 & -40 & -40 & 0 & - & 2.12@xmath630.23 & 44.0 + b3 0754 + 394 & 0.096 & 3840 & 1430 & 870 & 57.3 & 46.8 & 38.1 & 0.7 & 1.7 & 0.8 & -170 & -130 & 40 & 0.6 & 2.16@xmath630.33 & 43.9 + kuv 07549 + 4228&0.210 & 4620 & 1520 & 780 & 57.5 & 74.2 & 48.5 & 0.8 & 2.4 & 1.3 & 170 & 70 & -100 & 0.3 & 2.14@xmath630.22 & 44.9 + pg 0804 + 761 & 0.100 & 5490 & 2380 & 950 & 65.4 & 85.1 & 11.3 & 0.2 & 0.3 & 1.3 & -300 & -60 & 240&-0.2 & 1.30@xmath630.07 & 44.7 + rx j08297 + 3252 & 0.127 & 3420 & 820 & 710 & 33.7 & 30.9 & 44.2 & 1.3 & 3.2 & 0.9 & 150 & 150 & 0 & 1.3 & 2.13@xmath630.42 & 43.9 + rx j08307 + 3405 & 0.063 & 4500 & 1090 & 680 & 37.7 & 21.6 & 36.9 & 1.0 & 4.1 & 0.6 & 110 & 140 & 30 & 1.6 & 1.79@xmath630.26 & 43.6 + us 1329 & 0.249 & 5180 & 1280 & 760 & 112.4 & 48.2 & 55.8 & 0.5 & 2.7 & 0.4 & 170 & 120 & -50 & - & 1.69@xmath630.15 & 45.2 + vii zw 244 & 0.131 & 4510 & 2000 & 770 & 60.1 & 88.0 & 15.2 & 0.3 & 0.7 & 1.5 & -440 & -190 & 250 & 0.8 & 1.61@xmath630.17 & 44.2 + kuv 08482 + 4146 & 0.134 & 4130 & - & 740 & 62.2 & 39.4 & 11.4 & 0.2 & - & 0.7 & - & -520 & - & 1.2 & 1.47@xmath630.26 & 44.0 + mark 1220 & 0.064 & 4050 & 770 & 930 & 30.7 & 23.9 & 25.8 & 0.8 & 6.1 & 0.8 & -90 & 140 & 220 & - & 1.53@xmath630.31 & 43.5 + 4c 73.18 & 0.303 & 5670 & 1770 & 870 & 53.6 & 0.0 & 10.3 & 0.2 & 0.6 & 0.0 & -60 & 50 & 110 & - & 1.02@xmath630.16 & 45.1 + rx j20389 + 0303&0.137 & 4410 & 1410 & 750 & 65.5 & 67.7 & 31.3 & 0.3 & 0.7 & 1.0 & 40 & -100 & -140 & 1.3 & 1.81@xmath630.74 & 44.1 + s5 2116 + 81 & 0.086 & 4130 & - & 740 & 122.7 & 22.5 & 92.6 & 0.8 & - & 0.2 & - & 0 & - & 1.8 & 0.71@xmath630.14 & 44.0 + ms 21283 + 0349 & 0.094 & 4900 & 1330 & 660 & 105.4 & 50.9 & 54.8 & 0.5 & 2.5 & 0.5 & -370 & -310 & 60 & 0.4 & 2.14@xmath630.22 & 44.6 + ii zw 136 & 0.063 & 3770 & 980 & 760 & 81.4 & 72.5 & 17.4 & 0.2 & 0.9 & 0.9 & 250 & 120 & -130 & 0.9 & 2.46@xmath630.13 & 44.7 + ms 22102 + 1827 & 0.079 & 2950 & 870 & 900 & 27.3 & 68.9 & 6.2 & 0.1 & 0.3 & 2.5 & -120 & -120 & 0 & 1.3 & 2.49@xmath630.27 & 44.0 + rx j22186 + 0802 & 0.120 & 2530 & 780 & 810 & 22.5 & 36.1 & 23.7 & 1.1 & 3.5 & 1.6 & -140 & 90 & 220 & 1.6 & -&- + rx j22198 + 0935 & 0.060 & 2520 & 910 & 740 & 15.1 & 7.9 & 11.4 & 0.8 & 1.2 & 0.5 & -160 & -100 & 60 & - & 1.63@xmath630.45 & 43.1 + pg 2233 + 134 & 0.325 & 4750 & 1340 & 740 & 41.1 & 67.8 & 10.4 & 0.3 & 0.4 & 1.7 & -570 & -410 & 160 & - & 0.90@xmath630.62 & 44.3 + akn 564 & 0.025 & 3860 & 970 & 720 & 31.6 & 23.1 & 50.6 & 1.6 & 2.2 & 0.7 & -100 & -150 & -50 & 1.4 & 2.12@xmath630.04 & 44.5 + kuv 22497 + 1439 & 0.135 & 4140 & 1560 & 1040 & 47.5&110.1 & 4.2 & 0.1 & 0.2 & 2.3 & -470 & 0 & 470 & 0.6 & 2.04@xmath630.20 & 44.5 + rx j2256.6 + 0525 & 0.065 & 4240 & 1230 & 730 & 82.9 & 37.0 & 53.5 & 0.6 & 1.3 & 0.4 & -160 & -50 & 110 & 1.1 & 1.09@xmath630.20 & 43.8 + kaz 320 & 0.034 & 4430 & 1230 & 830 & 48.3 & 17.8 & 53.8 & 1.1 & 1.2 & 0.4 & 90 & 220 & 130 & 1.3 & 0.62@xmath630.12 & 43.1 + rx j23171 + 0838 & 0.165 & 5440 & 910 & 690 & 85.7 & 21.1 & 31.1 & 0.4&10.1 & 0.2 & 250 & 60 & -190 & 0.7 & 1.47@xmath630.37 & 44.2 + lllllllllllll ( 1)fwhm(h@xmath34)& & 0.002 & 0.032 & -0.32 & 0.35 & -0.59 & -0.05 & 0.34 & -0.20 & -0.04 & 0.38 & 0.48 + & & & & ( 0.0020)&(0.0009 ) & ( 0.0000 ) & & ( 0.001 ) & & & ( 0.0003)&(0.0000 ) + ( 2)fwhm(h@xmath21 ) & & & 0.46 & 0.30 & -0.263 & 0.23 & -0.294 & -0.15 & 0.23 & -0.104 & -0.66 & 0.11 + & & & ( 0.0000 ) & ( 0.004 ) & & & ( 0.005 ) & & & & ( 0.0000 ) & + ( 3)fwhm([oiii ] ) & & & & -0.03 & 0.004 & -0.02 & -0.08 & 0.02 & -0.06 & -0.166 & -0.15 & -0.01 + & & & & & & & & & & & & + ( 4)ew feii & & & & & -0.59 & 0.81 & -0.546 & -0.471 & 0.41 & -0.129 & -0.544 & 0.06 + & & & & & ( 0.0000 ) & ( 0.0000 ) & ( 0.0000 ) & ( 0.0000)&(0.0001 ) & & ( 0.0000 ) & + ( 5)ew [ oiii ] & & & & & & -0.64 & 0.73 & 0.33 & -0.39 & 0.195 & 0.67 & 0.05 + & & & & & & ( 0.0000 ) & ( 0.0000 ) & ( 0.0019)&(0.0002 ) & & ( 0.0000 ) & + ( 6)feii / h@xmath34 & & & & & & & -0.23 & -0.64 & 0.34 & -0.055 & -0.60 & 0.24 + & & & & & & & & ( 0.0000)&(0.0011 ) & & ( 0.0000 ) & + ( 7)[oiii]/h@xmath34 & & & & & & & & 0.009 & -0.33 & 0.27 & 0.45 & -0.314 + & & & & & & & & & ( 0.0017 ) & & ( 0.0000)&(0.0038 ) + ( 8)@xmath64 & & & & & & & & & -0.24 & 0.147 & 0.375 & 0.14 + & & & & & & & & & & & ( 0.0003 ) & + ( 9)@xmath65 & & & & & & & & & & -0.236 & -0.37 & 0.36 + & & & & & & & & & & & ( 0.0004)&(0.001 ) + ( 10)@xmath66 & & & & & & & & & & & 0.109 & -0.29 + & & & & & & & & & & & & ( 0.0069 ) + ( 11)[oiii]/h@xmath21 & & & & & & & & & & & & 0.1 + & & & & & & & & & & & & + ( 12)log @xmath67l@xmath68 & & & & & & & & & & & & +

the broad mechanism for radiation emission from a hot tenuous plasma is simple . thermal kinetic energy of free electrons in the plasma is transferred by collisions to the internal energy of impurity ions , @xmath0 where @xmath1 denotes an excited state and @xmath2 the ground state of the impurity ion . this energy is then radiated as spectrum line photons which escape from the plasma volume @xmath3 where @xmath4 is the emitted photon energy and @xmath5 its frequency . similarly ions in general increase or decrease their charge state by collisions with electrons @xmath6 where @xmath7 denotes the next ionisation stage of impurity ion @xmath2 . the situation is often referred to as the _ coronal picture_. the coronal picture has been the basis for the description of impurities in fusion plasmas for many years . however , the progress towards ignition of fusion plasmas and to higher density plasmas requires a description beyond the coronal approximation . models of finite density plasmas which include some parts of the competition between radiative and collisional processes are loosely called _ collisional - radiative_. however , collisional - radiative theory in its origins ( bates , 1962 ) was designed for the description of dynamic plasmas and this aspect is essential for the present situations of divertors , heavy species , transport barriers and transient events . the present work is centred on _ generalised collisional radiative _ ( @xmath8 ) theory ( mcwhirter and summers , 1984 ) which is developed in the following sections . it is shown that consideration of relaxation time - scales , metastable states , secondary collisions etc . - aspects rigorously specified in collisional - radiative theory - allow an atomic description suitable for modelling the newer areas above . the detailed quantitative description is complicated because of the need to evaluate individually the many controlling collisional and radiative processes , a task which is compounded by the variety of atoms and ions which participate . the focus is restricted to plasmas which are optically thin and not influenced by external radiation fields , and for which ground and metastable populations of ions dominate other excited ion populations . the paper provides an overview of key methods used to expedite this for light elements and draws illustrative results from the ions of carbon , oxygen and neon . the paper is intended as the first of a series of papers on the application of collisional - radiative modelling in more advanced plasma scenarios and to specific important species . the practical implementation of the methods described here is part of the adas ( atomic data and analysis structure ) project ( summers , 1993 , 2004 ) . illustrations are drawn from adas codes and the adas fundamental and derived databases . the lifetimes of the various states of atoms , ions and electrons in a plasma to radiative or collisional processes vary enormously . of particular concern for spectroscopic studies of dynamic finite density plasmas are those of translational states of free electrons , atoms and ions and internal excited states ( including states of ionisation ) of atoms and ions . these lifetimes determine the relaxation times of the various populations , the rank order of which , together with their values relative to observation times and plasma development times determines the modelling approach . the key lifetimes divide into two groups . the first is the _ intrinsic _ group , comprising purely atomic parameters , and includes metastable radiative decay , @xmath9 , ordinary excited state radiative decay @xmath10 and auto - ionising state decay ( radiative and auger ) , @xmath11 . the intrinsic group for a particular ion is generally ordered as @xmath12 with typical values @xmath13 where @xmath14 is the ion charge . the second is the _ extrinsic _ group , which depends on plasma conditions - especially particle density . it includes free particle thermalisation ( including electron - electron @xmath15 , ion - ion @xmath16 and ion - electron @xmath17 ) , charge - state change ( ionisation @xmath18 and recombination @xmath19 ) and redistribution of population amongst excited ion states ( @xmath20 ) . the extrinsic group is ordered in general as @xmath21 with approximate expressions for the time constants given by @xmath22(1/(z+1)^2)(kt_e / i_h)^{1/2}(\rm{cm}^{-3}/n_e)~~s \nonumber\\ \tau_{ion } & \sim & [ 10 ^ 5 - 10 ^ 7](z+1)^4(i_h / kt_e)^{1/2}e^{\chi / kt_e}(\rm{cm}^{-3}/n_e)~~s \nonumber\\ \tau_{i - i } & \sim & [ 7.0\times10 ^ 7](m_i / m_p)^{1/2 } \nonumber\\ & & ( kt_e / i_h)^{3/2}(1/z^4)(\rm{cm}^{-3}/n_i)~~s \\ \tau_{i - e } & \sim & [ 1.4\times10 ^ 9](m_i / m_p)^{1/2 } \nonumber\\ & & ( ( kt_e / i_h)+5.4\times10^{-4}(kt_i / i_h)(m_p / m_i))^{3/2 } \nonumber\\ & & ( 1/z^2)(\rm{cm}^{-3}/n_i)~~s \nonumber\\ \tau_{e - e } & \sim & [ 1.6\times10 ^ 6](kt_e / i_h)^{3/2}(\rm{cm}^{-3}/n_i)~~s . \nonumber \end{aligned}\ ] ] the ion mass is @xmath23 , the proton mass @xmath24 , the ionisation potential @xmath25 , the ion density @xmath26 , the electron density @xmath27 , the ion temperature @xmath28 , the electron temperature @xmath29 and the ionisation energy of hydrogen is @xmath30 . @xmath20 may span across the inequalities of equation [ eqn : eqn6 ] and is discussed in a later paragraph . from a dynamic point of view , the intrinsic and extrinsic groups are to be compared with each other and with timescales , @xmath31 , representing plasma ion diffusion across temperature or density scale lengths , relaxation times of transient phenomena and observation times . for most plasmas in magnetic confinement fusion and astrophysics @xmath32 where @xmath33 represents the relaxation time of ground state populations of ions ( a composite of @xmath19 and @xmath18 ) and it is such plasmas which are addressed in this paper . these time - scales imply that the dominant populations of impurities in the plasma are those of the ground and metastable states of the various ions . the dominant populations evolve on time - scales of the order of plasma diffusion time - scales and so should be modelled dynamically , that is in the time - dependent , spatially varying , particle number continuity equations , along with the momentum and energy equations of plasma transport theory . illustrative results are shown in figure [ fig : fig1]a . the excited populations of impurities and the free electrons on the other hand may be assumed relaxed with respect to the instantaneous dominant populations , that is they are in a _ quasi - equilibrium_. the quasi - equilibrium is determined by local conditions of electron temperature and electron density . so , the atomic modelling may be partially de - coupled from the impurity transport problem into local calculations which provide quasi - equilibrium excited ion populations and emissivities and then effective source coefficients ( collisional - radiative coefficients ) for dominant populations which must be entered into the plasma transport equations . the solution of the transport equations establishes the spatial and temporal behaviour of the dominant populations which may then be re - associated with the local emissivity calculations for matching to and analysis of observations . for excited populations , @xmath20 plays a special and complicated role due to the very large variation in collisional excitation / de - excitation reaction rates with the quantum numbers of the participating states . in the low density coronal picture @xmath34 and redistribution plays no part . critical densities occur for @xmath35 and for @xmath36 and allow division of the ( in principle ) infinite number of excited populations into categories including _ low levels _ , _ high singly excited levels _ and _ doubly excited levels _ for which important simplifications are possible . these are examined in section [ sec : sec2 ] . light element ions in fusion plasmas are generally in the singly excited state redistibutive case , approaching the doubly excited redistribution case at the higher densities . highly ionised ions of heavy species in fusion plasmas approach the coronal picture . illustrative results on critical densities are shown in figure [ fig : fig1]b . finally , because of the generally short @xmath15 compared with other timescales ( including those of free - free and free - bound emission ) , it is usually the case that the free electrons have close to a maxwellian distribution . this assumption is made throughout the present paper , but is relaxed in the next paper of the series ( bryans , 2005 ) . the basic model was established by bates ( 1962 ) . the ion in a plasma is viewed as composed of a complete set of levels indexed by @xmath37 and @xmath38 and a set of radiative and collisional couplings between them denoted by @xmath39 ( an element of the _ collisional - radiative matrix _ representing transition from @xmath38 to @xmath37 ) to which are added direct ionisations from each level of the ion to the next ionisation stage ( coefficient @xmath40 ) and direct recombinations to each level of the ion from the next ionisation stage ( coefficient @xmath41 ) . thus , for each level , there is a total loss rate coefficient for its population number density , @xmath26 , given by @xmath42 following the discussion in the introduction , it is noted that populated metastable states can exist and there is no real distinction between them and ground states . we use the term _ metastables _ to denote both ground and metastables states . metastables are the dominant populations and so only recombination events which start with a metastable as a collision partner matter . we condider the population structure of the @xmath14-times ionised ion , called the recombined or child ion . the @xmath43-times ionised ion is called the recombining or parent ion and the @xmath44-times ionised ion is called the grandchild . the metastables of the recombined ion are indexed by @xmath45 and @xmath46 , those of the recombining ion by @xmath47 and @xmath48 and those of the grandchild by @xmath49 and @xmath50 . therefore the ion of charge state @xmath14 has metastable populations @xmath51 , the recombining ion of charge @xmath43 has metastable populations @xmath52 and the grandchild ion of charge @xmath44 has metastable populations @xmath53 . we designate the remaining excited states of the @xmath14-times ionised ion , with the metastables separated , as _ ordinary _ levels for which we reserve the indices @xmath37 and @xmath38 and populations @xmath26 and @xmath54 . there are then , for example , direct recombination coefficients @xmath55 from each parent metastable into each child ordinary level and direct ionisation coefficients from each child ordinary level to each parent metastable @xmath56 such that @xmath57 . also there are direct ionisation coefficients @xmath58to the metastables of the child from the metastables of the grandchild . then the continuity equations for population number densities are @xmath59 = \left[\begin{array}{llll } \scriptnew{c}_{\mu \mu ' } & n_e~\scriptnew{r}_{\mu,\sigma } & 0 & 0 \\ n_e\scriptnew{s}_{\rho \mu ' } & c_{\rho \sigma } & c_{\rho j } & n_e~r_{\rho \nu ' } \\ 0 & c_{i \sigma } & c_{i j } & n_e~r_{i \nu ' } \\ 0 & n_es_{\nu \sigma } & n_es_{\nu j } & \scriptnew{c}_{\nu \nu ' } \end{array } \right ] \left[\begin{array}{l } n^-_{\mu ' } \\ n_{\sigma } \\ n_{j } \\ n^+_{\nu ' } \end{array } \right ] \nonumber \\\end{aligned}\ ] ] where the equations for the @xmath44-times and @xmath43-times ionised ions have been simplified by incorporating their ordinary population contributions in their metastable contributions ( shown as script capital symbols ) as the immediate focus is on the @xmath14-times ionised ion . this incorporation procedure is shown explicitly in the following equations for the @xmath14-times ionised ion through to equations [ eqn : eqn16 ] and may be done for each ionisation stage separately . note additionally the assumption ( made by omission of the ( 3,1 ) partition element , where 3 denotes the row and 1 the column ) that state - selective ionisation from the stage @xmath44 takes place only into the metastable manifold of the stage @xmath14 . from the quasi - static assumption , we set @xmath60 and then the matrix equation for the ordinary levels of the @xmath14-times ionised ion gives @xmath61 where we have used summation convention on repeated indices . substitution in equations [ eqn : eqn10 ] for the metastables of the @xmath14-times ionised ion gives @xmath62n^-_{\mu ' } \nonumber \\ & & + \left [ c_{\rho \sigma}-c_{\rho j}c^{-1}_{ji}c_{i \sigma}\right ] n_{\sigma } \nonumber \\ & & + n_e\left [ r_{\rho \nu'}- c_{\rho j}c^{-1}_{ji}r_{i \nu'}\right]n^+_{\nu'}. \end{aligned}\ ] ] the left - hand - side is interpreted as a total derivative with time - dependent and convective parts and the right - hand - side comprises the source terms . the terms in square brackets in equations [ eqn : eqn12 ] give the effective growth rates of each metastable population of the @xmath14-times ionised ion driven by excitation ( or de - excitation ) from other metastables of the @xmath14-times ionised ion , by ionisation to the @xmath43-times ionised ion and excitation to other metastables of the @xmath14-times ionised ion ( a negatively signed growth ) and by recombination from the metastables of the @xmath43-times ionised ion . these are called the @xmath8 coefficients . following burgess and summers ( 1969 ) , who used the name ` collisional - dielectronic ' for ` collisional - radiative ' when dielectronic recombination is active , we use the nomenclature @xmath63 for the @xmath8 recombination coefficients which become @xmath64 the @xmath8 metastable cross - coupling coefficients ( for @xmath65 ) are @xmath66 /n_e.\ ] ] note that the on - diagonal element @xmath67 /n_e$ ] with @xmath68 is a total loss rate coefficient from the metastable @xmath45 . substitution from equation [ eqn : eqn11 ] in the equations [ eqn : eqn10 ] for the metastables of the @xmath69-times ionised ion gives @xmath70 n_{\sigma } \nonumber \\ & & + \left [ c_{\nu \nu'}- n_e^2s_{\nu j}c^{-1}_{ji}r_{i \nu'}\right]n^+_{\nu'}. \end{aligned}\ ] ] the @xmath8 ionisation coefficients resolved by initial and final metastable state are @xmath71\ ] ] and note that there is contribution to cross - coupling between parents via recombination to excited states of the @xmath14-times ionised ion followed by re - ionisation to a different metastable , @xmath72.\ ] ] consider the sub - matrix comprising the ( 2,2 ) , ( 2,3 ) , ( 3,2 ) and ( 3,3 ) partitions of equations [ eqn : eqn10 ] . introduce the inverse of this sub - matrix as @xmath73 = \left[\begin{array}{ll } c_{\rho \sigma } & c_{\rho j } \\ c_{i \sigma } & c_{i j } \\ \end{array } \right]^{-1}\end{aligned}\ ] ] and note that the inverse of the ( 1,1 ) partition @xmath74^{-1 } \equiv \scriptnew{c}_{\rho \sigma}= \left [ c_{\rho \sigma}-c_{\rho j}c^{-1}_{ji}c_{i \sigma}\right].\ ] ] this compact representation illustrates , that the imposition of the quasi - static assumption leading to elimination of the ordinary level populations in favour of the metastable populations , may be viewed as a _ condensation _ in which the influence of the ordinary levels is _ projected _ onto the metastable levels . the metastables can be condensed in a similar manner onto the ground restoring the original ( ground states only ) collisional - radiative picture . the additive character of the direct metastable couplings @xmath75 means that these elements may be adjusted retrospectively after the main condensations . there are two kinds of derived coefficients associated with individual spectrum line emission in common use in fusion plasma diagnosis . these are _ photon emissivity coefficients _ ( @xmath76 ) and _ ionisation per photon ratio _ ( @xmath77 ) . the reciprocals of the latter are also known as _ photon efficiencies_. from equations [ eqn : eqn11 ] , the emissivity in the spectrum line @xmath78 may be written as @xmath79 this allows specification of the _ excitation _ photon emissivity coefficient @xmath80 and the _ recombination _ photon emissivity coefficient @xmath81 the ionisation per photon ratios are most meaningful for the excitation part of the emissivity and are @xmath82 each of these coefficients is associated with a particular metastable @xmath46 , @xmath83 or @xmath84 of the @xmath85 , @xmath86 or @xmath87 ions respectively . the radiated power in a similar manner separates into parts driven by excitation and by recombination as @xmath88 called the _ low - level line power coefficient _ and @xmath89 called the _ recombination - bremsstrahlung power coefficient _ where it is convenient to include bremsstrahlung with @xmath90 . note that in the generalised picture , additional power for the @xmath14-times ionised ions occurs in forbidden transitions between metastables as @xmath91 for the @xmath14-times ionised ion . in the fusion context , this is usually small . radiated power is the most relevant quantity for experimental detection . for modelling , it is the _ electron energy loss function _ which enters the fluid energy equation . the contribution to the total electron energy loss rate for the @xmath14-times ionised ion associated with ionisation and recombination from the @xmath43-times ionised ion is @xmath92 where the @xmath93 and @xmath94 are absolute energies of the metastables @xmath45 of the @xmath14-times ionised ion and of the metastables @xmath47 of the @xmath43-times ionised ion respectively . thus the electron energy loss is a derived quantity from the radiative power coefficients and the other generalised collisonal - radiative coefficients . note that cancellations in the summations cause the reduction to relative energies and that in ionisation equilibrium , the electron energy loss equals the radiative power loss . in the following sections , it is shown how the quasi - static assumption and population categorisation allow us to solve the infinite level population structure of each ion in a manageable and efficient way . such solution is necessary for low and medium density astrophysical and magnetic confinement fusion plasmas . this is unlike the simpler situation of very dense plasmas where heavy level truncation is used , because of continuum merging . the handling of metastables in a generalised collisonal - radiative framework requires a detailed specific classification of level structure compatible with both recombining and recombined ions . for light element ions , russell - saunders ( l - s ) coupling is appropriate and it is sufficient to consider only terms , since although fine structure energy separations may be required for high resolution spectroscopy , relative populations of levels of a term are close to statistical . so the parent ion metastables are of the form @xmath95 with @xmath96 the configuration and the recombined ion metastables are of the form @xmath97 with the configuration @xmath98 and the excited ( including highly excited ) terms are @xmath99 . @xmath100 and @xmath101 denote individual electron principal quantum number and orbital angular momentum , @xmath102 and @xmath103 denote total orbital and total spin angular momenta of the multi - electron ion respectively in the specification of an ion state in russell - saunders coupling . the configuration specifies the orbital occupancies of the ion state . for ions of heavy elements , relative populations of fine structure levels can differ markedly from statistical and it is necessary to work in intermediate coupling with parent metastables of the form @xmath104 and recombined metastables of the form @xmath105 with the configuration @xmath98 and the excited ( including highly excited ) levels @xmath106 . there is a problem . to cope with the very many principal quantum shells participating in the calculations of collisional - dielectronic coefficients at finite density necessitates a grosser viewpoint ( in which populations are _ bundled _ ) , whereas for modelling detailed spectral line emission , the finer viewpoint ( in which populations are fully _ resolved _ ) is required . in practice , each ion tends to have a limited set of low levels principally responsible for the dominant spectrum line power emission for which a bundled approach is too imprecise , that is , averaged energies , oscillator strengths and collision strengths do not provide a good representation . note also that key parent transitions for dielectronic recombination span a few ( generally the same ) low levels for which precise atomic data are necessary . in the recombined ion , parentage gives approximate quantum numbers , that is , levels of the same @xmath100 ( and @xmath101 ) divide into those based on different parents . lifetimes of levels of the same @xmath100 but different parents can vary strongly ( for example through secondary autoionisation ) . also the recombination population of such levels is generally from the parent with which they are classified . we therefore recognise three sets of non - exclusive levels of the recombined ion : * metastable levels - indexed by @xmath45 , @xmath46 . * low levels - indexed by @xmath37 , @xmath38 in a resolved coupling scheme , being the complete set of levels of a principal quantum shell range @xmath107 , including relevant metastables and spanning transitions contributing substantially to radiative power or of interest for specific observations . * bundled levels - segregated according to the parent metastable upon which they are built and possibly also by spin system - which can include _ bundle - nl _ and _ bundle - n_. viewed as a recombining ion , the set ( i ) must include relevant parents and set ( ii ) must span transitions which are dielectronic parent transitions . time dependence matters only for the populations of ( i ) , high precision matters only for groups ( i ) and ( ii ) and special very many level handling techniques matter only for group ( iii ) . to satisfy the various requirements and to allow linking of population sets at different resolutions , a series of manipulations on the collisional - radiative matrices are performed ( summers and hooper , 1983 ) . to illustrate this , suppose there is a single parent metastable state . consider the collisional - radiative matrix for the recombined ion and the right hand side in the bundle - n picture , and a partition of the populations as @xmath108 $ ] with @xmath107 and @xmath109 . elimination of the @xmath110 yields a set of equations for the @xmath111 . we call this a ` condensation ' of the whole set of populations onto the @xmath100 populations . the coefficients are the effective ionisation coefficients from the @xmath100 , the effective cross - coupling coefficients between the @xmath100 and the effective recombination coefficients into the @xmath100 , which now include direct parts and indirect parts through the levels @xmath112 . exclusion of the direct terms prior to the manipulations yields only the the indirect parts . call these @xmath113 and @xmath114 . we make the assumption that @xmath113 and @xmath114 may be expanded over the resolved low level set ( see section [ sec : sec2.2 ] ) to give the expanded indirect matrix @xmath115 and @xmath116 where @xmath37 and @xmath38 span the resolved low level set ( ij ) . these indirect couplings are then combined with higher precision direct couplings @xmath117 and @xmath118 so that @xmath119 and @xmath120 the procedure is shown schematically in figure [ fig : fig2 ] . the process may be continued , condensing the low level set onto the metastable set . the generalised collisional - dielectronic coefficients are the result . the time dependent and/or spatial non - equilibrium transport equations which describe the evolution of the ground and metastable populations of ions in a plasma use these generalised coefficients . following solution , the condensations can be reversed to recover the complete set of excited populations and hence any required spectral emission . the progressive condensation described above can be viewed as simply one of a number of possible paths which might be preferred because of special physical conditions or observations . for light element ions , four types of bundling and condensation are distinguished in this work : * ground parent , spin summed bundle - n @xmath121 lowest n - shell . * parent and spin separated bundle - n @xmath121 lowest spin system n - shell ( the _ bundle - ns population model _ ) . * low ls resolved @xmath121 metastable states ( the _ low - level population model _ ) . * parent and spin separated bundle - n @xmath121 low ls resolved @xmath121 metastable states . type ( a ) corresponds to the approach used in summers ( 1974 ) . type ( c ) corresponds to the usual population calculation for low levels in which ( consistent ) recombination and ionisation involving excited states are ignored . it establishes the dependence of each population on excitation for the various metastables only . type ( d ) , effectively the merging of ( b ) and ( c ) in the manner described earlier , is the principal procedure to be exploited in this work for first quality studies . details are in the following sub - sections . .[table : tab1]bundle-@xmath122 calculation pathways . the parent / spin system weight factor is defined in equation [ eqn : eqn32 ] and [ eqn : eqn33 ] . @xmath123 indicates the number of metastables of the recombined ion associated with the parent / spin system . the _ parent index _ , sequentially numbering the different parents shown in brackets , is used as the reference for tabulation of coefficients . [ cols="^,^,^,^,^,^ " , ] figure [ fig : fig5 ] illustrates the behaviour of low level populations . the graph is of the parameter @xmath124 from equation [ eqn : eqn20 ] . energy levels for high bundle-@xmath122 levels and their a - values , maxwell averaged collision strengths , radiative recombination coefficients , dielectronic recombination coefficients and ionisation coefficients are generated from a range of parametric formulae and approximations described in earlier works ( burgess and summers , 1976 ; summers , 1977 ; summers and hooper , 1983 ; burgess and summers , 1987 ; summers , 2004 ) . high quality specific data when available are acccessed from archives and substituted for the default values . this is a systematic procedure for dielectronic coefficients ( see sections [ sec : sec2.3.1 ] and [ sec : sec5 ] below ) . for low levels , a complete basis of intermediate coupling energy level , a - value and scaled born approximation collision data , spanning the principal quantum shell range @xmath125 is generated automatically using the cowan ( 1981 ) or autostructure ( badnell , 1986 ) procedures . this is called our _ baseline _ calculation . these data are merged with more restricted ( in level coverage ) but similarly organised highest quality data from archives where available ( e.g. r - matrix data such as ramsbottom ( 1995 ) ) . the data collection is compressed by appropriate summing and averaging to form a complete ls term basis and augmented with comprehensive high quality ls resolved dielectronic recombination , radiative recombination and collisional ionisation coefficients mapped from archives ( see sections [ sec : sec2.3.1 ] , [ sec : sec2.3.2 ] and [ sec : sec5 ] below ) . the radiative data has its origin in the work of burgess and summers ( 1987 ) . the final data collection is called a _ specific ion file _ ( adas data format _ adf04 _ ) . the detailed content is examined in section [ sec : sec3 ] . details of state selective dielectronic recombination and ionisation coefficients are given in the following sub - sections . state selective dielectronic recombination coefficients are required to all resolved low levels and to all bundle-@xmath122 shells for the various initial and intermediate state metastable parents @xmath47 for @xmath8 modelling . these are very extensive data and have been prepared for the present @xmath8 work through an associated international ` dr project ' summarised in badnell ( 2003 ) [ hereafter called _ dr - paper i _ ] and detailed in subsequent papers of the series . based on the independent particle , isolated resonance , distorted wave ( ipirdw ) approximation , the partial dielectronic recombination rate coefficient @xmath126 from an initial metastable state @xmath47 into a resolved final state @xmath37 of an ion @xmath127 is given by @xmath128 where @xmath129 is the statistical weight of the @xmath130-electron doubly - excited resonance state @xmath38 , @xmath131 is the statistical weight of the @xmath132-electron target state and the autoionization ( @xmath133 ) and radiative ( @xmath134 ) rates are in inverse seconds . the suffix @xmath135 is used here to denote a parent ion state . @xmath136 is the energy of the continuum electron ( with angular momentum @xmath101 ) , which is fixed by the position of the resonances . note that the parent states @xmath135 are excited , that is they exclude the metastable parents @xmath83 . the code autostructure ( badnell , 1986 ; badnell and pindzola , 1989 ; badnell , 1997 ) was used to calculate multi - configuration ls and intermediate coupling energy levels and rates within the ipirdw approximation . the code makes use both of non - relativistic and semi - relativistic wavefunctions ( pindzola and badnell , 1990 ) and is efficient and accurate for both the resolved low level and high-@xmath100 shell problems . lookup tables ( see section [ sec : sec5 ] ; _ adf09 _ ) are prepared comprising state selective recombination coefficients at a standard set of @xmath14-scaled temperatures , for each metastable parent , to all ls resolved terms of the recombined ion with valence electron up to @xmath100-shell @xmath137 ( normally @xmath138 ) and to bundle-@xmath122 levels of a representative set of @xmath100-shells ( usually spanning @xmath139 to 999 ) . these bundle-@xmath122 coefficients are simple sums over orbital states @xmath140 and so apply at zero density . this provides an extensive , but still economical , tabulation . in _ dr paper - i _ , we introduced an associated code , the ` burgess bethe general program ' bbgp . bbgp is used here as a support function in a model for the @xmath101-redistribution of doubly - excited states which provide a correction to the accurate , but unredistributed , dielectronic data . the redistributed data ( regenerated in the same _ adf09 _ format ) are normalised to ipirdw at zero density . the procedure is similar to that for singly excited systems . for @xmath141-averaged levels , the number densities expressed in terms of their deviations , @xmath142 from saha - boltzmann , and referred to the initial parent @xmath83 , are given by @xmath143^{3/2 } \frac{\omega_{p , nl}}{\omega_{\nu'}}e^{-e / kt_\rme}b_{p , nl}\ , . \label{eq22}\end{aligned}\ ] ] thus , in the bbgp zero - density limit , with only resonant capture from the @xmath47 parent balanced by auger breakup and radiative stabilization back to the same parent , we have @xmath144 in the extended bbgp model , we include resonant - capture from initial metastables other than the ground , dipole - allowed collisional redistribution between adjacent doubly - excited @xmath101-substates of the same @xmath100 by secondary ion- and electron - impact , and losses by ` alternate ' auger break - up and parent radiative transition pathways . the population equations for the @xmath101-substates of a doubly - excited @xmath100-shell become @xmath145 @xmath146 denotes resonance capture coefficients . @xmath147 denotes the number of parent metastables which are the starting point for resonance capture whereas @xmath148 denotes the number of true excited parent states . ion impact redistributive collisions are effective and are represented in the equations as @xmath149 ion contributions . the density corrected bundle-@xmath122 recombination coefficients are then @xmath150 details are given in _ dr paper - i_. figure [ fig : fig6 ] shows the effects of redistribution of the doubly excited @xmath142 factors in the recombination of @xmath151 . autoionisation rate calculations in ls coupling exclude breakup of non - ground metastable parent based ` singly excited ' systems by spin change . such breakup only occurs in intermediate coupling but can not be ignored for a correct population assessment even in light element ions . these rates are computed in separate intermediate coupling dielectronic calculations . these rates are main contributors to the parent metastable cross - coupling identified earlier ( cf . the @xmath152 $ ] intermediate states in figure [ fig : fig4]b leading back to the @xmath153 parent ) . direct ionisation coefficients for excited n - shells in bundle-@xmath122 population modelling are evaluated in the exchange - classical - impact parameter ( ecip ) approximation ( burgess and summers , 1976 ) . the method which merges a symmetrised classical binary encounter model with an impact parameter model for distant encounters is relatively simple and of moderate precision , but has a demonstrated consistency and reliability for excited n - shells in comparison with more eleborate methods . there is a higher precision requirement for state selective ionisation from metastable and low levels . such ionisation includes direct and excitation / autoionisation parts , but the latter only through true excited parents . stepwise ionisation is handled in the collisional - radiative population models . there are extensive ionisation cross - section measurements , but these are in general unresolved and so are of value principally for renormalising of resolved theoretical methods . the most powerful theoretical methods used for excitation , namely r - matrix with pseudostates ( rmps ) ( bartschat and bray , 1996 ; ballance , 2003 ) and convergent - close - coupling ( ccc ) ( bray and stelbovics , 1993 ) , have some capability for ionisation , but at this stage are limited to very few electron systems as is the time - dependent - close - coupling ( tdcc ) ( pindzola and robicheaux , 1996 ; colgan , 2003 ) ionisation method . for @xmath8 calculations , we have relied on the procedures of summers and hooper ( 1983 ) for initial and final state resolution of total ionisation rate coefficients and on the distorted wave approximation . the distorted wave method is our main method for extended @xmath8 studies for many elements and most stages of ionisation . rmps and tdcc studies are directed mostly at the neutral and near neutral systems where the distorted wave method is least reliable ( loch , 2005 ) . we use the configuration average distorted ( _ cadw _ ) wave approach of pindzola , griffin and bottcher ( 1986 ) . it has reasonable economy of computation while allowing access to complex , multi - electron ions , highly excited states , excitation / autoionisation and radiation damping . it expresses the configuration averaged excitation cross - section for @xmath154 as @xmath155 and the configuration averaged ionisation cross - section for @xmath156 as @xmath157 where @xmath158 are squared two - body coulomb matrix elements , @xmath159 denotes the average cross - section and @xmath160 , @xmath161 and @xmath162 denote average initial , final and ejected electron momenta respectively in the configuration average picture . the configuration average direct ionisation cross - sections and rate coefficients may be unbundled back to resolved form using angular factors obtained by sampson and zhang ( 1988 ) . note that the excitations described here are to auto - ionising levels and so resolved auger yields may be used which are the same as those in the dielectronic calculations of section [ sec : sec2.3.1 ] above . in fact the ionisation coefficient calculation results are structured and archived in adas ( format _ adf23 _ ) in a manner very similar to state selective dielectronic recombination . extensive studies have been carried out on light and heavy elements ( colgan , 2003 ; loch , 2003 ) . for bundle-@xmath122 modelling , the expected fundamental data precision is @xmath163 30% for excitation and ionising collisional rate coefficients , @xmath163 10% for a - values and state selective recombination coefficients and @xmath163 1% for energies . cccccccc ion & e & a & & @xmath164 & @xmath165 & @xmath103 + & . & & @xmath166 & @xmath167 & & & + @xmath168 & a & b & b & b , c & b & b & b + @xmath169 & a & b & b & b , c & b & b & b + @xmath170 & a & b & a , b & b , c & b & b & b + @xmath171 & a & a , b & b & b , c & b & b & b + @xmath172 & a & a & b & b , c & b & b & b + @xmath173 & a & a & b & b & b & b & b + @xmath174 & a & b , c & b & b & b & b & b + @xmath175 & a & b , c & b & b & b & b & b + @xmath176 & a & b , c & a , b & b & b & b & b + @xmath177 & a & b & a & a & b & b & b + @xmath178 & a & b & a & a & b & b & b + @xmath179 & a & b & b & b & b & b & b + @xmath180 & a & a & b & b , c & b & b & b + @xmath181 & a & a & b & b , c & b & b & b + @xmath182 & a & b , c & b & c & b & b & b + @xmath183 & a & c & b & b & b & b & b + @xmath184 & a & b & a & b & b & b & b + @xmath185 & a & b & a & b & b & b & b + @xmath186 & a & b & a & b & b & b & b + @xmath187 & a & b & a & b & b & b & b + @xmath188 & a & b & a & b & b & b & b + @xmath189 & a & b & b & b & b & b & b + @xmath190 & a & a & b & c & b & b & b + @xmath191 & a & a & b & c & b & b & b + for low level modelling many sources are used . a rating is given for the classes of fundamental data for the ions of carbon , oxygen and neon in table [ table : tab4 ] which is based on the following considerations . for energy levels , categories are _ a _ spectroscopic , _ b _ @xmath163 0.5% and _ c _ @xmath163 1.0% . c _ is anticipated from ab initio multi - configuration structure calculations , _ b _ from such calculations with extended optimising and _ a _ reflects direct inclusion of experimental energies from reference sources . for a - values , categories are _ a _ @xmath163 5% , _ b _ @xmath163 10% and _ c _ @xmath163 25% . c _ is anticipated from our baseline calculations , _ b _ from optimised multi - configuration structure calculations with extended optimising and _ a _ from specific studies in the literature . for electron impact maxwell averaged collision strengths , @xmath192 , the categories are _ a _ @xmath163 10% , _ b _ @xmath163 20% and _ c _ @xmath163 35% . c _ is from our baseline calculations , _ b _ from distorted wave calculations and _ a _ from specific r - matrix calculations , equivalent methods or experiment . for radiative recombination , the categories are _ a _ @xmath163 5% , _ b _ @xmath163 10% and _ c _ @xmath163 50% . c _ is from scaled methods using hydrogenic matrix elements , _ b _ from distorted wave one - electron wave functions in an optimised potential using spectroscopic energies and _ a _ from specific r - matrix calculations and experiment . b _ is the baseline in adas . for dielectronic recombination , the categories are _ a _ @xmath163 20% , _ b _ @xmath163 30% and _ c _ @xmath163 45% . c _ is from bbgp approximations , _ b _ from ls - coupled calculations using autostructure from the dr project and _ a _ from ic - coupled calculations using autostructure with parent and lowest resonance energy level adjustments from the dr project . b _ is the baseline in adas . it is to be noted that the theoretical relative precision which is consistent with the variation between the three categories is 15% better but dielectronic recombination comparisons with experiment still show unexplained discrepancies at the 20% level so the present categories are safe . for ionisation , the categories are _ a _ @xmath163 10% , _ b _ @xmath163 25% and _ c _ @xmath163 40% . c _ is from ecip approximations , _ b _ from configuration average with angular factor term resolution and _ a _ from rmps , tdcc calculations and experiment . b _ is the baseline in adas . as indicated in section [ sec : sec2.3 ] , these various data are assembled in an _ adf04 _ file which is sufficient to support the primary low - level population calculation . the tabulations are at a set of temperatures arising from a fixed set of z - scaled temperatures ( see section [ sec : sec2.1.1 ] ) which spans the full range to asymptotic regions of reaction data . collision data are converted to these ranges using _ c - plots _ ( burgess and tully ( 1992 ) and this procedure also flags data errors or queries . the precision of the specific ion file determines the achievable precision of all derived populations , emissivities and collisional - radiative coefficients ( whiteford , 2005 ; omullane , 2005 ) . the assembling of data in the _ adf04 _ is systematic and orderly . a comment section at the end of the file details the assembly steps , implementer , codes used and dates . this includes baseline and supplementation files , merging , ls compression , dielectronic recombination data inclusion etc . also there is extended detail of orginal data sources and a history of updates . a given _ adf04 _ file represents a ` snapshot ' of the state of available knowledge at the time . it is subject to periodic review and adas codes ( see section [ sec : sec5 ] ) are designed to enable easy reprocessing of all derived data following fundamental data update . the grading for each ion given in table [ table : tab4 ] is justified and supported by the comments from its _ adf04 _ file . a summary from @xmath193 is given in the following paragraphs in illustration . for @xmath188 , the low levels span 44 terms , including up to the @xmath194 shell built on the parent @xmath195 and @xmath196 shell built on the @xmath197 . only the @xmath195 parent is treated as a metastable from the @xmath8 point of view . the intermediate coupled baseline data set is _copmm@xmath19810-ic@xmath198ne6.dat_ with preferred supplementary energy a - value and @xmath192 data merged from _ copjl#be - ic@xmath198ne6.dat_. ionisation potentials and energy levels are from the nist standard reference database apart from @xmath199 ( kelly , 1987 ) and @xmath200 levels by ( ramsbottom , 1995 ) . the categorisation is _ a_. a - values were drawn from the opacity project ( opacity project team , 1996 ; tully , 1991 ) as justified ( for n iv and o v ) by wiese , fuhr and deters ( 1996 ) and supported / adjusted by @xmath163 3% by fleming ( 1996a , b ) , jnsson ( 1998 ) , froese fischer , godefeid and olsen ( 1997 ) , froese fischer , gaigalas and godefried(1997 ) , nussbaumer and storey ( 1979 ) , sampson , goett and clark ( 1984 ) and ramsbottom ( 1995 ) . b _ is safe . @xmath192s are taken from ramsbottom ( 1995 ) . these are for a 26 ls eigenstate multichannel r - matrix calculation . the categories assigned are _ a _ and _ b_. radiative and dielectronic recombination and ionisation all follow the baselines in adas , that is categories _ b _ although the summed and averaged ionisation rate coefficient ( over metastables ) is normalised to experiment . figure [ fig : fig7 ] shows @xmath201 coefficients for the c ii 858 spectrum line . the coefficients depend on both electron temperature and electron density in general . a common practice in spectral analysis is to observe principally the strongest resonance line of an ion . such emission is driven largely from the ground state ( figure [ fig : fig7]a ) and because of the large a - value , the density sensitivity occurs at relatively high density . thus such resonance line emission at moderate to low densities mostly reflects temperature and the distribution of the ionisation stages . comparison in near equilibrium ionisation balance plasmas of line ratios from the same ion , by contrast , is mostly directed at electron density and relies on the presence of metastables and spin changing collisional processes to confer the sensitivity . as shown earlier , the balance of the dominant ground and metastable populations is disturbed in dynamic plasmas and so density sensitivity may be modified by the dynamic state . the distinction of the metastable driven @xmath201 in @xmath8 modelling , as illustrated in figure [ fig : fig7]b , allows more complete diagnostic study and the possibility of separation of the two effects . in strongly recombining plasmas ( most commonly photo - ionised astrophysical plasmas ) the direct contribution of recombination to the emission may dominate the excitation part . the @xmath202 , as illustrated in figure [ fig : fig7]c , is then required . in the fusion context , multi - chordal spectral observations are important for the study of impurity transport especially near sources . visible and quartz uv observations are convenient and this places a requirement for @xmath201s from higher quantum shells . it is this primarily which defines the span of our _ low - levels _ for population modelling . emission from higher @xmath100-shells is significantly affected by cascading from yet higher levels . the full machinery of projection as described in section [ sec : sec2.1 ] is necessary for our global ambition of 20% precision for emissivity coefficients . figure [ fig : fig8 ] illustrates the @xmath8 recombination coefficients . at low electron density , radiative and dielectronic recombination dominate . for capture from metastables , alternate auger branching can largely suppress the dielectronic part of the surfaces . thus figure [ fig : fig8]a shows the characteristic exponential rise at the temperature for excitation of the main parent ( core ) transition of dielectronic recombination and then the subsequent fall - off ( as @xmath203 ) , in contrast with [ fig : fig8]c . at moderate densities , suppression of the high @xmath100-shell populations , principally populated by dielectronic recombination , through re - ionisation occurs and the coefficient falls in the dielectronic recombination region . at very high density , three - body recombination becomes effective , preferentially beginning at the lower electron temperatures . it is evident that recombination is less effective from the metastable at relevant ionisation balance temperatures . models which ignore the role of capture from the metastable parent ( which may be the dominant population ) can lead to substantial errors in recombination coefficients , while simply excluding all capture from the metastable can not deliver the precision sought for current modelling . for light elements in astrophysical plasmas , the zero - density coronal assumption for the recombination coefficient is still frequently made . this can not be justified even at solar coronal densities . figure [ fig : fig9 ] illustrates the @xmath8 ionisation coefficients . at low electron density , the coefficient is dominated by direct ionisation , including excitation / auto - ionisation , from the driving metastable . relatively high electron densities are required before the stepwise contribution begins . it is primarily the excitation to , and then further excitation and ionisation from , the first excited levels which controls this . the ground and metastable resolved coefficients both show the same broad behaviour as the usual ( stage to stage ) collisonal - radiative coefficient , tending to a finite limit at very high density . it is to be noted that the @xmath204 coefficients are required to be able to construct a meaningful stage to stage collisional - radiative ionisation coefficient from the generalised progenitors . it remains the case that most plasma modelling ( certainly in the fusion area ) is not adjusted to the use of the generalised coefficients as source terms . reconstruction of stage to stage source terms ( at the price of a reduction in modelling accuracy ) from the generalised coefficients is still is a requirement and is addressed more fully in section [ sec : sec5 ] . the generalised coefficients may be used to establish the equilibrium ionisation balance for an element in which the dominant ground and metastable populations are distinguished , that is the fractional abundances @xmath205}}{n^{[tot ] } } \right ) : \sigma=1,\cdots , m_z ; z=0 , \cdots , z_0\ ] ] where @xmath206 is the number of metastables for ionisation stage @xmath14 and @xmath207}=\sum_{z=0}^{z_0 } n^{[z]}=\sum_{z=0}^{z_0}\sum_{\sigma=0}^{m_z } n_{\sigma}^{[z]}\ ] ] in equilibrium . writing @xmath208}$ ] for the vector of populations @xmath209}$ ] , the equilibrium population fractions are obtained from solution of the matrix equations . @xmath210 } & n_e~\underline{\scriptnew{r}}^{[1 \rightarrow 0 ] } & 0 & . \\ n_e\underline{\scriptnew{s}}^{[0 \rightarrow 1 ] } & \underline{\scriptnew{c}}^{[1,1 ] } & n_e~\underline{\scriptnew{r}}^{[2 \rightarrow 1 ] } & . \\ 0 & n_e\underline{\scriptnew{s}}^{[1 \rightarrow 2 ] } & \underline{\scriptnew{c}}^{[2,2 ] } & . \\ 0 & 0 & n_e\underline{\scriptnew{s}}^{[2 \rightarrow 3 ] } & . \\ . & . \end{array } \right ] \left[\begin{array}{l } \underline{n}^{[0 ] } \\ \underline{n}^{[1 ] } \\ \underline{n}^{[2 ] } \\ \underline{n}^{[3 ] } \\ . \end{array } \right]_{equil}=0.\end{aligned}\ ] ] these in turn may be combined with the @xmath211 and @xmath90 to obtain the equilibrium radiated power loss function for the element as @xmath212 } & = & \sum_{z=0}^{z_0 } p^{[z]}\left ( \frac{n^{[z]}}{n^{[tot ] } } \right ) _ { equil } \nonumber \\ & = & \sum_{z=0}^{z_0}\sum_{\sigma=0}^{m_z } ( \scriptnew{plt}^{[z]}_{\sigma}+\scriptnew{prb}^{[z]}_{\sigma})\left ( \frac{n^{[z]}_{\sigma}}{n^{[tot]}}\right ) _ { equil}.\end{aligned}\ ] ] the equilibrium fractional abundances and equilibrium radiated power function are illustrated in figure [ fig : fig10 ] . it is useful at this point to draw attention to emission functions which combine emission coefficients with equilibrium fractional abundances . they are commonly used in differential emission measure analysis of the solar atmosphere ( lanzafame , 2002 ) where they are called @xmath213 functions . in solar astrophysics , it is assumed that the @xmath213 are functions of the single parameter @xmath29 ( either from a zero - density coronal approximation or by specification at fixed density or pressure ) and usually the abundance of hydrogen relative to electrons @xmath214 is incorporated in the definition . for finite density plasmas , in the generalised collisonal - radiative picture , we define @xmath215 functions , parameterised by @xmath29 and @xmath27 as @xmath216}_{j \rightarrow k } & = & \sum_{\sigma}\scriptnew{pec}^{[z](exc)}_{\sigma , j \rightarrow k}\left ( \frac{n_{\sigma}^{[z]}}{n^{[tot ] } } \right ) _ { equil } \nonumber \\ & & + \sum_{\nu'}\scriptnew{pec}^{[z](rec)}_{\nu',j \rightarrow k}\left ( \frac{n_{\nu'}^{[z+1]}}{n^{[tot ] } } \right ) _ { equil}. \end{aligned}\ ] ] these are strongly peaked functions in @xmath29 . the precision of the present modelling and data , including the full density dependence , is in principle sufficient to allow bivariate differential emission measure analysis ( judge , 1997 ) although this remains to be carried out . the general behaviour of a @xmath215 function is illustrated in figure [ fig : fig10]c . the organisation of the main calculations and data flow is shown in figure [ fig : fig11 ] . the calculations executed for the paper have been implemented in general purpose codes and attention has been given to the precise specification of all data sets and the machinery for accessing and manipulating them . this includes initial data , intermediate and driver data as well as the final @xmath8 products and follows adas project practice . population structure modelling requires an initial input dataset of energy levels , transition probabilities and collisional rates of adas data format _ adf04 _ which is complete for an appropriate designated set of low levels . for the light elements , best available data were assembled and verified as described in section [ sec : sec3 ] . these data are substitutes for more moderate quality , but complete , _ baseline _ data prepared automatically ( see _ gcr - paper iii _ ) . the _ adf04 _ files are required for ls - terms . in practice , we find it most suitable to prepare the data for lsj levels and then bundle back to terms . for @xmath8 modelling , state selective recombination and ionisation coefficients must be added to form the fully specified _ the ls - coupled dielectronic coefficients are mapped in from very large comprehensive tabulations of data format _ _ prepared as part of the _ dr project _ ( badnell , 2003 ) . the resulting _ adf04 _ files are the adas preferred data sets and are available for elements helium to neon . the two population codes , called adas204 and adas208 , work together . adas204 is the bundle-@xmath122 model . data on the metastable parent structure , quantum defects , auto - ionisation thresholds and autoionsation rates are required . these data may be extracted from the _ adf04 _ files and it has been helpful to make the driver dataset preparation automatic . high quality shell selective dielectronic data are essential and this is part of the provision in the _ adf09 _ files described in the previous paragraph . adas204 provides complete population solutions , but extracts from these solutions the condensed influence on the low n - shells , as _ projection matrices _ , for connection with the calculations of adas208 . it is to be noted that the main on - going development is refinement of atomic collision rates between the key low levels . the projection matrices are not subject to frequent change and so are suitable for long - term archiving ( _ adf17 _ ) . adas208 is the low - level resolved population model which delivers the final data for application . it draws its key data from the fully configured _ _ file and supplements these with projection data . the evaluation of the population structure takes place at an extended set of @xmath14-scaled electron temperatures and densities ( see section [ sec : sec2 ] ) and this means that the resulting @xmath8 coefficient data are suited to interpolation along iso - electronic sequences . thus the initial tabulation of @xmath8 coefficients is in iso - electronic datasets . it is convenient to implement the gathering and mapping from iso - electronic to iso - nuclear in a separate step , which also supports the merging back to the unresolved stage - to - stage picture if required . the production of @xmath217 and @xmath218 coefficients is directly to iso - nuclear oriented collections . the number of @xmath217s from the population calculations can , in principle , be very large . we restrict these by a threshold magnitude and to particular important spectral regions . it is straightforward to rerun adas208 to generate @xmath217 or @xmath218 coefficients alone in spectral intervals of one s choice . the separation of the adas204 and adas208 tasks and the ease of modifying data within an _ adf04 _ file means that ` what - if ' studies on the sensitivity of the derived data to fundamental data uncertainly can readily carried out . special adas codes enable detailed study of cumulative error and dominating sources of uncertainty such that error surfaces , for example for a @xmath217 as a function of electron temperature and density , may be generated . such error ( uncertainty ) analysis for theoretical derived coefficients and its utilisation in the confrontation with diagnostic experiments is the subject of a separate work ( see omullane , 2005 ) . the requirements for precise modelling of spectral emission and the relating of ionisation stages in thermal plasmas have been considered . collisional - radiative methodologies have been developed and extended to enable the full role of metastables to be realised , so that this generalised ( @xmath8 ) picture applies to most dynamically evolving plasmas occurring in magnetic confinement fusion and astrophysics . the procedures are valid up to high densities . the studies presented in the paper explore the density effects in detail within the @xmath8 picture and show that the density dependencies of excited ion populations and of effective rate coefficients can not be ignored . specific results are presented for light elements up to neon and the computations are carried out in an atomic basis of terms ( ls - coupled ) . such modelling will remain sufficient up to about the element argon , beyond which a level basis ( intermediate coupling ) becomes necessary . heavier elements will be examined in further papers of this series . considerable attention has been given to the generation and assembly of high quality fundamental data in support of the @xmath8 modelling . also datasets of fundamental and derived data have been specified precisely and codes have been organised following the principles of the adas project . the product of the study is the preferred adas data for the light element ions at this time . the atomic data and analysis structure , adas , was originally developed at jet joint undertaking . bates d r , kingston a e and mcwhirter r w p 1962 _ proc . soc . a _ * 267 * 297 .

what is the most broad - ranging currently active field of applied probability ? measuring breadth by the number of different academic disciplines where it appears , it seems hard to beat a field appearing in physics , computer science and electrical engineering , economics and finance , psychology and sociology , population genetics and epidemiology and ecology . curiously , the field has no good name , though readers of _ bernoulli _ will most likely know it , under the continuing influence of liggett s 1985 book @xcite , as _ interacting particle systems_. it turns out that mathematically similar toy models of different real - world entities have been repeatedly re - invented in the different disciplines mentioned above , and literally thousands of papers have been written since 2000 on discipline - dependent variant models intended as more realistic . rather than attempting a brief overview of the whole field as it stands today ( on which i am certainly not an expert ) , let me take an opposite approach , imagining starting with a blank slate . i will set out in the next section a particular conceptual setting what sorts of things are we trying to model ? then a particular technical setup , and then indicate what sort of mathematical results one might seek . given this manifesto , the rest of the article describes some basic models , gives pointers to the existing literature and presents a few open problems suggested by our setting . the article is based on a short course given in summer 2012 at warwick and cornell , and more extensive informal material is available at @xcite . i must emphasize that the article is intended as an introduction to the topics it mentions , not as an authoritative survey , which would require ten times the length ! in particular none of the mathematics outlined here is both new and non - straightforward , except perhaps some open problems . the particular setting we envisage is of the human social network . people meet or communicate , and explicitly or implicitly receive information from the other , and this interaction may change the `` state '' ( opinion or possessions or ) of the individuals . now that sentence skips over many issues . `` information '' is a complicated concept @xcite , and formalizing it as one of a set of prespecified states hardly captures the nuances of human experience . and obviously we receive information in many ways other than one - on - one interaction with other individuals . but the fact that humans do a lot of different things via interactions emphasizes the point that we are interested in a variety of different rules , a point that would be less visible if we took some real or hypothetical physical system as a prototype setting . social networks ( whose mathematical modeling and analysis long predates the facebook era ) are usually modeled as unweighted graphs vertices are individuals , edges indicate some specified relationship . in most contexts it would be more realistic to use _ weighted _ graphs , the weight indicating strength of relationship . this is seldom done because of increased model complexity and the fact that the meaning of `` strength of relationship '' is context - dependent . but we will give it the specific meaning of `` frequency of meeting '' , formalized via the rates @xmath0 below . the story above suggests the following technical framework , in which randomness arises primarily via assumed random times of meetings . because the framework can be used outside the social network setting , i will use the impersonal word `` agent '' in what follows . consider @xmath1 _ agents _ and a nonnegative array @xmath2 , indexed by unordered pairs @xmath3 , which is irreducible ( i.e. , the graph of edges corresponding to strictly positive entries is connected ) . assume * each unordered pair @xmath4 of agents with @xmath5 meets at the times of a rate-@xmath0 poisson process , independent for different pairs . call this collection of poisson processes the _ meeting process_. where needed we write @xmath6 for the set of agents . now introduce a set @xmath7 of states , and consider a process @xmath8 in which each agent @xmath9 is in some state @xmath10 at time @xmath11 , and the state can only change when agent @xmath9 meets some other agent @xmath12 , at which time the two agents states are updated according to some rule ( typically deterministic , but might be random ) which depends only on the pre - meeting states @xmath13 . so in the deterministic case , there is a given arbitrary function @xmath14 and when agents in states @xmath15 and @xmath16 meet their states are updated as @xmath17 we call such a process an _ fmie ( finite markov information exchange ) process _ and emphasize that a specific such process has two levels . the meeting process is determined by the meeting rate matrix @xmath2 ( which we loosely call the `` geometry '' , imagining that we tend to meet more frequently with individuals who are `` close '' in some sense ) . the information update process is determined by some rule . then the different named fmie models we shall describe ( such as the voter model ) correspond to different rules for the informational superstructure ; we regard such models as defined over arbitrary geometries . of course much of the beauty and utility of mathematics is in providing abstractions that encompass apparently different things , and as stated at the start fmie models have been used in many disciplines . let me briefly mention three that are relevant to our models later . conceptually closest is the mathematical modeling of epidemics , whose long history is indicated by the 1957 monograph @xcite . a typical classical model like the sir ( susceptible infected recovered ) model corresponds to a 3-state fmie if the relevant distributions are exponential . here of course the vertices are people or animals and the edges relate to physical proximity . classical models considered homogeneously mixing populations ( i.e. , the complete graph geometry ) but there is extensive post-2000 literature on more general geometries . a quite different context , from theoretical computer science , concerns distributed randomized algorithms . here agents are processors , communicating with nearby ( direct connections in a network ) processors , and performing computations based on data transferred . see @xcite for the particular topic of _ gossip algorithms_. what are called _ interacting particle systems ( ips ) _ arose historically as statistical physics models of phenomena such as magnetism , and mathematicians became interested in particular in rigorous study of phase transitions , in the setting of infinite geometries , specifically the infinite @xmath18-dimensional lattice . this literature long ago branched out to study many `` toy models '' not so directly related to any specific physical phenomenon . our development is in the same `` toy models '' spirit , and the models naturally suggested by the social networks context partly overlap and partly differ from those naturally suggested by statistical physics . in particular , the abstraction of what we are studying as `` information flow through networks '' seems more helpful than any physical visualization . a broader but less focused view of the whole `` statistical physics of social dynamics '' topic can be found in @xcite . the concept of an _ ips _ is generally discussed via example rather than explicit definition . for instance liggett @xcite takes as examples the voter model , contact process , exclusion process , glauber dynamics for the ising model , and the potlatch process . our mathematical setup of _ fmie process _ in section [ sec : tech ] is intended as a more precisely circumscribed class of processes to study . for instance , only the first three processes on liggett s list are fmie ( for the contact process , add an artificial agent with rate - one meetings with each other agent , to effect recovery ) . what we call the meeting process is often called the _ graphical representation _ in ips . our emphasis on a _ finite _ number of agents gives a somewhat different perspective than in classical ips , in which the finite setting was considered only occasionally , e.g. , @xcite . our viewpoint is by analogy with the modern literature on mixing and hitting times for finite markov chains @xcite ; study how the behavior of a model depends on the underlying geometry , where `` behavior '' means quantitative aspects of finite - time behavior . more precisely , we typically study @xmath19 limits in a family of finite - agent processes , rather than @xmath20 limits in a single infinite - agent process . the averaging process studied in section [ sec : ave ] provides a prototype for the style of results we would like to establish for a given fmie model ; see also the discussion of possible @xmath20 limits in section [ sec : scc ] . the fmie setup allows an arbitrary update function @xmath21 at ( [ def - f ] ) and an arbitrary `` geometry '' of meeting rates @xmath2 . one can hardly hope to say anything of substance in this generality ( we will mention various `` general principles '' but these are rather trite , once said ) . at a pragmatic level , if seeking analytic results one chooses an update function such that the fmie process can be analyzed on the complete graph ( `` mean - field '' , roughly ) geometry , otherwise it will likely be difficult to handle other geometries . and obversely , one chooses a geometry in which one can analyze the two basic processes of section [ sec:2basic ] , otherwise it will likely be difficult to handle more complicated update rules . from our viewpoint , much of the literature since 2000 consists of inventing and investigating a large number of increasingly elaborate update rules , while considering only a small number of `` standard geometries '' , indicated below . keep in mind that it is easy to invent and simulate models , but hard to give rigorous proofs or to relate convincingly to real - world data . here we briefly describe four of the `` standard geometries '' on which models are often studied . recall @xmath1 is the number of agents . note that in a realistic social network there is no reason why @xmath22 , which is the rate at which agent @xmath9 meets other agents , should be constant in @xmath9 . but many simple models do have this property , in which case we may rescale time to fit a _ standardization convention _ @xmath23 as in the first two examples below . here @xmath24 . here @xmath25 for adjacent agents @xmath26 . the next two examples involve networks modelled as _ random _ graphs . formally this adds another level of randomness , though informally we can just think of being given a typical realization . in such ( non - regular ) graphs one typically can not satisfy ( [ standardization ] ) and instead the default convention is to take rates @xmath27 for each edge @xmath28 . here one starts with the @xmath18-dimensional discrete torus ( or similar non - random graph ) and then adds random other edges in some specified way , for instance independently with probabilities proportional to @xmath29 . there are several ways ( in particular the _ configuration model _ @xcite ) to define graphs corresponding to the intuitive notion of being `` completely random '' subject to the constraint that , as @xmath19 , the distribution of vertex - degrees converges to a prescribed distribution . such models are tractable because their @xmath19 local weak limit is just a modified galton - watson tree . there is a huge `` complex networks '' literature ( see , e.g. , the monographs @xcite ) devoted to inventing and studying network models , and i will not write out a long list here . let me instead mention three other mathematically natural geometries which have _ not _ been studied as much as one might have expected ( and two more will be mentioned in section [ sec : infinite ] ) . start with the torus @xmath30 and add rates @xmath31 for non - adjacent @xmath4 . this is a way to interpolate between the torus and the complete graph . note the distinction between this and the `` small worlds '' model , in which the rate is non - vanishing for a few random edges . given a model whose behavior is understood on the two - dimensional lattice , one could investigate the effect of `` disorder '' by taking instead a proximity graph over a poisson process of points in the plane @xcite . this is a standard example in markov chain theory . here we discuss two fmie processes that are `` basic '' in two senses . they fit the definition of fmie process but are much simpler than ( and therefore unrepresentative of ) a typical fmie process ; yet many other fmie processes can be regarded as being built over one of these base processes , with extra structure added . for each fmie , we give only a verbal description , which the reader can readily formalize . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ there is one token . when the agent @xmath9 holding the token meets another agent @xmath12 , the token is passed to @xmath12 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this can evidently be viewed as a 2-state fmie process . the natural aspect to study is @xmath32 the agent holding the token at time @xmath11 . this @xmath33 is the ( continuous - time ) markov chain with transition rates @xmath2 . we call this the _ associated _ markov chain associated with the underlying geometry . let me digress to mention , by analogy , a nuance . a google scholar search on `` exact phrase @xmath34 galton - watson ; year @xmath34 19651969 '' returns only papers studying @xmath35 population size in generation @xmath36 of a galton - watson branching process . changing to `` year @xmath34 20052009 '' , half the papers talk about the galton - watson * tree * , which of course has more structure than the population size process one can study cousins , for instance . analogously , the token process has more structure than the associated markov chain one can study the time until agent @xmath9 meets someone who has previously held the token , for instance . for finite - state markov chains , we have available both a classical general theory and a long catalog of explicit calculations in specific geometries . for our purposes , the main point is that several fmie models ( classically the voter model and the exclusion process , but also the averaging process ) have close connections with the associated markov chain , and parts of the `` standard modern theory '' @xcite of finite chains can be used in their analysis . for readers more familiar with discrete - time chains , let me emphasize that ( because @xmath37 ) the stationary distribution of the associated chain is always uniform ; this does not assume the standardization ( [ standardization ] ) . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ initially one agent is infected . whenever an infected agent meets another agent , the other agent becomes infected . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in classical language , this is the si epidemic with @xmath38 distributions for infection times . but again there is a nuance . suppose that on each undirected edge @xmath28 of the geometry we put a single @xmath38 r.v . @xmath39 , independent over edges , and regard @xmath39 as the length of the edge . then there is an induced distance ( @xmath34 minimum route length ) between any pair of agents . this model of random distances is what i will call the _ first passage percolation ( fpp ) _ model . note this differs from classical lattice fpp , which assumes i.i.d . variables with arbitrary distribution . here is the nuance . in the pandemic process started with agent @xmath9 , write @xmath40 for the time when agent @xmath12 is infected . in the fpp process write @xmath41 for the distance between @xmath9 and @xmath12 . from the memoryless property of the exponential , we see that @xmath42 and indeed that @xmath43 for each @xmath9 . but pandemic and fpp are different processes in that @xmath44 whereas we have only @xmath45 . ( one can fix this particular issue by re - interpreting , in fpp , the time @xmath39 as the elapsed time , after one end is infected , until the other end is infected , but this misses the overall point that the distributions of the whole arrays @xmath46 are different . ) the pandemic model is `` basic '' in a very specific sense , in that it exhibits the fastest possible spread of information in any fmie model . that is , in an arbitrary fmie process @xmath8 over a given geometry , the value of @xmath47 can only be influenced by the value of @xmath48 if @xmath49 . here of course we are exploiting the natural coupling of two arbitrary fmie processes over the same geometry , which simply uses the same realization of the meeting process . albeit obvious , one can view this as an instance of the following . it is often useful to consider the natural coupling of two fmie processes . models of real - world epidemics are of course much more elaborate . regarding our simple pandemic model , it is analogous to the markov chain in two ways : * there is extensive literature doing calculations in specific geometries ; i show a little in section [ sec : pandemic ] . * several other fmies can be `` built over '' the pandemic model see section [ sec : built_over_p ] . but it differs it that there seems no existing `` general theory '' . see a conjecture in section [ sec : wlln ] . in these remarks , we are envisaging the finite - state case . time - asymptotics for fixed @xmath1 are not the issue ; in most models it s easy to see what s going on . informally there are four qualitatively different possible time - asymptotic behaviors , and each of the models we treat fits readily into one of these . * absorption in a random one ( of a small number of ) `` ordered '' configurations . [ pandemic , averaging , voter , deference ] . * absorption in a random one ( of a large number of ) `` disordered '' configurations . [ compulsive gambler ] . * * convergence to the unique stationary distribution ( perhaps on a subset of configurations ) . [ token , interchange , fashionista ] . the elided third item is the most general possibility for a finite - state markov chain ( applied to the configurations of a fmie process ) ; there are one or more _ strongly connected components _ ( sccs ) of the directed transition graph , and the limit distribution is a mixture over the stationary distributions on each scc . while it s easy to invent artificial fmie models with this behavior , i do nt know any `` natural '' such model . the following question is less easy that it first appears . the natural conjecture ( involving subsets of @xmath6 closed under @xmath21 , and functions whose sum is conserved by the interaction ) is false . consider the fmie process over the complete graph on @xmath1 agents ( for large @xmath1 ) defined by a deterministic update function @xmath21 as at ( [ def - f ] ) , what are the sccs of the fmie process ? as a prototype fmie model for which one can say something quantitative about its behavior over a general geometry we take the following . curiously , though it appears as an ingredient in more elaborate models , the only place i have seen the basic process itself is in the `` gossip algorithms '' literature @xcite which derives a version of the `` global bound '' , proposition [ pgcon ] . a careful treatment of the results below is written out in @xcite here i give a compressed version . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ initially each agent has some ( real - valued ) amount of money . whenever two agents meet , they divide their money equally . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is rather obvious that the individual values @xmath10 converge as @xmath50 to the average of initial values , so what is there to say ? write @xmath51 for the initial configuration @xmath52 , that is agent @xmath9 has unit money and other agents have none , and write @xmath53 for the transition probabilities of the associated markov chain . [ l1 ] for the averaging process with initial configuration @xmath51 we have @xmath54 . more generally , from any deterministic initial configuration @xmath55 , the expectations @xmath56 evolve exactly as the dynamical system @xmath57 where @xmath58 is the generator of the associated markov chain . the key point behind this results is that we can rephrase the dynamics of the averaging process as @xmath59 in informal language , this implies that the motion of a random penny which at a meeting of its owner agent is given to the other agent with probability @xmath60 is as the associated markov chain at half speed , that is with transition rates @xmath61 . to continue , we need to recall some background facts ( e.g. , @xcite , chap . 3 ) about reversible chains , here specialized to our continuous - time setting of the associated markov chain , that is with uniform stationary distribution ( @xmath37 ) . a function @xmath62 has ( with respect to the uniform distribution ) average @xmath63 , variance @xmath64 and @xmath65 norm @xmath66 defined by @xmath67 the @xmath65 norm will be used in several different ways . for a possible time-@xmath11 configuration @xmath68 of the averaging process , the quantity @xmath69 is a number , and so the quantity @xmath70 appearing in the proposition below is a random variable . the _ dirichlet form _ of the associated markov chain is @xmath71 where @xmath72 indicates summation over _ unordered _ pairs . the spectral gap of the chain , defined as the gap between eigenvalue @xmath73 and the second eigenvalue of @xmath58 , is characterized as @xmath74 the proof of the following result uses only the ingredients above and the averaging process dynamics . [ pgcon ] from an initial configuration @xmath75 with average zero , the time-@xmath11 configuration @xmath8 of the averaging process satisfies @xmath76 where @xmath77 is the spectral gap of the associated mc . to introduce the next result , thinking heuristically of the agents who agent @xmath9 most frequently meets as the `` local '' agents for @xmath9 , it is natural to guess that the configuration of the averaging process might become `` locally smooth '' faster than the `` global smoothness '' rate implied by proposition [ pgcon ] . in this context we may regard the dirichlet form @xmath78 as measuring the `` local smoothness '' , more accurately the local roughness , of a function @xmath79 , relative to the local structure of the particular meeting process . the next result implicitly bounds @xmath80 at finite times by giving an explicit bound for the integral over @xmath81 . note that , from the fact that the spectral gap is strictly positive , we can see directly that @xmath82 exponentially fast as @xmath20 ; proposition [ p - smooth ] is a complementary non - asymptotic result . [ p - smooth ] for the averaging process with arbitrary initial configuration @xmath83 , @xmath84 this looks slightly magical because the identity does not depend on the particular rate matrix @xmath58 , but of course the definition of @xmath85 involves @xmath58 . the next result illustrates the following . notions of _ duality _ are one of the interesting and useful tools in classical ips , and equally so in the social dynamics models we are studying . there is a formal definition of duality in @xcite , section 2.3 , but we find the concept easier to understand via examples . the duality between the voter model and coalescing chains ( recalled in section [ vmdual ] ) is the simplest and most striking example . the relationship we give in proposition [ p : adual ] for the averaging model is perhaps more representative of the general style of duality relationships . we repeat the `` random penny '' argument used in proposition [ l1 ] . now there are two pennies , and at any meeting there are independent decisions to hold or pass each penny . analyzing how the two pennies move leads to a certain coupling ( @xcite , sec . 2.4 ) of two ( half - speed ) associated markov chains @xmath86 the motion of the two pennies is coupled because , at a meeting of their two owners , one or both may be passed . the coupled process has the following relationship with the averaging process . [ p : adual ] for the averaging model @xmath8 started from a configuration @xmath83 which is a probability distribution over agents , and for each @xmath11 , @xmath87 where @xmath86 denotes the coupled process started from random agents @xmath88 chosen independently from @xmath83 . on a simple geometry where one can calculate @xmath89 via proposition [ l1 ] , one could seek to use proposition [ p : adual ] to investigate second - order properties of the averaging process , and some open problems of this type are mentioned in @xcite . finally , parallel to proposition [ pgcon ] which builds upon a standard @xmath65 bound for reversible markov chains is the following result @xcite , inspired by the theory surrounding log - sobolev inequalities @xcite . for a configuration @xmath90 which is a probability distribution write @xmath91 for the entropy of the configuration . consider the averaging process where the initial configuration is a probability distribution . by concavity of the function @xmath92 it is clear that in the averaging process @xmath93 can only increase , and hence @xmath94 a.s . ( recall @xmath95 is the entropy of the uniform distribution ) . [ pentropy ] for the averaging process whose initial configuration is a probability distribution @xmath83 , @xmath96 where @xmath97 is the log - sobolev constant of the associated markov chain . this model has a random update rule , most elegantly implemented via a `` directed '' convention in the meeting model : when agents @xmath4 meet , choose a ( uniform ) random direction and indicate it using an arrow @xmath98 or @xmath99 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ initially each agent has a different `` opinion '' agent @xmath9 has opinion @xmath9 . when @xmath9 and @xmath12 meet at time @xmath11 with direction @xmath98 , then agent @xmath12 adopts the current opinion of agent @xmath9 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this model and many variants have acquired a substantial literature in the finite setting ( as well as the classical infinite lattice setting ) so what s written here is far from complete . in the classical setting , one traditionally assumed only two different initial opinions . in our finite - agent case , it seems more natural to take the initial opinions to be all different . for notation , write @xmath100 so @xmath101 is a random partition of @xmath6 . note that @xmath102 may be empty , or may be non - empty but not contain @xmath9 . the number of different remaining opinions can only decrease with time . ultimately the process must be absorbed in one of the @xmath1 `` everyone has same opinion '' configurations . this fits one of our section [ sec : scc ] general possibilities for a fmie process : absorption in a random one ( of a small number of ) `` ordered '' configurations . a natural quantity of interest is this * consensus time * @xmath103 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ initially each agent has a token agent @xmath9 has token @xmath9 . at time @xmath11 each agent @xmath9 has a ( maybe empty ) collection ( cluster ) @xmath104 of tokens . when @xmath9 and @xmath12 meet at time @xmath11 with direction @xmath98 , then agent @xmath9 gives his tokens to agent @xmath12 ; that is , @xmath105 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ so @xmath106 is a random partition of @xmath6 . a natural quantity of interest is the * coalescence time * @xmath107 regarding each non - empty cluster as a particle , the particles move as the associated markov chain at half - speed ( rates @xmath61 ) , moving independently until two particles meet and thereby coalesce . note this factor @xmath60 in this section . for fixed @xmath11 , @xmath108 . in particular @xmath109 . it is important to realize that they are different as processes . for fixed @xmath9 , note that @xmath110 can only change by @xmath111 , but @xmath112 jumps to and from @xmath73 . the proof in the classical lattice setting @xcite ( which refers to coalescing _ random walks _ extends unchanged to our setting of general symmetric rates @xmath2 . one simply considers a realization of the ( directed ) meeting process on a fixed time interval @xmath113 $ ] , uses this to construct the coalescing chains process at time @xmath11 , then observes that reversing the direction of time and the arrows gives a construction of the voter model at time @xmath11 . almost all of the literature on the finite voter model has focussed on estimating @xmath109 in terms of the geometry , and i will show some of this work in sections [ sec : vmcg][sec : king2 ] . but there are other questions one can ask about the finite - time behavior , to be mentioned in section [ sec : pre - asy ] . there are two ways to analyze @xmath114 on the complete graph , both providing some bounds on other geometries . the first uses coalescing chains . one aspect of * kingman s coalescent * @xcite is the continuous - time mc on states @xmath115 with standardized rates @xmath116 . for that chain , the mean hitting time to state @xmath117 is @xmath118 and in particular @xmath119 . now when we consider coalescing chains on the complete @xmath1-graph , the number of clusters evolves as the continuous - time mc on states @xmath120 with rates @xmath121 . so @xmath122 and in particular @xmath123 the second way is to consider the voter model directly , specifically the variant of the voter model with only 2 opinions . consider the number @xmath124 of agents with the first opinion . on the complete @xmath1-graph , @xmath124 evolves as the continuous - time mc on states @xmath125 with rates @xmath126 this process arises in classical applied probability ( e.g. , as the moran model in population genetics ) . we want to study the absorption time @xmath127 by general birth - and - death formulas , or by comparison with simple rw , @xmath128 this is maximized by @xmath129 , and @xmath130 now we can couple the original voter model ( @xmath1 different initial opinions ) with the variant with only 2 opinions , initially held by @xmath131 and @xmath132 agents . ( just randomly assign these two opinions , initially . ) from this coupling , we see @xmath133 where the fraction is the chance that two specified agents are assigned different opinions , initially . in particular , the latter with @xmath129 implies @xmath134 comparing with the correct asymptotics ( [ voter - kn ] ) , we see that although constant is not precise , the order of magnitude is correct . consider a general geometry @xmath2 , and write @xmath135 . suppose the flow rates satisfy , for some constant @xmath136 , @xmath137 the right side is normalized so that on the complete graph this holds with @xmath138 . we can repeat the analysis above the process @xmath124 now moves at least @xmath139 times as fast as on the complete graph , and so @xmath140 this style of argument can be used for many fmie processes . bottleneck statistics give crude general bounds . often the speed at which the asymptotic behavior sets in depends on some quantification of the connectivity of the geometry . if for some @xmath141 the quantity @xmath142 is small , it indicates a possible `` bottleneck '' subset of size @xmath141 . it is convenient to combine the family @xmath143 into a single statistic , but the appropriate way to do this is ( fmie ) model - dependent . in the voter model case above , we used @xmath144 quantities like @xmath139 are descriptive statistics of a weighted graph , often given names like `` isoperimetric constants '' , but `` bottleneck statistics '' seems a more informative name . the best known use of bottleneck statistics is cheeger s inequality ( which bounds the spectral gap of a markov chain in terms of @xmath139 ) and its generalizations using other bottleneck statistics see @xcite for an detailed treatment . such techniques are used extensively in the theory of algorithms , where @xmath139 is called _ conductance _ see , e.g. , @xcite for analysis of a gossip algorithm . estimating @xmath145 is clearly related to study of the _ meeting time _ @xmath146 of two independent copies of the chain started at @xmath9 and @xmath12 , a topic that arises in other contexts . recall that on the complete graph the mean coalescence time is asymptotically twice the mean meeting time . also note that under enough symmetry ( e.g. , continuous - time random walk on the discrete torus ) the relative displacement between the two copies evolves as the same random walk run at twice the speed , and study of @xmath146 reduces to study of @xmath147 . all this suggests there is a close connection between the coalescence time and hitting times for the associated chain . the following result was conjectured long ago but only recently proved . for any chain , define a statistic @xmath148 [ t_oliveira ] there exist numerical constants @xmath149 such that , for any finite irreducible reversible markov chain , @xmath150 the proof involves intricate analysis that i will not try to outline here . if the stationary distribution @xmath151 contains exponentially rare states then the theorem is uninformative . but in our setting of fmie processes , the associated chain always has _ uniform _ stationary distribution . one can create examples of geometries in which the bound is not the correct order of magnitude , but i suspect that in any `` natural '' geometry theorem [ t_oliveira ] does give the correct order . a different style of general result given recently in @xcite shows that on a regular graph @xmath152 where @xmath77 is the spectral gap of the associated mc . but this bound is of correct order only for expander - like graphs . return to the setting of section [ vmdual ] chains are associated chains run at half speed . write @xmath153 for the mean meeting time from independent uniform starts . in a sequence of chains with @xmath19 , impose a condition such as the following . for each @xmath154 @xmath155 assuming ( [ vmau ] ) , under what further conditions can one prove @xmath156 that such a result is anticipated has been folklore since 1989 , when cox @xcite proved ( [ opkingman ] ) for the torus @xmath157^d$ ] in dimension @xmath158 . ( in that geometry , @xmath159 for @xmath160 , where @xmath161 is the mean number of visits of standard random walk on @xmath162 to the origin . ) see @xcite for a physicist s view . one could try to prove this in two stages . * stage 1 . * for fixed @xmath141 , show that the mean time for @xmath141 initially independent uniform walkers to coalesce should be @xmath163 . * stage 2 . * show that for @xmath164 slowly , the time for the initial @xmath1 walkers to coalesce into @xmath165 clusters is @xmath166 . it is likely that stage 1 can be proved without further assumptions , as follows . from old results on mixing times @xcite , condition ( [ vmau ] ) is enough to show that the variation distance mixing time @xmath167 . so as a prototype use of @xmath168 by considering time intervals of length @xmath169 , for @xmath170 , the events `` a particular pair of walkers meets in the next @xmath169-interval '' are approximately independent . this makes the `` number of clusters '' process behave ( to first order ) as the kingman coalescent as the number decreases from @xmath141 to @xmath117 . this approach is implemented in @xcite for @xmath171 , but it should hold for general geometries . on the other hand , stage 2 requires some different assumptions to control short - time behavior . see @xcite for a recent extensive study of this problem , and the current best results . as mentioned earlier , almost all of the literature on the finite voter model has focussed on estimating @xmath172 in terms of the geometry . but there are other aspects of finite - time behavior one might study , and we can uncover some such aspects by comparison with what we have studied in the averaging process . recall we start with all opinions different . if the proportions of agents with the various opinions are written as @xmath173 , the statistic @xmath174 is one measure of concentration versus diversity of opinion . this suggests studying @xmath175 . duality implies @xmath176 where @xmath177 is the meeting time for independent chains with uniform starts . one can study the distribution of @xmath177 in special geometries , and more generally can often apply the explicit exponential distribution approximation from @xcite , prop . * 2 . * a corresponding `` local '' measure of concentration versus diversity is the probability that agents @xmath178 chosen with probability @xmath179 ( `` random neighbors '' ) have same opinion at time @xmath11 . refinements of this `` clustering '' behavior have been studied in the classical two - dimensional lattice setting @xcite . * the diversity statistic @xmath174 emphasizes large clusters ( large time ) ; the statistic @xmath180 emphasizes small clusters ( small time ) . so one could consider @xmath181 this has apparently not been studied . certain special families of geometries have @xmath19 local weak limits ( which are infinite rooted random networks ) , and aspects of fmie models on the finite geometries can often be related to the behavior of the model on the infinite geometry.=1 two familiar examples of such geometries are for the torus @xmath30 , the @xmath182 limit is the infinite lattice @xmath162 ; for the `` random graphs with prescribed degree distribution '' geometry , the limit is a galton - watson tree with a size - biased offspring distribution after the first generation ; and two less familiar examples are for the complete graph with random edge - weights , the limit is the pwit @xcite , an infinite - degree tree where the weights on the edges at a given vertex form a rate-@xmath117 poisson process on @xmath183 ; a more elaborate graph @xcite built over ( iii ) , designed to have a @xmath19 local weak limit which is more complex ( not just a tree ) yet still fairly tractable . the relation between finite and infinite geometries is simplest for the `` epidemic '' ( fpp ) type models later , but also can be used for markov chain related models , starting from the following . for a large finite - state markov chain whose behavior near a state @xmath9 can be approximated be a transient infinite - state chain , we have @xmath184 where @xmath185 is defined in terms of the approximating infinite - state chain as @xmath186 . the approximation ( an instance of the poisson clumping heristic @xcite ) comes from the finite - state mean hitting time formula ( @xcite , lemma 2.11 ) via a `` interchange of limits '' procedure which requires ad hoc justification . in the context of the voter model , if there is a local weak limit of the finitie geometries , the local transience principle implies that the property @xmath187 is essentially equivalent to transience of the chain on the limit infinite geometry . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ initially each agent has some ( nonnegative real - valued ) amount of money . whenever two agents meet , they instantly play a fair game in which one agent acquires the combined money . in other words , if one has @xmath188 and the other has @xmath189 then the first acquires all @xmath190 with chance @xmath191 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ note that on the complete graph geometry , this process is just an augmentation of the kingman coalescent process . on a general geometry , a configuration in which the set of agents with non - zero money forms an `` independent set '' ( no two are adjacent in the weighted graph ) is obviously an absorbing configuration , and conversely . this process has apparently not been studied . i mention it because it provides a simple example of the `` disordered '' limit behavior mentioned in section [ sec : scc ] . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ initially each agent has a different token . when two agents meet , they exchange tokens . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ i will call this the interchange process ; the `` two type '' version is the classical exclusion process . though this model is not particularly natural in the social networks context , the result below ( which established a longstanding conjecture ) illustrates a result for which our general `` weighted graphs '' setting is the ( mathematically ) correct setting . the motion of an individual token is as the associated markov chain , so has some spectral gap @xmath192 . the whole interchange process is itself a reversible markov chain , so has a spectral gap @xmath193 which by a general `` contraction '' fact ( @xcite , sec . 4.6.1 ) satisfies @xmath194 . for every geometry we have @xmath196 . the proof involves intricate analysis that i will not try to outline here . recall the discussion in section [ sec : pan ] . for the purposes of this section , the pandemic process is the same as the fpp process ; we will use the language of `` infectives '' and `` time '' ( rather than `` percolation distance '' ) and write @xmath197 for the time taken for an epidemic started at @xmath9 to reach @xmath12 . in this section , we will recall the well - understood behaviors on the lattice ( section [ sec : fpp_torus ] ) and the complete graph ( section [ sec : fpp_complete ] ) . each type of behavior raises questions about the extent to which similar behavior remains true in more general geometries ( sections [ sec : wlln ] and [ sec : fpp_other ] ) which have not been systematically studied . we quote the classical result for fpp on the infinite lattice , with brief remarks . consider the fpp process on the edges of @xmath162 with @xmath198 edge - lengths , starting from the origin . write @xmath199 for the infected set at time @xmath11 , and @xmath200 for its fattening ( replace each point by a cube of side @xmath117 ) . there is a non - random closed convex set @xmath201 such that , for each @xmath202 , @xmath203 * the key proof ingredient is the _ subadditive ergodic theorem_. * the result hold for more general iid edge - times . * @xmath204 is _ not _ the unit ball in euclidean norm . * for later use write @xmath205 for the ( @xmath18-dimensional ) `` volume '' of @xmath204 . * understanding the asymptotic variance of @xmath41 is a famous hard problem ; see , e.g. , @xcite . consider now the fpp process on the discrete torus @xmath30 . we use our standardization convention ( [ standardization ] ) , so that rates @xmath25 for each edge @xmath28 . the following three `` first - order '' results are immediate from the shape theorem . asymptotics are as @xmath182 for fixed @xmath18 . @xmath206 the same holds ( in @xmath207 ) for @xmath208 itself . second , the time @xmath209 until all agents are infected satisfies @xmath210 for a constant @xmath211 defined in terms of the shape of @xmath204 . third , the proportion @xmath212 of infected agents satisfies @xmath213 for a certain function @xmath214 defined in terms of the shape of @xmath204 . having a limit `` shape '' represented by @xmath204 is very special to the lattice setting . what about the general setting of rates @xmath2 ? suppose we have a result of the kind @xmath215 then the random set @xmath216 of infectives at @xmath11 from initial @xmath9 is approximately the deterministic set @xmath217 . so , if the essence of a `` shape theorem '' is a deterministic approximation to the infected set , it is essentially just a wlln ( weak law of large numbers ) . a precise conjecture is formulated below . everything depends on @xmath1 ( not written ) and limits are as @xmath19 . the conclusion we want is @xmath218 we work in the fpp model , writing @xmath219 for the time to traverse edge @xmath220 . the obvious obstacle to the desired conclusion is that there may be a set @xmath221 ( with @xmath222 ) such that , in the random percolation path @xmath223 , the time taken to traverse the edge @xmath224 is not @xmath225 . one could conjecture this is the only obstacle ; here is a slightly weaker conjecture . [ conj-1 ] with arbitrary rates @xmath2 , if @xmath226 then ( [ conj-1 ] ) holds . it is intuitively clear ( and not hard to prove ) that condition ( [ tconc ] ) is necessary . proving the conjecture would hardly be a definitive result ( because condition ( [ tconc ] ) is not a condition directly on @xmath227 but would represent a start on understanding when the wlln holds . here we have rates @xmath228 and we start with one infective . this is a fundamental stochastic process whose basic properties have been rediscovered many times from different viewpoints , but there seems no authoritative survey . our focus is on the randomly - shifted logistic limit result , proposition [ plogistic ] . we give an outline taken from @xcite ( a similar treatment is in @xcite ) , though essentially the same result can be found in older genetics literature . there are two reasons for this particular focus . from this one result , we can read off many results like ( [ dgg ] , [ dgl ] ) , a fact which has not been appreciated in the combinatorics - inspired literature . and secondly , it plays a central role is investigating fmie processes build over pandemic , as we shall see in section [ sec : built_over_p ] . write @xmath229 time at which @xmath131 people are infected . the observation @xmath230 is a basic starting place for analysis . for instance , the time @xmath231 until the whole population is infected is just @xmath232 , and so @xmath233 and the time @xmath234 until a random person ( amongst the @xmath235 initially uninfected ) is infected is @xmath236 this analysis can be pursued to get distributional results , but let us first look at different methods that indicate the behavior without extensive calculation . the idea is to analyze separately the _ initial phase _ when @xmath237 people are infected , and the _ pandemic phase _ when @xmath238 people are infected . the `` deterministic , continuous '' analog of our `` stochastic , discrete '' model of an epidemic is the _ logistic equation _ @xmath239 for the proportion @xmath240 of a population infected at time @xmath11 . a solution is a shift of the basic solution @xmath241 setting @xmath242 for the number of individuals infected in our stochastic process , then scaling to the proportion @xmath243 , we find @xmath244 by including variance estimates , one could prove the following formalization of the idea that , during the pandemic phase , @xmath245 behaves as @xmath240 to first order . precisely , for fixed @xmath246 and @xmath247 we have ( in probability ) @xmath248 that is , we `` start the clock '' when a proportion @xmath246 of the population is infected . this can now be reformulated more cleanly in terms of the time @xmath249 at which half the population is infected : @xmath250 for later use , consider the times at which each process equals @xmath251 , where @xmath252 slowly . we find @xmath253 which rearranges to @xmath254 now turn to the initial phase of the epidemic . on a fixed initial time interval @xmath255 $ ] , the process @xmath242 of number of infectives converges in distribution to the process @xmath256 for which the times @xmath257 satisfy @xmath258 and this is the classic _ yule process _ , for which it is well known that @xmath259 now calculate informally , for @xmath260 , @xmath261 in other words @xmath262 where @xmath263 has the gumbel distribution @xmath264 . comparing with ( [ dn7 ] ) gives ( [ gng ] ) below . [ plogistic ] for the simple epidemic process on the complete @xmath1-vertex graph , there exist random @xmath265 such that @xmath266 where @xmath21 is the logistic function and @xmath263 has gumbel distribution . one could prove this by formalizing the arguments above , but there is a more efficient though less illuminating way . the result is essentially equivalent to the assertion @xmath267 the basic observation ( [ knk ] ) allows one to write down an explicit expression for the fourier transform of the left side , and one just needs to work through the calculus to check it converges to the fourier transform of the right side . ( transform methods are of course very classical in applied probability , and their use in epidemic models goes back at least to @xcite . ) the same transform argument ( then taking a different limit regime ) can be used to prove the following result ( e.g. , @xcite ) for the time @xmath268 until the entire population is infected : @xmath269 where @xmath270 are independent with gumbel distribution . while not obvious from the statement of the proposition above , it can in fact be deduced as follows . from ( [ dnun ] ) , the time @xmath271 until half the population is infected satisfies @xmath272 then symmetry ( under @xmath273 ) of the transition rates ( [ knk ] ) implies that @xmath268 is essentially distributed as the sum of two independent copies of @xmath271 . reconsider the time @xmath274 until a random person is infected . proposition [ plogistic ] implies it has asymptotic distribution @xmath275 where @xmath276 has logistic distribution function , independent of the gumbel time - shift @xmath263 . so we expect @xmath277 now @xmath278 by symmetry ( @xmath279 ) and one can calculate ( or look up ) that @xmath280 euler s constant , so the limit theorem is consistent with the exact formula ( [ euler ] ) . the extent to which different aspects of the `` mean - field '' behavior in section [ sec : fpp_complete ] extend to more general geometries is not well understood . for the `` random graphs with prescribed degree distributions '' ( with appropriate moments finite ) geometry , one expects to see qualitatively similar behavior ( the pandemic phase to have duration @xmath281 , centered at time @xmath282 ) , and almost precise analogs have indeed been proved e.g. , @xcite , theorem 3.1(b ) gives the analog of ( [ dgl ] ) . there has been extensive subsequent work on related problems and related `` locally tree - like '' geometries , but these proofs are substantially more technical than for the complete graph . and even on the hamming cube @xmath283 the behavior of fpp is not well understood @xcite . to think about possible `` general theory '' , let me suggest a loose analogy with the established theory of random walk on infinite graphs @xcite . in that theory , most geometries fall into one of two categories , exemplified by @xmath162 and @xmath18-trees , according to whether @xmath284 properties involving spectral radius , isoperimetric inequalities , amenability , ( sub)linear rate of escape can then be related to this categorization . that theory formalizes a distinction ( informally , between `` finite dimensional '' and `` infinite dimensional '' ) from the random walk viewpoint . so by analogy , we have seen that the finite pandemic process behaves differently on @xmath30 and on the complete graph . can we select one specific aspect of the different behavior as a useful basis for classifying geometries ? write @xmath285 . one obvious distinction between these two geometries is that on @xmath30 the means @xmath286 vary ( depending on distance between @xmath9 and @xmath12 ) but on the complete graph , the @xmath286 are the same ( to first order ) for most pairs @xmath28 . so we can define a property ( for a sequence of weighted graphs @xmath2 ) which combines the idea of the wlln ( [ conj-1 ] ) with the idea of the @xmath286 not varying much : @xmath287 this property ( which holds for the `` random graphs with prescribed degree distribution '' geometry @xcite ) is an attempt to formalize the notion of a geometry being `` locally infinite - dimensional '' from the fpp viewpoint . study equivalences between this property and the `` geometry '' of @xmath2 . two brief comments . we ca nt hope to establish ( [ conj - infdim ] ) using only isoperimetric inequalities they can only give order of magnitude . there is a loose analogy with the cutoff phenomenon in finite markov chains @xcite , in that we expect the property to hold in `` highly symmetric '' geometries . but that analogy is discouraging in that there is no useful general condition for cut - off . consider the geometry @xmath58 with agents @xmath171 and meeting rates at rate @xmath117 an agent meets a uniform random neighbor ; at rate @xmath288 an agent meets a uniform non - neighbor . here over this range the geometry qualitatively interpolates between the complete - graph case and the nearest - neighbor lattice case . so we expect the center of the pandemic window to interpolate between orders @xmath290 and @xmath141 , and the width of the pandemic window to interpolate between orders @xmath117 and @xmath141 . for the pandemic model over this geometry , it turns out that an analog of the `` randomly - shifted logistic limit '' holds , albeit with a different deterministic function @xmath21 in place of the logistic , and different scaling . this is rather counter - intuitive we envisage infectives as growing patches of mold in a petri dish , one might expect the detailed geometry of overlaps to play a role . but the key qualitative property is that , during the pandemic phase , the new infectives arise from `` colonies '' started earlier in the pandemic phase , not in the initial phase . this enables us to derive the following equation for the standardized function @xmath21 describing proportion of infectives r over the pandemic window : @xmath291 see @xcite for heuristics and @xcite for careful treatment of a closely related model . there is a large literature modeling spread of opinions , adoption of innovations etc via social networks , and some of the models fit our fmie setting . instead of describing some standard model let me describe two new fmie models suggested by the `` social networks '' context . mathematically they are variants of pandemic , and we will see in sections [ sec : deference][sec : fashionista ] that on the complete graph their behavior can be derived from the precise understanding of pandemic given by proposition [ plogistic ] . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the agents are labelled @xmath117 through @xmath1 , and agent @xmath9 initially has opinion @xmath9 . when two agents meet , they adopt the same opinion , the smaller of the two opinion - labels . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ clearly opinion 1 spreads as pandemic , so the `` ultimate '' behavior of deference is not a new question . we will consider the complete graph geometry , and study @xmath292 , where @xmath293 is the proportion of the population with opinion @xmath131 at time @xmath11 . we can rewrite proposition [ plogistic ] as @xmath294 where @xmath21 is the logistic function and @xmath295 , where @xmath296 arises as the limit r.v . ( [ yule_lim ] ) associated with the yule process . the key insight is that opinions @xmath117 and @xmath297 and and @xmath131 combined behave as one infection in pandemic , except for a different initial phase , and so the sum @xmath298 behaves as in ( [ xkf ] ) but with a different random shift @xmath299 . because the initial phase is just a collection of @xmath131 independent yule processes , we see that the sum has a representation of format ( [ xkf ] ) with @xmath300 here the @xmath301 are i.i.d . @xmath198 . so the @xmath19 limit behavior of the deference model in the complete graph geometry is , over the pandemic phase , @xmath302 the deference model envisages agents as `` slaves to authority '' . here is a conceptually opposite `` slaves to fashion '' model , whose analysis is mathematically surprisingly similar . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ at the times of a rate-@xmath77 poisson process , a new fashion originates with a uniform random agent , and is time - stamped . when two agents meet , they each adopt the latest ( most recent time - stamp ) fashion . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is easy to see that , for the random partition of agents into `` same fashion '' , there is a stationary distribution . for the complete graph geometry , we can copy the analysis of the deference model . combining all the fashions appearing after a given time , these behave as one infection in pandemic ( over the pandemic window ; newly - originating fashions have negligible effect ) , hence their proportion behaves as a random time - shift of the logistic curve @xmath21 . so when we study the vector @xmath303 of proportions of agents adopting different fashions @xmath131 , we expect @xmath19 limit behavior of the form @xmath304 where @xmath305 are the points of some stationary process on @xmath306 . knowing this form for the @xmath19 asymptotics , we can again determine the distribution of @xmath307 by considering the initial stage of spread of a new fashion . it turns out that @xmath308 where @xmath309 are the times of a rate-@xmath77 poisson process on @xmath310 . the second expression makes it clear that @xmath307 is a stationary process . here is the outline argument for ( [ fash - limit ] ) . consider the recent fashions at time @xmath311 adopted by small but non - negligible proportions of the population . more precisely , consider fashions originating during the time interval @xmath312 $ ] , where @xmath313 slowly . for a fashion originating at time @xmath314 , the time-@xmath73 set of adopting agents will be a subset of the corresponding epidemic process , which we know has proportional size @xmath315 where @xmath316 has @xmath198 distribution . the times @xmath317 of origination of different fashions form by assumption a rate-@xmath77 poisson process , and after we impose i.i.d . shifts @xmath318 we note ( as an elementary property of poisson processes ) that the shifted points @xmath319 still form a rate-@xmath77 poisson process , say @xmath320 , on @xmath310 . so the sizes of small recent fashion groups ( that is letting @xmath321 ) , for which overlap between fashions becomes negligible , are approximately @xmath322 . summing over @xmath323 gives @xmath324 and we end up with the representation ( [ fash - limit ] ) . on the torus @xmath171 the stationary distribution of fashionista is qualitatively similar to the _ dead leaves model _ @xcite and , as outlined in @xcite , one can apply simple scaling arguments to analyze the diversity statistic @xmath325 and show @xmath326 the corresponding result on the complete graph is @xmath327 and ( [ fash - limit ] ) provides a ( complicated ) explicit expression for the right side . in particular , we can see the @xmath328 behavior is @xmath329 for constant @xmath330 . studying @xmath331 in general geometries is an * open topic * , which opens up other questions to study about short - term behavior of pandemic . there is a huge literature attempting more realistic epidemic models , which i will not try to outline here . the fmie - type model that has been most extensively studied in the mathematical literature is the contact process ( the sis model with exponential infection times ) , studied classically on @xmath162 @xcite . a nice survey , in the style of this article , of recent rigorous work on the contact process on preferential attachment , small worlds and power - law degree geometries can be found in @xcite , and the latest technical work in @xcite . the current form of conjecture [ conj-1 ] emerged from conversations with jian ding . and i thank an anonymous referee for many helpful expository suggestions . research supported by nsf grant dms-07 - 04159 .

suppression of the low - energy part of bremsstrahlung from ultrarelativistic electrons passing through an amorphous scattering medium is the celebrated lpm - effect @xcite , named after landau and pomeranchuk who predicted it qualitatively @xcite , and migdal @xcite who provided an accurate theory for bremsstrahlung in a semi - infinite uniform medium . it has also been studied for finite - thickness targets @xcite , and for the opposite limit of thin target , which is of intrinsic interest @xcite . the theoretical predictions were accurately tested experimentally only within the last two decades @xcite , and the theoretical and experimental state of the art is reviewed in @xcite . a natural further step , both for practice and for theory , is to study composite targets . one of the simplest configurations thereof is a sequence of thin plates with a gap between them comparable to the photon formation length @xmath1 the latter quantity depends on the electron energy @xmath2 and photon energy @xmath0 in the window covered by the detector , @xmath3 standing for the electron mass . the problem of structured target as applied to the gamma - quantum emission , was first discussed by blankenbecler @xcite , who treated it at the basis of the model for multiple scattering devised in @xcite . subsequently , other authors @xcite extended their formalisms to handle this problem . an experimental verification of the existence of interference effects in bremsstrahlung on compound targets has recently been undertaken at cern @xcite . it was emphasized @xcite that high electron energy allows achieving macroscopic values for the photon formation length ( [ l_f - def ] ) , and thereby study its manifestations simply by tuning the distance between the scattering plates by a micrometer screw . but even for cern energies , @xmath4 gev , the coefficient in eq . ( [ l_f - def ] ) amounts only @xmath5 m , wherefore to make @xmath6 practically macroscopic ( @xmath7 mm ) , one needs to consider sufficiently small photon energies compared to @xmath2 , which then renders the problem essentially classical . at the same time , condition @xmath8 can be safely fulfilled , wherewith the contribution from transition radiation may be neglected . the experiment @xcite conducted at such conditions favored the theory @xcite , but also found some departures from it , and , surprisingly , from other theoretical predictions , which were expected to be more accurate . to clarify the origin of remaining discrepancies , it may be instructive to recollect underlying assumptions for the theories at the market . it is fair to say that blankenbecler and drell @xcite in essence extended landau and pomeranchuk s approach @xcite to cover the case of finite - thickness target , and incorporate quantum effects . at that , a simplified model of scattering in a medium was adopted , in which transverse dimensions of the target random field exceeded that of the electron wave packet , allowing to exactly integrate over electron impact parameters and to reduce the photon emission amplitude to an integral over particle trajectory , like in classical mechanics . a strong simplification was made then , in the spirit of @xcite , by replacing the medium average of an oscillatory integrand by the corresponding oscillatory function of averaged variables . that yielded the angle - integral spectrum in the form of a double time integral , which for most cases of interest had to be computed numerically . along these lines , blankenbecler @xcite evaluated the spectrum of radiation on @xmath9 plates , and discovered an additional maximum , or shoulder " in the region where the gap width becomes commensurable with the radiation coherence length . the approach of blankenbecler and drell has the merit of being simple and qualitatively correct . however , it suffers from a lack of accuracy due to the oversimplified averaging ( the neglect of fluctuations ) , which may be essential for comparison with experiments . an attempt to improve it was undertaken in blankenbecler s later paper @xcite , where a correlation of the amplitude and the phase of the integrand in the double time integral was taken into account , while other fluctuations were neglected , as before . as we will show , however , the correction so evaluated improves the accuracy only for the case of weak scattering in the plates , whilst in the opposite case of strong ( non - dipole ) scattering , in which interference effects are actually the strongest , this correction is rather counterproductive . almost simultaneously with @xcite , zakharov @xcite advanced with a calculation of the lpm effect in finite targets including proper averaging . it was built upon techniques originally developed for particle physics , notably the impact parameter representation . the merit of the latter is that it naturally invokes fourier transforms of scattering - angle distribution functions , which from solution of the kinetic equation express as simple exponentials of the fourier - transformed scattering differential cross - section . the main attention in papers @xcite was payed to a single finite - thickness target , while for structured target , the results were presented only graphically . zakharov noticed that the interference features in his calculated spectrum were less pronounced than those in @xcite , albeit attributed that to the crudeness of the blankenbecler - drell model for scattering medium , rather than to an oversimplification of their averaging procedure . shortly after , baier and katkov @xcite proposed a technique which allowed them to handle the case of @xmath9 scattering plates analytically . to this end , an alternative form of the double time integral representation for the radiation spectrum was adopted , and radiation spectra for @xmath9 plates were calculated by mathematical induction , assuming all the plates to be identical , equidistant , and the scattering angle distributions in them to be gaussian . the influence of the transition radiation was taken into account too , which can be valuable for lower - energy experiments . baier and katkov ultimately restricted their analysis to the case of plates infinitesimally thin compared to the gap width , and demonstrated that instead of one maximum , the radiation spectrum features a sequence of maxima and minima of decreasing amplitude . the common drawback shared by all the abovementioned approaches is that the resulting expressions for the radiation spectrum are rather unwieldy , and relations of the spectral features with physical parameters in the problem remain obscure . what is more serious , however , is that it led to a number of qualitative controversies in the literature , the most noticeable among which are the following : * it was suggested in @xcite that the position of the greatest spectral maximum corresponds to a situation when the photon formation length ( [ l_f - def ] ) matches the distance between the plate centres . but in that case , the resonance condition ought to involve an additional factor @xmath10 , as in coherent bremsstrahlung ( see @xcite ) . however , therewith it would definitely contradict the experiment @xcite . * in the integral expressing the radiation spectrum , the oscillating function typically has the argument @xmath11 , where @xmath12 is the electron s lorentz factor , and @xmath13 the radiation angle . the use of the latter expression suggests that for the distance @xmath14 between the plate centers , the main spectral maximum should be located at @xmath15 involving factor @xmath10 , as expected , but also a denominator @xmath16 , which might compensate it . practically , though , it is not a priori obvious , with respect to which direction the radiation angles should be counted , and what are their typical values . if one estimates them as @xmath17 , where @xmath18 is the rms scattering angle in the plates @xcite , then for @xmath19 the compensation between the @xmath10 factor and the denominator can be significant . but numerically , such estimates still contradict the experimental findings @xcite . * concerning the comparative strength of the contributions from the enhancement region and the suppression region at low @xmath0 , it is unobvious whether their integrated effect must be zero , as suggested in @xcite , or the suppression must dominate in the total energy losses , as one might anticipate from the migdal s theory @xcite of radiation in a thick target , where no enhancement manifests itself in the spectrum at all . in the absence of cogent explanation for those controversies , the state of the theory can hardly be regarded as satisfactory . besides that , it is becoming desirable further to consider the case of plates of unequal thickness , and to fully scrutinize the role of non - gaussian ( coulomb ) tails in scattering , insofar as approaches of baier and katkov , as well as blankenbecler and drell rested on gaussian description of scattering . given rather fundamental significance of the present problem , it makes sense at present to reappraise its theoretical treatment . in this paper we shall derive physically more transparent representations for the radiation spectrum , which will help clarify the matter as a whole , and at the same time improve the numerical accuracy . to simplify the problem , we will consider the case of two _ thin _ scattering plates ( foils ) , which allows avoiding double time integral representations for the radiation spectrum . we will also limit ourselves to classical electrodynamics , and neglect the transition radiation , which are good approximations for the performed experiments @xcite . on the other hand , we will investigate in more detail the influence of coulomb corrections . to alleviate the connection between the theory and the experiment , we introduce appropriate visibilities of the interference pattern . we will be able to resolve the controversies alluded above , and to correct abundant numerical errors existing in the literature . ultimately , we compare our predictions with the recent experimental data , and show that there is no significant discrepancy between the experiment and the theory . in classical electrodynamics , the spectral - angular energy distribution of radiation ( with vector @xmath20 marking the direction of photon emission ) expresses in a known way through the charged particle trajectory @xmath21 , @xmath22 @xcite : @xmath23}\frac{d}{dt } \frac{\bm{n}\times\bm{v}(t)}{1-\bm{n}\cdot \bm{v}(t ) } \right|^2,\ ] ] @xmath24 being the electron charge . for our present study , it will suffice to consider the simplest case when the electron experiences 2 abrupt scatterings , at instants @xmath25 and @xmath26 : @xmath27 integration over the time in eq . ( [ init - t - int ] ) trivially yields [ eq2 ] @xmath28,\label{expand - sum - sq - j}\end{aligned}\ ] ] with @xmath29 @xmath30 @xmath31 eq . ( [ expand - sum - sq - ja ] ) implies that if scatterings are well separated , the spectral radiation amplitude can be viewed as a sum of spectral amplitudes of radiation at single scatterings ( with a distance- and @xmath0-dependent phase shift @xmath32 between them ) . that may be conceived as the absence of effects of temporarily nonequilibrium proper field on the radiation amplitude level . however , observed in practice is the radiation intensity , which is proportional to the amplitude square . it differs from a sum of intensities on separate plates , engaging certain dependence on @xmath0 through the cosine of the phase shift between the scatterings . an important generic property of eq . ( [ expand - sum - sq - j ] ) is that the @xmath0-integral of the interference term therein strictly vanishes : @xmath33 by virtue of the relation @xmath34,\ ] ] and strictly non - zero value of the argument of the @xmath35-function for @xmath36 . the absence of interference in the total radiative energy loss at successive random scatterings has a simple physical reason : equivalently , it can be expressed as a _ time _ integral of the particle acceleration squared , in which time - separated scatterings obviously give non - interfering contributions @xcite : @xmath37 hence , if a suppression occurs in one region of the spectrum ( typically at lowest @xmath0 where it is associated with the lpm effect ) , it should be accompanied by a commensurable _ enhancement _ in some other region ( which is natural to call the anti - lpm effect ) . the established property is undoubtedly general and must hold as well for a number of scatterings greater than 2 . but it may contradict our experience that in a target with @xmath35-correlated scatterings , interference effects to the spectrum are purely suppressive , and there is seemingly no enhancement region at all @xcite . the contradiction is resolved if one takes into account that in condensed matter , distances between atoms are very short indeed , being usually far shorter than photon formation length . specifically , if we consider inequality @xmath38 where _ quantum _ arguments stipulate that @xmath39 , condition ( [ aa - gg - lf ] ) is fulfilled already at @xmath40 mev . in the latter ( quite ubiquitous ) case , the radiation spectrum reaches its quantum end at @xmath41 sooner than the radiation enhancement can develop . therefore , the spectral enhancement has no room to develop , and the suppression dominates both locally and globally @xcite . one could imagine that for gaseous targets , where distances between atoms are much greater than @xmath42 , the situation might change , but in the latter case , it is essential that @xmath43 should be regarded as a random variable . we will return to effects of randomization of @xmath43 in sec . [ sec : randomization ] , and demonstrate that the anti - lpm effect still manifests itself on a scale of photon formation length comparable with the range of _ anti_-correlation , which is of order of atomic size , again . meanwhile , we will concentrate on studying the case of radiation on two scattering foils with definite and macroscopic separation . before we proceed , however , there is another observation worth making . for bremsstrahlung in a sufficiently thin layer of substance , a fair approximation is the smallness of scattering angles ( dipole limit ) , when @xmath44 . in that case , the structure of currents ( [ j21-def])([j32-def ] ) appreciably simplifies . linearizing @xmath45 in @xmath46 and @xmath47 in @xmath48 , integrating over radiation angles , and averaging ( [ expand - sum - sq - j ] ) over scattering angles @xmath49 , @xmath50 yields @xmath51 with @xmath52 @xmath53 , and @xmath54 . since in most cases , successive scatterings are not causally connected , the correlator in ( [ g ] ) reduces to a product of mean scattering angles : @xmath55 for amorphous targets , mean deflection angles are obviously zero , wherefore the whole interference in the spectrum in this approximation vanishes . to preserve the interference effects between the targets , one thus needs either to render both scatterings asymmetric are challenging to manufacture . more compact deflectors could be created with the aid of lasers , as proposed in @xcite , or bent crystals . ] , or to arrange _ non - dipole _ conditions for radiation in the amorphous target , which may be simpler to bring into practice , and will be the subject of our study in what follows . to make our presentation self - contained , before addressing interference effects in a composite target , let us recapitulate the properties of bremsstrahlung on a single amorphous foil . this section gives the account of relevant formulae , highlighting the differences between gaussian and molire averaging procedures , as well as between frameworks of different authors . we will also propose two our own recipes for incorporating coulomb corrections , which give a reasonable balance between precision and simplicity for the addressed case . radiation spectrum at scattering to a definite angle of arbitrary size ( which is small by absolute value , but may be sizable compared with the typical radiation angle @xmath56 ) expresses by an integral ( see , e.g. , @xcite ) @xmath57 where @xmath58 small - angle expansion of ( [ f - def ] ) can be obtained as @xmath59 series ( [ f - dipole - series ] ) has finite ( actually , unit ) convergence radius , which is natural in view of the presence in ( [ f - def ] ) of the structure @xmath60 . the large - angle asymptotics of ( [ f - def ] ) is logarithmic : @xmath61 the logarithm here stems from the fact that although for a large - angle scattering , the radiation cones from the initial and final electron lines are well separated , the region between them is filled by an enhanced radiation , owing to the interference between the cones . it is integration over this inter - jet domain which gives rise to the large logarithm in ( [ f - la ] ) . ( a ) green solid curve , @xmath62 [ eq . ( [ fsigma - change - var ] ) or ( [ k1 ^ 2gauss ] ) ] ; red dashed curve , @xmath63 [ eq . ( [ f - def ] ) or eq . ( [ fb(t)=f(sigma2 ) ] ) ] ; red dot - dashed curve , the corrected blankenbecler - drell approximation ( [ ftilde - blankenbecler - thintarget ] ) . ( b ) red curve ( upper ) ; molire averaging for gold , blue curve , molire averaging for carbon ; dashed curve , gaussian averaging by eq . ( [ highland - approx ] ) . ] eqs . ( [ i1 ] ) , ( [ f - def ] ) may be used to derive radiation spectrum in a thin layer of substance [ under condition when a target thickness is much smaller than the photon formation length ( [ l_f - def ] ) in the considered range of @xmath0 ] . to this end , ( [ i1 ] ) must be averaged with the appropriate distribution in scattering angles in the target @xcite . in the simplest approximation , this distribution can be taken to be gaussian . averaging over scattering angles with the gaussian distribution with rms scattering angle @xmath18 , @xmath65 one gets @xmath66 where @xmath67 inserting here ( [ f - dipole - series ] ) , and integrating termwise yields @xmath68 with @xmath69 . series ( [ f - aver - series ] ) diverges for any finite @xmath70 ( not surprisingly since it was derived by integrating a series beyond its convergence domain ) ; nonetheless , the sequence of its terms may be used as an asymptotic expansion . the first two terms in eq . ( [ f - aver - series ] ) are @xmath71 note that the coefficient at the leading - order term is the same as in ( [ f - dipole ] ) , but the coefficient at the next - to - leading order term is twice greater . to obtain large-@xmath70 asymptotics of @xmath62 , it is advantageous to change in ( [ fsigma ] ) the integration variable to @xmath72 : @xmath73 the remaining integral in ( [ fsigma - change - var ] ) is a bounded function , ranging from 0 to 1 . at large @xmath70 , this term vanishes due to the pre - factor @xmath74 , so the corresponding asymptotics of @xmath75 is determined by that of macdonald function at the origin : @xmath76 . employing this in ( [ fsigma - change - var ] ) , we get @xcite @xmath77 it is instructive to compare result ( [ i1sigma])([fsigma ] ) with representations derived within frameworks of blankenbecler and drell @xcite , and baier and katkov @xcite , under the same conditions of ( geometrically ) thin target and gaussian scattering . blankenbecler s result [ see @xcite , eq . ( 30 ) ] for the thin - target limit reads [ f - blankenbecler - thintarget ] @xmath78 evaluation of this integral yields @xmath79 where @xmath63 is given by eq . ( [ f - def ] ) . hence , the correspondence between @xmath80 and @xmath70 is @xmath81 and the approximation of blankenbecler consists in substituting the rms scattering angle to the non - averaged ( fixed - scattering - angle ) radiation spectrum : @xmath82 this correspondence is what one could expect , given that blankenbecler and drell @xcite , similarly to landau and pomeranchuk @xcite , replaced an average of an oscillatory function in the integrand of a double time integral by the oscillatory function with average amplitude and phase . as is seen from comparison of asymptotics ( [ fsigma - la ] ) and ( [ f - la ] ) , however , such a simple - minded replacement deviates from the exact result only by @xmath83 . yet another possible source of inaccuracy , though , may stem from the fact that for numerical calculations , blankenbecler adopted relation @xmath84 ( with @xmath85 the target thickness and @xmath86 the radiation length ) . the coefficient in ( [ blankenbecler - q2def ] ) differs by a constant factor @xmath87 from that in the rossi formula @xmath88 accordingly , blankenbecler defined the scaled target thickness [ see @xcite , eq . ( 10.1 ) , @xcite , eq . ( 16 ) ] @xmath89 consistent with the correspondence rule ( [ t = sigma2/6 ] ) . but physically , the coefficient in the relation between @xmath90 and @xmath91 should vary logarithmically with @xmath85 . more accurate description of multiple coulomb scattering will be furnished in sec . [ subsec:1platemoliere ] . in @xcite , blankenbecler derived a correction to eq . ( [ f - blankenbecler - thintarget - int ] ) [ see @xcite , eq . ( 109 ) ] : @xmath92 ^ 2}-1\bigg).\end{aligned}\ ] ] at small @xmath80 , i.e. , small @xmath70 , it rather neatly reproduces the next - to - leading order term in the exact small - angle expansion ( [ fsigma - dipole ] ) , provided one uses here the correspondence ( [ t = sigma2/6 ] ) : @xmath93 however , at large @xmath70 ( or @xmath80 ) , approximation ( [ ftilde - blankenbecler - thintarget ] ) fails by a factor of 2 [ see fig . [ fig : f](b ) , red dot - dashed curve ] , and can not be regarded as tenable there . baier and katkov s result for the case of one foil @xcite reads missing in @xcite . ] @xmath94 ^ 2}\right\}.\ ] ] evaluation of this integral re - obtains expression ( [ fsigma ] ) , once one identifies @xmath95 and @xmath96 baier and katkov stated their own prescription for computation of @xmath97 , which will be compared with our prescription in the next subsection . for very thin targets , the distribution in scattering angles can significantly deviate from a gaussian , while in the extreme @xmath98 , it must turn proportional to the differential cross - section of single scattering . to be more accurate , one may utilize the general ( molire ) solution of the kinetic equation in terms of a fourier - bessel integral @xcite : @xmath99},\ ] ] where @xmath100 is the scattering differential cross - section . it may be noted that at substantial thicknesses , ( [ moliere - distr - def ] ) is approximable by a gaussian plus corrections @xcite , but we will not resort to such approximations here , dealing directly with the exact integral representation ( [ moliere - distr - def ] ) . according to eqs . ( [ expand - sum - sq - j ] ) , ( [ j21-def ] ) , @xmath101 where the last integral effectively extends over the transverse plane of small angle differences . since the two terms in parentheses in ( [ 80 ] ) only differ by a shift in the @xmath102 plane , it may be expedient to expand them into fourier integrals , so that the angular shift converts to a phase factor : @xmath103 considering that @xmath104 is the photon s transverse momentum , @xmath105 can be interpreted as the photon s impact parameter relative to the parent electron . inserting ( [ wf - angle - wf - imp - par ] ) to eq . ( [ 80 ] ) obtains @xmath106 ^ 2\left|1-e^{i(\bm{v}_1-\bm{v}_2)\cdot \bm{r } } \right|^2\nonumber\\ & \,&\label{40a}\\ & = & \frac{4e^2}{\pi\gamma^2}\int_0^{\infty } dr r k_1 ^ 2(r/\gamma)\left[1-j_0\left(\chi r\right)\right].\label{k1 ^ 2(1-j)}\end{aligned}\ ] ] it can be checked that evaluation of the integral in ( [ k1 ^ 2(1-j ) ] ) reproduces the explicit form ( [ i1 ] ) , ( [ f - def ] ) . the reader familiar with the light - cone formalism @xcite will recognize @xmath107 as being ( the low-@xmath0 approximation for ) the electron - photon component of the electron wave function , with the vector index accounting for the photon polarization . in this spirit , @xmath108 amounts the complete wave function after electron scattering to a definite angle , and for @xmath109 this wave function differs from zero . correspondingly , the integral of its modulus square gives the total probability of emission of a photon with energy @xmath0 , i.e. , the radiation spectrum . interpretation of this kind may be noteworthy for intuitive understanding of the final results , albeit in this paper we will refrain from using it technically , and adhere to the standard formulation of classical electrodynamics . to proceed , when expression ( [ k1 ^ 2(1-j ) ] ) is averaged over scattering angles with a gaussian distribution ( [ gauss - distr ] ) , it becomes @xmath110 on the other hand , averaging with molire distribution ( [ moliere - distr - def ] ) leads to the form @xmath111}\right\},\ ] ] essentially coinciding with the result of zakharov @xcite ( see also @xcite ) . it is also in the spirit of glauber form @xcite for scattering of a high - energy composite quantum system . ( it is not crucial that in our case the scattering is multiple , since it conserves the impact parameter , anyway . ) to use formula ( [ k1 ^ 2(1-edsigma ) ] ) , we need further to specify the differential cross - section of scattering on one atom . the simplest model thereof is @xmath112 where @xmath113 , and @xmath114 , with @xmath115 playing the role of the atomic screening radius . inserting ( [ dsigma - model ] ) to ( [ k1 ^ 2(1-edsigma ) ] ) gives @xmath116}\right\}\!.\ ] ] at practice , parameter @xmath117 , so for any value of @xmath118 , for description of radiation it is always legitimate to expand @xmath119 in vicinity of the origin : @xmath120 . to determine the precise value of @xmath121 , one may note that in the limit of small @xmath85 , @xmath122 on the other hand , by definition of the radiation length ( in the approximation of complete screening ) , that must equal in eq . ( [ 43x0 ] ) . ] @xmath123 parameter @xmath121 can thus be related to phenomenological parameters @xmath86 and @xmath124 via @xmath125 in this paper , our approach will be simply to employ relation ( [ x0-rho1 ] ) in eq . ( [ exp - k1 ] ) . the factor @xmath118 in the exponent can be expressed in terms of the ratio @xmath126 : @xmath127 therewith , we have only one adjustable parameter @xmath128 ( instead of two , for radiation with and without atom ionization or excitation ) . that parameter depends on the radiation length , which may include all the molecular binding and crystal structure effects . at practice , though , @xmath86 itself is often inferred from calculations by formulas for bremsstrahlung on a free atom , neglecting the inter - atomic effects @xcite . utilizing the commonly used ( calculated @xcite ) values for the radiation length , for gold we obtain @xmath129 , @xmath130 , whereas for carbon , @xmath131 , @xmath132 . note that if the screening radius is estimated from relation @xmath133 , for gold it amounts @xmath134 , while for carbon , @xmath135 . the presented estimates for @xmath115 are generally greater than those following from the prescription used by zakharov @xcite @xmath136 for gold , eq . ( [ zakh - a ] ) gives @xmath137 , while for carbon , @xmath138 , the latter deviating distinctly from our above estimate . fortunately , though , the sensitivity to this parameter is only logarithmic . [ fig : f](b ) overlays the predictions of eq . ( [ exp - k1 ] ) for carbon and for gold . along with those , we plot formfactor ( [ fsigma ] ) for gaussian averaging with the rms scattering angle evaluated by the widely used empirical formula - dependence of the coefficient were obtained @xcite , but to keep the treatment simple , and expose numerical differences between the approaches , we will confine ourselves here to parameterization ( [ highland - approx ] ) . ] @xcite @xmath139 formula ( [ highland - approx ] ) is known to work with an accuracy better than 10% for @xmath140 . from fig . [ fig : f](b ) we conclude that for bremsstrahlung on one plate , the agreement is good in the whole range of @xmath85 , including @xmath141 . that is natural since it can be shown that under some approximations , formula ( [ highland - approx ] ) can actually be derived from the rigorous representation ( [ exp - k1 ] ) ( see appendix [ appendix ] ) . correspondence between values of rms scattering angle calculated by our eq . ( [ 51 ] ) and by baier and katkov s eq . ( [ bk - rms - prescription ] ) . red curve for carbon ; blue curve for gold . the auxiliary dashed line marks the diagonal @xmath142 . ] finally , our eq . ( [ 51 ] ) approximately accounting for coulomb corrections in multiple scattering can also be compared with the recipe of baier and katkov @xcite : @xmath143.\end{aligned}\ ] ] predictions of eq . ( [ 51 ] ) and ( [ bk - rms - prescription ] ) are confronted in fig . [ fig : rms ] , and are reasonably close , although some differences are perceptible . we are now in a position to address the case of two plates . in eq . ( [ expand - sum - sq - j ] ) , the first two terms are individual contributions from each plate , described by formulas of sec . [ sec:1plate ] , while the nontrivial third term of eq . ( [ expand - sum - sq - j ] ) involves the product of currents induced by scattering in different plates . hence , averaging of those currents over corresponding scattering angles proceeds independently : @xmath144 as was pointed out in the end of sec . [ sec:2scat ] , product ( [ av - prod = prod - av ] ) receives a suppression after integration over azimuthal directions of @xmath20 . therefore , it is reasonable to carry out the averaging in two steps : first , to average over azimuths of @xmath20 , which does not need a reference to a specific form of the ( axially - symmetric ) scattering angle distribution function . at the second step , one implements convolutions with specific angular distribution functions . granted that the currents entering eq . ( [ eq2 ] ) have similar structure , it suffices to consider a generic form @xmath145 where @xmath146 may equal @xmath147 or @xmath148 . averaging of expression ( [ j - def ] ) over azimuths of the scattering angle , say , wrt direction @xmath149 ( it makes no difference since ultimately all directions are to be integrated over ) proceeds as follows . decomposing @xmath146 into components parallel and orthogonal to @xmath149 , @xmath150 with @xmath151 , we compute the azimuthal average : @xmath152 now we can insert @xmath153 , and omit the component parallel to @xmath20 , which does not contribute to the vector product with @xmath20 appearing in eq . ( [ av - prod = prod - av ] ) . as a result , the azimuthally averaged radiation amplitude assumes the form @xmath154 with @xmath155 in the small - angle approximation , @xmath156 , @xmath157 , @xmath158 , @xmath159 , we obtain @xmath160[\gamma^{-2}+(\theta+\chi)^2]}}\right)\nonumber\\ & \equiv&\frac{1}{2(1+\theta^2)}-\frac{1 } { 4\theta^2}\!\left(\!1\!+\!\frac{\theta^2-x-1}{\sqrt{(x+1-\theta^2)^2 + 4\theta^2}}\!\right),\quad\end{aligned}\ ] ] @xmath161 . function ( [ j - scalar - def ] ) is everywhere positive . at small @xmath162 , it expands as @xmath163\!\!\bigg\}\nonumber\\ & \underset{\theta\ll 1}\simeq&\frac12-\frac1{2(1+x)^2}+\mathcal{o}(\theta^2),\end{aligned}\ ] ] where @xmath164 are legendre polynomials @xcite arising as coefficients in the expansion @xmath165 . before proceeding to averaging over moduli of scattering angles , i.e. , @xmath166 , it is also instructive to examine the asymptotics of @xmath167 as a function of @xmath166 . at small @xmath166 , @xmath168\,\ , \label{j - quadrupole - approx - a}\\ & \underset{x\ll1}\simeq&\frac{1}{(1+\theta^2)^3}x+\mathcal{o}(x^2).\label{j - quadrupole - approx}\end{aligned}\ ] ] here the leading term is proportional to @xmath169 , whereas in the dipole approximation the current is @xmath170 , but the latter contribution vanishes after azimuthal averaging [ see eq . ( [ g ] ) ] . thus , the leading term ( [ j - quadrupole - approx ] ) corresponds to a quadrupole " approximation . on the other hand , at large @xmath171 and fixed @xmath13 , @xmath172 factor @xmath173 here describes the radiation from the @xmath149 half - line alone . as for the radiation from the @xmath146 half - line , it is smeared by the azimuthal averaging , but in any case , the inter - jet region extends only out to polar angle @xmath174 . it is noteworthy that function @xmath167 features no enhancement around @xmath175 , which could be expected due to the presence of the @xmath146-jet . so , after the azimuthal averaging , the latter jet does not manifests itself as a peak . for @xmath176 , the radiation is suppressed because radiation amplitudes from @xmath149 and @xmath146 lines strongly cancel . ( a ) loglog plot of the azimuthally averaged electromagnetic current at one scattering , eq . ( [ j - scalar - def ] ) , for scattering angles @xmath177 , 0.3 , 1 , 3 , 6 , 10 , 15 ( solid curves , bottom to top ) . dashed curves , approximation ( [ j - quadrupole - approx ] ) ; ( b ) same for the gaussian - weighted azimuthally averaged electromagnetic current at one scattering , eq . ( [ j - scalar - av - chi ] ) . dashed curves , approximation ( [ g - small - sigma ] ) . dot - dashed curves , approximation ( [ gsigma - la ] ) . ] employing representation eq . ( [ approx - aver - j ] ) in eq . ( [ av - prod = prod - av ] ) , we have @xmath178 where @xmath179 and we invoked the identity @xmath180 . the minus sign in eq . ( [ av - g ] ) , i.e. , the negativity of interference between the currents , retraces to the general property of saturation of radiation . in eq . ( [ expand - sum - sq - j ] ) , though , factor ( [ interf - jj ] ) is multiplied by an @xmath0-dependent factor @xmath181 , and the corresponding product is yet to be integrated over polar radiation angles . given that @xmath181 is sign - alternating , the spectrum may then actually oscillate with the increase of @xmath182 , not only experience a suppression . combining with the previously computed bremsstrahlung contributions from individual plates , which are @xmath0-independent , we get @xmath183.\,\,\end{aligned}\ ] ] the dependence of function ( [ didomega - combined ] ) on @xmath182 is shown in fig . [ fig : exact - oscill ] . as anticipated , it exhibits an oscillatory behavior . oscillations of spectrum of the bremsstrahlung on two foils , computed for gaussian scattering - angle distribution functions with rms @xmath184 ( thin red curve ) , 1 ( yellow ) , 2 ( green ) , 5 ( blue ) . at @xmath185 , the spectrum must rise again , and reach the bethe - heitler constant value ( see sec . [ sec : plate - formfactor ] ) . ] to determine the magnitude and the phase of the oscillations , note that at @xmath186 , the integral in the last term of ( [ didomega - combined ] ) is dominated by the vicinity of the lower endpoint , and its evaluation with @xmath187 yields the asymptotics @xmath188.\,\,\end{aligned}\ ] ] according to this result , maxima of @xmath189 should appear at @xmath190 with @xmath124 integer , independently of shapes of distribution functions @xmath191 . that simple prediction is confirmed by fig . [ fig : exact - oscill ] . a suggestive geometrical interpretation for the rule ( [ secondary - max ] ) , in the spirit of wave optics , is that the maxima correspond to situations when the distance between the plates equals an integer number of full coherence lengths @xmath192 . concerning the main maximum ( which accounts for the bulk of the anti - lpm enhancement ) , however , there is no possibility to associate it with the case of unit wavelength shift as proposed in @xcite . for @xmath193 , of course , formula ( [ di - largelambda ] ) does not apply , but from the point of exact formula ( [ didomega - combined ] ) , the main maximum of @xmath194 emerges when in the integrand of the last term , the maximum of owes in particular to the factor @xmath195 , which is due to the cancellation of interference effects at small @xmath162 see the end of sec . [ sec:2scat ] . ] @xmath196 , achieved at some @xmath197 , coincides with the extremum of @xmath198 : @xmath199 whereat @xmath200 . this situation has to be discriminated from the case @xmath201 in eq . ( [ di - largelambda ] ) , where the existence of a maximum for @xmath202 is irrelevant because there are several oscillations within its width , and the region of dominant contribution shifts to the endpoint . in the latter case , the half - wavelength shift corresponds to a spectral _ minimum _ next to the main maximum . as opposed to that , case ( [ denom=1+theta^2 ] ) with the account of the denominator @xmath203 can be called the inclined half - length , which is numerically close to quarter - wavelength . it is worth stressing in this regard that although there is a denominator in relation ( [ denom=1+theta^2 ] ) depending on @xmath70 , but it obeys a relation @xmath204 . clearly , even at large @xmath70 , the contributing @xmath162 in eq . ( [ didomega - combined ] ) are @xmath205 , i.e. , @xmath206 with respect to the direction @xmath149 . therefore , the use of the in - medium coherence length @xcite @xmath207 in the resonance condition @xmath208 is _ unjustified_. in fact , employment of ( [ l - f - in - medium ] ) in application to the present problem led in @xcite to a large underestimate of the location of the first spectral maximum . finally , one must pay attention to the pronounced spectral minimum at @xmath209 , which must be associated with zero wavelength shift . this limit is known to correspond to a situation when the radiation amplitude completely factorizes from the scattering one , and so radiation on two plates becomes the same as on one plate with the aggregate thickness . such a relation may be not immediately transparent from eq . ( [ didomega - combined ] ) , although proves by straightforward integration ; we will also supply a simpler derivation thereof in sec . [ subsec : v0 ] . to deduce all the consequences from the averaging , let us now assume @xmath210 in ( [ av - g ] ) to be a gaussian weighting distribution ( [ gauss - distr ] ) . the averaging of @xmath167 with such a distribution simplifies after an integration by parts : @xmath211^{3/2}}.\qquad\label{j - scalar - av - chi}\end{aligned}\ ] ] the behavior of function ( [ j - scalar - av - chi ] ) is illustrated in fig . [ fig : j](b ) for several values of @xmath70 . it is largely similar to that of @xmath167 shown in fig . [ fig : j](a ) , but is smoother at @xmath212 , the latter property being natural inasmuch as the unit step function in ( [ j - largeangle - approx ] ) is smoothened out by the averaging over @xmath171 . numerically , eq . ( [ didomega - combined ] ) with entries ( [ fsigma - change - var ] ) , ( [ j - scalar - av - chi ] ) proves to be equivalent to eqs . ( 2.34 ) , ( 2.36 ) of @xcite under the correspondence rule ( [ b - sigma2 ] ) , although analytically , the equivalence of both approaches is challenging to verify . our representation seems to be better suited for the analytic study of @xmath70- and @xmath182-dependences . with its aid , we will now derive small- and large-@xmath70 asymptotics of function @xmath213 , which will be useful in what follows . at small @xmath70 , plugging expansion ( [ j - quadrupole - approx - a ] ) to eq . ( [ j - scalar - av - chi - a ] ) , and integrating termwise , we obtain @xmath214\label{g - small - sigma - series}\\ & \underset{\sigma\ll 1}{\simeq}&\sigma^2(1+\theta^2)^{-3}+\mathcal{o}(\sigma^4).\label{g - small - sigma}\end{aligned}\ ] ] similarly to eq . ( [ f - aver - series ] ) , expansion ( [ g - small - sigma - series ] ) diverges , but again , it remains sensible as an asymptotic series . its leading term ( [ g - small - sigma ] ) coincides with the asymptotics of non - averaged expression ( [ j - quadrupole - approx ] ) , granted that there the @xmath171-dependence factorizes . the behavior of approximation ( [ g - small - sigma ] ) is shown in fig . [ fig : j](b ) by dashed curves . on the other hand , at significant @xmath70 , it is legitimate to expand @xmath215 to taylor series in vicinity of point @xmath216 and integrate termwise : @xmath217^{3/2}}\nonumber\\ & \,&\qquad\qquad\qquad\qquad\times \left(1+\frac{\theta^2 - 1-x}{\sigma^2}\right)\nonumber\\ & = & \frac{1}{1+\theta^2}e^{-\frac{\theta^2 - 1}{\sigma^{2}}}\left(\frac12-\frac1{\sigma^2}\right)\nonumber\\ & \approx&\frac{1}{2(1+\theta^2)}e^{-\frac{1+\theta^2}{\sigma^{2}}}.\end{aligned}\ ] ] according to fig . [ fig : j](b ) ( dot - dashed curves ) , approximation ( [ gsigma - la ] ) works well for @xmath218 . at large @xmath70 and fixed @xmath162 , the exponential here tends to unity , and form ( [ j - largeangle - approx ] ) is retrieved . if instead of a gaussian we prefer to use molire weighting distribution ( [ moliere - distr - def ] ) , it may be more convenient to return to eq . ( [ expand - sum - sq - j ] ) for the radiation spectrum . in the last line of eq . ( [ expand - sum - sq - j ] ) , the product of currents consists of 4 terms , which have similar structure . it suffices to calculate only one of those , @xmath219,\end{aligned}\ ] ] while the other terms can be reconstructed by replacements @xmath220 or / and @xmath221 . changing the integration variable to @xmath222 , inserting representations ( [ wf - angle - wf - imp - par ] ) for the first factor in the first line of ( [ 89 ] ) , and a complex conjugate representation for the second factor , one brings it to the form of an integral over impact parameters : @xmath223 here @xmath224 may be regarded as green s function for a two - dimensional free schrdinger equation on the light front : @xmath225\!s_{\omega}\!\left(\bm{r}_1-\bm{r}_3,t_{21}\right)\!\theta(t_{21})\nonumber\\ = -i\delta(\bm{r}_1-\bm{r}_3)\delta(t_{21}),\qquad\end{aligned}\ ] ] and it evaluates in closed form : @xmath226.\ ] ] note that this function depends only on the difference between @xmath227 and @xmath228 , just as in ( [ didomega - combined ] ) the cosine depends , besides @xmath229 , only on the angle between @xmath149 and @xmath20 . restoring the rest of the interference terms by substitutions @xmath221 or @xmath220 , and combining them , we get @xmath230 convolving this with the molire distribution function , and integrating over azimuths of @xmath227 and @xmath228 , leads to the result @xmath231}\right\ } \nonumber\\ \times\int_0^{\infty } d \rho_3\rho_3 k_1(\rho_3 ) \left\{1-e^{-n_2l_2 \int d\sigma(\chi)[1-j_0(\rho_3\gamma\chi)]}\right\ } \nonumber\\ \times \cos\left(\frac{\rho_1 ^ 2+\rho_3 ^ 2}{4\omega}-\omega\right)j_1\left(\frac{\rho_1\rho_3}{2\omega}\right),\quad\end{aligned}\ ] ] where for convenience we rescaled the impact parameter variable to @xmath232 . note that after integration over the azimuth , the dependence only on the impact parameter difference is somewhat disguised . at large @xmath182 , upon approximations @xmath233 , @xmath234 , ( [ impact-2scat ] ) factorizes : @xmath235}\right\ } \nonumber\\ \times\int_0^{\infty } d \rho_3\rho_3 ^ 2 k_1(\rho_3 ) \left\{1-e^{-n_2l_2 \int d\sigma(\chi)[1-j_0(\rho_3\gamma\chi)]}\right\}.\,\,\end{aligned}\ ] ] this form proves to be equivalent to eq . ( [ di - largelambda ] ) , by virtue of the identity @xmath236}\right\}\quad\label{80a}\\ = \frac14\int d^2\chi \frac{dw_{\text{m}}}{d^2\chi } \int_0^{\infty } d \rho\rho^2 k_1(\rho ) \left[1-j_0(\rho\gamma\chi)\right]\nonumber\\ = \int d^2\chi \frac{dw_{\text{m}}}{d^2\chi } \left[\frac12-\frac1{2(1+x)^2}\right ] \equiv\left\langle g\right\rangle_{\text{m}}(0)\quad\end{aligned}\ ] ] [ cf . ( [ g(0 ) ] ) ] . the spectrum of radiation on two plates that we have evaluated , even under the simplification of a negligible plate thicknesses , is a function of 3 variables : @xmath237 , @xmath237 , and @xmath182 . in limiting cases , though , it reduces to simpler functions , which will be scrutinized below . the visibility of the low-@xmath0 dip , eq . ( [ v - through - f ] ) . ( a ) for gaussian averaging and arbitrary @xmath237 , @xmath238 . ( b ) for gaussian averaging and @xmath239 . solid black curve , exact expression ( [ visibility ] ) . dashed black curve , its asymptotics ( [ v - largesigma ] ) . purple dot - dashed curve , the visibility in the blankenbecler - drell approximation . ( c ) for molire averaging . blue curve , carbon ; red curve , gold ; black dashed curve , gaussian averaging with rms scattering angle computed by formula ( [ highland - approx ] ) . ] for experimental tests , it is imperative to estimate visibilities of gross features of the spectrum , and find their optimum by choosing appropriate foil materials and their thicknesses . the interference term reaches its largest absolute value at @xmath209 . therefore , it is natural to consider the characteristic quantity @xmath240 adopting the terminology from optics , this may be called the visibility of the low-@xmath0 dip . with the aid of eq ( [ didomega - combined ] ) , it can be explicitly written as [ visibility ] @xmath241 as we already mentioned at the end of sec . [ subsec : aggr - spectr ] , here the spectrum must reduce to that from a single plate of aggregate thickness . this can now be proven rather easily , based on the impact parameter representation ( [ impact-2scat ] ) . there , @xmath242 , and terms @xmath243 , @xmath244 cancel with some of the interference terms , leaving @xmath245 . thereby , @xmath246 indeed reduces to the form @xmath247 at small @xmath70 , @xmath248 i.e. , essentially it is proportional to the minimal among @xmath249 , @xmath250 . at large @xmath237 , @xmath238 , @xmath251 thus , in principle , asymptotically @xmath246 tends to @xmath252 . that is natural , inasmuch as radiation at one plate saturates as a function of its thickness , and hence the radiation at any two strongly scattering foils must be about twice stronger than on one of them . but practically , such a saturation is achieved only logarithmically , and is too remote , so the second term in eq . ( [ v - largesigma ] ) is usually significant . the behavior of ( [ visibility ] ) and asymptotic approximation ( [ v - largesigma ] ) is illustrated in fig . [ fig : v](b ) by black solid and dashed curves . for comparison , the purple dot - dashed curve also shows the corresponding result for blankenbecler s theory , when in eq . ( [ v - through - f ] ) @xmath75 is replaced by @xmath253 . notably , that curve intersects with the exact result at @xmath254 , and keeps close to it at greater @xmath70 . but at @xmath255 the difference is relatively large . it is also interesting to note that the blankenbecler - drell prediction for @xmath246 is lower than the exact one , i.e. , the neglect of fluctuations leads to an underestimation of the interference effect in radiation . that means that under the conditions of cancellation of dipole contributions in the interference , fluctuations help creating uncompensated contributions . likewise , it is useful to compare in this limit the predictions of gaussian and molire averaging [ see fig . [ fig : v](c ) ] . it appears that the gaussian averaging prediction becomes inaccurate at @xmath256 , just where the accuracy of approximation ( [ highland - approx ] ) itself becomes worse than 10% . in the opposite limit of large @xmath0 , according to eq . ( [ di - largelambda ] ) , the visibility may be characterized by the ratio @xmath257 at small @xmath70 , this quantity behaves as @xmath258 whereas at large @xmath237 , @xmath238 , it logarithmically decreases : @xmath259 the dependence of @xmath260 on both parameters @xmath237 and @xmath238 is shown in fig . [ fig : vinfty](a ) . its maximum is achieved at @xmath261 [ see fig . [ fig : vinfty](b ) ] . there , the function still does not exceed 0.7 . yet , in ( [ di - largelambda ] ) it is multiplied by @xmath262 with @xmath263 , hence , actual visibility in the high-@xmath0 region is on the level of a few percent , demanding formidable measurement statistics . for comparison , in fig . [ fig : vinfty](b ) by dot - dashed curve we show the behavior of eq . ( [ v_infty - def ] ) after replacement of averaged radiation formfactors by non - averaged functions of averaged argument : @xmath264 which we presume to correspond to blankenbecler - drell approximation . function ( [ 89 ] ) intersects with the exact result at @xmath265 , and keeps close to it at greater @xmath70 , but at @xmath266 the difference is rather large . same as fig . [ fig : v ] , for visibility of secondary minima and maxima , eq . ( [ v_infty - def ] ) . the dashed curve in ( b ) obeys eq . ( [ vinfty - asympt ] ) for @xmath239 . ] fig . [ fig : vinfty](c ) compares predictions of eq . ( [ v_infty - def ] ) when averaging is performed with the molire distribution , by eqs . ( [ 80a ] ) , ( [ exp - k1 ] ) ( the blue solid curve for carbon , the red solid curve for gold ) , and with gaussian distribution , eqs . ( [ j - scalar - av - chi ] ) and ( [ fsigma - change - var ] ) ( black dashed curve ) . there is a distinct difference between the predictions around the maximum , which is achieved at @xmath267 , i.e. , @xmath268 . the sign of this difference tells that in the region of substantial photon energies , the account of fluctuations of scattering leads to a lowering of the interference effect in radiation , contrary to the situation with @xmath246 in the previous section . that admits rather simple explanation : fluctuations always tend to suppress the radiation , but at @xmath209 , where the interference effect itself is suppressive , that worked towards increasing the effect , whereas in the large-@xmath0 region it just lowers both amplitudes @xmath269 and @xmath270 in the interference term . for a generic case , integral ( [ didomega - combined ] ) can only be evaluated numerically . but for limiting cases ( [ j - quadrupole - approx ] ) and ( [ j - largeangle - approx ] ) , the integration is analytically manageable . for completeness , in the remainder of this section , we will collect formulae for the corresponding limiting spectral shapes , and compare them with results of @xcite , obtained for same limits , at @xmath271 . to begin with , consider the case when scattering angles in both of the foils are small . inserting ( [ g - small - sigma ] ) to ( [ didomega - combined ] ) , we get @xmath272 where @xmath273 and @xmath274 is the quadrupole - quadrupole interference function . the latter function achieves its first ( anti - lpm ) maximum at @xmath275 ( see fig . [ fig : g - qq ] ) . according to eqs . ( [ qq ] ) , ( [ g - qq - def ] ) , the visibility of spectral fringes is @xmath276 . it must be noted that our function ( [ g - qq - def ] ) appears to differ from function @xmath277 obtained in @xcite , eq . the comparison of those functions in fig . [ fig : g - qq ] shows ample differences except in the limits @xmath278 and @xmath279 . we believe therefore that the corresponding result of @xcite is in error . ( while contributions @xmath280 from @xmath281 and @xmath282 must cancel ) . we did not follow this procedure , wherefore we can not indicate the source of the error . ] the next case to be considered is when scattering angles on each of the foils are @xmath283 . inserting eq . ( [ gsigma - la ] ) to ( [ didomega - combined ] ) , obtains @xmath284,\end{aligned}\ ] ] where @xmath285\nonumber\\ & = & e^{-\xi}\cos\omega-\mathfrak{re}\left[(1+\xi+i\omega)e_1\left(\xi+i\omega\right)\right],\nonumber\\\end{aligned}\ ] ] with @xmath286 the exponential integral . term @xmath287 in the argument of @xmath288 is negligible if @xmath289 ; then our eq . ( [ ll ] ) reduces to eq . ( 2.45 ) of @xcite . this limiting function has an anti - lpm maximum at @xmath290 ( see fig . [ fig : gll ] ) , which agrees with the conclusions of blankenbecler - drell and baier - katkov that the maximum is achieved at @xmath291.m . ] however , for small @xmath182 , term @xmath292 in the argument of @xmath288 is crucial . there , @xmath293 , wherewith @xmath294 corrections proportional to @xmath292 , especially in the prefactor of the logarithm in ( [ g_ll - approx ] ) , can be important at moderately small @xmath295 . neglecting @xmath296 or @xmath297 under the logarithm sign in ( [ g_ll - approx ] ) , we arrive correspondingly at eq . ( 2.42 ) or ( 2.43 ) of @xcite . in fig . [ fig : gll ] we compare the behavior of function ( [ e1 ] ) ( solid curve ) , with that of approximation ( [ g_ll - approx ] ) ( dotted curve ) , and of an approximation resulting when we put in eq . ( [ e1 ] ) @xmath298 ( dashed curve ) . the latter two approximations are seen to work in complementary regions . interference function for the case of double small - angle - scattering . solid curve , eq . ( [ g - qq - def ] ) close to the red curve of fig . ( [ fig : exact - oscill ] ) . blue dot - dashed curve , baier - katkov s function @xmath299 @xcite . ] finally , we examine the case when one of the foils scatters weakly , whereas another one scatters the electrons through angles @xmath283 . inserting the corresponding limiting forms ( [ g - small - sigma ] ) and ( [ gsigma - la ] ) to ( [ didomega - combined ] ) , we arrive at @xmath300\right\},\ ] ] where @xmath301 this function has a maximum at @xmath302 . notably , there is no logarithmic dependence in the interference term , and its amplitude does not depend on @xmath238 , which corresponds to strict saturation of radiation . correspondingly , the visibility in this case is @xmath303 , i.e. , even lower than for the case of double small - angle scattering . interference function for the case of double large - angle - scattering . solid curve , eq . ( [ e1 ] ) close to the blue curve of fig . [ fig : exact - oscill ] . dashed curve , @xmath304 . dotted curve , asymptotics ( [ g_ll - approx ] ) . ] hitherto we presumed negligible thicknesses of both targets , but at a sufficiently large @xmath0 , finite target thickness must inevitably show up . indeed , when @xmath305 decreases down to the value of @xmath85 , the target geometry gets resolved . thereat , the radiation spectrum must additionally rise from the doubled but logarithmically saturated values to the completely saturation - free bethe - heitler value . in principle , our framework requires only minor modification to reflect the mentioned rise : it suffices to multiply the radiation spectrum computed above by the plate formfactor , which smoothly interpolates between unity at @xmath306 and migdal s function @xmath307 at @xmath308 ( specifically , @xmath309 ) . the implementation of such an interpolation , however , is beyond our scope in the present paper . the best way to test our equations is to confront them with experimental data . measurements of bremsstrahlung spectra from 178 gev electrons on a sequence of plates were performed in @xcite with two equal 26 @xmath310 m thick golden foils ( @xmath311 mm ) separated by a gap of variable width . the scattering strength parameter for one such a foil , according to eq . ( [ highland - approx ] ) , estimates as @xmath312 , so the scattering angles there were rather large . in fig . [ fig : experiment ] we compare predictions of our eq . ( [ didomega - combined ] ) and eq . ( [ impact-2scat ] ) with the experiment @xcite , letting @xmath43 be equal to the distance between plate centres , @xmath14 . the agreement may be regarded as fair . some possible experimental inaccuracies at @xmath313 mev were discussed in @xcite , but they do not seem to cause large deviations . since our gaussian and molire averaging procedures give numerically close results , and on the other hand , it was found in sec . [ sec:1plate ] that our prescriptions for scattering parameters are close to those of baier - katkov and zakharov , we expect that baier - katkov s and zakharov s predictions for the radiation spectrum must be in agreement with the experiment @xcite , too , even though the qualitative inferences based on the baier - katkov theory seem to disagree with it . shapes of spectral oscillations for conditions of experiment @xcite ( @xmath314 m ) . ( a ) @xmath315 m ; ( b ) @xmath316 m ; ( c ) @xmath317 m . dark blue curves , gaussian averaging , with @xmath70 calculated by eq . ( [ highland - approx ] ) . green curves , molire averaging for gold . ] as for comparison of experimental results with predictions of blankenbecler , recalling our results of sec . [ sec : analysis ] , it should be noticed that the use of plates with scattering strength @xmath318 closely corresponds to the point of intersection of visibilities in bd approximation with exact ones [ see fig . [ fig : vinfty](b ) ] . therefore , blankenbecler s predictions may agree with particular experiment @xcite rather nicely , too . basic agreement of experimental results with @xcite ( and disagreement with @xcite ) was actually reported in @xcite . however , for foils several times thinner that currently used ( or made of a lighter material ) , we predict a growing inaccuracy of blankenbecler - drell predictions . at the same time , for weaker scattering foils , the visibility of _ secondary _ spectral minima and maxima can yet increase by about a factor of 2 [ see figs . [ fig : vinfty](b),(c ) ] . having dwelt enough on the case of bremsstrahlung on two targets separated by a fixed distance , it would be instructive also to generalize it so that it could be compared with the conventional case of bremsstrahlung in an extended random medium . of course , presently we possess equations for radiation only at two scatterings , but arguably , the greatest contribution to the interference must come from nearest scatterings . therefore , we can exploit previous formulae , if @xmath43 is regarded as a random variable distributed within some range @xmath319 . rather realistically , it can be modeled by an exponential @xmath320 , corresponding to the probability of avoiding an encounter with the next atom in a uniform gas with the mean interatomic distance @xmath321 . to preserve the anti - lpm effect , though , we must prevent @xmath43 from tending to zero , as was pointed out in sec . [ sec:2scat ] . for simplicity , we shall assume the shape of the anticorrelation to be described also by an exponential , but shorter - range one , thus choosing the distribution function @xmath322 parameter @xmath323 provides the ` repulsion ' distance between the scatterers . weighting eq . ( [ didomega - combined ] ) with ( [ dwdt21 ] ) , we get @xmath324\!\!\bigg\}.\end{aligned}\ ] ] the @xmath0-depending factor in the integrand is now a difference of lorentzians ( bell - shaped functions of definite sign ) having different heights and widths determined by @xmath321 or @xmath325 , but equal areas . the first term in the brackets is negative , and can be associated with lpm - like uniform suppression , whereas the second term is positive , representing the anti - lpm enhancement , which is now strictly positive and non - oscillatory . the shapes of radiation - angle - integral spectra are illustrated in fig . [ fig : oscill - aver - t21 ] for same values of @xmath70 as in fig . [ fig : exact - oscill ] , and ( for the sake of illustration ) a moderately large ratio @xmath326 . it may be observed that with the increase of @xmath70 , the shape of lpm - like suppression sharpens towards @xmath327 , although it does not tend to the thick - target dependence @xmath328 . obviously , to get the @xmath329 behavior , more than two scatterings need to be involved . spectra of bremsstrahlung on two plates with a randomized distance between them , eq . ( [ i - aver - t21 ] ) , for @xmath330 and same parameter values for scattering strengths as in fig . [ fig : exact - oscill ] . ] finally , we note that in atomic matter , scales @xmath321 and @xmath325 should vastly differ . to estimate @xmath321 , it may suffice to note that according to migdal s calculation , typical @xmath0 at which lpm effect develops are @xmath331 , where @xmath332 is the range at which the angle of multiple scattering becomes comparable with the radiation angle @xmath56 . hence , @xmath333 , and owing to the presence of factor @xmath334 , this range is at least by 4 orders of magnitude greater than typical interatomic distance ( a few units of @xmath42 ) , which must constitute the scale for @xmath325 . therefore , anti - lpm enhancement in atomic matter must be virtually invisible indeed , reconciling its principal existence with the conventional theories of the lpm effect . put differently , @xmath335 can be effectively sent to zero , justifying the formal use of the @xmath35-correlated model of multiple scattering employed by migdal @xcite . in course of our study , we arrived at the following conclusions : * for scattering foils with macroscopic and fixed separation , the lpm - like suppression of bremsstrahlung at lowest @xmath0 is accompanied by an adjacent enhancement , which can be called the anti - lpm effect . the latter effect may be associated with an inclined - half - wavelength resonance [ see eq . ( [ denom=1+theta^2 ] ) ] , but not with the unit wavelength resonance , as was previously suggested in @xcite . in classical electrodynamics , lpm and anti - lpm effects are predicted to have equal strength , in the sense that their @xmath0-integrated effect is zero ( see the end of sec . [ sec:2scat ] ) . however , in a random and weakly scattering media , like the ordinary atomic matter , the anti - lpm enhancement is expected to smear over a broad spectral interval , and thus become faint . * the shape of the anti - lpm maximum depends on the strength of the scattering in each plate ( see fig . [ fig : exact - oscill ] ) . it assumes universal forms for limits of strong or weak scattering : for small scattering angles in both plates , the shape of the oscillatory pattern in the spectrum is described by eq . ( [ g - qq - def ] ) . for large scattering angles in both plates , it is described by function ( [ e1 ] ) ( unifying different limiting cases described in @xcite ) . for the mixed case when one plate scatters strongly , and another one weakly , the spectrum shape obeys eq . ( [ g_ql - def ] ) , where the dependence of the interference term on @xmath70 strictly saturates . * at asymptotically large @xmath0 , the dominant contribution to the interference term stems from photon emission angles close to @xmath149 , such that @xmath336 . therewith , the interference term scales as @xmath337 , with the coefficient proportional to the product of @xmath167-amplitudes of radiation emitted strictly parallel to @xmath149 [ see eq . ( [ di - largelambda ] ) ] . granted this universality , the locations of _ secondary _ maxima and minima are virtually independent of the scattering strength in the plates . we have also established that if the rms scattering angle is calculated with a proper account of coulomb corrections , the predictions of gaussian averaging are pretty close to those of the more precise molire averaging , and practically comply with the results of zakharov @xcite , and baier and katkov @xcite . for 2 scattering plates , the equation for molire averaging is basically as simple as that for gaussian averaging . finally , we note that satisfactory agreement of predictions of classical electrodynamics with experiment was found . for future experiment planning , it may be useful to refer to visibilities of the interference minima and maxima at low and at large @xmath0 , computed in sec . [ sec : analysis ] ( figs . [ fig : v ] and [ fig : vinfty ] ) . it may be added that prospects of observing anti - lpm effect in radiation may be not restricted to artificial assemblies of plates , provided one finds a case when @xmath338 , i.e. , every constituent of the matter can scatter the radiating projectile relativistically . conditions for such a situation may be sought , for instance , in nuclear matter . this work was supported in part by uffr project no . to clarify the success of the gaussian approximation in conjunction with interpolation ( [ highland - approx ] ) for the rms scattering angle , consider the large-@xmath339 limit of eq . ( [ exp - k1 ] ) : @xmath340}\right\}\nonumber\\ & \simeq&2\int_0^{\rho_0 } \frac{d\rho}{\rho}\left\{1-e^{-\frac12\rho^2\gamma^2\chi^2_c\left(\ln\frac{2}{\gamma\chi_1\rho}+\frac12-\gamma_{\text{e}}\right)}\right\}\nonumber\\ & \,&+2\int_{\rho_0}^{\infty } d\rho \rho k_1 ^ 2(\rho)\nonumber\\ & \simeq&\int_0^{\rho_0 ^ 2 } \frac{d\rho^2}{\rho^2}\left\{1-e^{-\frac14\rho^2\gamma^2\chi^2_c\left[\ln\left(\frac{\chi^2_c}{\chi^2_1}\ln\frac{\chi^2_c}{\chi^2_1}\right)+1 - 2\gamma_{\text{e}}\right]}\right\}\nonumber\\ & \,&+2\left(\ln\frac2{\rho_0}-\frac12-\gamma_{\text{e}}\right)\nonumber\\ & \simeq&\ln\left\{\gamma^2\chi^2_c\left[\ln\left(\frac{\chi^2_c}{\chi^2_1}\ln\frac{\chi^2_c}{\chi^2_1}\right)+1 - 2\gamma_{\text{e}}\right]\right\}-1-\gamma_{\text{e}}. \nonumber\\\end{aligned}\ ] ] comparing it with eq . ( [ fsigma - la ] ) , we establish the correspondence @xmath341.\ ] ] if we neglect here @xmath342 , and substitute @xmath118 from eq . ( [ chi2c / chi21 ] ) , it casts as next , one can put @xmath344 , @xmath345 , @xmath346 , and @xmath347 can vary from @xmath348 ( if @xmath349 ) to @xmath350 ( if @xmath351 ) . as a result , @xmath70 can be approximated by @xmath352 with @xmath353 , and @xmath354 . those numbers comply with the coefficients in eq . ( [ highland - approx ] ) , which we have thereby derived ab initio . parameter @xmath310 can also be compared with rossi s @xmath355 mev . product @xmath356 becomes twice smaller than @xmath357 at @xmath358 , which then exactly corresponds to approximation ( [ blankenbecler - q2def ] ) . in general case , of course , it is more reliable to use formula with an explicit @xmath359 dependence , like ( [ highland - approx ] ) .

the discovery of a primordial magnetic field ( pmf ) would have a profound impact on our understanding of the early universe @xcite and would help explain the origin of the observed magnetic fields in galaxies and clusters @xcite . several observational probes are currently being investigated and the cosmic microwave background ( cmb ) is a promising tool to discover and study the pmf on the largest cosmic scales . a pmf can affect the thermal distribution and the polarization of the cmb . the most relevant cmb signature depends on the form of the magnetic field spectrum and also on the level of instrumental noise for different observations . for instance , current cmb bounds of a few ng @xcite on a scale - invariant pmf @xcite derive from temperature anisotropies sourced by the magnetic stress - energy . comparable bounds were recently obtained in @xcite using the lyman-@xmath0 forest spectra @xcite . as the stress - energy is quadratic in the magnetic field strength , improving this bound by a factor of @xmath1 would require a @xmath2 fold improvement in the accuracy of the spectra . on the other hand , faraday rotation ( fr ) of cmb polarization , being linear in the magnetic field strength , offers an alternative way to improve cmb bounds on a scale - invariant pmf by an order of magnitude @xcite , and possibly more , with the next generation cmb experiments . even current cmb data is close to providing competitive bounds on scale - invariant pmf through their fr signature @xcite . in contrast , a pmf produced causally in an aftermath of the electroweak or qcd phase transition @xcite would have a blue spectrum @xcite with most of the power concentrated near a small cutoff scale set by the plasma conductivity at recombination . it has been suggested @xcite that such small - scale fields can appreciably alter the recombination history via enhancement of small scale baryonic inhomogeneities . consequently , the strongest cmb constraint would come from an overall shift in the distance to last scattering . according to @xcite , upcoming cmb experiments can rule out causally produced fields with a comoving strength larger than @xmath3 . given the promise of fr to significantly improve the bounds on scale - invariant pmf @xcite , it is important to assess the strength of the rotation induced by the magnetic fields in our own galaxy . a rotation measure ( rm ) map of the milky way has recently been assembled by oppermann et al @xcite based on an extensive catalog of fr of compact extragalactic polarized radio sources . they also presented the angular power spectrum of rm and find that it is very close to a scale- invariant spectrum at @xmath4 . 1 of @xcite shows that the root - mean - square rm away from the galactic plane is about @xmath5-@xmath6 . additional insight into galactic rm on smaller scales can be gained from haverkorn et al @xcite who studied stokes parameters generated by diffuse polarized sources residing inside the milky way . their power spectra generally agree with the scale - invariance of the large scale tail of the power spectrum of @xcite , but also show evidence of a break in the spectrum , indicating a power law suppression of the rm power on small scales . such a break in the rm spectrum is also present in the model of minter and spangler @xcite who , by fitting to a small set of rm data from extragalactic sources , argued that the spectrum should be set by kolmogorov turbulence on small scales . the approximately scale - invariant shape of the galactic rm spectrum can obscure fr constraints on scale - invariant pmf . based on calculations in @xcite , one can estimate that a typical galactic rm of @xmath6 corresponds to an _ effective _ , or `` energy equivalent '' ( see eq . ( [ beff ] ) for the definition ) pmf strength of @xmath7 ng . this is well below the bounds from planck @xcite obtained using non - fr diagnostics and , as we confirm in this paper , below the levels detectable by planck via the mode - coupling statistics induced by fr . however , as shown in @xcite , future cmb experiments capable of detecting the weak lensing b - mode will be able to constrain a scale - invariant pmf of @xmath8 ng strength at @xmath9 ghz . operating at lower frequencies and combining fr information from several channels may further improve the constraints . clearly , with such high sensitivities the contribution of the galactic rm to the total fr signal will become important . in this paper , we investigate the imprint of the galactic rm on cmb observables , and its impact on detectability of the pmf via the eb and tb mode - coupling correlations . we start by introducing the necessary concepts and reviewing the known galactic rm measurements in sec . [ frcmb ] . in sec . [ sec : detection ] , we estimate detectability of the galactic rm by upcoming and future cmb experiments and forecast future bounds on the scale - invariant pmf under various assumptions . we conclude with a summary in sec . [ sec : summary ] . a cmb experiment measures stokes parameters in different directions on the sky , with parameters @xmath10 and @xmath11 quantifying linear polarization . if cmb photons pass through ionized regions permeated by magnetic fields , the direction of linear polarization is rotated by an angle @xcite @xmath12 where @xmath13 is the direction along the line of sight , @xmath14 is the differential optical depth , @xmath15 is the observed wavelength of the radiation , @xmath16 is the `` comoving '' magnetic field strength ( the physical field strength scales with the expansion as @xmath17 ) and @xmath18 is the comoving length element along the photon trajectory . the rotation measure , @xmath19 , is a frequency independent quantity used to describe the strength of fr . under the rotation of the polarization vector , the two stokes parameters transform as @xmath20 \includegraphics[height=0.46\textwidth]{clbb } \caption{the cmb b - mode spectrum from faraday rotation sourced at $ 30$~ghz by a scale - invariant primordial magnetic field of 1 ng strength ( red solid ) , by the full sky galactic magnetic field ( blue dot ) , and by the galactic field with planck 's sky mask ( $ f_{\rm sky}=0.6 $ ) . the black short - dash line is the input e - mode spectrum , the black dash - dot line is the contribution from inflationary gravitational waves with $ r=0.1 $ , while the black long - dash line is the expected contribution from gravitational lensing by large scale structure . } \label{fig : cl } \end{figure } the quadratic estimator method is based on the assumption of statistical isotropy of primordial perturbations . this implies statistical independence of primordial $ e_{l m}$ and $ b_{l m}$ for different $ lm$ pairs , { \it e.~g . } $ \langle e_{l m}^ * e_{l'm ' } \rangle = \delta_{ll ' } \delta_{mm ' } c_l^{ee}$ for the primordial e - mode . furthermore , if primordial fields are gaussian , all of their correlation functions can be expressed in terms of the power spectra . fr introduces correlations between unequal $ lm$ and generates connected four - point correlations . the corresponding non - gaussian signal can be used to extract the rotation angle . namely , given a cmb polarization map , one constructs quantities such as \cite{kamionkowski:2008fp , gluscevic:2009 mm } \be { \hat d}_{ll'}^{lm,{\rm map}}={4\pi \over ( 2l+1)(2l'+1 ) } \sum_{mm ' } b_{lm}^{\rm map}e_{l'm'}^{\rm map * } \xi_{lml'm'}^{lm } \label{eq : dll } \ee which is the minimum variance unbiased estimator for $ d_{ll'}^{lm } = 2 \alpha_{lm}c_l^{ee}h_{ll'}^l$ as shown in \cite{pullen:2007tu}. then , each pair of $ l$ and $ l'$ provides an estimate of $ \alpha_{lm}$ via \be [ { \hat \alpha}_{lm}]_{ll ' } = { { \hat d}_{ll'}^{lm,{\rm map } } \over 2c_l^{ee}h_{ll'}^l } \ . \label{alphallpr } \ee the minimum variance estimator $ { \hat \alpha}_{lm}$ is obtained from an appropriately weighted sum over estimators for each $ ll'$ pair . if all such pairs were statistically independent , the weighting would be given by the inverse variance of each estimator . however , one has to account for the correlation between the $ ll'$ and $ l'l$ pairs and a detailed derivation of the appropriate weighting can be found in \cite{kamionkowski:2008fp}. quantities such as $ { \hat d}_{ll'}^{lm,{\rm map}}$ can also be constructed from products of t and b or , more generally , all possible quadratic combinations , { \it i.e. } \{eb , be , tb , bt , te , et , ee\}. one can construct an estimator $ { \hat \alpha}_{lm}$ that utilizes all these combinations while accounting for the covariance between them \cite{gluscevic:2009mm}. in this paper , we opt to consider eb / be and tb / bt separately , since one of them is typically much more informative than other combinations . for further details of the method the reader is referred to \cite{kamionkowski:2008fp , gluscevic:2009mm}. we will assume a `` stochastic '' primordial magnetic field such that its value at any given location is described by a probability distribution function . this implies that one can only predict statistical properties of fr , such as the rotation power spectrum , $ c_l^{\alpha \alpha}$. the estimator for the power spectrum directly follows from the estimator of $ \alpha_{lm}$ and derives from four - point correlations , such as ebeb . in the next section , we will examine detectability of the rotation spectrum caused by the galactic and primordial magnetic fields . we note that the form of the quadratic estimator allows for contributions from the monopole ( $ l=0 $ ) and dipole ( $ l=1 $ ) of the fr field which , in principle , should not be ignored . however , the monopole is generally not expected for fr , since it would imply a non - zero magnetic charge enclosed by the cmb surface , while the dipole , corresponding to a uniform magnetic field , is strongly constrained for the pmf and is relatively unimportant for the galactic rm . in addition to mode - coupling correlations of eb and tb type , fr also contributes to the b - mode polarization spectrum , $ c_l^{bb}$. in fig.~\ref{fig : cl } we show the b- mode spectrum at $ 30 $ ghz due to fr by a pmf with a scale - invariant spectrum and effective field strength of $ 1~{\rm ng}$.the b - mode spectrum due to fr by the galactic field is also shown for two choices of sky cuts . for reference , we show the $ e$ mode auto - correlation spectrum which acts as a source for the fr b modes , the b modes from inflationary gravitational waves with $ r=0.1 $ , and the expected contribution from weak lensing ( wl ) . as can be seen from fig.~ \ref{fig : cl } , the shape of the fr induced b mode spectrum largely mimics that of the e- mode , and the galactic contribution has a very similar shape to that of the scale- invariant pmf . a more detailed discussion of the fr induced b - mode spectrum can be found in \cite{pogosian:2011qv}. as shown in sec.~\ref{sec : detection } , the fr signatures of the galactic or scale - invariant pmf are always less visible in the b - mode spectrum compared to the eb quadratic estimator , with the latter promising to provide the tightest constraints . we note that for causally generated pmf , which have blue spectra with most of the power concentrated at the dissipation scale , the signal - to - noise ratio ( snr ) for detecting fr in bb is higher than that in the quadratic estimators \cite{yadav:2012uz } , although the resulting constraints are not competitive with other probes of causal fields . in what follows , we will focus on fr signatures of scale - invariant pmf only . \subsection{a scale - invariant primordial magnetic field } \label{pmf } assuming statistical homogeneity and isotropy , the magnetic field correlation function in fourier space is~\cite{1975mit .. bookr .... m } \begin{eqnarray } \langle { b}^ * _ i ( { \bm k } ) { b } _ j ( { \bm k } ' ) \rangle & = & \left [ \left(\delta _ { ij } - { k_i k_j \over k^2 } \right ) s ( k ) + i \epsilon_{ijl } { k_l \over k } a(k ) \right ] \nonumber \\ & & \hskip 2 cm \times ( 2 \pi)^3 \delta^3 ( { \bm k } -{\bm k } ' ) \end{eqnarray } where repeated indices are summed over , $ b_i({\bm k})$ denotes the fourier transform of the magnetic field at wave vector $ { \bm k}$ , $ k=|{\bm k}|$ , and $ s(k)$ and $ a(k)$ are the symmetric and antisymmetric ( helical ) parts of the magnetic field power spectrum . the helical part of the power spectrum does not play a role in faraday rotation ; only the symmetric part of the spectrum is relevant for us . on scales larger than the inertial scale , { \it i.e. } $ k < k_i$ , $ s(k)$ will fall off as a power law . on scales smaller than a dissipative scale , { \it i.e. } $ k > k_d$ , the spectrum will get sharply cut - off . a second power - law behavior is possible for $ k_i < k < k_d$. however , we shall assume that $ k_i \approx k_d$ , so that the magnetic field spectrum is given by a single power law behavior at all scales larger than the dissipation scale . these features can be summarized by \cite{pogosian:2011qv } \begin{eqnarray } s(k ) = \begin{cases } \omega_{b\gamma } { \rho}_\gamma \frac{32\pi^3 n}{k_i^3 } \left ( \frac{k}{k_i } \right ) ^{2n-3 } , & \mbox{$0<k < k_{\rm diss}$}\\ 0 , & \mbox{$k_{\rm diss } < k$ } \end{cases } \label{eq : pb } \end{eqnarray } where $ \omega_{b\gamma}$ is the ratio of cosmological magnetic and photon energy densities , and $ \rho_\gamma$ is the photon energy density . for convenience , we define an `` effective '' , or `` energy equivalent '' , magnetic field strength \begin{equation } b_{\rm eff } \equiv ( 8\pi \omega_{b\gamma } \rho_\gamma ) ^{1/2 } = 3.25\times 10^{-6}\sqrt{\omega_{b\gamma } } ~{\rm gauss}. \label{beff}\ ] ] a uniform magnetic field of strength @xmath21 will have the same energy density as a stochastic magnetic field with spectrum in eq . ( [ eq : pb ] ) . scale invariance for @xmath22 corresponds to @xmath23 , in which case the expressions in eq . ( [ eq : pb ] ) are not well defined . one reason is that the energy density diverges logarithmically for a strictly scale invariant field . we get around this issue by only considering approximately scale invariant magnetic fields for which @xmath24 is small . note that , for scale invariant fields , @xmath21 is equal to the field strength smoothed over a scale @xmath25 ( if it is larger than the dissipation scale ) , or @xmath26 , frequently used in the literature ( _ e.g. _ in @xcite ) . as we have already noted , the galactic rm spectrum is approximately scale invariant . the rm induced by a scale invariant pmf is also scale invariant @xcite . hence the amplitudes of the galactic and pmf rotation measure spectra can be related directly for a scale invariant pmf . for a scale - invariant angular spectrum @xmath27 , the quantity @xmath28 is constant . hence , one can define an amplitude @xmath29 for the range of @xmath30 over which the rm spectrum is approximately scale - invariant . one can see from fig . [ fig : rmcl ] that this holds for the galactic @xmath19 for @xmath31 . similarly , one can define an amplitude a_= the relation between the faraday rotation angle in radians and the rm is given by @xmath32 where @xmath33 . for example , a characteristic galactic rm of @xmath34 at @xmath35 gives a rotation angle of @xmath36 rad , or @xmath37 . the amplitudes of @xmath38 in radians and @xmath39 in @xmath40 are therefore also related via a_= 10 ^ -4 ^-2_30 a_rm m^2 . [ alpharm ] to gain some intuition into the relative importance of the galactic rm prior to more detail forecasts in sec [ detect - pmf ] , let us estimate the @xmath21 of scale - invariant pmf that gives rm comparable to that of our galaxy . for this , we can compare the amplitude @xmath38 from pmf to the amplitude for galactic spectra at the same frequency . the relation between pmf and @xmath38 involves integration along the line of sight ( eq . ( [ alpha - fr ] ) ) . rather than trying to find an approximate analytical solution , we use exact spectra found numerically in @xcite . from fig . 1 of ref . @xcite , one can infer that for a scale invariant pmf with @xmath41 , @xmath42 rad@xmath43 at @xmath35 . we can then use eqs . ( [ alpharm ] ) and eq . ( [ beff ] ) to relate @xmath21 to @xmath39 : b_eff 0.021 a_rm ng m^2 rad . [ eq : beffrm ] from fig . [ fig : rmcl ] , we can see that @xmath44 for the galactic rm spectrum obtained without any sky cuts . it follows from eq . ( [ eq : beffrm ] ) that this corresponds to a scale - invariant pmf with @xmath45 . such pmf are well below levels detectable by planck through their fr signal , but can be detected at high significance by future sub - orbital and space - based experiments @xcite . we can also estimate the peak value of the b - mode spectrum , @xmath46 , in ( @xmath47k)@xmath43 , for a given @xmath39 and frequency . from fig . [ fig : cl ] , we see that @xmath48k)@xmath49 for a scale invariant pmf of @xmath50 ng strength . using this fact along with eq . ( [ eq : beffrm ] ) , we can deduce p_b ^2 ( k)^2 ^2 ( k)^2 . thus , for a typical galactic @xmath39 of @xmath51 rad / m@xmath43 , at @xmath51 ghz , we should expect a peak in the b - mode spectrum of about @xmath52 ( @xmath47k)@xmath43 , which agrees with the blue dotted line in fig . [ fig : cl ] corresponding to the unmasked galactic rm . in practice , the galactic plane is masked in all cmb experiments , _ e.g. _ the mask applied to planck s @xmath51 ghz map only uses @xmath7 of the sky . this significantly reduces the amplitude of the galactic rm spectrum and , as a consequence , the b - mode spectrum . this is shown as a blue dashed line in fig . [ fig : cl ] . one can see that the galactic b - mode is at least one , and more likely two , orders of magnitude below the weak lensing signal , so there is almost no hope of seeing it even at a frequency as low as @xmath51 ghz . going a factor of @xmath53 higher in frequency , which is more typical in cmb studies , results in another factor of @xmath54 decrease in the b - mode spectrum . one of the conclusions of this section is that the galactic rm is practically invisible in the bb auto - correlation . hence , we should look at the mode - coupling estimators , as they are known to be much more sensitive to fr @xcite . as shown in @xcite , future experiments can easily probe @xmath55 strength pmf fields with the mode - coupling estimators . sky cut . the bottom panel also shows lines of symmetric cuts corresponding to @xmath56 and @xmath57 . , scaledwidth=85.0% ] , obtained from the rm map of oppermann et al @xcite with different cuts . shown are the rm spectra corresponding to , from top to bottom , a scale - invariant pmf of @xmath50 ng , galaxy with no sky cut , with a mask used by planck for their @xmath58 ghz map , a planck mask for the @xmath51 ghz map , and symmetric cuts corresponding to @xmath56 and @xmath57 . , scaledwidth=40.0% ] a rotation measure ( rm ) map of milky way has been put together by oppermann et al @xcite based on an extensive catalog of faraday rotation data of compact extragalactic polarized radio sources . the rm is strongest along the galactic plane , but remains significant at the poles . however , the latitude dependence of the rms rm disappears at latitudes higher than about @xmath59 , where the dependence flattens out at a rms value of about @xmath60 rad / m@xmath43 . oppermann et al @xcite also presented the angular power spectrum of rm calculated after factoring out the latitude dependence . they find that @xmath61 at @xmath62 , which is very close to a scale - invariant spectrum . some insight into galactic rm on smaller scales is provided by haverkorn et al @xcite . they studied stokes parameters , generated by diffuse polarized sources residing inside the galaxy , at several frequencies around 350 mhz . they focused on two separate patches away from the galactic plane , each covering about @xmath63 sq . the rotation measures in both were less than @xmath60 rad / m@xmath43 . however , since they are looking at sources inside the galaxy , they only see a fraction of the total rm along each line of sight . thus , their numbers can not be directly compared to rm of @xcite . still their power spectrum gives a sense for the scale dependence of the total rm , and generally agrees with the flat nature of the large scale tail of the power spectrum of @xcite . at a scale of about @xmath64 ( @xmath65 ) , they see evidence of a break in the spectrum , indicating a power law suppression of the rm power on small scales . such a break in the spectrum is also present in the model of minter and spangler @xcite , whose analysis is based on a small set of rm data from extragalactic sources . they argued that on small scales the spectrum should be due to kolmogorov turbulence . the corresponding scaling of @xmath66 would be @xmath67 . to summarize , based on the evidence in @xcite , if we were to measure rm in a few degree patches near galactic poles , we would expect a rm spectrum that is roughly scale invariant at @xmath68 turning into a @xmath67 tail on smaller scales . as we will show in the next section , the quadratic estimator of the fr angle is largely uninformative for @xmath69 . hence , we do not need a rm map with a resolution higher than that of oppermann et al @xcite . note that the cmb polarization even in a small patch around the poles is affected by very long wavelength modes of variations of the rm . in terms of eq . ( [ eq : blm ] ) , lowest multipoles of @xmath70 contribute significantly to the much higher multipoles of @xmath71 . thus , to predict the observability of the fr in cmb in a small patch one can not simply use the high @xmath72 portion of the rm spectrum . we should use the full rm spectrum obtained from the rm after applying a mask normally used to block the galactic plane . in fig . [ fig : rmmap ] we show the rm map for our galaxy obtained from the faraday rotation of extragalactic sources in units of rad / m@xmath73 by oppermann et al @xcite . the top panel is the full map . the mid - panel shows the uncertainty in rm in rad / m@xmath73 . most of the signal and uncertainty come from the galactic disk . the bottom panel shows the rm map with part of the sky removed with a mask used for planck s mask @xmath51 ghz map @xcite . the bottom panel also shows lines of symmetric latitude cuts corresponding to sky fractions of @xmath57 and @xmath74 . in fig . [ fig : rmcl ] we present the power spectra @xmath75 of the galactic rm under different symmetric sky cuts and two planck mask cuts : one for @xmath51 ghz with @xmath76 , and one for @xmath58 ghz with @xmath77 . the figure also shows the rm spectrum from a scale - invariant pmf with @xmath78 ng . to calculate the galactic rm spectra , we first use healpix @xcite to find _ lm = d w ( ) r ( ) y^*_lm ( ) , [ tildealm ] where @xmath79 is the rm field , and @xmath80 is the mask which is @xmath81 if a pixel is masked and @xmath50 if it is not . we then evaluate c^rm_l = 1 f_sky(2l+1 ) _ m =- l^l a^*_lm a_lm , [ eq : clrm ] which are the so - called `` pseudo-@xmath82 '' rescaled by @xmath83 . the commonly used procedure for estimating @xmath82 s from a partial sky is detailed in @xcite . it assumes statistical isotropy and involves calculating a conversion matrix that relates `` pseudo-@xmath82 '' , obtained from a partial sky , to the `` full - sky '' @xmath82 . in eq . ( [ eq : clrm ] ) , as a crude approximation , we ignore the mode - coupling and replace the matrix with a rescaling by the @xmath83 factor . note that the galactic rm map is not isotropic and the `` full - sky '' spectrum reconstructed from a given patch is not the same as the actual spectrum evaluated from the full - sky . thus , our estimates in sec . [ sec : detection ] of the detectability of the galactic rm spectrum , and of the detectability of the pmf with the galactic rm as a foreground , are specific to particular masks . the error bars in fig . [ fig : rmcl ] receive contributions from the uncertainty in the measurements of the rm map and from the cosmic variance . we are interested in answering two questions : 1)_how well can cmb polarization experiments measure the galactic rm spectrum ? _ 2)_what bounds on a scale - invariant pmf can be placed from measurements of fr ? _ both contributions , the galactic fr and the primordial fr , contribute in the same way to the quadratic estimators discussed in sec . [ basics ] . let @xmath84 be the estimator of the rotation angle that one reconstructs from linear combinations of quadratic terms @xmath85 . the variance in @xmath84 , for a statistically isotropic rotation angle , is defined as @xmath86 with @xmath87 given by @xcite ^2_l = c_l^ + n_l^eb , [ eq : variance ] where @xmath88 is the rotation power spectrum , and @xmath89 is the part of the variance that does not depend on the rotation angle and contains contributions from variances of individual estimators @xmath90 ( see eq . ( [ eq : dll ] ) ) , while accounting for their covariance . it is given by @xcite ( n_l^eb)^-1 = 1 _ ( 2 + 1)(2+1)[c^ee_ h^l_]^2 _ ^ee c_^bb , [ eq : ebnoise ] where @xmath91even , and ^xx_^prim_+f_dl[c^xx]^wl_+i^xx _ , [ clvariance ] is the measured spectrum , with @xmath92 standing for either ee or bb , that includes the primordial contribution , @xmath93^{\rm prim}$ ] , the weak lensing contribution , @xmath93^{\rm wl}$ ] , and the noise term associated with the experiment , @xmath94 . we allow for the possibility of partial subtraction of the weak lensing contribution by introducing a de - lensing fraction , @xmath95 , in ( [ clvariance ] ) . the efficiency of de - lensing depends on the method used to de - lens @xcite , as well as the noise and the resolution parameters of the experiment . in @xcite it was found that the quadratic estimator method of de - lensing @xcite , which is similar but orthogonal @xcite to the one we use to extract the fr rotation , can reduce the lensing contribution to bb by up to a factor of @xmath96 as it reaches the white noise limit . on the other hand , the so - called iterative method @xcite can further reduce the lensing contribution by a factor of @xmath60 or larger . in our analysis we have only considered two cases one with no de - lensing ( @xmath97 ) and one with a factor of @xmath9 reduction of the lensing b - mode ( @xmath98 ) . we adopt the latter optimistic assumption with the aim of illustrating the relative significance of lensing contamination for an experiment of given noise and resolution . note that the numerator in eq . ( [ eq : ebnoise ] ) contains the unlensed e - mode spectrum , while the denominator includes the lensing contribution to e- and b - mode spectra . predictions for both are readily obtained from camb @xcite for a given cosmological model . the instrumental noise term , @xmath94 , accounts for the detector noise as well as the suppression of the spectrum on scales smaller than the width of the beam @xcite : i^ee_=i^bb_= ^2_p ( ^2 ^2_fwhm / 82 ) , [ eq : polnoise ] where @xmath99 quantifies detector noise associated with measurements of cmb polarization and @xmath100 is the full width at half maximum of the gaussian beam . in the case of tb , the expression for the variance analogous to eq . ( [ eq : ebnoise ] ) is @xcite ( n_l^tb)^-1 = 1 _ ( 2 + 1)(2+1)[c^te_ h^l_]^2 _ ^tt c_^bb . [ eq : tbnoise ] eq . ( [ variance - alm ] ) assumes that the distribution of the rotation angle is isotropic , and the angular brackets denote an ensemble average over many realizations of @xmath0 as well as realizations of the primordial cmb sky . this is a well - motivated assumption for the pmf and for the galactic rm near galactic poles . away from the poles , the galactic rm shows a strong dependence on the galactic latitude @xcite . this means that the forecasted variance can differ from the actual uncertainty for larger values of @xmath83 , although we expect it to be of the correct order of magnitude . for the galactic rm map , we take the rotation angular spectrum to be @xmath101 , with @xmath102 calculated from the rm map of @xcite after applying sky cuts , as plotted in fig . [ fig : rmcl ] and explained in sec . [ galaxy ] . we will consider several sets of experimental parameters corresponding to ongoing , future and hypothetical experiments . namely , we consider planck s @xmath51 ghz lfi and @xmath9 ghz hfi channels based on actual performance characteristics @xcite , polarbear @xcite at @xmath103 ghz with parameters compiled in @xcite , quiet phase ii @xcite at @xmath104 ghz using the parameters compiled in @xcite , @xmath51 , @xmath105 , @xmath58 and @xmath9 ghz channels of a proposed cmbpol satellite @xcite , as well as an optimistic hypothetical sub - orbital and space - based experiments at @xmath51 and @xmath103 ghz . the assumed sky coverage ( @xmath83 ) , resolution ( @xmath100 ) , and instrumental noise ( @xmath99 ) parameters are listed in tables [ table : galaxy ] and [ table : pmf ] . for an optimistic @xmath51 ghz sub - orbital experiment with ( solid blue ) and without ( blue dot ) de - lensing , as well as for a hypothetical future @xmath51 ghz space probe with ( red short dash ) and without ( red long dash ) de - lensing ( @xmath98 ) . the spiky nature of the plot is due to the spikiness of the rm spectra , as seen in fig . [ fig : rmcl ] . experimental parameters assumed in this plot are given in the text and in table [ table : galaxy].,scaledwidth=45.0% ] [ cols="^,^,^,^,^,^,^",options="header " , ] -1 cm the snr of the detection of the primordial fr rotation angle is given by ( s n ) ^2_xb = _ l=1^l_max f_sky ( 2l+1 ) [ c_l^,pmf]^2 2[c_l^,pmf + f_dgc_l^,g+n^xb_l]^2 . [ eq : ebsnrp ] where @xmath106 stands for either e or t , depending on whether one uses the eb or tb estimator , @xmath107 is the primordial contribution , while @xmath108 is the `` de - galaxing '' factor , _ i.e. _ the fraction of the galactic rotation angle spectrum that is known from other sources and can be subtracted . as in the case of the galactic rm , the snr receives a non - zero contribution from all multipoles @xmath30 , including the smallest , even for a small sky coverage . in the case when a large fraction of the sky is available , there is an optimal cut of the sky , @xmath109 , for which the s / n is maximal . this is because the galactic rm contribution is weaker when large parts of the sky are cut , but there are more available modes when more of the sky is kept . in what follows , we will state when values of @xmath109 are different from @xmath83 . in fig . [ fig : dsnpmg ] , we plot contributions per @xmath110 to the net snr in detection of the fr from a scale - invariant pmf with @xmath111 ng for the same two hypothetical experiments considered in fig . [ fig : dsngal ] . in addition to the four cases considered in fig . [ fig : dsngal ] , the green dot - dash line shows the case when the galactic contribution is partially subtracted for the space - based probe with @xmath112 . a few observations can be made from this plot . first is that both experiments can detect a @xmath8 ng pmf at high significance , which constitutes a big improvement on prospects of detecting a pmf in cmb via non - fr signatures @xcite . second is that de - lensing makes a significant difference for both the space - based and sub - orbital experiments . third is that subtracting the galactic rm moderately improves the detection at multipoles in the @xmath113 range . finally , we can see that most of the information comes from @xmath114 , meaning that future cmb experiments can reconstruct fr maps corresponding to pmf of @xmath8 ng strength up to a degree resolution . in table [ table : pmf ] we present our forecasts for @xmath115 bounds on @xmath21 that can be expected from various ongoing and future cmb experiments . we present results obtained from the eb estimator , because they lead to the strongest bounds , except for the case of planck s @xmath51 ghz channel , for which the tb estimator gives better constraints . for each experiment , we check the effect of de - lensing ( + dl ) with @xmath98 as well as further partial subtraction of the galactic rm ( + dl+dg ) with @xmath112 . the optimal sky cuts are indicated when relevant . one of the interesting facts one can extract from this table is that very competitive bounds can be placed by polarbear and quiet , while even @xmath116 fields can be constrained in principle by future experiments . to show the dependence of the pmf bound on the instrumental noise and resolution , we plot contours of constant @xmath115 bounds on @xmath21 in fig . [ fig : contour ] . the three panels show the cases with and without de - lensing and with an additional partial subtraction of the galactic rm . an optimal sky cut was used for each set of parameters . we have studied the detectability of a scale - invariant pmf due to fr of the cmb using polarization correlators in the presence of the foreground imposed by the milky way magnetic field . we have found that the galactic rm is not a serious barrier to observe a scale - invariant pmf and that the eb quadratic correlator is in general more powerful than the bb corelator . we have also studied the detectability of the pmf as a function of the sky coverage . we find that suborbital experiments can be almost as effective as space borne experiments , and the obstruction caused by the galaxy is relatively weak if the observed patch is near the poles . also , as the mode - coupling correlations of cmb are mostly sourced by the largest scale features ( low @xmath30 ) of the rotation measure , a full sky cmb map is not necessary to access the pmf on the largest scales . our results on the sensitivities of various experiments to the galactic magnetic field are summarized in table i. similarly in table ii we show the effective magnetic field strength that can be constrained by those experiments , taking into account the lensing of the cmb and the fr due to the milky way . the dependence of the observable pmf to the parameters of the experiment are shown in fig . [ fig : contour ] . here we have focussed on the observability of cmb fr at a single frequency . cross - correlating polarization maps at multiple frequencies with comparable sensitivity to fr can further boost the significance of detection . we will address this possibility in a future publication . also , frequency dependence of polarization can be used to separate out fr induced b - modes from those induced by gravitational waves . we thank the two anonymous referees for their careful reading of the initially submitted version and constructive suggestions that helped to improve the paper . we are grateful to niels oppermann and collaborators @xcite for making their rotation measure maps publicly available at @xcite . some of the results in this paper have been derived using the healpix @xcite package . lp and tv benefited from previous collaborations with amit yadav . we thank steven gratton , yin - zhe ma , phil mauskopf and vlad stolyarov for discussions and helpful comments . sd is supported by a sese exploration postdoctoral fellowship ; lp is supported by an nserc discovery grant ; tv is supported by the doe at arizona state university . d. grasso and h. r. rubinstein , phys . rept . * 348 * , 163 ( 2001 ) [ astro - ph/0009061 ] . l. m. widrow , rev . phys . * 74 * , 775 ( 2002 ) [ astro - ph/0207240 ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1303.5076 [ astro-ph.co ] . d. paoletti , f. finelli and f. paci , mon . not . * 396 * ( 2009 ) 523 [ arxiv:0811.0230 [ astro - ph ] ] . d. paoletti and f. finelli , arxiv:1208.2625 [ astro-ph.co ] . m. s. turner and l. m. widrow , phys . d * 37 * , 2743 ( 1988 ) . b. ratra , astrophys . j. * 391 * , l1 ( 1992 ) . t. kahniashvili , y. maravin , a. natarajan , n. battaglia and a. g. tevzadze , astrophys . j. * 770 * , 47 ( 2013 ) [ arxiv:1211.2769 [ astro-ph.co ] ] . r. a. c. croft , d. h. weinberg , m. bolte , s. burles , l. hernquist , n. katz , d. kirkman and d. tytler , astrophys . j. * 581 * , 20 ( 2002 ) [ astro - ph/0012324 ] . t. kahniashvili , y. maravin and a. kosowsky , phys . d * 80 * , 023009 ( 2009 ) [ arxiv:0806.1876 [ astro - ph ] ] . a. yadav , l. pogosian and t. vachaspati , phys . d * 86 * , 123009 ( 2012 ) [ arxiv:1207.3356 [ astro-ph.co ] ] . t. vachaspati , phys . b * 265 * , 258 ( 1991 ) . r. durrer and c. caprini , jcap * 0311 * , 010 ( 2003 ) [ astro - ph/0305059 ] . k. jedamzik and g. sigl , phys . d * 83 * , 103005 ( 2011 ) [ arxiv:1012.4794 [ astro-ph.co ] ] . k. jedamzik and t. abel , arxiv:1108.2517 [ astro-ph.co ] . n. oppermann , h. junklewitz , g. robbers , m. r. bell , t. a. ensslin , a. bonafede , r. braun and j. c. brown _ et al . _ , arxiv:1111.6186 [ astro-ph.ga ] . m. haverkorn , p. katgert and a. g. de bruyn , astron . astrophys . * 403 * , 1045 ( 2003 ) [ astro - ph/0303644 ] . l. pogosian , a. p. s. yadav , y. -f . ng and t. vachaspati , phys . d * 84 * , 043530 ( 2011 ) [ erratum - ibid . d * 84 * , 089903 ( 2011 ) ] [ arxiv:1106.1438 [ astro-ph.co ] ] . a. kosowsky and a. loeb , astrophys . j. * 469 * , 1 ( 1996 ) [ astro - ph/9601055 ] . d. d. harari , j. d. hayward and m. zaldarriaga , phys . d * 55 * , 1841 ( 1997 ) [ astro - ph/9608098 ] . m. kamionkowski , phys . lett . * 102 * , 111302 ( 2009 ) [ arxiv:0810.1286 [ astro - ph ] ] . s. yadav , r. biswas , 1 , m. su and m. zaldarriaga , phys . d * 79 * , 123009 ( 2009 ) [ arxiv:0902.4466 [ astro-ph.co ] ] . v. gluscevic , m. kamionkowski and a. cooray , phys . d * 80 * , 023510 ( 2009 ) [ arxiv:0905.1687 [ astro-ph.co ] ] . v. gluscevic , d. hanson , m. kamionkowski and c. m. hirata , arxiv:1206.5546 [ astro-ph.co ] . s. yadav , m. shimon and b. g. keating , phys . d * 86 * , 083002 ( 2012 ) [ arxiv:1207.6640 [ astro-ph.co ] ] . w. hu and t. okamoto , astrophys . j. * 574 * , 566 ( 2002 ) [ astro - ph/0111606 ] . a. r. pullen and m. kamionkowski , phys . d * 76 * , 103529 ( 2007 ) [ arxiv:0709.1144 [ astro - ph ] ] . monin a. s. , iaglom a. m. , statistical fluid mechanics : mechanics of turbulence . volume 2 , cambridge , mass . , mit press , ( 1975 ) . k. m. gorski , e. hivon , a. j. banday , b. d. wandelt , f. k. hansen , m. reinecke and m. bartelman , astrophys . j. * 622 * , 759 ( 2005 ) [ astro - ph/0409513 ] . e. hivon , k. m. gorski , c. b. netterfield , b. p. crill , s. prunet and f. hansen , astrophys . j. * 567 * , 2 ( 2002 ) [ astro - ph/0105302 ] . c. m. hirata and u. seljak , phys . d * 67 * , 043001 ( 2003 ) [ astro - ph/0209489 ] . u. seljak and c. m. hirata , phys . d * 69 * , 043005 ( 2004 ) [ astro - ph/0310163 ] . http://camb.info/ a. lewis , a. challinor and a. lasenby , astrophys . j. * 538 * , 473 ( 2000 ) [ arxiv : astro - ph/9911177 ] . j. r. bond , g. efstathiou and m. tegmark , mon . not . astron . soc . * 291 * , l33 ( 1997 ) [ astro - ph/9702100 ] . a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1303.5062 [ astro-ph.co ] . b. keating , s. moyerman , d. boettger , j. edwards , g. fuller , f. matsuda , n. miller and h. paar _ et al . _ , arxiv:1110.2101 [ astro-ph.co ] . y. -z . ma , w. zhao and m. l. brown , jcap * 1010 * , 007 ( 2010 ) [ arxiv:1007.2396 [ astro-ph.co ] ] . d. samtleben [ quiet collaboration ] , nuovo cim . b * 122 * , 1353 ( 2007 ) [ arxiv:0802.2657 [ astro - ph ] ] . d. baumann _ et al . _ [ cmbpol study team collaboration ] , aip conf . proc . * 1141 * , 10 ( 2009 ) [ arxiv:0811.3919 [ astro - ph ] ] .

the problem of reconstructing binary matrices from their row and column sums is a classical inverse task in combinatorics ; see @xcite . even though the term was introduced much later it can be seen as a root of _ discrete tomography _ ; see @xcite , and the collected editions @xcite , @xcite . of particular relevance for the present paper are two well - known observations : ( 1 ) the question of consistency of the data i.e. , the question whether there exists a matrix whose row and column sums coincide with the given data , can be solved in polynomial time . ( 2 ) typically , its row and column sums do not determine the underlying matrix uniquely . see @xcite for characterizations of the ` rare ' cases of uniqueness . in the present paper we address the second issue by adding additional window constraints , specifying ( or giving bounds on ) how many points there are in certain minors of the matrix . these constraints come up naturally in _ dynamic discrete tomography _ @xcite . an application of particular relevance in physics is that of particle tracking ; see @xcite , @xcite . here , the positions of particles over time are to be reconstructed from two ( sometimes more ) high speed camera images . window constraints are a natural way of modeling additional physical information for instance on the speed of the particles ; see @xcite and the background literature given there for further information . here we are taking a complexity theoretical view commemorating observation ( 1 ) . in fact , when adding various kinds of window constraints , we are interested in the boundary between polynomial - time solvability and @xmath1-hardness . intuitively speaking , we ask which kinds of additional constraints can be added without imposing a significant extra computational effort . or , phrased differently , what is the computational price to pay for reducing the number of solutions by utilizing additional window information . we will focus on the effect of three different parameters : @xmath2 corresponds to the _ size _ of the window , @xmath3 is the _ number of @xmath4 s _ in the nonzero minors , and @xmath5 specifies the allowed positions of @xmath4 s in the windows , referred to as the _ pattern _ of the window . the choice for these parameters will specify the given problem rec@xmath6 which is formally introduced in section [ sec - results ] . hence @xmath7 @xmath8 and @xmath5 are given beforehand ( i.e. , are not part of the input ) . as it will turn out , the problem exhibits various unexpected complexity jumps . omitting technical details ( which are all given in section [ sec - results ] ) some of these jumps can be summarized as follows ; see table [ table : sumtable1 ] . for @xmath9 the problems are in @xmath10 regardless on how the other parameters are set ; see theorem [ thm : tract]([thm : tract : i ] ) . for @xmath11 it depends on @xmath3 and @xmath5 whether the problems are in @xmath10 or @xmath1-hard ; see theorems [ thm : tract]([thm : tract : ii]),([thm : tract : iii ] ) and [ thm : np ] . for @xmath11 , some values of @xmath3 render the problems tractable while others make them @xmath1-hard ; see theorems [ thm : tract]([thm : tract : ii ] ) and [ thm : np]([thm : np : ii ] ) even if the patterns are not restricted at all . adding a pattern constraint may turn an otherwise @xmath1-hard problem into a polynomial time solvable problem ; see theorems [ thm : np]([thm : np : ii ] ) and [ thm : tract]([thm : tract : iii ] ) . the reverse complexity jump , however , can also be observed ; see theorems [ thm : tract]([thm : tract : ii ] ) and [ thm : np]([thm : np : i ] ) . the present paper is organized as follows : section [ sec - results ] will introduce our notation and state the main results . the proofs of our tractability results are given in section [ sec - tractability ] while the @xmath1-hardness results are proved in section [ sec - intractability ] . section [ sect : finalremarks ] contains some final remarks . in particular , it provides a ( potentially also quite interesting ) extension of our main problem and collects related implications of our main results . we begin with some standard notation . let @xmath12 @xmath13 and @xmath14 denote the set of integers , natural numbers , and non - negative integers , respectively . for @xmath15 set @xmath16 , @xmath17 @xmath18:=\{1,\dots , k\}$ ] , and @xmath18_0:=[k]\cup\{0\}.$ ] the support of a vector @xmath19 is defined as @xmath20 . with @xmath21 we denote the all - ones vector of the corresponding dimension . the cardinality of a finite set @xmath22 is denoted by @xmath23 we will use the notational convention that specific settings of variables and parameters are signified by a superscript @xmath24 in this paper we often refer to points ( or variables ) @xmath25 where @xmath26 are points of the integer grid @xmath27 . then , @xmath28 and @xmath29 denote the @xmath30- and @xmath31-coordinates of @xmath26 , respectively . in the grid @xmath32\times[n],$ ] the set @xmath33 $ ] and @xmath32\times\{j\}$ ] is called _ column @xmath34 _ and _ row @xmath35 _ respectively . for @xmath36 we refer to @xmath37 $ ] and @xmath32\times\{(j-1)k+1,\dots , jk\}$ ] as _ vertical strip @xmath34 ( of width @xmath2 ) _ and _ horizontal strip @xmath38 ( of width @xmath2 ) _ , respectively . of course , the standard matrix notation can be obtained from our notation by the coordinate transformation @xmath39 generally , any subset of the @xmath32\times[n]$ ] grid is called a _ window_. windows of the form @xmath40\times[c , d])\cap\mathbb{z}^2,$ ] with @xmath41 and @xmath42 @xmath43 are called _ boxes . _ defining for any @xmath36 and @xmath44 the set of _ ( lower - left ) corner points _ @xmath45\times [ n])\cap(k\mathbb{n}_0 + 1)^2,$ ] we call any box @xmath46 ^ 2 $ ] with @xmath47 a _ block _ ; see figure [ fig : ill ] for an illustration . ( here we depict the structure both as point set and as pixels . in the following we restrict the figures to pixel images , which we find more intuitive . ) the blocks form a partition of @xmath32\times[n],$ ] i.e. , @xmath48\times[n].$ ] in the main part of this paper , we consider such non - overlapping blocks , which play also a role in super - resolution imaging @xcite . in section [ sect : finalremarks ] we consider also other windows , which may be positioned at other places than those defined by @xmath49 next we introduce three patterns that we will study in detail since they exhibit already the general complexity jump behavior we are particular interested in . the first pattern @xmath50 is unconstrained , i.e. , does not pose any additional restrictions on the positions of @xmath4 s . the second pattern @xmath51 forces all elements in the @xmath52 block to be @xmath53 except possibly for the two entries in the lower - left and upper - right corner . the third pattern @xmath54 excludes all patterns that admit more than one @xmath4 in each row of the @xmath52 block ( see also figure [ fig : c1 ] ) . here are the formal definitions . for @xmath55 let @xmath56_0 ^ 2}$ ] denote the power set of @xmath57_0 ^ 2.$ ] then we set @xmath58_0 ^ 2},\\ p(k,1)&:=\{\{(0,0)\},\{(k-1,k-1)\}\},\\ p(k,2)&:=\{m\in2^{[k-1]_0 ^ 2 } \::\ : |m\cap\left([k-1]_0\times\{j\}\right)|\leq 1 \textnormal { for all } j\in [ k]_0.\}\end{aligned}\ ] ] further , for @xmath59 and @xmath60,j\in[n]}$ ] we set @xmath61 for @xmath62 and @xmath63 we define now the following problem rec@xmath64 rec@xmath65 + @xmath66 & \textnormal{task:}\quad & & \omit \rlap{find $ \xi_{p , q}\in\{0,1\},$ \:$(p , q)\in[m]\times[n],$ with } \\ \displaystyle & & & \sum_{\mathclap{p\in[m ] } } \xi_{p , q}&&=r_q , & & q\in[n ] , & & \textnormal{(row sums ) } \\ \displaystyle & & & \sum_{q\in[n ] } \xi_{p , q}&&=c_p , & & p \in [ m ] , & & \textnormal{(column sums)}\\ \displaystyle & & & \sum_{\mathclap{(p , q)\in b_k(i , j ) } } \xi_{p , q } & & \leq v(i , j ) , & & ( i , j)\in c(m , n , k ) , \quad & & \textnormal{(block constraints)}\\ \displaystyle & & & \textnormal{pat}_k(x , i , j)&&\in p(k , t),\qquad & & ( i , j)\in c(m , n , k),\qquad & & \textnormal{(pattern constraints),}\\[1.2ex ] & & & \omit \rlap{or decide that no such solution exists.}\\\end{aligned}\ ] ] in other words , we ask for @xmath67-solutions that satisfy given row and column sums , _ block constraints _ of the form @xmath68 with given @xmath69 and _ pattern constraints _ that restrict the potential locations of the 1 s in each block . we remark that the @xmath70 @xmath71 can be viewed as some part of prior knowledge . in particle tracking @xcite , for instance , the @xmath72 may reflect prior knowledge about physically meaningful particle trajectories ; see also @xcite . our main results show that the computational complexity of @xmath73 may change drastically when @xmath7 @xmath8 or @xmath5 is varied . [ thm : tract ] a. rec@xmath74 for any @xmath75 and @xmath76 [ thm : tract : i ] b. rec@xmath77 for any @xmath78 [ thm : tract : ii ] c. rec@xmath79 for any @xmath80 and @xmath81 [ thm : tract : iii ] [ thm : np ] a. rec@xmath82-hard for any @xmath78 [ thm : np : i ] b. rec@xmath83-hard for any @xmath78 [ thm : np : ii ] the most notable changes are summarized in table [ table : sumtable1 ] . some of these changes may at first glance seem somewhat counterintuitive . for instance , restricting the solution space via pattern constraints turns the @xmath0-hard problem rec@xmath84 @xmath85 into the polynomial time solvable problem rec@xmath86 figure [ fig : c1 ] depicts the possible types of blocks in rec@xmath87 conversely , additional pattern constraints convert the tractable problem rec@xmath88 into the @xmath0-hard problem rec@xmath89 @xmath90 ccc & @xmath91 & @xmath0-hard + varying @xmath2 & rec@xmath92 & rec@xmath84 + & rec@xmath93 & rec@xmath89 + varying @xmath3 & rec@xmath94 & rec@xmath84 + varying @xmath5 & rec@xmath95 & rec@xmath84 + & rec@xmath88 & rec@xmath89 + + -entries ( black pixels ) in each block of any solution of a @xmath96 instance.,scaledwidth=50.0% ] this section contains the proofs of our tractability results stated in theorem [ thm : tract ] . clearly , a polynomial - time algorithm for rec@xmath97 can be given along the classical lines for deciding whether given row and column sums of a binary matrix are consistent . in the following we give one of the various arguments . the problem obviously reduces to that of reconstructing a @xmath67-matrix from given row and column sums with some of the matrix entries fixed to @xmath98 it is well known that the coefficient matrix is totally unimodular , see , e.g. , @xcite , ( * ? ? ? the problem can thus be solved efficiently , e.g. , by linear programming . next we turn to the proof of theorem [ thm : tract]([thm : tract : ii ] ) . for @xmath99 let @xmath100)\cap i|\qquad \textnormal{and}\qquad \rho_j(i):=|([i]\times\{j\})\cap i|\ ] ] denote the number of blocks @xmath101 with @xmath102 @xmath103 and @xmath104 that lie in the same vertical and horizontal strip , respectively , as the block @xmath105 moreover , for @xmath106 we set @xmath107\times[n],\\ \pi_x(i)&:=\{i\in[m]:\exists j\in[n]:(i , j)\in i\},\\ \pi_y(i)&:=\{j \in [ n]:\exists i \in [ n]:(i , j)\in i\}.\end{aligned}\ ] ] note that @xmath108 and @xmath109 denotes the projection of @xmath110 onto the first and second coordinate , respectively . we use a ( slightly generalized version of a ) result of @xcite on the following problem . dr@xmath111 + @xmath112_0 , & & \textnormal{(row sum measurem . ) } \\ & & & \omit \rlap{$\displaystyle c_{i+l}\in \mathbb{n}_0,$ } & & & & i\in\pi_x(i ) , \:\ : l\in [ k-1]_0 , & & \textnormal{(column sum measurem.)}\\[1.2ex ] & \textnormal{task:}\quad & & \omit \rlap{find $ \xi_{p , q}\in\{0,1\ } , \:(p , q)\in g(i)$ with } \\ \displaystyle & & & \sum_{\mathclap{p:(p , j)\in g(i)}}\xi_{p , j+l}&&=r_{j+l } , & & j\in\pi_y(i ) , \:\ : l\in [ k-1]_0,&&\textnormal{(row sums ) } \\ \displaystyle & & & \sum_{\mathclap{q:(i , q)\in g(i)}}\xi_{i+l , q}&&=c_{i+l } , & & i\in\pi_x(i ) , \:\ : l\in [ k-1]_0,&&\textnormal{(column sums)}\\ \displaystyle & & & \sum_{\mathclap{(p , q)\in b_k(i , j)}}\xi_{p , q}&&=1,&\:\ : & ( i , j)\in i , & & \textnormal{(block constraints),}\\ & & & \omit \rlap{or decide that no such solution exists.}\end{aligned}\ ] ] [ prop : mono ] a. [ mono2 ] an instance @xmath113 of dr@xmath111 is feasible if , and only if , for every @xmath114 we have @xmath115 b. [ mono1 ] dr@xmath116 as a service to the reader we remark that a solution @xmath117 @xmath118 for a given instance of dr@xmath111 is obtained by setting for every @xmath114 and @xmath119 @xmath120 an illustration is given in figure [ fig : example1 ] ( taken from @xcite ) . with @xmath121 in gray color . ( right ) solution defined by .,scaledwidth=45.0% ] now we provide a polynomial - time algorithm for rec@xmath122 let @xmath123 denote an instance of rec@xmath122 we proceed in two steps . first , we solve the following reconstruction - from - row - and - column - sums instance of finding @xmath124 @xmath71 satisfying the constraints @xmath125\cap(k\mathbb{n}_0 + 1),\\ \sum_{j:(i , j)\in i } \eta_{i , j}&=\sum_{l=0}^{k-1}c_{i+l } , & & i\in [ m]\cap(k\mathbb{n}_0 + 1),\\ \eta_{i , j}&=0 , & & ( i , j)\in \{(a , b)\in c(m , n , k):v(a , b)=0\}. \end{aligned}\ ] ] if no solution exists , we report infeasibility of @xmath126 the idea behind solving first is that in this way we locate boxes that we want to contain a 1 . in a second step , if is feasible , we invoke dr@xmath111 from proposition [ prop : mono ] for the instance @xmath127 where @xmath128 and @xmath129 if there is no solution we report infeasibility of @xmath130 otherwise , we return the solution of dr@xmath131 it remains to be shown that this solves @xmath113 correctly . to this end , suppose that @xmath113 has a solution @xmath132 @xmath133\times[n].$ ] then , is also feasible ( a solution is given by @xmath134 @xmath47 ) and @xmath135 satisfies the corresponding instance of dr@xmath131 conversely , let @xmath136 @xmath71 denote a solution to and let @xmath110 be defined as in . for every @xmath114 we have clearly @xmath115 by proposition [ prop : mono ] there is thus a solution to dr@xmath131 this solution satisfies all row and column sum constraints of @xmath113 by definition . further since @xmath137 by definition of , there are 1 s in the solution only in those blocks @xmath138 @xmath71 for which @xmath139 holds . the block constraints are thus also satisfied , which proves the claim . the integer linear program can be solved in polynomial time as the coefficient matrix is well known to be totally unimodular ; see , e.g. , @xcite , ( * ? ? ? the set @xmath110 in is determined in polynomial time , and dr@xmath140 by proposition [ prop : mono]([mono1 ] ) . hence , rec@xmath141 next we turn to the proof of theorem [ thm : tract]([thm : tract : iii ] ) . it relies on the following result . [ prop : totallyunimodular ] let @xmath142 be the node - edge incidence matrix of a ( simple ) bipartite graph , and let @xmath143 denote a binary @xmath144 matrix with the following properties . a. for every @xmath145 $ ] there exists an index @xmath146 $ ] with @xmath147 ; [ lem1:prop2 ] b. the supports of any two distinct vectors @xmath148 and @xmath149 of @xmath150 with @xmath151 for some @xmath152 $ ] do not intersect . [ lem1:prop3 ] then , @xmath153 is totally unimodular . it suffices to prove the assertion under the additional assumption that latexmath:[\[\label{eq : h } all vectors @xmath149 whose support is a singleton can be added later since appending any subset of rows of the @xmath155-identity matrix to a totally unimodular matrix yields again a totally unimodular matrix . as the matrix @xmath156 is the node - edge incidence matrix of a bipartite graph , let @xmath157 denote a corresponding partition of ( the indices of ) the vertices of the graph i.e. , a partition of @xmath158 $ ] . clearly , for every @xmath159 with @xmath160 we have @xmath161 it suffices to prove that every collection of rows of @xmath162 can be split into two parts such that the sum of the rows in one part minus the sum of the rows in the other part is a vector with entries in @xmath163 ; see e.g. ( * ? ? ? * theorem 19.3 ) . hence , let @xmath164 $ ] and @xmath165 $ ] denote such a collection of row indices of @xmath156 and @xmath166 respectively . for @xmath167 $ ] we set @xmath168 of course , the sets @xmath169 and @xmath170 form a partition of @xmath171 i.e. , @xmath172 also , it follows from ( i ) , ( ii ) , , , and the fact that the graph is simple that @xmath173 i.e. , the sets constitute a partition of @xmath174 . by ( i ) , ( ii ) , , and we have @xmath175 hence @xmath176 similarly , @xmath177 therefore , @xmath178 i.e. , the collection @xmath110 and @xmath174 of rows of @xmath156 and @xmath166 respectively , is split into two parts such that the sum of the rows in one part minus the sum of the rows in the other part is a vector with entries in @xmath163 . now we present a polynomial - time algorithm for rec@xmath179 when @xmath180 . we claim that the problem can be formulated as that of finding an integer solution to @xmath181 with a totally unimodular matrix @xmath182 and integral right - hand side @xmath183 showing that the problem is solvable in polynomial time ; see @xcite , ( * ? ? ? 16.2 ) . assembling the variables @xmath184 for @xmath185\times[n]$ ] into an @xmath186-dimensional vector and rephrasing the row and column sum constraints in matrix form @xmath187 where @xmath188 contains the @xmath189 s and @xmath190 contains the @xmath191 s , @xmath156 is the node - edge incidence matrix of a bipartite graph , and hence totally unimodular . as the pattern constraints ensure that no row of a block contains two 1 s , and since @xmath192 or @xmath193 we can ( equivalently ) replace the block and pattern constraints by the _ box constraints _ @xmath194_0,\ ] ] where @xmath195\times\{j+l\}).$ ] we can rephrase the box constraints in matrix form @xmath196 with a binary matrix @xmath150 and a vector @xmath197 containing the respective right - hand sides . since each box @xmath198 is contained in a row of @xmath32\times [ n]$ ] the condition in lemma [ prop : totallyunimodular ] is satisfied . also , any two of these boxes are disjoint since the blocks are disjoint . hence the condition in lemma [ prop : totallyunimodular ] is satisfied . therefore , by lemma [ prop : totallyunimodular ] , the matrix @xmath199 is also totally unimodular . appending @xmath200 and the identity matrix @xmath201 yields again a totally unimodular matrix , hence @xmath202 is totally unimodular . since the corresponding right - hand side vector @xmath203 is integral , the instance can be solved by linear programming . this section contains the proof of theorem [ thm : np ] . we begin with the @xmath0-hardness of @xmath204 for @xmath80 . it suffices to show the result for @xmath205 since the @xmath0-hardness for larger @xmath2 can be inferred from that for @xmath205 by setting the row and column sums @xmath206 and @xmath207 to zero for every @xmath47 and @xmath208 see @xcite . so , let @xmath209 we use a transformation from the following problem 3-color tomography , which is @xmath0-hard by @xcite . ( note that the third color is just `` blank . '' ) 3-color tomography + @xmath210.$}\\[1.2ex ] & \textnormal{task:}\quad & & \omit \rlap { find $ \xi_{p , q}^{(a)}\in\{0,1\},\:\:$ $ ( p , q)\in[m]\times[n],$ $ a \in [ 2],$ with } \\ \displaystyle & & & \sum_{p\in[m ] } \xi^{(a)}_{p , q}&&=r_q^{(a)},\qquad & & q\in[n ] , a\in[2],&&\textnormal{(row sums ) } \\ \displaystyle & & & \sum_{q\in[n ] } \xi^{(a)}_{p , q}&&=c_p^{(a ) } , & & p\in[m ] , a\in[2],&&\textnormal{(columns sums ) } \\ \displaystyle & & & \xi_{p , q}^{(1)}+\xi_{p , q}^{(2)}&&\leq1 , & & ( p , q)\in [ m]\times[n]\qquad&&\textnormal{(disjointness condition),}\\[1.2ex ] & & & \omit \rlap{or decide that no such solution exists.}\end{aligned}\ ] ] let @xmath211 denote an instance of 3-color tomography . we are setting up an instance @xmath127 of our reconstruction problem by defining the row and column sums @xmath212 and @xmath213 , respectively . the block and pattern constraints ensure that there are only three possibilities for setting the 1 s in each block . they are shown in figure [ fig:3colors ] . instances.,scaledwidth=30.0% ] note that the types can be viewed as representing three colors , which are counted by different row and column sums . the row and column sums with odd indices count type 1 blocks , the other sums count type 2 blocks . setting @xmath214,\:\ : & a\in[2 ] , \\ \displaystyle c_{2p+a-2}&:=c^{(a)}_p,\quad & & p\in[m],\:\ : & a\in[2],\end{aligned}\ ] ] and using the correspondences @xmath215,\\ \displaystyle { \xi_{p , q}^{(1)}}^*={\xi_{p , q}^{(2)}}^*=0 \quad & & \leftrightarrow \quad ( \xi_{2p-1,2q-1}^*,\xi_{2p-1,2q}^*,\xi_{2p,2q-1}^*,\xi_{2p,2q}^ * ) \textnormal { is of type~3},&&\end{aligned}\ ] ] we conclude that @xmath113 admits a solution if , and only if , @xmath127 admits a solution . next we turn to the @xmath1-hardness of rec@xmath84 for @xmath216 . again it suffices to prove the result for @xmath209 the general structure of the proof follows that of the @xmath0-hardness result under data uncertainty given in @xcite . however , some adaptations and additions are required as will be detailed below . as in @xcite we give a transformation from the @xmath0-hard problem 1-in-3-sat @xcite , which asks for a satisfying truth assignment that sets exactly one literal true in each clause of a given boolean formula in conjunctive normal form where all clauses contain three literals ( involving three different variables ) . for a given instance of 1-in-3-sat a `` circuit board '' is constructed that contains an initializer , several connectors , and clause chips . the general structure of the circuit board from the proof of @xcite and its modification , which contains additional connectors and initializers , are outlined in figure [ fig : p2layoutleq ] . the initializers contain , for every variable @xmath217 @xmath218,$ ] so - called _ @xmath219-chips _ , while the connectors contain _ @xmath220-chips . _ the clause chips are more complex as they consist of two _ collectors _ , two _ verifiers _ , and a _ transmitter_. the initializer holds a truth assignment for the variables @xmath221 of the given instance @xmath113 of 1-in-3-sat . the boolean values true and false are encoded in @xmath219-chips , @xmath218,$ ] by the type 1 and 2 blocks , respectively , that are shown in figure [ fig : p2types ] . -chips.,scaledwidth=17.0% ] the proof in @xcite shows how the truth assignment is transmitted through the circuit board , and how the verifier chips indeed check the feasibility of the given 1-in-3-sat instance . figure [ fig : p2example2 ] depicts a specific problem instance and a corresponding solution for the original construction of @xcite . the proof in @xcite employs three different types of block constraints . for every block @xmath222 @xmath223 there is one of the following constraints : @xmath224 the names in parenthesis are those from @xcite ( even though @xmath225 may seem more natural here in the third case ) . as for binary variables , the constraint @xmath226 is equivalent to @xmath227 it suffices to adapt the construction in such a way that no @xmath228-block constraints are required . in fact , the @xmath228-block constraints in the original construction are only employed for the @xmath220-chips , @xmath218,$ ] contained in the connectors . for each @xmath220-chip , @xmath218,$ ] in a connector , we replace the @xmath228-block constraints by a @xmath229-block constraint . then , between any two clause chips we insert an additional copy of a connector and an initializer each . these copies are placed as indicated in figure [ fig : p2layoutleq](b ) . the additional row and column sums are prescribed again to values @xmath4 and @xmath230 in alternation . the new copies ensure that each @xmath220-chip , @xmath218,$ ] is contained in a horizontal or vertical strip that contains a @xmath219-chip of an initializer . the corresponding row or column sums have values @xmath4 and @xmath231 and since we have @xmath229-block constraints for both chips , we can distribute the @xmath232 ones only in such a way that each of the two chips contains exactly two 1 s . in other words , each @xmath220-chip , @xmath218,$ ] is required to contain exactly two 1 s . figure [ fig : change ] shows the solution of our correspondingly adapted circuit board from figure [ fig : p2example2 ] . a solution @xmath233 is shown ( non - zero components @xmath234 are depicted as black pixels ) , representing the boolean solution @xmath235 , scaledwidth=60.0% ] with this adaptation the proof from @xcite carries over . as a service to the reader we explain how the truth assignment is transmitted in our particular example ; see figure [ fig : transmission ] . the formal proof that , in particular , this transmission and , in general , the complete transformation works as needed , follows exactly as in @xcite . ( for the example shown in figure [ fig : change ] ) . , scaledwidth=60.0% ] suppose the @xmath236-chip of the initializer in the lower right corner of the circuit board shown in figure [ fig : transmission ] is of type 1 . the truth assignment is vertically transmitted from the initializer to a connector , which negates the boolean value as the x - rays are set to @xmath4 and @xmath230 in alternation . then , this boolean value is horizontally transmitted to an initializer ( again negated , so that we have the original values back ) , then vertically to a connector ( again negated ) . from here the boolean value is transmitted horizontally to the first clause chip , where it enters and exits , yet again negated , through a collector . in the part between the two collectors , the boolean values for the literals appearing in the first clause take a path through the verifiers that ensure that exactly one literal is set to true ; the values for the other literals are just transmitted ( again with negation ) . this process continues in a similar way for the remaining clauses . we end up with feasible solution for our instance of rec@xmath84 if , and only if , there is a satisfying truth assignment for our given instance of 1-in-3-sat . let us first point out that the question of inferring information about an otherwise unknown object from observations `` under the microscope , '' i.e. , through certain windows , is fundamental for a wide range of applications . for corresponding results on the scanning of binary or integer matrices ; see e.g. @xcite . in our terminology , @xcite studies the task of reconstructing binary matrices from their number of @xmath4 s in each windows of a fixed size . in addition to the patterns @xmath247 , and @xmath54 we consider , for @xmath216 , a fourth one @xmath248_0 ^ 2 } \::\ : |m\cap\left([k-1]_0\times\{j\}\right)|\geq k-1 \textnormal { for all } j\in [ k]_0\}.\ ] ] note that @xmath249 can be viewed as the colored - inverted version of @xmath250 @xmath254 & \textnormal{task:}\quad & & \omit \rlap{find $ \xi_{p , q}\in\{0,1\},$ \:$(p , q)\in[m]\times[n],$ with } \\ \displaystyle & & & \sum_{\mathclap{p\in[m ] } } \xi_{p , q}&&=r_q , & & q\in[n ] , & & \textnormal{(row sums ) } \\ \displaystyle & & & \sum_{q\in[n ] } \xi_{p , q}&&=c_p , & & p \in [ m ] , & & \textnormal{(column sums)}\\ \displaystyle & & & \sum_{\mathclap{(p , q)\in w_{k}(i , j ) } } \xi_{p , q } & & \sim_{i , j } v(i , j ) , & & ( i , j)\in c(m , n , k , l ) , \quad & & \textnormal{(window constraints)}\\ \displaystyle & & & \textnormal{pat}_{k}(x , i , j)&&\in p(k , t ) , \qquad \qquad \qquad & & ( i , j)\in c(m , n , k , l),\qquad & & \textnormal{(pattern constraints),}\\[1.2ex ] & & & \omit \rlap{or decide that no such solution exists.}\end{aligned}\ ] ] tables [ tablep01 ] and [ tablenp01 ] below list tractability and intractability results for @xmath253 , respectively , which are simple corollaries to our main theorems [ thm : tract ] and [ thm : np ] . in fact , they are either just reinterpretations of these results or rely , in addition , on one or two of the following simple transformation principles ( t1 ) , ( t2 ) , or ( t3 ) . ( t1 ) is the trivial observation that rather than reconstructing the ones in an image , we may reconstruct the zeros instead . more precisely , this process of `` color inversion '' works as follows . let @xmath255 denote the number of positions in the window @xmath256 . then , we associate a given instance @xmath113 of @xmath253 with the instance @xmath127 of @xmath257 where @xmath258 @xmath259 and @xmath260 , @xmath261 @xmath262,$ ] @xmath263 @xmath264,$ ] @xmath265 and , for all @xmath266 , @xmath267 is replaced by @xmath268 clearly , the problems @xmath253 and @xmath257 lie in the same complexity class . the transformation ( t2 ) adds empty rows and columns to an instance @xmath113 of @xmath269 to extend @xmath0-hardness results for @xmath205 to higher values of @xmath2 . more precisely , let @xmath270 @xmath271 and set for @xmath272\times[n])\cap c(m , n,2)$ ] and @xmath273 $ ] @xmath274,\\ 0&:&\textnormal{otherwise,}\end{array}\right.\\ c'_{\frac{k}{2}(i-1)+l}&:=\left\{\begin{array}{lll}c_{i+l}&:&l\in[2],\\ 0&:&\textnormal{otherwise,}\end{array}\right.\\ ( \sim'_{i , j},v'(k(i-1)/2 + 1,k(j-1)/2 + 1)))&:=(\sim_{i , j},v(i , j)).\end{aligned}\ ] ] this defines an instance @xmath127 of @xmath275 with row sums @xmath276 column sums @xmath277 and block constraints @xmath278 @xmath279 clearly , the instance @xmath113 admits a solution if , and only if , @xmath127 admits a solution ( by filling / extracting the @xmath280-blocks of @xmath113 into / from the @xmath52-blocks of @xmath127 ) . as this is a polynomial - time transformation , @xmath0-hardness of @xmath269 implies @xmath0-hardness of @xmath281 in the final transformation ( t3 ) additional rows and columns are filled by ones . for a given instance @xmath113 of @xmath282 and @xmath80 , let @xmath283 , @xmath284 , and set for @xmath272\times[n])\cap c(m , n,2)$ ] and @xmath273 $ ] @xmath274,\\ m'&:&\textnormal{otherwise,}\end{array}\right.\\ c'_{\frac{k}{2}(i-1)+l}&:=\left\{\begin{array}{lll}c_{i+l}&:&l\in[2],\\ n'&:&\textnormal{otherwise,}\end{array}\right.\\ ( \sim'_{i , j},v'(k(i-1)/2 + 1,k(j-1)/2 + 1)))&:=(\sim_{i , j},k^2+v(i , j)-4).\end{aligned}\ ] ] this defines an instance @xmath127 of @xmath285 with row sums @xmath276 column sums @xmath277 block constraints @xmath278 @xmath286 @xmath287 @xmath288 and @xmath289 clearly , the instance @xmath113 admits a solution if , and only if , @xmath127 admits a solution ( by filling / extracting the @xmath280-blocks of @xmath113 into / from the @xmath52-blocks of @xmath127 ) . as this is a polynomial - time transformation , @xmath0-hardness of @xmath269 implies @xmath0-hardness of @xmath290 c@3exc@3exc@3exc@3exc@3exc@3exl@3ex@xmath2&@xmath291&@xmath292&@xmath293&@xmath5&@xmath174&reference + 1&@xmath14&@xmath294&@xmath294&@xmath295_0$]&@xmath296&thm . [ thm : tract]([thm : tract : i ] ) + 2&@xmath297&@xmath298&@xmath299_0$]&0&@xmath300&@xcite + @xmath301&@xmath302&@xmath303&@xmath304&@xmath53&@xmath305&thm . [ thm : tract]([thm : tract : ii ] ) + @xmath301&@xmath297&@xmath306&@xmath304&@xmath53&@xmath305&thm . [ thm : tract]([thm : tract : ii ] ) & ( t1 ) + @xmath301&@xmath307&@xmath294&@xmath297&@xmath308&@xmath305&thm . [ thm : tract]([thm : tract : iii ] ) + @xmath301&@xmath294&@xmath309_0$]&@xmath303&@xmath230&@xmath305&thm . [ thm : tract]([thm : tract : iii ] ) & ( t1 ) + + c@4exc@4exc@4exc@4exc@4exc@4exl@4ex@xmath2&@xmath291&@xmath292&@xmath293&@xmath5&@xmath174&reference + @xmath301&@xmath294&@xmath297&@xmath310&0&@xmath296&@xcite [ tablenp01:14 ] + @xmath301&@xmath302&@xmath294&@xmath294&1&@xmath305&thm . [ thm : np]([thm : np : i ] ) + @xmath301&@xmath311&@xmath294&@xmath294&0&@xmath305&thm . [ thm : np]([thm : np : ii ] ) [ tablenp01:17 ] + @xmath301&@xmath312&@xmath294&@xmath294&0&@xmath305&thm . [ thm : np]([thm : np : ii ] ) & ( t3 ) + @xmath301&@xmath294&@xmath313&@xmath294&0&@xmath305&thm . [ thm : np]([thm : np : ii ] ) & ( t1 ) + @xmath301&@xmath294&@xmath314&@xmath294&0&@xmath305&thm . [ thm : np]([thm : np : ii ] ) & ( t1 ) & ( t2 ) + + let us finally point out , that the tables [ tablep01 ] and [ tablenp01 ] can be largely extended . in particular , we can consider reconstruction problems for different types of windows which are even allowed to vary for each @xmath315\times[n]$ ] . while some of the complexity results in this more general setting may seem rather marginal , other may be interesting for certain applications ; see @xcite . as this would go far beyond the scope of the present paper we refrain , however , from carrying the results to the extremes here . p. gritzmann . on the reconstruction of finite lattice sets from their x - rays . in _ discrete geometry for computer imagery ( eds . : e. ahronovitz and c. fiorio ) , dcgi97 , lecture notes on computer science 1347 , springer _ , pages 1932 , 1997 . c. kisielowski , p. schwander , f. h. baumann , m. seibt , y. kim , and a. ourmazd . an approach to quantitative high - resolution transmission electron microscopy of crystalline materials . , 58(2):131155 , 1995 . p. schwander , c. kisielowski , f. h. baumann , y. kim , and a. ourmazd . mapping projected potential , interfacial roughness , and composition in general crystalline solids by quantitative transmission electron microscopy . , 71(25):41504153 , 1993 . j. zhu , j. gao , a. ehn , m. aldn , z. li , d. moseev , y. kusano , m. salewski , a. alpers , p. gritzmann , and m. schwenk . measurements of 3d slip velocities and plasma column lengths of a gliding arc discharge . , 106(4):04410114 , 2015 .

physicists have long realized the importance of the earth s magnetic field and that this field originates in the earth s fluid core . the geodynamo is the process by which the rotating , convecting , electrically conducting molten iron in the earth s fluid core maintains the geomagnetic field against ohmic decay . the convective processes in the core that produce the velocity fields required for dynamo action are a combination of thermal and compositional convection . the full dynamo problem requires the examination of the full three dimensional partial differential equations governing convective , incompressible magnetohydrodyamics ( mhd ) . in the past decades computer models have been used to simulate the actual geodynamo . see , for example , the review articles by roberts et al on modeling the geodynamo @xcite and references therein . however , as glatzmaier et al @xcite remark , no three dimensional dynamo model has been run at the spatial resolution required to simulate the broad spectrum of turbulence which surely exists in the earth s fluid core . current computers and numerical methods require the imposition of diffusivities that are several orders of magnitude larger than those which are realistic . it is therefore reasonable to attempt to gain some insight into the geodynamo by considering a reduction of the full mhd equations to a system that is more tractable , but one that retains many of the essential features of the problem . the magnetogeostrophic equation proposed by moffatt and loper @xcite is one such model . the physical postulates of this model are the following : slow cooling of the earth leads to slow solidification of the liquid metal core onto the solid inner core and releases latent heat of solidification that drives compositional convection in the fluid core . the arguments for the appropriate ranges of the characteristic length , velocity , and perturbation density are based on these physical postulates . we first present the full coupled three - dimensional mhd equations for the evolution of the velocity vector @xmath0 , the magnetic field vector @xmath1 and the buoyancy field @xmath2 in the boussinesq approximation and written in the frame of reference rotating with angular velocity @xmath3 . the physical forces governing this system are coriolis forces , lorentz forces , and gravity . following the notation of moffatt and loper @xcite we write the equations in terms of dimensionless variables . the orders of magnitude of the resulting non - dimensional parameters are motivated by the physical postulates of the moffatt and loper model : @xmath4 = - \nabla p + { e}_2 \cdot \nabla b + r_m b\cdot\nabla b + n^2 \theta { e}_3 + \eps \delta u \label{eq:1.1 } \\ & r_m \left [ \partial_t b + u \cdot \nabla b - b\cdot \nabla u\right ] = { e}_2 \cdot \nabla u + \delta b \label{eq:1.2 } \\ & \partial_t \theta + u \cdot \nabla \theta = \eps_\kappa \delta \theta + s \label{eq:1.3 } \\ & \nabla \cdot u = 0 , \nabla \cdot b = 0 . \label{eq:1.4}\end{aligned}\ ] ] here @xmath5 is a given smooth function that represents the forcing of the system . the mathematical statement of the geodynamo problem asks whether there are initial data for the mhd system for which the evolution of the perturbation of the magnetic field @xmath6 grows for sufficiently long time . this can be interpreted as a question of the existence of instabilities of . the notation in is the following . the cartesian unit vectors are denoted by @xmath7 , and @xmath8 . for simplicity , we have assumed that the axis of rotation and the gravity @xmath9 are aligned in the direction of @xmath8 . we have assumed that the magnetic field @xmath1 consists of the sum of an underlying constant field @xmath10 and a perturbation @xmath6 . our choice of @xmath11 as the underlying magnetic field is consistent with the models where the magnetic field is believed to be predominantly toroidal ( cf . the dimensionless parameters are @xmath12 where the typical velocity scale @xmath13 is chosen to be @xmath14 here @xmath15 is a typical amplitude of the buoyancy @xmath16 and @xmath17 is a characteristic length scale . @xmath18 is the sum of the fluid and magnetic pressures , @xmath19 is the kinematic viscosity , @xmath20 is the magnetic diffusivity , and @xmath21 is the molecular diffusivity of the compositional variation that creates an ambient density @xmath22 . the usual estimate for @xmath20 in the fluid core is @xmath23 . the values of @xmath19 and @xmath21 are speculative , but likely to be extremely small . for a more detailed discussion of plausible ranges of the physical parameters that are appropriate for the geodynamo , we refer the reader to the book of ghil and childress @xcite . for the regions in the earth s fluid core modeled by moffatt and loper , it is argued that the dimensionless parameters are the following orders of magnitude . the inverse elsasser number @xmath24 will be of order unity , or somewhat less in the condition of the fluid core , while the rossby number @xmath25 is of order @xmath26 . the value of @xmath27 is somewhat uncertain since the viscosity @xmath19 is a matter of guess work , but plausibly @xmath28 in the core conditions . the inverse peclet number @xmath29 is also likely to be extremely small , perhaps of the order of @xmath30 . according to moffatt and loper @xcite , the magnetic reynolds number @xmath31 is relatively small in the range of @xmath17 considered . with these estimates of the sizes of the parameters the dominant balance of the leading order terms in equations and give the following reduced system : @xmath32 along with the incompressibility condition @xmath33 the _ linear _ system of equations determine the differential operators that relate the vector fields @xmath0 and @xmath6 with the scalar buoyancy @xmath2 . these operators encode the vestiges of the physics in the problem , namely coriolis force , lorentz force , and gravity . vector manipulations of give the expression @xmath34 u = n^4 ( { e}_3 \cdot \nabla ) \delta ( { e}_3 \times \nabla \theta ) + n^2 ( { e}_2 \cdot \nabla)^2 \nabla \times ( { e}_3 \times \nabla \theta ) \label{eq:1.8}.\end{aligned}\ ] ] the sole remaining nonlinearity in the system comes from the coupling of with the equation for the time evolution of the @xmath2 . we call this nonlinear active scalar equation the _ magnetogeostrophic _ ( mg ) equation . it is the subject of this article . we study the properties and behavior of the active scalar equation in @xmath35 @xmath36 \label{eq : mg:2}\end{aligned}\ ] ] via an examination of the fourier multiplier operator @xmath37 obtained from , which relates @xmath38 and @xmath16 . more precisely , @xmath39 = ( { \widehat}{m}_j { \widehat}{\theta})^{\vee}\ ] ] for @xmath40 , and the explicit expressions for the components of @xmath41 as functions of the fourier variable @xmath42 are @xmath43 where @xmath44 on @xmath45 we define @xmath46 , for all @xmath47 , since for self - consistency of the model we assume that @xmath16 and @xmath38 have zero vertical mean , i.e. @xmath48 . several important properties of the @xmath49 s are immediately obvious . firstly , the functions are strongly anisotropic with respect to the dependence on the integers @xmath50 , and @xmath51 . this is a consequence of the interplay of the three above mentioned physical forces . secondly , the function @xmath52 is an even function of @xmath50 , and @xmath51 . thirdly , we have @xmath53 for any @xmath54 ( we use the standard summation convention on repeated indices ) , which is built into the system since the velocity field is divergence free @xmath55 lastly , although the symbols @xmath49 are bounded ( by a constant that depends only on @xmath56 ) in regions of fourier space where @xmath57 , this is not the case in the `` curved '' regions where @xmath58 , and @xmath59 with @xmath60 . in such regions the symbols are unbounded as @xmath61 with @xmath62 that _ is independent of @xmath56 _ , and this bound is sharp . hence the mg equation falls in the class of abstract active scalar equations ( see below ) where the relation between the divergence - free velocity field @xmath38 and the scalar @xmath16 is given as a singular operator of order @xmath63 . these properties of @xmath41 make the mg equation interesting and challenging mathematically , as well as having a clear physical basis in its derivation from the mhd equations . , @xmath64 , and @xmath65 , with @xmath66 . the range of the frequencies is @xmath67 $ ] and @xmath68$].,title="fig:",scaledwidth=30.0% ] , @xmath64 , and @xmath65 , with @xmath66 . the range of the frequencies is @xmath67 $ ] and @xmath68$].,title="fig:",scaledwidth=30.0% ] , @xmath64 , and @xmath65 , with @xmath66 . the range of the frequencies is @xmath67 $ ] and @xmath68$].,title="fig:",scaledwidth=30.0% ] in section [ sec : diffusive ] we state results proved in @xcite for the diffusive mg equation , i.e. , the case when @xmath69 . we recall that in the fluid core @xmath29 is positive but probably extremely small . in this case the mg equation is global well posed . to prove this fact we use the control on @xmath70 , a quantity that is non - increasing along solutions , and which gives a bound on the drift velocity @xmath38 in the space @xmath71 . the latter information on the drift is `` critical '' for the natural scaling of the _ linear _ drift - diffusion equation @xmath72 indeed , for @xmath73 divergence - free , we proved in @xcite that the solution @xmath16 of becomes hlder continuous in finite time , by appealing to the classical ideas of de giorgi @xcite , in the spirit of @xcite . this result was also established independently in @xcite . see also @xcite , the more recent @xcite , and references therein , for regularity issues of linear parabolic equations with singular coefficients and divergence - free drift . the hlder continuity of solutions to the linear equation may then be bootstrapped to show that the solutions of the dissipative mg equations become smooth in arbitrarily short finite time @xcite . having established the global well - posedness of for any @xmath69 , we can proceed to examine the stability of equilibria . we show that the existence of an unstable eigenvalue for the linearized equations implies nonlinear ( lyapunov ) instability of the nonlinear mg equation ( see also @xcite ) . we then exhibit the existence of an unstable eigenvalue in a particular example , using the method of continued fractions cf . the magnitude of this eigenvalue is @xmath74 which , in the context of the fluid core , is extremely large . thus the mg equation can be used to model very rapid exponential growth in time of @xmath16 , and hence , through , of the perturbation velocity and magnetic fields , i.e. a dynamo type instability . we emphasize that this `` mg - dynamo '' instability requires the presence of all three forces coriolis , lorentz , and gravity because the dimensionless parameter @xmath24 ( defined above in ) must be strictly in the range @xmath75 . in section [ sec : inviscid ] we examine the non diffusive mg equation and set @xmath76 ( cf . the situation is then dramatically different from the case @xmath69 . without the laplacian to control the unbounded operator @xmath37 , the mg equation is _ ill - posed _ in the sense of hadamard in sobolev spaces . such a result utilizes in an essential way both the unboundedness of @xmath52 and the fact that it is an even function of @xmath77 . the proof again uses the existence of an unstable eigenvalue of the linearized equation , however when @xmath76 it is possible to construct such an eigenvalue that is arbitrarily large and it then follows that there is no lipschitz solution map at time @xmath78 ( see also @xcite ) . in section [ sec : frac ] we describe results for the fractionally diffusive mg equation ( cf . @xcite ) , i.e. replace the laplacian by @xmath79 , for @xmath80 . this situation is non - physical but mathematically interesting in the context of the tools required to treat the fractional laplacian . we now have a dichotomy across the value @xmath81 . in the range @xmath82 the equation is _ locally _ well - posed , while it is hadamard _ lipschitz ill - posed _ for @xmath83 . at the critical value @xmath84 the problem is globally well - posed for suitably small initial data ( and source term ) , but is ill - posed for sufficiently large initial data . in terms of the non - dimensional parameters introduced above in , this dichotomy can be seen as global well - posedness for @xmath85 , and lipschitz ill - posedness for @xmath86 . a further feature of interest is that the anisotropy of the symbol can be explored to obtain an improvement in the regularity of the solutions when the initial data and source are _ supported on a plane in fourier space_. for such well - prepared data the local existence and uniqueness of solutions can be obtained for all values @xmath80 , and the global existence holds for all initial data when @xmath87 . we note that some of the theorems discussed in this article are valid for certain classes of partial differential equations cf . below , and in any spatial dimension @xmath88 . for such theorems , the three dimensional mg equation is a particular example of an active scalar equation for which these results hold . we conclude the manuscript with section [ sec : concl ] , in which we present a few remaining open problems regarding the mathematical analysis of the mg equations . in this section we discuss the global regularity of solutions to and the lyapunov stability of steady states , in the case that @xmath69 . in order to show that there is no finite time loss of regularity of solutions to with @xmath69 let us first perform a rough derivative count . the dissipation is smoothing by two derivatives , while the nonlinearity is also losing two derivatives : the first coming from @xmath89 , while the second one derivative loss is coming from the constitutive law @xmath90 $ ] , in view of . in this sense we may call the problem `` critical '' ( we make this more precise below ) . since the operator @xmath37 relating @xmath38 and @xmath16 is an order one pseudo - differenential operator , one may view the mg equation as an example of a more general class of of active scalar equations : @xmath91 with @xmath92 a matrix of zero - order pseudo - differential operators , and we use the summation convention on repeated indices . we have dropped the smooth source term @xmath93 from since it is not an obstacle to the well - posedness theory , have re - scaled the time variable so that the diffusivity coefficient is @xmath63 instead of @xmath29 ( this is not an issue since we will construct global in time solutions ) , and the @xmath92 corresponding to are given by @xmath94 . we call this class of active scalar equation `` singular '' since the drift velocity is given in terms of the advected scalar by a constitutive law which is losing derivatives . we refer the interested reader in to the work of chae , constantin , cordoba , gancedo , and wu @xcite and also @xcite for further discussions about the well - posedness theory of singular active scalar equations . to make the idea of criticality more precise , we note that the active scalar equation has a natural scaling invariance built into it : if @xmath95 is a solution , then @xmath96 is _ also _ a solution , with corresponding drift given by @xmath97 . hence @xmath98 is the critical lebesgue space with respect to the natural scaling of the equation . additionally is a scalar equation , so that @xmath99 norms are non - increasing in time for @xmath100 , and hence the @xmath101 norm is the `` strongest '' a priori controlled norm along solutions . in this sense we say that is a `` critical '' equation . another active scalar equation derived from the navier - stokes equations for a rapidly rotating fluid in a thin shell is the surface quasi - geostrophic equation , introduced by constantin , majda , and tabak @xcite ( see also @xcite and references therein ) , which in the critically dissipative case reads @xmath102 and @xmath103 represents the @xmath104 riesz transform . there are strong analogies between the systems and . the velocity field is divergence free in both cases , and the @xmath101 norm is critical with respect to the natural scaling of the equation . the main difference is that in for sqg the dissipative operator is non - local , but the drift velocity is not more singular than the advected scalar . inspired by the de giorgi - type proof of global regularity for the critically dissipative sqg equation of cafarelli and vasseur @xcite ( see also @xcite for different proofs ) , to show that solutions of do not blow up in finite time , we first look at the associated linear problem , where the drift velocity is _ given _ and obeys the bounds that are a priori available for weak solutions of the nonlinear problem : @xmath105 . in @xcite we use a de giorgi iteration technique to prove that a weak solutions of the _ linear _ drift - diffusion equation @xmath106 instantly become hlder continuous . more precisely , the result is : [ thm : degiorgi ] let the initial data @xmath107 and @xmath108 be a weak solution of , where the drift velocity @xmath109 is given by for a suitable matrix @xmath110 . then , for any @xmath111 , there exists @xmath112 such that @xmath113 . the above result was obtained independently by sergin , silvestre , sverak , and zlatos in @xcite . in view of the constitutive a direct corollary of therem [ thm : degiorgi ] is that : weak solutions @xmath16 of are hlder continuous in positive time , i.e. they obey @xmath114 for any @xmath115 , and a suitable @xmath112 . in view of the above discussion about scaling , hlder regularity is sub - critical information for - and one may bootstrap to prove that the solution is classical : [ thm : higher ] let @xmath116 be a weak solution of the initial value problem associated to , with @xmath117 . then @xmath118 , for any @xmath119 , for some @xmath120 . the proof of theorem [ thm : higher ] relies on a suitable bootstrap procedure in besov spaces , in the spirit of @xcite . repeating the argument for the derivative of @xmath16 one may further obtain that the solution of is in fact @xmath121 smooth in finite time . we conclude this subsection with a very brief discussion of the proof of theorem [ thm : degiorgi ] . we refer the reader to @xcite for all details . along the lines of caffarelli and vasseur ( * theorem 3 ) for the critical sqg equation , the first step consists of showing that a weak solution is bounded in the spacial variable ( see ( * ? ? ? * appendix a ) for a construction of weak solutions which obey the energy inequality ) . this is achieved by proving that the function @xmath122 has zero energy if @xmath123 is chosen large enough . the key ingredient is the inequality @xmath124 which can be shown to hold for weak solutions . using that @xmath109 is divergence free one obtains the energy inequality @xmath125 for all @xmath126 and @xmath127 . for @xmath128 , and @xmath129 to be chosen sufficiently large , letting @xmath130 , @xmath131 , and @xmath132 it follows from that @xmath133 for some constant @xmath134 . letting @xmath135 be sufficiently large we thus obtain implies that @xmath136 exponentially as @xmath137 , and hence @xmath138 . having established the boundedness of the weak solutions , the next step is to prove hlder regularity , which requires a more delicate argument . the main obstruction to applying the classical parabolic de giorgi estimates via an @xmath99-based caccioppoli inequality ( @xmath139 ) , is that @xmath140 is the derivative of a @xmath141 function . by essentially using the divergence - free nature of @xmath109 , and appealing to the john - nirenberg inequality , we may however prove a suitable local energy inequality : for any @xmath142 and @xmath143 , we have @xmath144 where @xmath145 is a fixed positive constant . here we have denoted by @xmath146 the parabolic cylinder @xmath147\times b_\rho(x_0)$ ] for @xmath148 and @xmath149 . moreover , estimate also holds with @xmath16 replaced by @xmath150 . the bound may be in turn used , using a suitable iteration argument to obtain an improvement of oscillation result . assume that @xmath151 , where @xmath152 is arbitrary . then we have @xmath153 for some positive constant @xmath145 . we note that the prefactor of @xmath154 of the right side of . besides estimate , the main ingredient in the proof of hlder continuity is the following estimate on the growth of level sets of the solution . there exist @xmath155 , @xmath156 , and @xmath157 such that for @xmath158 if we know that @xmath159 $ ] we have @xmath160 , @xmath161 , @xmath162 , @xmath163 , and @xmath164 . to prove the above estimate one uses a suitable version of the local parabolic energy inequality . bounds and are the main ingredients in the de giorgi iteration procedure , which at the end of the day yields the improvement in oscillation @xmath165 for some @xmath80 , where @xmath166 \times b_4 $ ] , @xmath167 \times b_r$ ] , @xmath158 are arbitrary , @xmath168 , and @xmath169 is as above . since @xmath170 , is then classical to show that implies the solution @xmath16 is hlder continuous , with exponent that may be computed explicitly . in this subsection we consider the stability of steady states to . first we prove for steady states of the the more general class of equations , with drift velocity given by , that linear implies nonlinear instability . we then turn to the specific mg equations , for which we construct an explicit unstable eigenvalue of the linear operator . let @xmath171 be a steady solution of , i.e. @xmath172 where @xmath173 is a time independent force , @xmath69 , and @xmath92 are zero order pseudo differential operators . writing the solution of as @xmath174 we obtain that the perturbation temperature @xmath16 obeys @xmath175 where the _ dissipative linear _ operator @xmath17 is defined as @xmath176 with @xmath177 , and the _ nonlinear _ operator @xmath56 is given by @xmath178 recall the following notion of lyapunov stability : [ def : nonlinear : instability ] let @xmath179 be a pair of banach spaces . a steady state @xmath180 is called @xmath179 nonlinearly stable if for any @xmath148 , there exists @xmath181 so that if @xmath182 and @xmath183 , then = 1em there exists a global in time solution @xmath16 to the initial value problem with @xmath184 ; and we have the bound @xmath185 for a.e . @xmath186 . an equilibrium @xmath187 that is not stable in the above sense is called lyapunov unstable . in the above definition , the banach space @xmath188 is the space where a local existence theorem for the nonlinear equations is available ( e.g. @xmath189 ) , while @xmath190 is the space where the spectrum of the linear operator is analyzed , and where the instability is measured ( e.g. @xmath191 ) . the following result states that the linear lyapunov instability of a steady state @xmath180 implies the nonlinear one . [ thm : linear : nonlinear ] let @xmath180 be a smooth , mean zero solution of . if the associated linear operator @xmath17 , as defined in , has spectrum in the unstable region , then the steady state is @xmath192 lyapunov nonlinearly unstable , for arbitrary @xmath193 . the proof relies on a bootstrap argument that is by now well - known for fluids equations @xcite , and an a priori time independent bound on the @xmath194 norm of the solution , which we obtain from the de giorgi @xmath195 estimate . we only briefly sketch the argument given in full detail in @xcite , to which refer the interested reader . first , we fix a linearly unstable smooth eigenfunction @xmath196 , with eigenvalue of _ maximal _ real part @xmath197 . in order to prove theorem [ thm : linear : nonlinear ] , for @xmath198 $ ] we consider solutions @xmath199 of @xmath200 where @xmath201 and @xmath196 are as above , and show that there exits a constant @xmath202 such that for any @xmath198 $ ] we may find a time @xmath203 with @xmath204 by definition [ def : nonlinear : instability ] , the bound shows that the trivial solution @xmath205 of is @xmath192 lyapunov unstable : indeed , an @xmath206 perturbation leads in finite time to an @xmath207 size solution , in finite time . in order to prove we use the following classical idea : the initial data being an eigenfunction of @xmath17 , will grow exponentially in time , while the nonlinear term may be suitably controlled iso that it can not cancel this exponential growth . to make this idea concrete , it is convenient to define @xmath208 , where @xmath209 . due to the shift , the resolvent of @xmath210 contains the full right - half of the complex plane , and being a bounded perturbation of the laplacian we also have that @xmath211 generates an analytic semigroup over @xmath212 , and @xmath213 is smoothing by two derivatives . then from the duhamel formula and the interpolation bound @xmath214 , we obtain the estimate @xmath215 the key step which allows us to close the estimates is proving the existence of a constant @xmath216 such that @xmath217 for all sufficiently small @xmath27 . the proof of estimate relies on the @xmath195 bound which may be obtained from theorem [ thm : degiorgi ] and a suitable bootstrap procedure to reach the @xmath218 regularity level ( see ( * ? ? ? * appendix a ) ) . lastly , let @xmath219 and @xmath220 be the maximal time such that @xmath221 from and one may then obtain @xmath222 and then in turn that @xmath223 a lower bound on @xmath224 , similar to but with the linear term minus the @xmath225 power term , may be obtained from the duhamel formula . in turn , this implies that @xmath226 and concludes the proof of theorem [ thm : linear : nonlinear ] . in order to apply theorem [ thm : linear : nonlinear ] to the ( nonlinear ) mg equations , we need to find a steady state @xmath180 for which the corresponding linear operator @xmath17 defined in , has unstable spectrum . to achieve this we use an idea that dates back at least to @xcite for navier - stokes and @xcite for euler . first , note that @xmath227 , with @xmath228 smooth , is automatically a steady state , with associated velocity @xmath229 , if we assume that the source term is given by @xmath230 . we thus consider the source term to equal @xmath231 , for which we have a steady state solution of given by @xmath232 for the above choice of the steady states , the equation @xmath233 becomes @xmath234 , \label{eq : eigenvalue}\end{aligned}\ ] ] where @xmath235 is the fourier multiplier defined earlier in . next , we make the _ ansatz _ that the temperature @xmath16 which solves has the form @xmath236 for some fixed @xmath50 in @xmath237 , to be chosen precisely later . using we may compute from the third component of the velocity field @xmath38 associated with @xmath16 given by , insert this into and obtain that @xmath238 and @xmath239 must solve @xmath240 where we have denoted @xmath241 and @xmath242 with @xmath56 being the non - dimensional constant in . an essential feature of the coefficients @xmath243 is that they form an increasing sequence which grows rapidly , as @xmath244 when @xmath245 . equation gives a recurrence relation for the sequence @xmath239 ( see e.g. @xcite ) : @xmath246 we emphasize that given any @xmath247 ( which then uniquely defines @xmath248 for all @xmath249 ) and given any @xmath250 , one can solve for all @xmath251 using the recursion relations . the key is however that only for suitable values of @xmath238 do the @xmath252 vanish sufficiently fast so that @xmath16 defined by is sufficiently smooth . solving for such @xmath238 amounts to solving the infinite continued fraction equation @xmath253 solving is the main difficulty in this argument , since we need not just show there is a solution , but also give bounds on it in terms of the @xmath243 . we omit these details and summarize the results obtained in ( * ? ? ? * lemma 2.6 ) , ( * ? ? ? * lemma 4.2 ) , and ( * ? ? ? * theorem 2.2 ) as follows . [ lem : contd : frac ] let @xmath254 be defined by and respectively , and let @xmath255 . there exits a real eigenvalue @xmath256 which solves , and we have the estimate @xmath257 moreover , the corresponding sequence @xmath239 decays exponentially in @xmath258 . a direct consequence of estimate is that the eigenvalue @xmath259 obeys the upper and lower bounds @xmath260 for @xmath261 fixed , we hence obtain from that there exists at most finitely many frequency pairs @xmath262 for which @xmath263 . moreover , one may show that in the parameter range of the geodynamo , i.e. @xmath264 , for suitable @xmath265 the left side of is positive ( see below ) , thereby proving : associated to the steady state @xmath180 defined in , the linearized mg operator @xmath17 defined in , has unstable spectrum . lastly , note that one may optimize in @xmath266 , and compute a lower bound for the largest eigenvalue of @xmath17 in terms of the physical parameters @xmath267 , and of @xmath268 . more precisely , we have that @xmath269 with the lower bound being attained e.g. for ( of course , we need to take integer part ) @xmath270 in particular , we note that as @xmath271 the lower bound given by the right hand side of blows up ( this will be important in sections [ sec : inviscid ] and [ sec : frac ] ) . moreover , for the above choice of @xmath272 and @xmath273 , we have @xmath274 as @xmath271 . this is in precise accordance to the fact that the absolute value of the fourier multiplier associated to @xmath37 grows at the fastest rate when @xmath275 as @xmath61 . see . combining the results in this subsection with theorem [ thm : linear : nonlinear ] , we have thus constructed an example of solutions to the mg equations which displays strong ( exponential ) growth in @xmath16 and hence in the velocity field @xmath38 by and also in the magnetic field @xmath276 through . the rate of this growth can be shown to be proportional to @xmath277 which is consistent with the dynamo instability scenario . as discussed in the introduction , in the fluid core the value of @xmath29 , given by the quotient of @xmath21 the molecular diffusivity coefficient , and @xmath278 , the product of the typical length scale and typical velocity in the fluid , is very small . therefore , from the physical point of view it is natural to study the non - diffusive ( unforced ) mg equation @xmath279 , \quad \nabla \cdot u = 0 \label{eq : inv:2}\end{aligned}\ ] ] or the corresponding non - diffusive singular active scalar equation when is replaced by , i.e. @xmath280 . as we shall see below , the system is _ ill - posed _ in sobolev spaces , in the sense that there is no operator which is lipschitz in time at @xmath78 . before we discuss the issue of ill - posedness , let us briefly discuss why the _ evenness _ of the operator @xmath37 breaks the classical proof of local existence in sobolev spaces for the class of singular active scalar equation . the standard energy - approach to local existence in @xmath281 would entail applying the @xmath282 to , and then take @xmath212 inner products with @xmath283 . morally speaking ( this is rigorous when @xmath284 is an integer in view of the leibniz rule ) we have only one term which seem to prevent closing the estimates at the @xmath281 level : @xmath285 indeed , the other term where more than @xmath286 derivatives lie on a single function is @xmath287 , but this term vanishes upon integrating by parts since @xmath288 . the only hope to treat the term @xmath289 would be to discover a commutator structure . however , since @xmath37 is _ not anti - symmetric _ , i.e. _ even _ in fourier space , we _ can not _ write @xmath290 , where @xmath291 ( -\delta)^{s/2 } \theta \ ; ( -\delta)^{s/2 } \theta dx = { \mathcal t } + { \mathcal c } \label{eq : s : good}\end{aligned}\ ] ] and @xmath292 $ ] is the usual commutator bracket . instead for the mg , we have @xmath293 , and we _ can not _ explore the fact that @xmath294 may be bounded in terms of @xmath295 ( if @xmath296 ) . this is the main reason why we do not seem to be able to close estimates at the sobolev level . on the other hand , it was observed in @xcite , see also @xcite , that if we replace @xmath90 $ ] ( cf . ) by a certain class of singular constitutive laws @xmath297 ( cf . ) , and the matrix @xmath92 is such that @xmath298 for any @xmath299 and @xmath300 ( here @xmath301 is the usual @xmath212 inner product ) , then the corresponding singular active scalar equation is locally well posed in @xmath281 for @xmath296 . the main reason is that under these assumptions the operator @xmath302 is anti - symmetric , and we are able to write @xmath303 , using the notation of and . the commutator term @xmath294 would enable one estimates to close at the @xmath281 level . lastly , we mention that for analytic initial data , the non - diffusive mg equation is indeed locally well - posed in the class of real - analytic functions , since each term in the equation loses at most one derivative . this was proven in ( * ? ? ? * theorem 2.1 ) by using a cauchy - kowalewski - type argument , in the spirit of the gevrey - class regularity proof of @xcite ( see also @xcite ) . in particular , this result justifies the short time numerical simulations of the mg equation are justified , as they only keep track of finitely many fourier modes ( hence the associated functions are real analytic ) . a cauchy problem for a certain partial differential equation is called _ well - posed in @xmath188 in the sense of hadamard _ , if for any initial data in @xmath188 , the problem has a unique solution in @xmath304 , with @xmath305 depending only on the initial data in the @xmath188-norm , and the solution map @xmath306 satisfies some continuity properties , for a suitable space @xmath307 . if one of these properties fail , the cauchy problem is called _ ill - posed_. in this section we discuss the failure of the solution map for the non - diffusive mg equation to be _ lipschitz continuous _ with respect to perturbations in the initial data ( in the topology of a sobolev space @xmath188 ) around a specific steady profile @xmath187 . such results have been recently obtained for several models in fluid dynamics , see e.g. @xcite and references therein . the proof consists of a linear and a nonlinear step . we construct a steady state and a sequence of eigenfunctions , with arbitrarily large eigenvalues , for the linearized mg equation around this steady state . once these eigenvalues are exhibited , one may use a fairly robust argument ( see , e.g. @xcite ) to show that this severe linear ill - posedness implies the lipschitz ill - posedness for the nonlinear problem . we emphasize that the key feature of the mg constitutive law which gives this ill - posednes is the evenness and unboundedness of the symbol associated to @xmath37 . in fact , the evenness also gives non - uniqueness for for bounded weak solutions of the non - diffusive mg equation , as was recently established by shvydkoy @xcite . let @xmath308 be a positive integer . we use the steady state @xmath309 with associated velocity field @xmath310 . similarly to , we consider the associated linearized ( non - diffusive ) mg operator defined by @xmath311 \partial_3 \th_0 \label{eq : steady : l}\end{aligned}\ ] ] where the @xmath235 operator is defined in terms of its fourier symbol . with this notation , we have [ thm : ill : lin ] the cauchy problem associated to the linearized @xmath312 equation @xmath313 is ill - posed in the sense of hadamard over @xmath212 . that is , for any @xmath314 , there exits a real - analytic initial datum @xmath315 , such that the ivp associated to does not have a solution @xmath316 , which obeys @xmath317 } \| \theta(t,\cdot ) \|_{l^2 } \leq k \|\theta(0,\cdot ) \|_{y } \label{eq : linear : well}\end{aligned}\ ] ] where @xmath318 is any sobolev space that embeds compactly in @xmath212 . the proof of the above theorem builds on the construction of unstable eigenvalues in the critically diffusive case , more specifically on lemma [ lem : contd : frac ] . as in section [ sec : unstable : eigenvalue : dissipative ] , cf . we make the ansatz that @xmath16 is given in terms of a fourier sine series @xmath319 with @xmath272 and @xmath273 to be chosen precisely later . note that since we are in the non - diffusive case , we need to set @xmath76 , and hence cf . we have @xmath320 for any @xmath321 . the value of @xmath243 remain however unchanged . using lemma [ lem : contd : frac ] , we obtain an eigenvalue @xmath322 , and an exponentially decaying sequence @xmath239 , such that cf . the bound @xmath323 holds . we set @xmath324 and @xmath325 in , where @xmath326 is a positive integer , and obtain the lower bound @xmath327 on the eigenvalue , associated to this choice of @xmath265 . here we also denote by @xmath329 the eigenfunction constructed in this way . using an additional argument , which requires suitably bounding the exponentially decaying sequence @xmath239 which defines our eigenfunction , one may in fact additionally normalize @xmath16 and prove ( cf . * lemma 2.7 ) , ( * ? ? ? * lemma 4.3 ) ) that @xmath330 where @xmath331 is a sufficiently large constant . for example , if @xmath332 with @xmath333 , we may explicitly compute a suitable such constant . we conclude the proof of theorem [ thm : ill : lin ] by using a contradiction argument . if such a solution would exist , for a fixed @xmath305 and @xmath334 , we pick @xmath335 large enough so that @xmath336 holds . the linear problem can be shown to have a unique @xmath212 solution , and hence by construction we have @xmath337 the assumed estimate is contradicted by our choice of @xmath338 and the lower bound . this concludes the proof of theorem [ thm : ill : lin ] . before stating our main ill - posedness result , let us recall the definition of lipschitz well - posedness . [ def : well ] let @xmath339 be sobolev spaces . the cauchy problem for the non - diffusive mg equation is locally lipschitz @xmath340 _ well - posed _ , if there exist continuous functions @xmath341 , so that for every pair of initial data @xmath342 there exist unique solutions @xmath343 of the initial value problem associated to , that additionally satisfy @xmath344 for every @xmath345 $ ] , where @xmath346 and @xmath347 . we note that defintion [ def : well ] allows the solution map to lose regularity . since the mg equation is first order , the typical pairs of spaces @xmath340 that we have in mind here are @xmath348 , and @xmath349 , with @xmath350 . the main result of this section is then : [ thm : nonlinear : ill ] the non - diffusive mg equations , are locally lipschitz @xmath340 _ ill - posed _ in sobolev spaces @xmath351 embedded in @xmath352 , in the sense of definition [ def : well ] . the proof of the above result is based on the severe linear ill - posedness obtained in theroem [ thm : ill : lin ] , and a standard perturbative argument ( * ? ? ? * theorem 2 ) or @xcite , which we breifly sketch below . let @xmath353 be the steady state defined above in , let @xmath354 \partial_r \th_0 $ ] be its associated linearized mg operator , and denote the mg nonlinearity by @xmath355 \partial_j \theta$ ] . by our assumptions on @xmath188 and @xmath318 , we have @xmath356 for some constant @xmath357 . also , let @xmath358 . in definition , we chose @xmath359 be the steady state , and let @xmath360 , for some smooth function @xmath361 to be chosen later . assume by contradiction that the problem is lipschitz locally well - posed in @xmath340 . one may then show that there exist @xmath362 and @xmath363 such that @xmath364 } \|\theta^{(1)}_\delta(t ) - \th_0\|_{x } \leq k \delta\ ] ] for _ any _ @xmath365 $ ] . writing the solution @xmath366 as an order @xmath367 perturbation around the steady state , @xmath368 we thus have @xmath369 } \|\psi_\delta(t)\|_x \leq k \label{eq : psi : bound:1}\end{aligned}\ ] ] and @xmath370 solves the cauchy problem @xmath371 the key observation is that when combined with , estimate yields @xmath372 for any @xmath373 $ ] . therefore , when passing @xmath374 , the nonlinearity in disappears , and we should obtain the linear limiting equation . indeed , from , , and the bounds , we obtain the necessary estimates to apply the aubin - lions compactness argument and show that @xmath375 strongly in @xmath376 and on this time interval the limiting function obeys @xmath377 but @xmath378 also inherits the bound from the sequence @xmath370 , which contradicts thereom [ thm : ill : lin ] choose @xmath361 an eigenfunction of @xmath17 as constructed in section [ sec : ill : linear ] , so that in time @xmath379 the exponential growth of the @xmath212 norm caused the solution to exceed the level @xmath380 . this concludes our sketch of the proof of theorem [ thm : nonlinear : ill ] . the fractionally diffusive mg equation , although non - physical , presents a very challenging and interesting mathematical problem . replacing the laplacian in ( [ eq : mg:1 ] ) by @xmath381 , for @xmath80 , we obtain the active scalar equation @xmath382 \label{eq : mgf:2},\\ & \theta(0,\cdot ) = \theta_0 \label{eq : mgf:3}\end{aligned}\ ] ] where the symbol of the fourier multiplier @xmath37 is given by ( [ eq : m1])([eq : d ] ) . note that here we work with the non - dimensional variables , so that the initial data and the source term are _ normalized to be of size unity_. the main feature of this situation is that the structure of @xmath383 ( anisotropy , evenness , and unboundedness ) and a relatively weak smoothing effect of the nonlocal operator @xmath384 produce a sharp dichotomy across the value @xmath84 , as follows . if @xmath82 the equation is _ locally _ well - posed , while it is hadamard _ lipschitz ill - posed _ for @xmath83 . at the threshold value @xmath84 the problem is globally well - posed for @xmath85 , respectively ill - posed for @xmath385 . furthermore , the anisotropy of the symbols @xmath383 can be used to improve the regularity of the solutions when the initial data and source have a suitably chosen fourier support . for such _ well - prepared _ data the local existence and uniqueness of solutions can be obtained for all values @xmath80 , and the global existence holds for all initial data when @xmath87 . we remark that in all above mentioned regimes , the problem is _ super - critical _ in the sense described in section [ sec : diffusive ] . in section [ sec : diffusive ] we stated the well - posedness result when the diffusion in the equation is given by the full laplacian . if the diffusion in the equation is introduced by the fractional operator @xmath384 , @xmath87 , then we have the following result . [ thm : fractional_local ] let @xmath87 , and fix @xmath386 . assume that @xmath387 and @xmath388 have zero - mean on @xmath35 . then there exists a time @xmath389 and a unique smooth solution @xmath390 of the cauchy problem ( [ eq : mgf:1])([eq : mgf:2 ] ) . under an additional assumption on the size of @xmath29 ( equivalently , the size of the initial data in the unscaled equations ) , we obtain a global existence result . [ thm : fractional_global ] let @xmath391 and @xmath93 be as in the statement of theorem [ thm : fractional_local ] , and let @xmath392 have zero - mean on @xmath35 , where @xmath393 . there exists a large enough number @xmath394 , such that if @xmath395 , then the unique smooth solution @xmath16 of the cauchy problem is global in time , i.e. @xmath396 . [ rem : rescale ] note that in @xcite we did not use the non - dimensional variables with which we are working here . therefore , the global well - posedness condition stated in @xcite is that the initial data and source , have to be _ sufficiently small _ , compared to the diffusivity coefficient @xmath21 . in this paper however , the data and source have been normalized to be of order unity , and the only coefficient remaining is @xmath29 . it is hence natural that the small data condition turns to a _ large enough _ @xmath29 condition . this remains valid throughout this section . the detailed proofs of theorems [ thm : fractional_local ] and [ thm : fractional_global ] are given in @xcite , and rely mainly on energy methods and the use of suitable commutator estimates . we omit these details here . in this subsection we consider the stability of steady states to . the construction follows the outline described in section [ sec : unstable : eigenvalue : dissipative ] . in this case , we choose @xmath397 and @xmath398 , for some integer @xmath399 . as before , the first step in proving that the cauchy problem is lipschitz ill - posed is to consider the linearized problem around the given steady state , and exhibit the existence of unstable eigenvalues with arbitrarily large real part . thus , we consider @xmath400 where the linear operator @xmath17 is given by @xmath401 we seek the temperature @xmath16 which solves ( [ eq : fractional_inst1 ] ) and has the form @xmath402 for some fixed @xmath403 to be chosen later . inserting ( [ eq : fractional_theta ] ) into ( [ eq : fractional_inst1 ] ) and using the definition of @xmath235 , we obtain that holds , but with @xmath404 denoted by @xmath405 and with @xmath243 given by above . as before , these assumptions lead to solving the continued fraction equation for some real @xmath406 . this is achieved by suitably modifying lemma [ lem : contd : frac ] to obtain : [ lemma : sn:5.2 ] let @xmath407 be defined by and respectively , and let @xmath408 . there exists a real eigenvalue @xmath409 , and a rapidly decaying sequence @xmath410 so that the function @xmath16 defined by solves . furthermore , we have the lower bound @xmath411 in particular , setting for instance @xmath412 and @xmath413 in , for each @xmath326 we obtain a real eigenvalue @xmath414 of @xmath17 , which is bounded from below as @xmath415 moreover , given a sobolev space @xmath318 that is compactly embedded in @xmath212 , we may ensure the existence of a constant @xmath416 such that the smooth eigenfunction @xmath329 corresponding to @xmath414 obeys . the immediate consequence of estimate is that due to @xmath417 , we have @xmath418 the above construction proves that the linearized problem is ill - posed in @xmath212 , as we have exhibited eigenvalues of arbitrarily large real part . following the argument outlined in section [ sec : ill : nonlinear ] , this type of strong ill - posedness for the linearized problem , automatically implies that the nonlinear problem is lipschitz ill - posed in sobolev spaces ( recall defintion [ def : well ] of lipschitz well - posedness ) . thus we obtain : [ thm : nonlinear : ill : frac ] the cauchy problem , with @xmath419 is locally lipschitz @xmath340 _ ill - posed _ in sobolev spaces @xmath351 embedded in @xmath420 . the value @xmath84 marks the threshold for proving the local well - posedness in sobolev spaces for . if the initial datum is small with respect to @xmath21 , or equivalently , if @xmath29 is sufficiently large ( see remark [ rem : rescale ] above ) , then one may use energy estimates to show that there exists a unique global smooth solution evolving from this data . in this direction we have : [ thm : gamma=1/2:small ] let @xmath421 and assume that @xmath387 and @xmath388 have zero - mean on @xmath422 . there exists a sufficiently large number @xmath394 , such that if @xmath395 , then the unique smooth solution @xmath423 of the cauchy problem is global in time . as in the case of @xmath87 , the proof is based on the classical kato - ponce commutator estimate . the proof however does not apply for the case of `` small '' @xmath29 ( large data ) , not even to prove the local existence of solutions . in the case where @xmath29 is suitably small , we may construct a steady state such that solutions evolving from initial data which is arbitrarily close to this steady state diverge from it at an arbitrarily large rate , i.e. the problem is hadamard ill - posed . more precisely , we have : [ thm : ill:1/2 ] let @xmath424 be sobolev spaces , let @xmath81 , and fix the integer @xmath425 . consider the steady state solution of given by @xmath426 , with source term @xmath427 , and assume that @xmath428 then , given any @xmath389 and any @xmath429 , there exists an initial condition @xmath430 and a corresponding solution @xmath431 of the cauchy problem , such that @xmath432 } \vert \theta^{(2)}(t,\cdot ) - \theta^{(1)}(t,\cdot ) \vert_{x } \geq 2 k \vert \theta^{(2)}_0 - \theta^{(1)}_0 \vert_{y}\end{aligned}\ ] ] where @xmath433 . that is , the problem is locally lipschitz ( x , y ) ill - posed when @xmath86 . to prove the above result , we follow the ill - posedness result in the case @xmath419 . the main difference is that while for @xmath419 by we have real eigenvalues of arbitrarily large real part , when @xmath434 the lower bound only blows up for @xmath29 sufficiently small , depending on @xmath24 and our choice of @xmath268 . this condition is rather unsatisfactory , since it implies @xmath271 as @xmath435 . in fact , we can establish a more robust smallness condition on @xmath29 , in the sense that it is independent of @xmath24 ( and @xmath268 ) . to see this , insert @xmath436 and @xmath437 in the lower bound , where @xmath438 . we obtain the lower bound @xmath439 therefore , under the assumption , which is independent of @xmath268 or @xmath56 , the right side of diverges to @xmath440 as @xmath441 and hence we again have a sequence of unstable eigenvalues of the linear problem , with arbitrarily large real part . combined with the nonlinear argument presented in section [ sec : ill : nonlinear ] this concludes the sketch of the proof of theorem [ thm : ill:1/2 ] . in this subsection we present results which take advantage of the curved nature ( paraboloid - type ) of the regions in frequency space where the symbol @xmath442 grows like @xmath443 . we observe that if the support of @xmath444 lies on a plane , then the operator @xmath37 behaves like an operator of order zero acting on this @xmath16 , and hence the velocity @xmath38 is as smooth as @xmath16 . this allows us to improve the well - posedness results in the range @xmath69 and @xmath445 , over the generic case when no conditions on the fourier transforms of the initial data ( and source term ) are imposed . to be more precise let us introduce some notation . for a rational number @xmath446 with @xmath447 , define frequency plane @xmath448 as @xmath449 one may show that whenever @xmath447 , latexmath:[\[\begin{aligned } \label{eq : m : plane : bound } this observation , combined with the fact that when the data @xmath15 and source @xmath93 have frequency support on @xmath451 , then so does the solution @xmath452 of the cauchy problem at all later times @xmath453 , means that the smoothing properties of the fractional diffusion term are stronger than they are for the generic data situation . hence it is possible to prove stronger regularity results , and in particular the _ local existence _ and uniqueness of sobolev solutions holds _ for all _ values of @xmath80 . [ thm : well - prepared : local ] let @xmath80 , and fix @xmath454 . assume that @xmath387 and @xmath388 have zero - mean on @xmath422 and are such that @xmath455 for some fixed @xmath456 and all @xmath457 . then there exists a @xmath389 and a unique smooth solution @xmath458 of the cauchy problem , such that @xmath459 for all @xmath460 . we note that the above theorem holds even if @xmath419 , a setting in which we know that for _ generic _ initial data the problem is ill - posed . moreover , one may also prove that if the initial data is small with respect to @xmath21 , or equivalently , @xmath85 in the non - dimensional setting of this manuscript , the solution obtained in theorem [ thm : well - prepared : local ] is global in time . the main difficulty in the proof of theorem [ thm : well - prepared : local ] is the construction and uniqueness of solutions , which live on the frequency plane @xmath451 . this relies on constructing of a suitable sequence of approximations supported on the fourier space on @xmath451 and showing its convergence using energy methods . the main obstacle one has to overcome is designing an iteration scheme which is both suitable for energy estimates , and is such that in each iteration step the frequency support of the approximation lies on @xmath451 . more explicitly , a scheme in which @xmath461 is given in terms of @xmath462 preserves the frequency support on @xmath451 , however is not good for closing the estimates at the level of sobolev spaces . on the other hand if we define @xmath461 in terms of @xmath463 , the energy estimates are now clear but it seems difficult to inductively prove that @xmath464 is supported on @xmath451 . therefore , we work with a regularized problem , where we introduce an artificial term @xmath465 . the advantage of considering a stronger dissipative problem is that we can _ construct _ smooth solutions which have frequency support lying on @xmath451 . for details of the construction we refer the interested reader to @xcite . lastly , we pass to the limit @xmath466 and obtain a solution of with desired properties . if the initial data @xmath15 and the source term @xmath93 have fourier support in @xmath451 , we show that for @xmath467 the local in time solution may be continued for all time . [ thm : well - prepared : global ] let @xmath468 and @xmath93 be as in the statement of theorem [ thm : well - prepared : local ] . if @xmath469 , the unique smooth solution @xmath16 of the cauchy problem is global in time . the proof hinges on the observation that the solution constructed in theorem [ thm : well - prepared : local ] satisfies @xmath470 . then , in view of and the hrmander - mikhlin theorem we have @xmath471 for all @xmath472 , and so the velocity @xmath38 is as smooth as the scalar @xmath16 . hence , the case @xmath473 is `` sub - critical '' for the a priori controlled quantities , and energy estimates in the spirit of those used to prove the global regularity of the sub - critical sqg equation @xcite , may be used to prove theorem [ thm : well - prepared : global ] . the case @xmath84 in theorem [ thm : well - prepared : global ] is more subtle , and is analogous with the problem of global regularity for the critical sqg equation @xcite . the anisotropic nature of the symbol @xmath37 , makes it more convenient to apply the de giorgi - inspired proof of @xcite in the context of the fractional mg equation with @xmath81 . the main ingredient is that if @xmath474 and @xmath475 , then @xmath476 . hence , considering first the linear problem with divergence - free @xmath141 drift we obtain hlder continuity of the solution . the later norm is sub - critical and one may bootstrap to higher regularity . we refer the interested reader to @xcite for details . we have proven that the magneto - geostrophic model proposed by moffatt and loper is capable of sustaining dynamo action . the mg - dynamo instability exhibited here relies crucially on the fact that the dimensionless parameter @xmath24 is neither @xmath477 , nor @xmath440 . the exponential growth is of order @xmath478 , for all @xmath479 , and @xmath480 . in terms of the physics in the model , this condition on @xmath24 requires that _ all three forces _ , coriolis , lorentz , and gravity , _ are non - zero _ , making the mg - instability somewhat different in nature from the usual rayleigh - taylor instability . at the level of mathematical analysis several problems remain to be addressed , including some well / ill - posedness issues . for the non - diffusive equation , _ is there a `` stronger '' form of ill - posedness such as norm inflation or even finite time blowup ? _ can we refine the well - posedness results for the fractionally diffusive problem in the range @xmath481 ? that is , _ can one establish the global existence of smooth solutions or on the contrary , prove finite time blowup ? _ several , interesting , mathematical issues arise when returning to the derivation of the mg equation from the full three dimensional mhd equations in the spirit of @xcite . in particular , we note that moffatt s and loper s initial discussion includes the kinematic viscosity @xmath19 , which is directly proportional to @xmath27 in . the values of @xmath27 and @xmath29 are highly conjectural , but both are believed to be _ extremely _ small . therefore , as _ opposed _ to the mg system , if we keep @xmath482 we derive from the system @xmath483 the presence of the laplacian term in makes the constitutive law relating @xmath38 and @xmath16 `` better '' , in the sense that the @xmath37 operator corresponding to , now depends both on @xmath56 and @xmath27 , and is in fact smoothing of degree @xmath484 ( see equations ( 29)(30 ) in @xcite ) . hence , in this case the constitutive law is even better than that for the two - dimensional euler equation in vorticity form , implying that : _ in the range @xmath482 , the system is globally well - posed , even if @xmath485_. this observation is interesting in view of the much more challenging problem of _ establishing properties of the solutions to in the vanishing parameters limits @xmath486 and @xmath271 _ , as proposed by moffatt and loper . the later question is directly related to the turbulent scaling regime for @xmath16 conjectured by moffatt in @xcite . `` turbulence '' in the framework of vanishing diffusivity @xmath29 for the forced system , and the presence of anomalous dissipation of energy for @xmath16 , or `` thetargy '' , may be plausible for very weak solutions ( below onsager critical ) of the non - diffusive equation . p. constantin , d. cordoba , and j. wu . on the critical dissipative quasi - geostrophic equation . , 50(special issue):97107 , 2001 . dedicated to professors ciprian foias and roger temam ( bloomington , in , 2000 ) . magnetostrophic turbulence and the geodynamo . in y. kaneda , editor , _ iutam symposium on computational physics and new perspectives in turbulence , nagoya , japan , september , 1114 , 2006 _ , volume 4 of _ iutam bookser . _ , pages 339346 . springer , dordrecht , 2008 .

perturbation theory remains the most accurate and reliable tool of theoretical analysis in elementary particle phenomenology . even the most promising approach based on the direct calculation of physical observables on the lattice still can not compete with perturbation theory supplemented with some non - perturbative additions as condensates or renormalons . the accuracy of experimental data for tests of the standard model permanently improves and demands the improvement of the accuracy of theoretical predictions @xcite . within perturbation theory this demand implies the calculation of feynman diagrams with an increasing number of loops ( see e.g. @xcite ) . at present , feynman diagrams of rather general form can be obtained analytically at most up to three loops . any of these calculations is still unique and involves sophisticated algorithms and an extensive usage of computer resources ( see e.g. @xcite ) . even the notion of the analytical calculation is changing now because the analytical results are expressed through rather special transcendental numbers which eventually have to be calculated numerically . the striking example is the calculation of massive three - loop bubbles @xcite where the analytical results for the diagrams are represented through the quantity which is given by a two - fold infinite sum ( see also @xcite ) . not all of the multiloop feynman diagrams are equally complicated . there are some topologies of @xmath3-loop diagrams or special kinematic regimes for the external momenta which are much simpler to compute than the general @xmath3-loop case ( see e.g. @xcite ) . the increasing complication of loop calculations within perturbation theory is more and more circumvented by using different kinds of expansion for the cases where the kinematics allows one to find a small parameter . heavy quark effective theory ( hqet ) and threshold expansions for heavy particle production are recent examples for such a situation . these expansions are mostly used for phenomenological applications where it even suffices to give approximate numerical values for the unexpanded multiloop integrals . in some cases one can obtain an analytical expansion that can be used for the simplification of the calculations or for the evaluation of the exact parameters of the effective lagrangians through matching conditions . in this paper we discuss a method to calculate the near threshold expansion of the spectral density for the class of multiloop sunset - type feynman diagrams termed water melon diagrams @xcite . these diagrams has recently drawn some attention as a laboratory for testing some advanced methods in loop calculations @xcite . however , water melon diagrams also have numerous phenomenological applications . the most important application of these diagrams ( as well as spectacle diagrams as their first generalization ) is the calculation of the effective potential in quantum field theory both at finite temperature and at zero temperature @xcite . among other applications one can name the sum rule analysis of baryons in qcd both in the massless approximation and with finite mass ( the leading order correlator being the standard sunset diagram ) @xcite or the treatment of baryons in the large @xmath4 limit of qcd @xcite which requires the calculation of @xmath5-loop water melon diagrams in the leading order . an interesting application is the calculation of dibaryon properties in operator product expansion within the sum rule approach which is important for understanding many - quark bags in the nucleus @xcite . the water melon diagrams emerge in chiral perturbation theory @xcite . the corresponding integrals appear for the correlator of the effective operators related to the mixing of neutral mesons in flavour dynamics @xcite and in the sum rule analysis of hybrid mesons @xcite . water melon diagrams constitute an important part of the contributions generated by the recurrence relations of the integration - by - part techniques for three - loop diagrams @xcite ( for the recent progress see e.g.ref . it is a rather hot topic and a dynamically developing area of loop calculations where new results are frequently reported ( see e.g.refs . @xcite ) . in our opinion , the method presented in ref . @xcite completely solves the problem of computing this class of diagrams . the method is simple and reduces the calculation of a @xmath3-loop water melon diagram to a one - dimensional integral which includes only well - known special functions in the integrand ( bessel functions of different kinds ) for any values of the internal masses . the technique is universal . the most interesting part of our analysis of water melon diagrams is the construction of the spectral decomposition of water melon diagrams , i.e. the determination of the discontinuity across the physical cut in the complex plane of the squared momentum . in this context a novel technique for the direct construction of the spectral density of water melon diagrams based on an integral transform in configuration space was presented in ref . @xcite . in the present paper we develop some explicit expansions for water melon diagrams and compare them with the exact results . the purpose of this paper is a three - fold one , namely : * to demonstrate the ease with which threshold expansions to arbitrary space - time dimensions and a general number of internal lines with arbitrary masses can be generated within a configuration space technique . the entire construction reduces to algebraic operations while the only integral encountered is a simple integral of the type of euler s gamma function . * to generate explicit forms of threshold expansions and to analyze their convergence properties . the explicit forms of threshold expansions for the simple sunset can be compared with threshold expansions that are obtained in momentum space . * to consider a case when one mass is much smaller than the others . this is an important generalization of the threshold expansion and can be analytically done within the configuration space technique . the present analysis of the threshold expansion of water melon diagrams shows the effectiveness of the configuration space technique for this topology class of feynman diagrams . on the other hand the obtained explicit results can be useful for a variety of applications which include the evaluation of water melon type diagrams . the technique can readily be generalized to some other simple topologies of @xmath3-loop diagrams . the paper is organized as follows . in sec . 2 we provide the tools necessary for the following calculations . sec . 3 deals with the threshold expansion . here we outline our main strategy for this expansion in configuration space and in the subsections that follow we give examples for the sunset diagram and the water melon diagram with four or more internal lines . in sec . 4 we concentrate on the strongly asymmetric case where one of the masses is much smaller than the others . we introduce the procedure of resummation of the contributions of this smallest mass and show explicitly how we get to a closed expression for the spectral density even in this case . we give examples for a strongly asymmetric mass arrangement for the water melon diagrams with two ( degenerate case ) , three , and four or more internal lines , and compare the resummed results with the pure threshold expansion as well as with the exact spectral density . in sec . 5 we discuss how to recover the non - analytic part of the polarization function through the spectral density near threshold . in sec . 6 we give our conclusions . we start with a brief outline of the technique which we use in this paper @xcite . the polarization function @xmath6 of a water melon diagram including @xmath3 internal lines with masses @xmath7 , @xmath8 in configuration space is given by the product @xmath9 the propagator @xmath10 of a massive line with mass @xmath11 in @xmath12-dimensional ( euclidean ) space - time is given by @xmath13 where we write @xmath14 . @xmath15 is a mcdonald function ( a modified bessel function of the third kind , see e.g. @xcite ) . in the limit @xmath16 the propagator in eq . ( [ prop ] ) simplifies to @xmath17 where @xmath18 is euler s gamma function . ( [ prod ] ) contains all information about the water melon diagrams and in this sense is the final result for the class of diagrams under consideration . it is of course known and was used since long ago @xcite . of some particular interest is the spectral decomposition of the polarization function @xmath6 which is connected to the particle content of a given model . the spectrum of particles is contained in the function @xmath19 related to the polarization function through the dispersion representation @xmath20 the dispersion representation of the polarization function in configuration space reveals the analytic structure of the polarization function @xmath6 . for applications , however , one may need the fourier transform of the polarization function @xmath6 given by @xmath21 ( we use the same notation for the function and its fourier transform because we think that this will cause no confusion ) . because of the lorentz invariance of the propagator the angular integration in eq . ( [ curr ] ) can be done explicitly in @xmath12-dimensional space - time ( for a generalization to tensor propagators see ref . the result reads @xmath22 where @xmath23 is the usual bessel function and @xmath24 is the rotationally invariant measure on the unit sphere in the @xmath12-dimensional ( euclidean ) space - time . note that the polarization function @xmath6 as well as its fourier transform @xmath25 are only functions of the absolute value @xmath26 and @xmath27 . the same is of course valid also for the other occurring functions . to simplify the notation we often write @xmath28 and @xmath29 for these absolute values . note that propagators of particles with non - zero spin in configuration space representation can be obtained from the scalar propagator by differentiation with respect to the space - time point @xmath30 . this does not change the functional @xmath30-structure and causes only minor modification of the basic technique ( for details see ref . @xcite and references therein ) . our final representation of the fourier transform of a water melon diagram is given by the one - dimensional integral @xmath31 which is a special kind of integral transformation with a bessel function as a kernel . this integral transformation is known as the hankel transform @xcite . the representation given by eq . ( [ exactpi ] ) is quite universal regardless of whether tensor structures are added or particles with vanishing momenta are radiated from any of the internal lines . note in this context that bessel functions are objects well - studied during the last two centuries . they therefore can be added to the list of elementary functions . because the dispersion representation of the polarization function in configuration space ( or the spectral density of the corresponding polarization operator ) has the form given in eq . ( [ convo1 ] ) , the analytic structure of the polarization function @xmath6 can be determined directly in configuration space without having to compute its fourier transform first . the transformation in eq . ( [ convo1 ] ) turns out to be a particular example of the hankel transform , namely the @xmath32-transform @xcite . as pointed out in ref . @xcite , the inverse @xmath32-transform in this case is given by @xmath33 where @xmath34 is a modified bessel function of the first kind and the integration runs along a vertical contour in the complex plane to the right of the right - most singularity of @xmath35 . the inverse transform given by eq . ( [ exactrho ] ) completely solves the problem of determining the spectral density of the general class of water melon diagrams with any number of internal lines and different masses by reducing it to the computation of a one - dimensional integral . for @xmath3 internal lines with equal masses @xmath11 the spectral density reads explicitly @xmath36 in contrast to our technique , in the standard , or momentum , representation the polarization function @xmath25 is calculated from a @xmath37-loop diagram with @xmath37 @xmath12-dimensional integrations over the entangled loop momenta which makes the computation difficult when the number of internal lines becomes large . with our method described in detail in ref . @xcite the @xmath38-dependence of the spectral density can be calculated by a one - fold numerical integration according to eq . ( [ exactrho ] ) . the numerical integration in eq . ( [ exactrho ] ) can be done for arbitrary space - time dimensions and a general number of lines with arbitrary masses . in this sense this is the most efficient representation for the spectral density of the water melon diagram . however , we can also develop an explicit expansion near the threshold with any desired accuracy . it does not require any complicated integrations at all . the corresponding expansion of eq . ( [ exactrho ] ) can then be compared with series expansions near the production threshold obtained with the traditional momentum space technique . we stress that we are only interested in the spectral density because it is the main object for physical applications ( see e.g. refs . @xcite ) . for practical reasons we start with eq . ( [ exactpi ] ) . the polarization function @xmath25 as written in eq . ( [ exactpi ] ) is , in general , uv divergent . the divergence can be subtracted by using the power series expansion of the weight function @xmath39 to an appropriate order which will be added and subtracted to this weight . this leads to a @xmath40-dependent power series of divergent subtraction terms plus an uv finite subtracted integral ( see e.g. @xcite ) . but because the subtraction terms do not contribute to the spectral density , we can avoid this subtraction at all . in order that the formally written expressions make sense they are supposed to be dimensionally regularized . we use the simplified or unorthodox dimensional regularization method for water melon diagrams ( see ref . @xcite ) . the threshold region of a water melon diagram is determined by the condition @xmath41 where @xmath42 is the euclidean momentum and @xmath43 is the threshold value for the spectral density . we introduce the minkowskian momentum @xmath44 defined by @xmath45 which is an analytic continuation to the physical cut . operationally this analytic continuation can be performed by replacing @xmath46 . to analyze the region near the threshold we use the parameter @xmath47 which takes complex values . the parameter @xmath48 is more convenient in euclidean domain while the parameter @xmath49 is the actual energy counted from threshold which is used in phenomenological applications . the spectral density as a function of @xmath50 is written as @xmath51 in the following . the analytic continuation of the fourier transform in eq . ( [ exactpi ] ) to the minkowskian domain has the form @xmath52 for the threshold expansion we have to analyze the large @xmath30 behaviour of the integrand . it is this region that saturates the integral in the limit @xmath53 or , equivalently , @xmath54 . it is convenient to perform the analysis in a basis where the integrand has a simple large @xmath30 behaviour . the most important part of the integrand is the bessel function @xmath55 which , however , contains both rising and falling branches at large @xmath30 . it resembles the situation with elementary trigonometric functions @xmath56 and @xmath57 to which the bessel function @xmath23 is rather close ( in a certain sense ) . indeed , @xmath57 ( or @xmath56 ) is a linear combination of exponentials , namely @xmath58 and has also both rising and falling branches at large pure imaginary argument : the exponentials show simple asymptotic behaviour @xmath59 at @xmath60 . the analogous statement is true for @xmath23 which can be written as a sum of two hankel functions , @xmath61 where @xmath62 . the hankel functions @xmath63 show simple asymptotic behaviour at infinity , @xmath64 accordingly we split up @xmath65 into @xmath66 with @xmath67 the two parts @xmath68 of the polarization function @xmath65 have completely different behaviour near threshold which allows one to analyze them independently . this observation makes the subsequent analysis straightforward . we first consider the contribution of the part @xmath69 . the behaviour at large @xmath30 is given by the asymptotic form of the functions which we simply write up to the leading terms as @xmath70 the large @xmath30 range of the integral ( above a reasonably large cutoff parameter @xmath71 ) has the general form @xmath72 where @xmath73 the right hand side of eq . ( [ pilap ] ) is an analytic function in @xmath48 in the vicinity of @xmath74 . it exhibits no cut or other singularities near the threshold and therefore does not contribute to the spectral density . we turn now to the second part @xmath75 . in contrast to the previous case , the integrand of this part contains @xmath76 which behaves like a rising exponential function at large @xmath30 , @xmath77 therefore the integral is represented by @xmath78 the function @xmath79 is non - analytic near @xmath74 because for @xmath80 the integrand in eq . ( [ pilam ] ) grows in the large @xmath30 region and the integral diverges at the upper limit . therefore the function which is determined by this integral is singular at @xmath80 ( @xmath81 ) and requires an interpretation for these values of the argument @xmath48 . in the complex @xmath48 plane with a cut along the negative axis the function is analytic . this cut corresponds to the physical positive energy cut . the discontinuity across the cut gives rise to the non - vanishing spectral density of the diagram . let us first discuss the analytic part @xmath69 of the diagram . this part reduces to a regular water melon diagram . indeed , using the relation @xmath82 between bessel functions of different kinds one can replace the hankel function @xmath83 with the mcdonald function @xmath84 . since the propagator of a massive particle ( massive line in the diagram ) is given by the mcdonald function up to a power in @xmath30 , this substitution shows that the weight function behaves like a propagator of an additional line with the `` mass '' @xmath44 . the explicit expression is given by @xmath85 @xmath86 is the polarization function of a new effective diagram which is equal to the initial polarization function multiplied by a propagator with @xmath44 as mass parameter . we thus end up with a vacuum bubble of the water melon type with one additional line compared to the initial diagram ( see fig . [ fig1 ] ) . these diagrams have no singular behaviour at the production threshold @xmath87 . as mentioned above , @xmath69 is analytic in @xmath48 near the origin @xmath74 and can therefore be omitted in the calculation of the spectral density . all derivatives of @xmath88 with respect to @xmath48 are represented as vacuum bubbles with one additional line carrying rising indices . such diagrams can be efficiently calculated within the recurrence relation technique developed in ref . @xcite . therefore one is left with the second part @xmath75 containing the non - analytical contributions in @xmath48 which leads to a non - vanishing spectral density . still the integral in eq . ( [ pm ] ) can not be done analytically . in order to obtain an expansion for the spectral density near the threshold in an analytical form we make use of the asymptotic series expansion for the function @xmath6 which crucially simplifies the integrands but still preserves the singular structure of the integral in terms of the variable @xmath48 . the asymptotic series expansion to order @xmath89 of the main part of each propagator , i.e. of the mcdonald function , is given by @xcite @xmath90,\quad ( \lambda , n):=\frac{\gamma(\lambda+n-1/2)}{n!\gamma(\lambda - n-1/2)}\ ] ] ( @xmath91 $ ] ) . therefore the asymptotic expansion of the function @xmath6 consists of an exponential factor @xmath92 and an inverse power series in @xmath30 up to an order @xmath93 which is closely related to @xmath89 . it is this asymptotic expansion that determines the singularity structure of the integral . we write the whole integral in the form of the sum of two terms , @xmath94 the integrand of the first term @xmath95 behaves as @xmath96 at large @xmath30 while the integrand of the second term accumulates all lower powers of the large @xmath30 expansion . note that only the large @xmath30 behaviour is essential for the near threshold expansion of the spectral density . this fact has been taken into account in eqs . ( [ pilap ] ) and ( [ pilam ] ) where we introduced a cutoff @xmath71 . however , from the practical point of view the calculation of the regularized integrals with an explicit cutoff is inconvenient . the final result of the calculation the spectral density of the diagram is independent of the cutoff , but the integration is technically complicated if the cutoff is introduced . however , in extending the integration over the whole region of the variable @xmath30 without using the cutoff one immediately encounters divergences at small @xmath30 because the asymptotic expansion is invalid in the region near the origin , so one is not allowed to continue the integration to this region . the standard way to cope with such a situation is to introduce dimensional regularization . it allows one to deal with divergent expressions at intermediate stages of the calculation and is technically simple because it does not introduce any cutoff and therefore does not modify the integration region drastically . note that dimensional regularization does not necessarily regularize all divergences in this case ( in contrast to the standard case of ultraviolet divergences ) but nevertheless suffices for our purposes . we use a parameter @xmath97 to regularize the divergences at small @xmath30 . this regularization prescription is an unorthodox version of the dimensional regularization ( see e.g.refs . @xcite ) . the first part @xmath95 in eq . ( [ pidias ] ) containing a difference of the polarization function and its asymptotic expansion as the integrand gives no contributions to the spectral density up to a given order of the expansion in @xmath48 . this is because the subtracted asymptotic series to order @xmath89 cancels the inverse power behaviour of the integrand to this degree @xmath89 . the integrand decreases sufficiently fast for large values of @xmath30 and the integral converges even at @xmath74 . therefore this term is inessential when the expansion of the spectral density is evaluated up to some order . one can readily determine the order of the expansion near @xmath74 at which a contribution to the spectral density appears in using some further simplifications of the integrand of the term @xmath95 in eq . ( [ pidias ] ) . namely , we replace the hankel function under the integration sign by its asymptotic series expansion . the resulting exponential factor @xmath98 can then be expanded in the parameter @xmath47 and integrated together with the finite inverse power series in @xmath30 . one obtains a finite power series in this parameter @xmath48 which leads to a non - regular term of order @xmath99 ( for instance , @xmath100 or @xmath101 ) . therefore the part @xmath95 is regular and gives no contribution to the spectral density up to the order @xmath99 . for this reason we concentrate on the expansion of the second part @xmath102 and find that only this part contains the contribution to the spectral density up to the @xmath89 . therefore the expansion of the spectral density at small @xmath50 is determined only by the integral @xmath102 of eq . ( [ pidias ] ) . this integral is still rather complicated to compute but we can go a step further in its analytical evaluation . indeed , since the singular behaviour of @xmath102 is determined by the behaviour at large @xmath30 , we can replace the first factor , i.e. the hankel function , in the large @xmath30 region by its asymptotic expansion up to some order @xmath89 . we use @xmath103\ ] ] ( cf . ( [ kas ] ) for the notation ) to obtain a representation @xmath104 the index `` _ das _ '' stands for `` double asymptotic '' and indicates that the integrand in eq . ( [ pidas ] ) consists of a product of two asymptotic expansions : one for the polarization function @xmath6 and another for the hankel function @xmath105 as weight ( or kernel ) . both asymptotic expansions are straightforward and can be obtained from standard handbooks on bessel functions . we therefore arrive at our final result : the integration necessary for evaluating the near threshold expansion of the water melon diagrams reduces to integrals of the type of euler s gamma function , i.e.integrals containing exponentials and powers . indeed , the result of the expansion in eq . ( [ pidas ] ) is an exponential function @xmath106 times a power series in @xmath107 , namely @xmath108 where @xmath109 has already been defined in eq . ( [ defa ] ) and the coefficients @xmath110 are simple functions of the momentum @xmath44 and the masses @xmath7 . the expression in eq . ( [ serx ] ) can be integrated analytically using @xmath111 the result is @xmath112 this expression is our final representation for the part of the polarization function of a water melon diagram necessary for the calculation of the spectral density near the production threshold . it is also one of the main results of our paper . next we discuss the general structure of the expression in eq . ( [ serint ] ) in detail . in the case where @xmath109 takes integer values , these coefficients result in @xmath113-divergences for small values of @xmath97 . the powers of @xmath48 in eq . ( [ serint ] ) have to be expanded to first order in @xmath97 and give @xmath114 because of @xmath115 @xmath116 in eq . ( [ serint ] ) contributes to the spectral density . for half - integer values of @xmath109 the power of @xmath48 itself has a cut even for @xmath117 . the discontinuity is then given by @xmath118 our method to construct a threshold expansion thus simply reduces to the analytical calculation of the integral in eq . ( [ pidas ] ) which can be done for arbitrary dimension and an arbitrary number of lines with different masses . in the next subsections we use our technique to work out some specific examples which demonstrate both the simplicity and efficiency of our method . the polarization function represented by the sunset diagram with three propagators with equal masses @xmath11 in @xmath119 space - time dimensions is given by @xmath120 the exact spectral density is given by the integral representation in eq . ( [ exactrho ] ) which for this particular case reads @xmath121 in order to obtain a threshold expansion of the spectral density in eq . ( [ exactrho4 ] ) we use eq . ( [ serint ] ) to calculate the expansion of the appropriate part of the polarization function . to illustrate the procedure we present the explicit shape of the integrand in eq . ( [ pidas ] ) which is given by an asymptotic expansion at large @xmath30 , @xmath122 from eq . ( [ ex ] ) we can easily read off the coefficients @xmath110 that enter the expansion in eq . ( [ serx ] ) . the spectral density is obtained by performing the term - by - term integration of the series in eq . ( [ ex ] ) and by evaluating the discontinuity across the cut along the positive energy axis @xmath81 . the result reads @xmath123 where the notation @xmath124 , @xmath125 is used . the simplicity of the derivation is striking . by no cost it can be generalized to any number of lines , arbitrary masses , and any space - time dimension . the standard equal mass sunset is chosen for the definiteness only . it also allows us to compare our results with results available in the literature . ( [ pidas430 ] ) reproduces the expansion coefficients @xmath126 obtained in ref . @xcite ( the fourth column in table 1 of ref . @xcite ) by a direct integration in momentum space within the technique of region separation @xcite . the case of the equal mass standard sunset diagram is the simplest one . there exists an analytical expression for the spectral density of the sunset diagram with three equal mass propagators in @xmath119 space - time dimensions in terms of elliptic integrals @xcite ( see also ref @xcite ) . this expression can be used for a comparison with our exact result in eq . ( [ exactrho4 ] ) or with the expansion in eq . ( [ pidas430 ] ) . however , we only present the result for @xmath127 in order to keep the resulting expressions in a reasonably short form ( cf.ref . @xcite ) . in @xmath127 space - time dimensions the spectral density for a sunset diagram with equal masses @xmath11 can be readily obtained . we just use the exact expression for the spectral density in the convolution representation @xcite and proceed towards @xmath128 equal masses . the convolution function for two spectral densities in @xmath127 dimensional space - time ( @xmath129 ) reads @xmath130 the two spectral densities one has to convolute are the spectral density of a correlator with two equal masses and the spectral density of a single massive line . while the latter is given by @xmath131 , the former can be obtained from eq . ( [ convolut ] ) by inserting @xmath132 , @xmath133 so the convolution leads to @xmath134 now we use the relation ( cf . ref . @xcite ) @xmath135 @xmath136 with @xmath137 and the definition of the complete elliptic integral of the first kind @xmath138 ( remark the difference in the definition ) for @xmath139 , @xmath140 , @xmath141 , and @xmath142 to perform the integration in eq . ( [ ellint ] ) . we obtain @xmath143 and finally end up with @xmath144 therefore the spectral density in terms of the energy @xmath50 reads ( see e.g.ref . @xcite ) @xmath145 by expanding the elliptic integral in terms of the threshold parameter @xmath50 one reproduces the threshold expansion in eq . ( [ pidas430 ] ) . the result for @xmath119 space - time dimensions is expressible by the elliptic integrals with some rational functions as factors that makes the result a bit longer . note that the representation in eq . ( [ ell ] ) is understood to be an analytical expression for the spectral density . however , it is a matter of taste whether the representation through the elliptic integrals as in eq . ( [ ell ] ) is considered simpler ( or in a more analytical form ) than the integral representation in eq . ( [ exactrho ] ) . the only objection against the latter which one can find in the literature is that the bessel functions are complicated ( see e.g. ref . but after more than a century of intensive investigation they are well - known and no more complicated than the square root of the fourth order polynomial which is used in eq . ( [ elldef ] ) to define the elliptic integral . the water melon diagrams with four or more propagators can not be easily done by using the momentum space technique because it requires the multiloop integration of entangled momenta . within the configuration space technique the generalization to any number of lines ( or loops ) is immediate by no effort . consider first a three - loop case of water melon diagrams ( also called banana diagrams @xcite or basketball diagrams @xcite ) . the polarization function of the equal mass water melon diagram with four propagators in @xmath119 space - time is given by @xmath146 the exact spectral density of this diagram can be obtained from eq . ( [ exactrho4 ] ) while the near threshold expansion can be found using eq . ( [ serint ] ) . we construct the expansion of the spectral density near threshold explicitly and compare it with the exact result . the expansion of the integrand ( cf . ( [ pidas ] ) ) reads @xmath147 after the integration and the calculation of the discontinuity we obtain the expansion of the spectral density in the form @xmath148 where @xmath124 and @xmath149 is the threshold value . one sees the difference with the previous three - line case . in eq . ( [ pidas440 ] ) the cut represents the square root branch while in the three - line case it was a logarithmic cut . one can easily figure out the reason for this by looking at the asymptotic structure of the integrand . for even number of lines ( i.e. odd number of loops ) it is a square root branch , while for an odd number of lines ( even number of loops ) it is a logarithmic branch . this is true in even space - time dimensions . in the general case the structure of the cut depends on the dimensionality of the space - time as well . the general formula reads @xmath150 for @xmath119 space - time dimension ( i.e. @xmath151 ) we can verify the result of ref . @xcite ( cf . ( [ genthr ] ) ) , @xmath152 numerically eq . ( [ pidas440 ] ) reads @xmath153 where we have written down the coefficients up to three decimal places . it is difficult to say anything definite about the convergence of this series . by construction it is an asymptotic series . however , we stress that the practical ( or explicit ) convergence can always be checked by comparing series expansions like the one shown in eq . ( [ sernum ] ) with the exact spectral density given in eq . ( [ exactrho4 ] ) by numerical integration . we conclude this part of the paper by the statement that the spectral density of the water melon diagram is most efficiently calculable within the configuration space technique . whether it is the exact result or the expansion , the configuration space technique can readily deliver the desired result . the exact formula in eq . ( [ exactrho4 ] ) as well as the threshold expansion obtained from it can be used to calculate the spectral density for an arbitrarily large number of internal lines . we stress that the case of different masses does not lead to any complications within the configuration space technique : the exact formula in eq . ( [ exactrho ] ) and/or the near threshold expansion work equally well for any arrangement of masses . we do not present plots for general cases of different masses because they are not very illustrative , showing only the common threshold . however , there is some interesting kinematic regime for different masses which is important for applications and which , to our best knowledge , have not been touched earlier . an analytical solution for the expansion of the spectral density in this regime is given in the next section . the threshold expansion for equal ( or close ) masses breaks down for @xmath155 . the example is shown in fig . [ fig2 ] for the @xmath119 proper sunset . however , if the masses are not equal , the region of the break - down of the expansion is determined by the mass with the smallest numerical value . the simplest example where one can see this phenomenon is the analytical expression for the spectral density of the simple loop ( degenerate water melon diagram ) with two different masses @xmath156 and @xmath157 . in @xmath119 space - time dimensions ( see e.g. ref . @xcite ) one has @xmath158 where @xmath159 . the threshold expansion is obtained by expanding the right hand side of eq . ( [ rho42 m ] ) in @xmath50 for small values of @xmath50 . if @xmath157 is much smaller than @xmath156 , the expansion breaks down at @xmath160 . the break - down of the series expansion can also be observed in more general cases . if one of the masses ( which we call @xmath161 ) is much smaller than the other masses , the threshold expansion is only valid in a very limited region @xmath162 . to generalize the expansion and extend it to the region of @xmath163 one has to treat the smallest mass exactly . in this case one can use a method which we call the resummation of the smallest mass contributions . below we describe the resummation technique . we start with the representation @xmath164 which is the part of the polarization function contributing to the spectral density . the integrand in eq . ( [ pipas ] ) has the form @xmath165 where the asymptotic expansions are substituted for all the propagators except for the one with the small mass @xmath161 . this is indicated by the index `` _ pas _ '' in eq . ( [ pipas ] ) which stands for `` partial asymptotic '' . the main technical observation leading to the generalization of the expansion method is that @xmath166 is still analytically computable in a closed form . indeed , the genuine integral to compute has the form @xmath167 where @xmath168 and @xmath169 . the integral @xmath166 in eq . ( [ pipas ] ) is thus expressible in terms of hypergeometric functions @xcite . for constructing the spectral density , being our main concern as mentioned before , one has to find the discontinuity of the right hand side of eq . ( [ intm0 ] ) . there are several ways to do this . we proceed by applying the discontinuity operation to the integrand of the integral representation of the hypergeometric function . the resulting integrals are calculated again in terms of hypergeometric functions . indeed , @xmath170 where @xmath171 . the final expression in eq . ( [ genuine ] ) completely solves the problem of the generalization of the near threshold expansion technique . for integer values of @xmath172 there are no singular gamma functions ( with negative integer argument ) . therefore we can lift up the regularization and set @xmath117 when using this expression . we thus have found a direct transition from the polarization function as expressed through the integral to the spectral density in terms of one hypergeometric function for each genuine integral . there is no need to use the recurrence relations available for hypergeometric functions . in the following subsections we give explicit examples for @xmath119 and compare with the exact result in eq . ( [ exactrho4 ] ) and the pure expansion near the threshold . in the following the standard threshold expansion without resummation is called the pure threshold expansion . we start with a ( over)simplified example of the two - line diagram with masses @xmath11 and @xmath173 in four space - time dimensions . in this example the expansion of the spectral density and its generalized expansion can be readily compared analytically with the exact result in eq . ( [ rho42 m ] ) . this is the feature that justifies our discussion in this section . the results for the spectral density of this diagram are shown in fig . the solid curve displays the exact result obtained by using eq . ( [ exactrho4 ] ) ( which reproduces the analytical expression in eq . ( [ rho42 m ] ) ) . we compare this result with the two expansions . the pure expansion of the spectral density near threshold ( the second order asymptotic expansion should suffice to show the general features in a short and concise form ) is given by @xmath174 where @xmath175 . as mentioned above , this series breaks down for @xmath176 ( see eq . ( [ rho42 m ] ) ) . if we look at the dotted curves in fig . [ fig3 ] this becomes obvious . here we have plotted the series expansions up to the fourth order with the mass arrangement @xmath177 . the dashed lines represent the resummation of the smallest mass contributions . the analytical expression for the spectral density of the polarization function in eq . ( [ pipas ] ) for the generalized asymptotic expansion based on eq . ( [ genuine ] ) is given by @xmath178 we have set the regularization parameter @xmath117 because the spectral density is finite . with @xmath117 the resulting expressions for the hypergeometric functions in eq . ( [ genuine ] ) simplify . the first term in the curly brackets of eq . ( [ pipas42 t ] ) is obviously equal to @xmath179 in this limit because the first parameter of the hypergeometric function vanishes for @xmath117 . however , we keep eq . ( [ pipas42 t ] ) in its given form to show the structure of the contributions . the generalized threshold expansion has the form @xmath180 where the functions @xmath181 represent effects of the resummation of the smallest mass and are not polynomials in the threshold parameter parameter @xmath50 . in the simple two - line case the hypergeometric functions reduce to elementary functions . for instance , @xmath182 higher order contributions are given by hypergeometric functions with larger numerical values of the parameters . they can be simplified by using gaussian recurrence relations for hypergeometric functions ( see e.g.ref . @xcite ) . the convergence of the expansion in eq . ( [ pipas42 t ] ) breaks down only at @xmath183 . the resummation leads to an essential improvement of the convergence in comparison with the pure threshold expansion . in fig . [ fig4 ] we show the same curves divided by the leading order term . this representation is more convenient for the diagrams which we will discuss in following subsections . note that eq . ( [ resumsm ] ) does not lead to the exact function in eq . ( [ rho42 m ] ) because terms of order @xmath184 stemming from the difference part @xmath95 of the correlator are missing . it simply corrects the behaviour of the coefficient functions by the small mass contributions . here we analyze the sunset diagram with two equal masses @xmath11 and a third mass @xmath173 ( @xmath177 ) . the exact result obtained by using eq . ( [ exactrho4 ] ) and normalized to the leading order term is shown in fig . [ fig5 ] as the solid curve . the pure expansion near the threshold reads @xmath185 it is shown by the dotted curves in fig . [ fig5 ] . in case of the resummation of the smallest mass contributions we obtain hypergeometric functions which do not obviously reduce to elementary functions in this case . the result for the spectral density within the asymptotic expansion up to the second order in eq . ( [ pipas ] ) is given by @xmath186 we see that the dashed curves in fig . [ fig5 ] that represent the result in eq . ( [ pipas43 t ] ) approximate the exact curve much better than the dotted curves . with this example we conclude our consideration of the strongly asymmetric case and at the same time show the way to the multi - line water melon diagrams which can be treated in an analogous manner . the result for the exact expression obtained by using eq . ( [ exactrho4 ] ) is shown in fig . [ fig6 ] as a solid line , normalized to the leading order term . the dotted lines represent the results for the pure expansion near threshold which is given by @xmath187 the asymptotic expansion to the second order in eq . ( [ pipas ] ) gives @xmath188 in fig . [ fig6 ] one can see how the expansion improves if the resummation of the smallest mass contributions ( displayed as dashed lines ) is performed . the result of this section is quite general and applicable to all cases of one small mass . for some particular arrangement of masses one can obtain even simpler expressions as discussed in the next section . in this section we obtain a result for the resummation of the smallest mass effects along a different route , namely , via the convolution of spectral densities . however , this method works in a narrower kinematic region than the method described in the previous section . in @xmath119 space - time dimensions , the convolution weight is given by @xmath189 the upper limit of the integration is determined by the requirement of positivity of the the square root argument . the zeros of the square root with respect to @xmath190 are given by @xmath191 , and the demand @xmath192 together with @xmath193 leads to @xmath194 or @xmath195 . the physical region is the latter one . with @xmath196 for the spectral density of the single small mass line we obtain @xmath197 where the low limit of integration is @xmath198 . we insert @xmath199 and @xmath200 and obtain @xmath201 where @xmath202 . for this function we use the threshold expansion in @xmath203 as expansion parameter . for small @xmath204 the threshold expansion inserted for @xmath205 is valid because @xmath204 implies @xmath206 . the described procedure can be extended to the case of a very light sub - block of the diagram , e.g. a light fish diagram . in this case we have to replace @xmath207 by the spectral density of the light sub - diagram which is well - known . the analytic structure of water melon diagrams is completely fixed by the dispersion representation . therefore we have focussed on the computation of the spectral density as the basic quantity important both for applications and the theoretical investigation of the diagram . however , with an analytical expression for the spectral density @xmath19 at hand we can readily reconstruct the non - analytic piece of the polarization function in momentum space by using the dispersion relation @xmath208 we rewrite this equation in terms of threshold parameters according to @xmath209 and @xmath199 and obtain @xmath210 uv singularities can be removed by subtraction or by dimensional regularization . we again use the unorthodox dimensional regularization prescription . for a general form of the threshold expansion @xmath211 we have to calculate integrals of the form @xmath212 only the powers @xmath213 contribute to the singular part of the polarization function . expressions like the one presented in eq . ( [ restore ] ) then allow one to restore that part of the polarization function @xmath65 which has singularities near the threshold . we have discussed the configuration space technique for the calculation of @xmath3-line two - point diagrams termed water melon diagrams . this technique allows one to calculate the spectral density for arbitrary space - time dimensions and any number of internal lines with arbitrary mass values . within this technique one can use either an exact representation as one - fold integral or an expansion near the production threshold . we have developed an efficient method for constructing the near threshold expansion of water melon diagrams that uses only asymptotic expansions of bessel functions which are well - known and simple functions . we have considered a strongly asymmetric mass arrangement where one mass is much smaller than the others . in this case the `` pure '' threshold expansion which is done in terms of the threshold parameter @xmath50 breaks down at the energies in the vicinity of the smallest mass . in order to extend the approximation to higher energies we have introduced a generalized threshold expansion which exactly resums all contributions of the smallest mass . we have presented closed expressions for the generalized expansion in several particular cases and demonstrated the improvement of the convergence gained by the resummation of the smallest mass contributions . the particular kinematic regime of this case could be treated analytically because it reduced to the evaluation of the one - fold integral in terms of hypergeometric functions . the discontinuity across the physical cut for the generalized expansion has been found in terms of hypergeometric functions as well . to conclude , we stress that the configuration space technique is a powerful and convenient tool for investigating different properties of water melon diagrams . the practical convenience of our method is demonstrated in fig . [ fig7 ] where we have plotted the spectral density for a four - line water melon diagram in @xmath214 , @xmath215 , and @xmath216 space - time dimensions . * acknowledgments * + we thank andrei davydychev and jrgen krner for interesting discussions and careful reading the manuscript . a.a . pivovarov acknowledges a valuable discussion with v.a . smirnov on the region separation technique in threshold expansions . the work of a.a . pivovarov is supported in part by the volkswagen foundation under contract no . i/73611 and by the russian fund for basic research under contract 99 - 01 - 00091 . s. groote gratefully acknowledges a grant given by the graduiertenkolleg `` eichtheorien experimentelle tests und theoretische grundlagen '' . 99 particle data group , `` review of particle properties '' , eur . j. * c3 * ( 1998 ) 1 k.g . chetyrkin , j.h . khn and a. kwiatkowski , phys . * 277 * ( 1996 ) 189 k.g . chetyrkin , j.h . khn and m. steinhauser , nucl . * b505 * ( 1997 ) 213 d.j . broadhurst , eur . j. * c8 * ( 1999 ) 311 o.m . ogreid , p. osland , `` more series related to the euler series '' , + report no . bergen-1999 - 04 , hep - th/9904206 r. coquereaux , phys . rev . * d23 * ( 1981 ) 2276 ; + v.i . zakharov , nucl . * b385 * ( 1992 ) 452 ; + d.j . broadhurst and a.g . grozin , phys . rev . * d52 * ( 1995 ) 4082 s. groote , j.g . krner and a.a . pivovarov , nucl . * b542 * ( 1999 ) 515 ; eur . j. * c11 * ( 1999 ) 279 ; phys . * 443 b * ( 1998 ) 269 f.a . berends , m. buza , m. bhm and r. scharf , z. phys . * c63 * ( 1994 ) 227 f.a . berends , a.i . davydychev and n.i . ussyukina , phys . lett . * 426 b * ( 1998 ) 95 s. coleman and e. weinberg , phys . * d7 * ( 1973 ) 1888 ; + r. jackiw , phys . rev . * d9 * ( 1974 ) 1686 ; + r. jackiw and s.templeton , phys . * d23 * ( 1981 ) 2291 a.k . rajantie , nucl . * b480 * ( 1996 ) 729 j.m . chung and b.k . chung , `` renormalization group improvement of the effective potential in massive @xmath217 theory : nnnlo logarithm resummation '' , report no . mit - ctp-2929 , hep - th/9911196 j.o . andersen , e. braaten and m. strickland , `` the massive thermal basketball diagram '' , hep - ph/0002048 j.m . chung and b.k . chung , phys . * d59 * ( 1999 ) 105014 d.j . gross , r.d . pisarski and l.g . yaffe , rev . phys . * 53 * ( 1981 ) 43 ; + t. appelquist and r.d . pisarski , phys . * d23 * ( 1981 ) 2305 b.l . ioffe , nucl . * b188 * ( 1981 ) 317 n.v . krasnikov , a.a . pivovarov and a.n . tavkhelidze , + z. phys . * c19 * ( 1983 ) 301 ; jetp lett . * 36 * ( 1982 ) 333 a.a . ovchinnikov , a.a . pivovarov and l.r . surguladze , + yad . fiz . * 48 * ( 1988 ) 562 ; int . j. mod . * a6 * ( 1991 ) 2025 s. groote , j. g. krner and a. a. pivovarov , phys . * d61 * ( 2000 ) 071501 g. t hooft , nucl . phys . * b72 * ( 1974 ) 461 e. witten , nucl . phys . * b160 * ( 1979 ) 57 s.a . larin _ et al . _ , yad . fiz . * 44 * ( 1986 ) 1066 s.a . kulagin and a.a . pivovarov , yad . * 45 * ( 1987 ) 952 p. post and k. schilcher , phys . * 79 * ( 1997 ) 4088 j. gasser and m.e . sainio , eur . j. * c6 * ( 1999 ) 297 s. narison and a.a . pivovarov , phys . * 327 b * ( 1994 ) 341 i.i . balitsky , d.i . dyakonov and a.v . yung , phys . * 112 b * ( 1982 ) 71 d.j . broadhurst , z. phys . * c54 * ( 1992 ) 599 p.a . baikov , `` advanced method of solving recurrence relations for multiloop feynman integrals '' , talk presented at the @xmath218 international workshop on new computing techniques in physics research ( aihenp 99 ) , heraklion , greece , 1216 april , 1999 n.e . ligterink , `` solving multiloop feynman diagrams using light front coordinates '' , report no . ect-99 - 016 , hep - ph/9911411 a.i . davydychev and v.a . smirnov , nucl . * b554 * ( 1999 ) 391 b. kastening , h. kleinert , `` efficient algorithm for perturbative calculation of multiloop feynman integrals '' , quant - ph/9909017 g.n . watson , `` theory of bessel functions '' , cambridge , 1944 e. mendels , nuovo cim . * 45 a * ( 1978 ) 87 c.s . meijer , proc . amsterdam akad . ( 1940 ) 599 ; 702 a. erdelyi ( ed . ) , `` tables of integral transformations '' , + volume 2 , bateman manuscript project , 1954 a.h . hoang _ et al . _ , `` top - antitop pair production close to threshold : synopsis of recent nnlo results , '' report no . slac - pub-8369 , hep - ph/0001286 ; + a. czarnecky and k. melnikov , phys . * 80 * ( 1998 ) 2531 ; + m. beneke , a. signer and v.a . smirnov , phys . * 80 * ( 1998 ) 2535 ; + k. melnikov and a. yelkhovsky , nucl . * b528 * ( 1998 ) 59 ; + a.a . penin and a.a . pivovarov , hep - ph/9904278 ; nucl . * b549 * ( 1999 ) 217 ; nucl . * b550 * ( 1999 ) 375 ; phys . lett . * 443 b * ( 1998 ) 264 ; + o.i . yakovlev , phys . lett . * 457 b * ( 1999 ) 170 ; + a.h . hoang and t. teubner , * d60 * ( 1999 ) 114027 ; + t. nagano , a. ota and y. sumino , phys . * d60 * ( 1999 ) 114014 n.n . bogoliubov and d.v . shirkov , `` quantum fields '' , benjamin , 1983 a.a . pivovarov , phys . * 236 b * ( 1990 ) 214 ; phys . lett . * 263 b * ( 1991 ) 282 v.a . smirnov , phys . * b404 * ( 1997 ) 101 ; + m. beneke and v.a . smirnov , nucl . * b522 * ( 1998 ) 321 ; + v.a . smirnov and e.r . rakhmetov , teor . mat . 120 * ( 1999 ) 64 a.p . prudnikov , yu.a . brychkov and o.i . marichev , + `` integrals and series '' , vol . 2 , gordon and breach , new york , 1990 b.almgren , arkiv fr fysik , * 38 * ( 1967 ) 161 ; + s.bauberger , f.a . berends , m. bhm and m. buza , nucl . * b434 * ( 1995 ) 383 s. groote , j. g. krner and a. a. pivovarov , phys . * d60 * ( 1999 ) 061701 a.a . pivovarov and v.f . tokarev , yad . fiz . * 41 * ( 1985 ) 524 i.s . gradshteyn and i.m . ryzhik , + `` tables of integrals , series , and products '' , academic press , 1994 m. abramowitz , i.a . stegun ( eds . ) , `` handbook of mathematical functions '' , dover publ . , new york , 9th printing , 1970

the experimental realization of quantum condensation of quasiparticles with finite lifetimes has opened the possibility to study situations where the steady states of the condensate are not just controlled by energetics , but involve also consideration of pumping and decay , leading to steady states with quasiparticle currents . the exploration of the possible behavior exhibited in these systems is driven particularly by recent experimental progress in microcavity polaritons@xcite and magnons@xcite ( as well as more exotic realizations in helium@xcite ) . microcavity polaritons are the quasiparticles that result from strong coupling between photons confined in planar semiconductor microcavities and excitons confined in quantum wells . the photons are confined by bragg reflectors to a two - dimensional cavity , and for small in - plane momentum have an effectively quadratic dispersion , with a mass of the order of @xmath0 @xmath1 of the free electron mass . from strong coupling between these photon states and exciton states in quantum wells , new normal modes arise : polaritons . these are hybrid light matter particles , which have both a light effective mass , as well as sufficient interactions to allow a quasi - equilibrium particle distribution to be established . however , the distribution is only quasi - equilibrium , since the photon component is not perfectly confined , so polaritons have a finite lifetime , and to maintain a steady state population , continuous pumping is required . even when the energy distribution of polaritons looks reasonably thermal , this does not mean effects of finite lifetime may be neglected , as one may have thermalized energy distributions on top of steady state flows driven by the pumping and decay . experimentally , it is observed that above a threshold pumping strength , an accumulation of low energy polaritons is accompanied by : a significant increase in temporal coherence , spatial coherence that extends over the entire cloud of polaritons@xcite , the appearance of quantized vortices@xcite , suggestions of changes to the excitation spectrum@xcite , and a robustness of polariton propagation to disorder@xcite . of these , the observation of spontaneous vortices@xcite in polariton condensates provides a strong hint that the steady states of the system are affected by the flow of particles . for a more general introduction to microcavity polaritons , their condensation , and the physical systems studied there exist several reviews or books , e.g. , refs . . the aim of this article is to investigate the rich varieties of behavior that come from combining spatial profiles driven by particle flow with the dynamics due to the spin degree of freedom that microcavity polaritons also possess . the effects of this spin degree of freedom have been previously considered in equilibrium : due to the weak attractive interaction between opposite spin polarizations , a condensate is expected to be linearly polarized@xcite , i.e. , to have an equal density of left- and right - circularly polarized components . applying a magnetic field can then lead to a phase transition , as the density of one circular polarization increases , and the other decreases , leading through an elliptically polarized phase to a single circularly polarized condensate@xcite . in the absence of further symmetry breaking , the linear or elliptically polarized phase has two gapless modes , corresponding to two broken symmetries : the overall polariton phase , and the direction of linear polarization ( i.e. , orientation of elliptical polarization ) . the transition to a circular polarized state is thus a phase transition , as such a phase has only a single broken symmetry , and a single gapless mode . assuming no other symmetry breaking terms , the linearly polarized condensate can have independent vortices of the left- and right - circularly polarized components@xcite ; such polarization - dependent phase winding has recently been observed@xcite . these two vortices are not completely independent . even in the absence of any further symmetry breaking , there is a short - range density - density interaction between vortices of opposite polarizations@xcite , however such a term does not depend on the circulations of the two vortices . if one also takes account of te - tm splitting of the electromagnetic modes@xcite , this leads to a term that splits linear polarization states according to whether the polarization is parallel or perpendicular to the polariton momentum ; as such , this provides a circulation dependent interaction between vortices of the two polarizations@xcite . in addition , real materials are expected to possess a small splitting between linear polarization states due to asymmetry of the quantum - well interfaces in non - centro - symmetric crystals@xcite ; such a splitting can also be controlled by applying electrostatic fields@xcite or stress@xcite . such terms will again induce interactions between vortices of the left- and right - circular polarization , and if strong enough , will lead to a pinning of the polarization of a polariton condensate , as has been observed in experiment @xcite . considering the splitting of linear polarizations as a phase - locking term between the two circular polarization components , and combining this with a magnetic field that favors one or other circular polarization , one has a josephson problem@xcite , where the energy favoring linear polarization is a josephson coupling , and the magnetic field leads to a bias between the two fields . @xcite have considered the interplay between these spinor josephson oscillations , and josephson coupling between two different spatial modes of a trapped polariton condensate , showing that complex behavior can arise in this four - mode system without pumping and decay . a similar problem has also been studied in the context of spinor condensates of cold atoms in double well potentials@xcite . if one does include pumping and decay then even the two - mode system ( i.e. , just the dynamics of the spin component ) can show a rich variety of behavior @xcite . due to dissipation , the dynamics settles on an attractor , which may either correspond to a phase - locked ( i.e. , linearly polarized ) state , or may correspond to a limit cycle , in which the phase difference between the two polarization components continually increases , with the cyclic nature arising from @xmath2 periodicity in the phase difference . there also exist parameter ranges where there is bistability , so that which attractor the system settles on depends on the initial conditions . in this article , we will both review this two - mode dynamics , and then extend to the case of including many spatial modes in addition to the spin dynamics . to describe multiple spatial modes in a trapped , pumped , decaying condensate , a mean - field approach to this problem leads to a complex gross pitaevskii equation@xcite . it is also possible to include fluctuations beyond mean - field theory within such a formalism by quantum stochastic approaches@xcite . as well as describing the spatial structure of nonequilibrium condensates , the complex gross pitaevskii equation has also been applied in a wide range of areas , see for example refs . . one application which is very closely connected to microcavity polaritons is the dynamics of lasers propagating through strongly nonlinear materials , where combinations of cubic and quintic nonlinearities can produce a state with a preferred density , sometimes referred to as `` liquid light''@xcite . the results in this article suggest that consideration of polarization dynamics in such materials may also provide interesting results . by considering the spinor complex gross pitaevskii equation to describe the spinor condensate in an harmonic trap , several new classes of attractors are seen to occur in addition to those present for the two - mode system . there are limit cycles describing small oscillations of the phase difference between the two components , and limit cycles with @xmath3 periodicity of the phase difference . in addition , the different dynamical attractors correspond to different spatial profiles , in which the presence or absence of vortex lattices can be influenced by the applied magnetic fields . as well as describing the steady state profiles , the complex gross pitaevskii equation can describe the normal modes for small fluctuations about such steady states . in the absence of a spatial trap , one notable consequence of including pumping and decay is that the long - wavelength modes of a dissipative condensate are modified , to become diffusive@xcite ; similar results are seen both for incoherent pumped condensates , and for optical parametric oscillation , where scattering of pumped polaritons into signal and idler states is modified by bosonic enhancement due to the population of the signal@xcite . from this standpoint , another purpose of this article is to address how the combination of pumping , decay , and symmetry - breaking terms modify the spectrum of the spinor condensate . while a diffusive mode for the overall polariton density and phase always exists , the long - wavelength modes for spin modes can show a range of behaviors : diffusive , linearly dispersing , or gapped , dependent on the balance of decay and symmetry - breaking terms . the remainder of this article is arranged as follows . in section ii we introduce the model of spinor condensates as a system of coupled gross pitaevskii equations with pumping and decay terms . by neglecting the spatial variations of polarized components we discuss the bifurcation diagrams for a two - mode system in section iii . the normal modes of the homogeneous system are obtained in section iv . in section v we study the stability of cross - polarized vortices . the detailed study of the dynamics of the full trapped system is performed in section vi . in section vii we discuss how various regimes can be detected in experiments . the conclusions in section viii summarize our findings . the model we consider in this article consists of the complex gross - pitaevksii equation ( cgpe ) as described in ref . , taking into account the two possible polarizations of polaritons , written in the basis of left- and right - circular polarized states , denoted by @xmath4 . in addition to the terms discussed in ref . , three new parameters arise from considering the spin degree of freedom in an applied magnetic field . firstly , in addition to an interaction with the total polariton density @xmath5 . secondly , there is a magnetic field term @xmath6 , which may naturally arise@xcite due to asymmetry at the quantum - well interfaces@xcite , or may be induced by electric fields@xcite or stress@xcite . put together , these yield the coupled cgpe : @xmath7 \psi_l + j_1 \psi_r,\end{gathered}\ ] ] with the analogous equation for @xmath8 following by replacing @xmath9 and @xmath10 . in this article , we will either neglect @xmath11 ( in sections [ sec : review - two - mode ] , [ sec : norm - modes - extend ] ) , or consider a harmonic trap , @xmath12 . in either case , we will rescale energies and lengths in terms of the harmonic - oscillator energy @xmath13 , and length @xmath14 such that @xmath15 . by additionally rescaling the density such that @xmath16 the cgpe can be written as : @xmath17 \psi_l + j \psi_r\end{gathered}\ ] ] with the definitions @xmath18 . if considering a harmonic trap , then @xmath19 , otherwise we will take @xmath20 . the dimensionless parameter @xmath21 describes the extent of spin anisotropy of the interaction ; using the estimate in ref . we take @xmath22 . from estimates of the polarization splitting in refs . , we take @xmath23 . using the estimates of trap frequency given in ref . , we will take @xmath24 , and we may write the dimensionless splitting @xmath25 ; in the numerical results , we will consider a variety of values of @xmath26 around this value . the spatially extended , harmonically trapped system is unstable unless pumping is restricted to a finite spot@xcite . for our studies of this system , we therefore consider a circular pumping spot with a cutoff radius @xmath27 : @xmath28 . for constant pumping strength , @xmath27 determines whether the system forms vortices spontaneously@xcite . in order to provide a basis from which to understand the behavior observed in the trapped system , it is first useful to review the results that occur in the `` two - mode model '' , where spatial variation of each component is neglected , so the coupled complex gross pitaevskii equation reduces to two equations for the variables @xmath29 . this model was studied by @xcite , where the conditions for the existence of a steady state ( i.e. , synchronized ) solution were found . as @xmath30 increased , @xcite found that a synchronized solution could persist for some range of @xmath30 , with an increasing phase and density difference between the two modes , until at some critical @xmath30 the steady state vanished . near the critical @xmath30 , the existence of bistability was noted . this section will both summarize these previous results , as well as analyze the behavior of the desynchronized solution , showing that the dynamics is analogous to that of the damped driven pendulum , or current - biased josephson junction@xcite . to discuss the dynamics , it is convenient to reparametrize using : @xmath31 and write coupled equations for @xmath32 . ( the global phase @xmath33 does not affect the dynamics . ) in visualizing the dynamics , one may consider a bloch vector , defined by @xmath34 , in terms of the above variables . since there is pumping and decay , the length of the bloch vector is not conserved , hence there is a dynamical equation for @xmath35 . the coupled equations have the form : @xmath36 writing the equations in this form firstly allows one to understand the basic effects of pumping and decay on the dynamics , and , secondly , will be shown below to reduce to the damped driven pendulum in a relevant limit . for a steady state , it is clear that the condition @xmath37 defines a bloch surface , @xmath38 . one may then study the dynamics of points that start on this surface to characterize the nature of the transition as @xmath30 increases . as shown in fig . [ fig : attraction - basin ] , each starting point @xmath39 is attracted either to a fixed point , or to a limit cycle . the fixed point is the synchronized solution ; the limit cycle is the desynchronized solution , in which the phase difference @xmath40 between the two spin components is changing this behavior is a cycle since the dynamics are periodic under @xmath41 . as @xmath30 increases , there is a lower critical @xmath42 below which the all initial conditions on the bloch surface flow to the synchronized solution , and an upper critical @xmath43 which corresponds to the previously found@xcite point beyond which synchronized solutions can no longer exist . , as indicated , and shows a colormap according to the final state found by starting from the given initial conditions @xmath44 . points that flow to the fixed point are colored gray , others are white . the fixed point is marked by a red cross , and the limit cycle ( when it exists ) by the magenta line.,width=307 ] the behavior seen in fig . [ fig : attraction - basin ] can be understood by noticing that the choice of parameters used puts one in the `` josephson regime '' of the josephson equations@xcite , i.e. , @xmath45 , and that in addition the typical dynamics obey @xmath46 . in this case , eq . ( [ eq:6 ] ) simply reduces to @xmath47 . then , by eliminating @xmath48 from eq . ( [ eq:4],[eq:5 ] ) , one finds : @xmath49 this equation describes a damped driven pendulum , or alternatively a current - biased josephson junction . , due to our choice of sign of @xmath26 , `` gravity '' for the pendulum acts to drive @xmath50 ) . the behavior of eq . ( [ eq:7 ] ) is well known ( see , e.g. , ref . ) ; for large damping @xmath51 there is a simple transition at @xmath52 between fixed points for @xmath53 and limit cycles above . for smaller damping ( explicitly , for for @xmath54 ) , there is a range of bistability , where both limit cycles and fixed points may be found@xcite . the inset of fig . [ fig : ft - two - mode ] shows the bifurcation diagram determined from eq . ( [ eq:4][eq:6 ] ) , and for comparison , the value @xmath43 and the critical damping that result from the approximations leading to eq . ( [ eq:7 ] ) . as well as explaining the classes of behavior seen , eq . ( [ eq:7 ] ) may be used to explain the critical behavior near @xmath42 . for @xmath51 sufficiently small that bistability exists , then near @xmath42 , the period of the limit cycle grows like @xmath55|$ ] . in the context of the current problem , this period relates to the average chemical potential difference between the two components , @xmath56 . for large @xmath30 , one eventually reaches @xmath57 . the evolution of @xmath58 is illustrated in fig . [ fig : ft - two - mode ] , which plots the spectral weight @xmath59 , choosing initial conditions so the limit cycle is obtained for all @xmath60 . , showing bifurcation of frequencies as @xmath30 is increased , and illustrating critical behavior @xmath61 . initial conditions for each @xmath30 are chosen so that a limit cycle ( i.e. , desynchronized solution ) is found for all @xmath62 . inset : regions of stability of the synchronized and desynchronized solutions found from eq . ( [ eq:4][eq:6 ] ) . the dotted line marks the approximate @xmath43 appropriate in the josephson regime , eq . ( [ eq:7 ] ) , and the @xmath63 marks the known bifurcation point@xcite in that regime.,width=307 ] before considering the interplay of spin dynamics with the spatial profile of a trapped condensate , one may first consider a simpler problem involving the interplay of spin and spatial dynamics the finite momentum normal modes of the pumped , decaying spinor condensate . the normal modes of the spinor condensate without pumping and decay were discussed in @xcite . in the equilibrium case , for small @xmath64 , there is an elliptical condensate ( i.e. , a finite density of both spin components ) and there are two gapless linear modes , describing excitations of the global phase , and the relative phase of the two components . when @xmath64 becomes large enough to cause a transition to a circularly polarized state , only a single gapless mode survives , describing phase modes of the condensed component . on the other hand , the presence of pumping and decay is known to replace the linear dispersion of the phase modes with a diffusive behavior at small momentum@xcite . the aim of this section is to see how these two effects are combined . the result is that while the global phase mode remains diffusive , the real part of the relative phase mode can be either gapped , linearly dispersing , or diffusive according to the value of @xmath30 chosen . in the following , we will first present numerical results for the normal - mode frequencies , and then discuss how these can be straightforwardly interpreted in the same josephson regime as discussed above . for the purpose of numerically calculating the normal modes , it is simplest to consider the bogoliubov parametrization of fluctuations at some wavevector @xmath65 ( allowing for decay ) as : @xmath66 substituting this into eq . ( [ eq:2 ] ) , and linearizing in @xmath67 yields the secular equation @xmath68 = 0 $ ] , where in the basis @xmath69 , the matrix @xmath70 is @xmath71 noting that for plane waves , the kinetic energy term @xmath72 , the matrix elements are : @xmath73 and similarly with @xmath74 . the normal modes calculated this way are shown in figs . [ fig : uniform - stability ] , [ fig : uniform - spectrum ] . figure [ fig : uniform - stability ] shows the modes at @xmath75 ( lower panels ) as well as the value of @xmath30 , and the densities @xmath76 ( upper panels ) corresponding to a solution with a given value of @xmath40 . [ this is to make use of the fact that the steady state of eqs . ( [ eq:4][eq:6 ] ) can be found explicitly for @xmath77 as a function of @xmath40 ( see ref . ) . ] , between the two circular polarization components . top left : densities of components in steady state solution . top right : detuning corresponding to given solution . bottom : real ( left ) and imaginary ( right ) parts of frequency for small fluctuations about the steady state . the gray shaded region is unstable to small fluctuations . the crosses mark the conditions used for the finite @xmath65 spectra in fig . [ fig : uniform - stability ] . plotted for @xmath78.,width=307 ] in fig . [ fig : uniform - stability ] , the range of @xmath40 is restricted to @xmath79 $ ] , since the range @xmath80 $ ] is equivalent to this , after swapping @xmath74 and @xmath81 . within this range , only values @xmath82 correspond to stable solutions . as only the modes at @xmath75 are shown , there is always a zero mode corresponding to global phase rotations . the other three modes divide into an overdamped mode ( largest imaginary part ) , and a pair of modes which can be either overdamped or underdamped , as seen by the bifurcation of the real part at @xmath83 . three values of @xmath40 ( corresponding to three different applied detunings @xmath30 ) are chosen to illustrate the overdamped , critically damped , and underdamped cases , and the normal modes with finite @xmath65 are shown for these points in fig . [ fig : uniform - spectrum ] . in the overdamped case ( for @xmath40 near @xmath84 ) , both the relative - phase and global - phase modes are diffusive at small @xmath65 . when underdamped , the relative - phase mode always has a nonzero real frequency . when critically damped , the real part of the relative - phase mode has a linear dispersion . we will next discuss how this behavior can be understood in the josephson regime . . in all cases the total density modes are diffusive , while the spin mode changes from diffusive ( top row , @xmath40 nearest @xmath84 ) , to linearly dispersive ( middle row ) to underdamped spin oscillations ( bottom row , @xmath40 nearest @xmath85).,width=307 ] to understand the modes qualitatively , it is clearest to work in the basis of @xmath86 . extending eqs . ( [ eq:4][eq:6 ] ) to include spatial gradients via a madelung transformation , one finds the equations : @xmath87 where we have introduced the shorthand @xmath88 , with @xmath89 . when considering fluctuations about the uniform state , the gradient terms simplify , as expressions such as @xmath90 are necessarily second order in the fluctuations , and so may be neglected . we will make the same approximations as led to eq . ( [ eq:7 ] ) ( i.e. , assume @xmath91 ) , and use the steady - state solution in this limit ( @xmath92 ) to simplify the equations for fluctuations . if one writes the fluctuations as @xmath93 then since these are real variables , the normal modes have the time dependence @xmath94 , giving a secular equation : @xmath95 = 0,\ ] ] where @xmath96 indicates an expansion in powers of momentum , to understand the small @xmath65 dispersion . the first two terms are : @xmath97 the nature of the @xmath75 normal modes is immediately clear : @xmath33 has no restoring force , and so has a zero frequency oscillation ; @xmath98 has a damping rate @xmath99 , and so describes a decaying mode ; @xmath48 and @xmath40 are mixed , and have modes with frequencies given by : @xmath100 noting that @xmath101 is negative , and that the prefactor is the same as that of @xmath102 in eq . ( [ eq:7 ] ) , one may write : @xmath103 where @xmath104 is a `` plasma frequency '' describing the restoring force as a function of angle . the normal modes are @xmath105 . the transition between under- and overdamping occurs because for @xmath106 the restoring force is large so @xmath107 , while as @xmath108 , the restoring force vanishes , and so there is an intermediate value of @xmath40 ( and hence @xmath30 ) at which @xmath109 , describing the critical damping . note that whatever the choice of parameters , an overdamped regime will always exist since @xmath110 must always tend to zero at @xmath111 . however if @xmath51 is sufficiently large the underdamped regime may vanish . the approximate expressions for the eigenvalues @xmath112 match the general form seen in the bottom panel of fig . [ fig : uniform - stability ] , it is apparent that the full problem has some additional variation of damping rate with @xmath40 , not captured by the approximations used in the above expressions . one may now explain the linear dispersion at this critical damping as arising from degenerate perturbation theory , allowing a @xmath113 perturbation to give a @xmath114 splitting . setting @xmath109 , and writing @xmath115 , then expanding eq . ( [ eq:11 ] ) to leading order in @xmath116 gives : @xmath117,\ ] ] hence describing modes @xmath118 . note that although these modes have a linear dispersion of the real part as a function of @xmath65 , they have a lifetime that remains finite as @xmath119 . as such , this may provide an example where linear dispersion of a given mode in a condensed system need not imply superfluidity of the associated density . further work is required to determine the current - current response function in this system , which will determine whether there is a difference of transverse and longitudinal response for spin currents , however superfluidity of the spin current is not expected here , since the spin orientation is locked by the josephson term @xmath26 . the linear dispersion arises only when the damping matches frequency of this phase - locking term , producing critical damping . when an harmonic trap is introduced , the existence of steady state currents can destabilize the thomas fermi like density profile , and lead to spontaneous rotating vortex lattices@xcite . before addressing how this instability interacts with the polarization degree of freedom , we first discuss the simpler question of the dynamics and stability of individual vortices of opposite polarization . one should note first that a single vortex in a pumped decaying condensate already has a more complicated structure than the vortex without pumping and decay the vortex becomes a spiral vortex , with both radial and azimuthal currents ; see appendix [ sec : spir - vort - single ] for further discussion . considering a spinor condensate system with vortices in both polarization components , there is an interplay between the weak attractive coupling that invites the vortices to have their cores aligned irrespective of their circulation , the josephson coupling that discourages the alignment of vortices of opposite circulation and the detuning that stabilizes such alignment . to illustrate these possibilities we consider several examples . the notation @xmath120 refers to condensates with vortices of topological charge @xmath121 ( @xmath122 ) in left ( right ) condensate . from eq . ( [ eq:2 ] ) , with @xmath123 , the following set of stability scenarios are found : @xmath124 : : : all @xmath125 and @xmath126 vortex complexes are dynamically stable . @xmath127 : : : solutions @xmath128 are stable , @xmath125 and @xmath129 are unstable . depending on the strength of @xmath26 , vortices may start precessing around the center of the trap , move beyond the boundary forming either a pair of rarefaction waves with opposite velocities or a complex @xmath128 , or disappear at the condensate s edge and bringing in two aligned pairs of vortices @xmath128 and @xmath130 . [ jjj ] illustrates these possibilities for @xmath131 and @xmath132 . @xmath133 : : : for a given @xmath26 , any sufficiently large @xmath30 allows the vortex complexes @xmath129 and @xmath125 to stabilize . fig . [ oppvort ] shows such stabilized @xmath129 complex for @xmath134 for @xmath135 and @xmath136 ( top row ) , @xmath137 ( second row ) , @xmath138 ( left bottom ) and @xmath139 ( right bottom ) . the initial state is @xmath140 where @xmath141 , @xmath142 and @xmath143 is the vortex core parameter ; see appendix [ sec : spir - vort - single ] . depicted are the density plots and streamlines of @xmath144 ( left panels in first and second row , bottom row ) and @xmath8 ( right panels in first and second row ) . luminosity is proportional to the magnitude of the density for density plots and to the velocity for streamlines . the size of the pumping is shown as a circle of radius @xmath145 . for @xmath136 the final state consists of two precessing vortices of opposite circulation . for @xmath137 the system evolves into a pair of two gray solitons . for @xmath138 the vortex of the negative circulation in @xmath8 leaves and a vortex of positive circulation enters the cloud . for @xmath139 both vortices leave the condensate and two pairs of vortices of opposite circulation enter . for @xmath138 and @xmath139 @xmath146 , so only @xmath144 is shown . , width=307 ] for @xmath147 and @xmath137 . depicted are the density plots and streamlines of @xmath144 ( left ) and @xmath8 ( right ) . luminosity is proportional to the magnitude of the density for density plots , and velocity is shown by streamlines . the size of the pumping is shown as a circle of radius @xmath145.,width=307 ] the stationary state shown in fig . [ jjj ] for @xmath137 is the vorticity - free , but not radially symmetric , solution of eq . ( [ eq:2 ] ) . these are analogous to the rarefaction solitary waves of the nonlinear schrdinger equation discovered by jones and roberts @xcite . in trapped 2d condensates these waves were also found @xcite : they form as two vortices of opposite circulation disappear at the boundary of the condensate . in the conservative gpe these waves propagate with velocities exceeding the velocity of any vortex pair . in spinor damped / driven condensates two such solutions induce flow in opposite directions forming a stationary complex . the plots of the real and imaginary parts of @xmath144 and @xmath8 shown in the left panel of fig . [ grey ] , and the density plots of fig . [ jjj ] can be compared to the bottom panel of fig . 9 and the left panel of fig . 10 of ref . . it also follows from simulations that @xmath148 , so the found stationary state satisfies a one component ginzburg landau equation @xmath149\psi + j \psi(x ,- y ) . \label{gl_new}\end{aligned}\ ] ] this also suggests a way to generate solitary waves in spinor condensates . the dark soliton ( obtained by phase imprinting , for instance ) will undergo a transverse snake instability and form two stationary pairs of vortices of opposite circulation , whereas starting with a @xmath129 complex one will obtain stationary rarefaction pulses . ( solid lines ) and @xmath8 ( dashed lines ) of the rarefaction wave complex as a stationary solution of eq . ( [ eq:2 ] ) with @xmath150 for @xmath151 and @xmath152 ( left ) and @xmath153 ( right ) plotted along the axis of density symmetry . the functions are antisymmetric with respect to the origin @xmath154 for @xmath152 , but not for @xmath155.,width=307 ] as @xmath30 increases , the rarefaction waves in two components lose their antisymmetry and the density minima move away from the center as the right panel of fig . [ grey ] illustrates . for intermediate @xmath30 , complexes combining a rarefaction wave in one component with a precessing vortex in the other component emerge ; for even larger @xmath30 two vortices of opposite circulation move around in the condensate . finally , for even larger @xmath30 , their cores overlap and vortices move to the center of the condensate . these possibilities are illustrated on figs . [ oppvort ] and [ possibledelta ] . the cross - polarized vortices deform the condensate to a slightly oblate form as seen in the density plots of fig . [ oppvort ] . the streamlines and the density plots indicate that the vortices coexist with rarefaction waves that rotate in the counterclockwise direction . ( top row ) and @xmath156 ( bottom row ) for the outcome of instability of cross - polarized vortices obtained by numerical integration of eq . [ eq:2 ] with @xmath150 , @xmath151 and various @xmath30 . the initial state is the same as in fig . luminosity is proportional to density . the size of the pumping is shown as a circle of radius @xmath145 . for @xmath153 two rarefaction pulses form a stationary complex . for @xmath157 a vortex of negative circulation in the @xmath98 component is coupled to a rarefaction pulse in the @xmath158 component . for @xmath159 the two vortices of opposite circulation do not align their cores . for @xmath157 and @xmath159 the complexes precess around the center with the individual vortices moving along an epitrochoid ; see also fig . [ motionvort ] . , width=307 ] the vortex trajectories with pumping , decay and a spin degree of freedom are nontrivial . for a one - component conservative gpe , a single vortex moves perpendicularly to the background density gradient due to the magnus force@xcite ( the speed of the motion is however nonuniversal , and depends on the global condensate shape @xcite ) . for a two - component conservative gpe with vortices in both components , there is an additional advection of each vortex by the flow pattern of the other component . including also pumping and decay , the trajectories of the vortices are yet more complicated , as illustrated in fig . [ motionvort ] for a @xmath129 complex with @xmath160 . both vortices move along trajectories closely resembling epitrochoids , and the distance between the vortices varies quasi - periodically with time . similarly complicated cycloid trajectories of vortices are known for two - layer fluids with one vortex in each layer such behavior has been seen for example in models of tropical vortices@xcite . obtained by numerical integration of eq . [ eq:2 ] with @xmath150 , @xmath151 and @xmath159 , and with pumping in a radius @xmath161 . these parameters are the same as for the density profiles in the right hand column of fig . [ possibledelta ] . the trajectory of the vortex of positive ( negative ) circulation in the left ( right ) component is shown as thick black ( thin blue ) line . the start ( end ) points of the time interval plotted are shown as red ( green ) filled circles . note that the vortex trajectories are not closed.,width=153 ] we study the time - dependent problem defined by eq . ( [ eq:2 ] ) with a harmonic trapping potential @xmath150 by numerical integration , using a fourth - order , finite - difference approximation in space and a fourth - order runge kutta method in time . pumping is restricted to a circular spot of radius @xmath27 centered at the bottom of the trap , as described at the end of section [ sec : model - spin - polar ] . in all of the following , we take the experimentally realistic parameters@xcite @xmath162 and @xmath163 . with this choice of parameters , the critical radius can be estimated from the thomas fermi approximation for the problem without polarization@xcite as @xmath164 . this section will present and discuss the transition between synchronized and desynchronized states , and the interplay with the instability to vortex lattice formation . section [ sec : critical - delta ] first addresses the value of @xmath30 for which a transition occurs , as one changes the size of the pumping spot , and coupling @xmath26 . section [ sec : attractors ] will then discuss in greater detail the nature of the new attractors near the critical @xmath30 that occur when one includes spatial degrees of freedom . let us start with the simplest trapped problem , taking @xmath165 and @xmath166 . we study the behavior of the system as the detuning @xmath30 is increased in the following way : starting at @xmath152 from a gaussian initial state , we let the system evolve until a steady solution is reached . we then increase @xmath30 in steps , taking for each new @xmath30 the final state at the previous @xmath30 as the new initial state . for each value of @xmath30 , the chemical potentials of the two polarization components are calculated ( see appendix [ app : numerical - precession ] for details ) . the result is shown in the left panel of fig . [ fig : chempotj1r3-j1r4 ] . the components stay synchronized , sharing a common chemical potential , up to @xmath167 . as the detuning is increased further , the components desynchronize and ac josephson oscillations occur between them ( fig . [ fig : acjosephson ] ) . note that as the components desynchronize , the chemical potentials show an oscillatory time dependence . [ fig : chempotj1r3-j1r4 ] shows a time average in this case . . for solutions with time - dependent chemical potentials , a time average is shown.,width=307 ] in the limit of large time . left : @xmath168 as a function of time . right : the chemical potentials of the two polarization components . ( time given in units of @xmath169 . blue and red color denote @xmath158 and @xmath98 components , respectively.),width=307 ] in terms of a suitably defined , spatially averaged phase difference @xmath40 between the components ( see appendix [ app : numerical - precession ] ) , the desynchronization transition corresponds to a transition from a fixed point to a limit cycle in the @xmath170 plane , in direct analogy to the two - mode problem ( see fig . [ fig : r3progression ] and compare with fig . [ fig : attraction - basin ] ) . and @xmath166 . left : synchronized solution for small detuning . right : desynchronized solution in the limit of large detuning ( corrected for numerical precession).,width=307 ] the right panel of fig . [ fig : chempotj1r3-j1r4 ] shows the chemical potentials of the two polarization components as the detuning is increased stepwise in a system pumped in a spot with radius @xmath171 . the progression from the synchronized solution at small detunings to the desynchronized solution in the limit of large detunings is now more complicated than for a small pumping spot . @xmath171 is too small a pumping radius for vortices to form spontaneously in the corresponding one - component system@xcite . accordingly , the spinor system with zero detuning develops circularly symmetric densities , identical in both components ( but with different phases for the two components ) . for small enough detunings , the system stays synchronized , the densities adjusting to accommodate the common , constant chemical potential ( left - most panels of fig . [ fig : r4progression ] ) . and @xmath165 . top : phase portraits for the phase difference @xmath40 ( a numerical precession has been removed from the @xmath172 plot ) . middle and bottom : total densities of the @xmath158 and @xmath98 components , respectively.,width=307 ] as the detuning increases beyond @xmath173 the rotational instability of the high - angular - momentum modes @xcite reappears , and a four - vortex lattice forms in both components . at this point , the smaller component is sufficiently depressed by the detuning that the effective critical radius is smaller than the pumping radius . the suppressed component becomes rotationally unstable and drags the larger component along . this transition to a vortex - lattice solution causes the ( common ) chemical potential to drop , as shown in fig . [ fig : chempotj1r3-j1r4 ] . while the two components remain on average phase locked , the drop in chemical potential at the formation of the vortex lattice is accompanied by a time - dependence of the chemical potentials , which oscillate together around a common mean . due to the amplitude of the oscillations being slightly larger in the larger component , the fixed point in the @xmath170 plane turns into a small limit cycle , but with only small variations of @xmath40 unlike the @xmath2 periodic cycles in the desynchronized phase . ( middle panels of fig . [ fig : r4progression ] . ) finally , as @xmath30 becomes large , the components desynchronize completely . this situation is shown in the right - hand panels of fig . [ fig : r4progression ] . the difference in space - averaged phase now traces out a limit cycle winding through the full @xmath2 . the overall progression from synchronized solutions at small @xmath30 to desynchronized solutions with josephson oscillations holds very generally for different values of the josephson coupling @xmath26 and also for both small ( no vortices ) and large ( vortex lattice ) pumping - spot sizes . the mechanism in both cases is the same : the synchronized solution is upheld by adjusting the densities and a steady interconversion current forms . after desynchronization , the densities revert to profiles as for the single component condensate , with the average density set by the balance of pumping and decay , but with an additional time - dependent interconversion current . [ fig : densityprof ] shows density profiles of typical synchronized solutions with and without vortices , and also a snapshot of a desynchronized solution . -axis for representative examples of synchronized and desynchronized solutions . left : synchronized solution without vortices ( @xmath174 ) . middle : synchronized solution with vortex lattice ( @xmath175 ) . right : desynchronized solution without vortices ( @xmath176 ) . dashed and solid lines indicate left and right - polarized components , respectively . @xmath165 for all three examples.,width=307 ] we also performed calculations where we reset the initial conditions to a gaussian density profile , with equal phase of the two polarization components for each value of @xmath30 . based on the results of the two - mode model shown in fig . [ fig : attraction - basin ] , such initial conditions would be expected to find the limit cycle whenever it exists , whereas the stepwise increase of @xmath30 discussed above is intended to follow the fixed point . in these calculations with resetting of initial conditions , desynchronized solutions generally develop at smaller @xmath30 , as one should expect if there is a region of coexistence of synchronized and desynchronized solutions . [ fig : deltac ] shows the highest @xmath30 yielding synchronized solutions for different @xmath27 and different @xmath26 , both for stepwise increased @xmath30 ( top left ) and calculation directly from gaussian initial conditions ( top right ) . comparing the two plots gives an estimate of the region of coexistence . an example ( @xmath177 ) showing the region of bistability is given in the bottom left panel . note that the difference between stepwise increased @xmath30 and reset to gaussian initial condition for each @xmath30 vanishes for large detunings . this need not be the case for larger @xmath27 or larger @xmath26 , where more than one desynchronized , meta - stable state ( distinguished by numbers of vortices ) may be possible at high @xmath30 . yielding synchronized solutions plotted as a function of @xmath27 for @xmath178 and @xmath179 . left : @xmath30 increased stepwise . right : resetting initial conditions for each @xmath30 . bottom left : bifurcation of chemical potentials ( plotted as in fig . [ fig : chempotj1r3-j1r4 ] ) for the two approaches , indicating the region of bistability for the example @xmath177 . bottom right : sketch of regions of stability of synchronized and desynchronized solutions for @xmath166 . compare with inset of fig . [ fig : ft - two - mode ] . , width=307 ] our calculations suggest that the spatially extended system allows for a rich variation of behaviors . notably , interconversion due to the josephson coupling between the components need not be uniform in space . the interconversion rate is easily calculated from the time - dependent local phase difference @xmath180 as @xmath181 familiar from the josephson effect as found in any textbook@xcite . note that eq . ( [ eq:5 ] ) describes the same physics for the two - mode model . [ fig : interconvmap ] shows a map of the interconversion rate @xmath182 at two close points in time in the large - time limit of the solution directly from gaussian initial conditions with @xmath183 . this solution is desynchronized and exhibits josephson oscillations . however , at any given time there is interconversion in both directions at different points in space . at two different , but close , times . @xmath183.,width=307 ] calculations directly from gaussian initial conditions at large detuning with a large pumping spot ( @xmath184 ) indicate that solutions are possible in which the polarization components develop counter - rotating vortex lattices . this leads to rapid density modulations , particularly around the edge of the cloud , and a corresponding pattern of interconversion in opposite directions . in this case , the josephson oscillations are suppressed and the total ( scaled ) mass @xmath168 exhibits rapid , small - amplitude oscillations . these effects are shown in fig . [ fig : counter - rot ] . , chemical potentials and interconversion rate @xmath182 for a desynchronized solution with counter - rotating components . bottom : densities of the @xmath158 and @xmath98 polarization components , respectively . ( note that the gray scale has been adjusted for each component separately to accentuate the density modulations . ) @xmath185.,width=307 ] the spatial degree of freedom allows a number of behaviors of the @xmath170 phase portrait that are not possible in the two - mode problem , which only allows fixed points ( synchronized solutions ) and limit cycles with winding number @xmath186 ( desynchronized solutions ) . in the spatially extended system , these behaviors are exemplified in fig . [ fig : r3progression ] for a system without vortices . both these two classes of attractor can also be seen when a vortex lattice exists . in addition , the spatial degree of freedom gives rise to several new classes . we have already noted in fig . [ fig : r4progression ] an example of a synchronized limit cycle . this can be distinguished from the desynchronized limit cycle by the winding number , @xmath187 , over one period . the synchronized limit cycle has winding number @xmath188 , and the desynchronized cycles have winding number @xmath186 . we also find an example of limit cycle with winding number @xmath189 , shown in fig . [ fig : phaseportraitzoo ] . this solution also exhibits another behavior not possible in the two - mode model : a retrograde loop . this solution appears in calculation directly from a gaussian initial condition at @xmath190 , which is barely above @xmath191 for @xmath177 . for larger @xmath30 the loop quickly becomes a cusp and then disappears . and retrograde loop . left panel shows a blow - up of the loop . a numerical precession has been removed from these plots . ( @xmath192 . ) bottom left : possibly quasi - periodic behavior similar to a limit cycle with winding number @xmath188 ( @xmath193 from stepwise increase ) . bottom right : possible chaotic attractor ( @xmath194 from stepwise increase).,width=307 ] the bottom panels of fig . [ fig : phaseportraitzoo ] show two other behaviors found when stepwise increasing the detuning in a system with a strong josephson coupling ( @xmath195 ) . both these solutions are basically synchronized ( the time average of the chemical potential is the same for both components ) . however , the large detuning causes the chemical potentials to differ at most instances in time , resulting in behaviors similar to the limit cycles with winding number @xmath188 , but which appear to have a chaotic attractor and/or quasi - periodic behavior . to directly observe rotating vortex lattices in experiments would require time - resolved measurements on timescales of the order of the trap frequency , which would be challenging with current experimental configurations . it is however possible to see signatures of a vortex lattice in the momentum- and energy - resolved photoluminescence spectrum , which can be directly measured in the far field . the spectral weight is given by the modulus squared of the fourier transform of the wavefunction : @xmath196 as an illustration of this , fig . [ fig : spectra ] shows the spectral weight as a function of @xmath197 . the vortex lattice as well as the desynchronization transition can be seen in the dispersion spectrum . [ fig : spectra ] shows the spectra of the two polarization components for two different solutions . the top panels show a synchronized solution that exhibits a vortex lattice without a central vortex . each ring of vortices in the lattices shows up as a side band in the spectrum . if there were no vortices , only the bottom band would be present . as a contrast , the bottom panels show a desynchronized solution that has a vortex lattice with a central vortex . the presence of the central vortex means that the spectral weight vanishes at @xmath198 . note also that the desynchronization causes the spectra of the two components to be shifted relative to each other , whereas in the synchronized case , the two spectra are identical . for two vortex - lattice solutions . left and right panels show left and right polarized components , respectively . top : synchronized solution without a central vortex . ( @xmath199 . ) bottom : desynchronized solution with a central vortex . ( @xmath200.),width=307 ] the interplay between the effects of pumping and decay in setting the density profile , and the dynamics of the polarization degree of freedom lead to a rich variety of possible nonequilibrium steady states , as well as dynamical attractors for the polarized polariton condensate in a harmonic trap . even neglecting spatial currents , the two - mode model shows nontrivial behavior as a function of applied magnetic field : with strong damping there is a simple transition between synchronized and desynchronized states , whereas for weak damping , a region of bistability exists , in which the state established depends on the initial conditions . allowing for spatial fluctuations in a homogeneous system , the plane - wave modes on top of the synchronized solution show a second class of transition : if the phase locking term for the spin degree of freedom is large enough , then at weak magnetic fields , this term will provide sufficient restoring force for spin fluctuations to give underdamped global spin oscillations . as the magnetic field increases , the restoring force for such spin waves decreases , eventually vanishing when the synchronized solution becomes unstable . before this instability occurs , there is a transition between underdamped and overdamped spin oscillations , and at this transition , the spin wave energies have a linear dispersion vs momentum . introducing vortices , the phase locking term favors co - alignment of vortices in the two spin polarizations , but at sufficiently large @xmath30 anti - aligned vortex pairs may become stable . the phase - locking term also makes it possible to obtain a stationary complex of rarefaction waves . the experimental realization of such solitary vorticity - free waves would demonstrate another aspect of superfluid behavior@xcite in the incoherently pumped polariton system . considering the transition between synchronized and desynchronized states in the inhomogeneous system , for sufficiently small pump spots , the transition is similar to the two - mode model . for larger pump spots however , an instability to vortex - lattice formation may preempt desynchronization , driven by the density imbalance required to sustain a synchronized solution . for both large and small pumping - spot radius , the phase portraits near the critical detuning can be more complicated , showing synchronized limit cycles with winding number 0 , desynchronized limit cycles with winding number 2 , and chaotic behavior . in conclusion , the results presented here illustrate the wide variety of dynamical behavior that can arise in spinor polariton condensates , and suggest that experimental efforts to investigate such behavior should be feasible . this appendix will discuss spiral vortices in the one - component version of eq . ( [ eq:2 ] ) , i.e. , neglecting @xmath26 and @xmath30 . it was shown in ref . that below the critical radius of pumping for the formation of a vortex lattice there exist states with one or a few spiral vortices . by combining vorticity with pumping and decay , one has both radial and azimuthal supercurrents , both of which modify the density profile , and so these currents interact . this is shown in fig . [ vortices ] , which shows both the radial velocity ( main figure ) for vortex solutions , and the density profile ( inset ) . the vortex solution requires vanishing density at the origin , this makes the region at small radii a region of net gain , and leads to an outward - flowing current . thus , in the vortex solutions , there exists both a local maximum of radial current , and a point at finite radius ( of the order of the healing length ) at which this current vanishes . , of the vortex solutions of eq . ( [ eq : gpe_rescale ] ) for @xmath201 . the numbers next to the lines indicate the winding numbers of the vortices ( 1 and 2 ) and the ground state ( 0 ) . the tf approximation of the gradient flow of the ground state for small @xmath51 and @xmath202 @xmath203 , is given by the dashed line . the inset compares the density , @xmath204 , of a vortex solution for @xmath205 ( @xmath206 , gray ( red ) line ) with the density of the gpe vortex ( @xmath207 ) with the same number of particles ( @xmath208 , @xmath209 , black line ) . both vortex solutions have topological charge @xmath186.,width=307 ] in polar coordinates @xmath210 the wavefunction of a spiral vortex of topological charge @xmath211 in the one - component polariton condensate takes the form @xmath212 $ ] , with the equations governing @xmath213 and @xmath214 given by @xmath215 around the center of the trap @xmath216 and @xmath217 can be found recursively in the form of the power series @xmath218 to the leading order @xmath219 showing that the stronger the pumping the larger the outward velocity is in the vortex core . only vortices of topological charge @xmath186 are dynamically stable . for these , @xmath220 , where the numerical value of the vortex core parameter @xmath221 was first calculated by pitaevskii @xcite for the straight line vortex of uniform gross pitaevskii condensate . we note that around and above critical detuning , solutions to eq . ( [ eq:2 ] ) may exhibit irregular density profiles and irregular vortex lattices . this happens most frequently when solutions are found starting from gaussian initial conditions , but it may also happen for large @xmath27 when @xmath30 is increased stepwise . as an example , we show a solution obtained from gaussian initial conditions at @xmath222 in fig . [ fig : irregular - density ] . this solution shows an irregular vortex lattice , reminiscent of turbulent behavior . in order to numerically find the chemical potential of a condensate with a rotating vortex lattice , it is useful to use : @xmath223 along with the equation : @xmath224 in many cases , eq . ( [ eq:15 ] ) with @xmath225 is sufficient to determine @xmath226 . however , when there is a vortex at the center of the cloud , the numerator and denominator both vanish , whereas one of @xmath227 will give a well defined value . in such cases , it is then necessary to use eq . ( [ eq:16 ] ) and the integral for @xmath228 to eliminate @xmath229 . for plotting the phase portraits , it is necessary also to calculate the associated phase , @xmath33 such that @xmath230 . this can similarly be defined as : @xmath231\right\}.\ ] ] however , when there is a central vortex , the same difficulty will arise . to avoid this problem , two possible approaches are used . in some most cases it is sufficient to calculate @xmath232 , and numerically integrate this equation to find @xmath233 . numerical errors in evaluating @xmath234 can however introduce unphysical precession into the plot of @xmath235 . to avoid this , in those cases where the nature of the portrait is simple , we use @xmath236 , with @xmath237 adjusted to remove the spurious precession . this method has been used in figs . [ fig : r3progression ] , [ fig : r4progression ] ( right hand panel ) and [ fig : phaseportraitzoo ] . where the phase portrait involves finer structure , such as the middle panel of fig . [ fig : r4progression ] , the phase portrait has instead been obtained by : @xmath238\right.\\ & - \left.\ln\left[\int e^{i s \vartheta } \psi_l(\mathbf{r})\,d^2r\right ] \right\ } , \end{split}\ ] ] with @xmath239 according to whether there is a central vortex . since the two components are corotating in this case , such a definition satisfies : @xmath240 as required .

superconducting transmission line resonators can be used as quantum data buses , quantum memories , and single microwave photon detectors @xcite . they usually work in the microwave regime and can also be used as quantum nodes in so - called quantum networks @xcite . it is well known that the entanglement is one of the most important resources for quantum information processing @xcite , and microwave photons play a critical role in quantum state control for solid state quantum devices . therefore , engineering arbitrarily entangled microwave photon states is a very fundamental issue for both solid state quantum information processing and quantum optics @xcite on superconducting quantum chips . usually , nonclassical photon states of a single - mode cavity field are generated through the interaction between the cavity field and the two - level atom . the methods of generating nonclassical photon states can be classified into two ways . one is to engineer appropriate hamiltonians in different evolution durations by tuning experimental parameters when the target state is being generated @xcite . the other one is to obtain the target state via appropriately designed measurements @xcite . the former one is deterministic , while the latter one is probabilistic and usually has a low probability to succeed . if the nonclassical state is generated using natural atomic systems , the latter method is usually more practical since most of parameters are not possible or not easy to be tuned . however , in artificial atomic systems , the former method is more appropriate because system parameters can be artificially controlled . for example , superconducting quantum circuits ( sqcs ) @xcite provide us a very convenient way to deterministically engineer nonclassical states of a single - mode microwave field by varying the system parameters @xcite . the method of deterministically generating entangled photon states using atomic systems can be tracked to that of generating entangled phonon states of two vibrational modes @xcite , in a trapped ion interacting with laser fields , by using different sideband transitions . however , the number of steps in such a method @xcite exponentially depends on the maximum phonon numbers . a few proposals were put forward to overcome the exponential dependence of the phonon number by introducing auxiliary atomic energy levels @xcite , using phonon number dependent interactions @xcite , or employing multiphonon transitions of high phonon numbers @xcite . these methods have successfully reduced the number of steps into quadratic polynomials of the maximum phonon numbers . the generation of entangled microwave photon states of two modes using superconducting qubit has been studied @xcite , where a classically driven superconducting qubit with time - dependent frequency is coupled to two microwave fields in two separated cavities . the interaction hamiltonian between the superconducting qubit and the cavity fields of two modes is described by the jaynes - cummings model . therefore , there is only single photon transition in each step . however , the photon - number - dependent stark effects @xcite induced by the qubit - field coupling make it possible to independently implement operations for photon states . thus , the number of steps also quadratically depends on the maximum photon number . it has been shown that the superconducting qubit and the cavity field can have both transverse and longitudinal couplings when the inversion symmetry of the qubit potential energy is broken @xcite . the longitudinal coupling can induce multiphoton transitions @xcite in different sidebands as in trapped ions @xcite and thus arbitrary photon states of a single - mode cavity field can be more conveniently engineered @xcite . motivated by studies @xcite , we study a method to generate entangled microwave photon states in two separated cavities coupled by a superconducting qubit using multiphoton transitions . we first show that the longitudinal couplings can induce two - mode multiphoton processes similar to those in trapped ions @xcite , and then study an efficient way to generate superposed two - mode photon states . the paper is organized as below . in sec . [ sec : inter - mode ] , an effective hamiltonian , similar to that of trapped ions with two vibrational modes @xcite , is derived , and then different sideband transitions are discussed . in sec . [ sec : algorithm ] , a new algorithm is introduced to generate arbitrary superpositions of two - mode photon states . in sec . [ sec : parsel ] , we discuss how to choose parameters to obtain a high fidelity of the target state . in sec . [ sec : simresults ] , we numerically study the effects of both imperfect control pulses and the environment on the generated target state . [ sec : discussions ] , the advantages and experimental feasibility of our method are discussed . finally , we summarize our results in sec . [ sec : conclusions ] . and the second one has the frequency @xmath0 . the coupling strength is @xmath1 ( @xmath2 ) between the qubit and the first ( second ) cavity field . the qubit is driven by a classical field ( in the middle with the black color ) with the frequency @xmath3 and rabi frequency @xmath4.,scaledwidth=46.0% ] as schematically shown in fig . [ fig : model ] , we study a system where a superconducting qubit ( sq ) , modeled as a two level system , is coupled to two single - mode microwave fields in two separated cavities and driven by a classical field . the system hamiltonian can be given by @xmath5 here , @xmath6 and @xmath7 are the free hamiltonians of the sq and the cavity fields , respectively . moreover , @xmath8 is the interaction hamiltonian between the sq and cavity fields , and @xmath9 is the interaction hamiltonian between the sq and the classical field . in the qubit basis , the qubit hamiltonian is given by @xmath10 with @xmath11 and @xmath12 @xmath13 . the parameter @xmath14 is the qubit frequency . the kets @xmath15 and @xmath16 denote the ground and excited states of the qubit , respectively . the free hamiltonian of two cavity fields is given by @xmath17 where @xmath18 ( @xmath19 is the annihilation ( creation ) operator of the @xmath20th cavity field with its frequency @xmath21 and @xmath22 . the interaction hamiltonian between the qubit and two cavity fields is @xmath23 where @xmath24 is the coupling strength between the @xmath20th cavity field and the qubit , and @xmath25 is a parameter which depends on the inversion symmetry of the qubit potential energy . similarly , the interaction hamiltonian between the qubit and classical field is given by @xmath26 where @xmath4 is the coupling strength ( or rabi frequency ) between the qubit and the driving field . the parameters @xmath3 and @xmath27 are the driving frequency and driving phase , respectively . in eqs . ( [ eq : hg_tilde ] ) and ( [ eq : hd_tilde ] ) , when the qubit potential energy possesses inversion symmetry , i.e. , @xmath28 , there are only transverse couplings between the qubit and cavity fields @xcite . if the rotating wave approximation is further made and there is no driving ( @xmath29 ) , eq . ( [ eq:1 ] ) is reduced to extensively studied jaynes - cummings model @xcite . when the qubit potential energy possesses a broken inversion symmetry @xcite , i.e. , @xmath30 , there are both transverse and longitudinal couplings between the qubit and microwave fields . the broken inversion symmetry of the qubit potential energy can be achieved when the bias charge for the charge qubit or the bias flux for the flux qubit is tuned off the optimal point @xcite . but for the phase qubit , the inversion symmetry of the potential energy is always broken @xcite . here , we will study a general method and not specify a particular qubit . we now change the qubit basis into the current basis of the flux qubit or the charge basis of the charge qubit . this is equivalent to diagonalizing the operator @xmath31 . in the new basis , the hamiltonian in eq . ( [ eq:1 ] ) becomes @xmath32 here , the hamiltonians @xmath33 , @xmath34 , and @xmath35 are given by @xmath36 with @xmath37 , and @xmath38 @xmath39 . hereafter , the parameters @xmath40 and @xmath41 are called transverse and longitudinal frequencies of the qubit , respectively . the kets @xmath42 * * and * * @xmath43 * * are persistent current states of the flux qubit or charge states of the charge qubit . here , * * @xmath44 * * is the rotation operator along the @xmath45-axis , with * * @xmath46 * * @xmath47 . the parameter @xmath48 results in longitudinal couplings between the qubit and microwave fields in eq . ( [ eq:1 ] ) . below , we will show that @xmath48 can induce two - mode multiphoton processes in the qubit , and then use these multiphton processes to generate arbitrary superpositions of two - mode photon states . to see how the multiphoton processes can be induced by the longitudinal coupling when @xmath48 , we now apply a unitary transformation @xmath49 , \label{eq : d}\ ] ] to the hamiltonian in eq . ( [ eq : h ] ) . then , we obtain an effective hamiltonian @xmath50 .\nonumber\end{aligned}\ ] ] it is clear that @xmath51 is the displacement operator @xcite of two - mode cavity fields . the displacement quantity is @xmath52 for the @xmath20th cavity field . hereafter , we will call the picture after the operator @xmath51 as the displacement picture . the ratios @xmath53 are called the lamb - dicke parameters in analogy to trapped ions @xcite . to understand the classical - field - assisted multiphoton transitions of two cavity fields in the qubit , we apply to eq . ( [ eq : heff ] ) a time - dependent unitary transformation @xmath54 , \label{eq : ud}\ ] ] with @xmath55 . then , another effective hamiltonian @xmath56 , \label{eq : heffd}\ ] ] can be derived , with the time - dependent term @xmath57 .\ ] ] here , @xmath58 is the bessel function of the first kind . equation ( [ eq : heffd ] * ) * shows that multiphoton transitions with different modes can be controlled by the classical field as in trapped ions @xcite . in the interaction picture with the free hamiltonian @xmath59 , equation ( [ eq : heffd ] ) becomes @xmath60 where @xmath61 is the coupling strength between the qubit and cavity field with each different transition process , and its algebraic form is@xmath62 t+in\phi\right\ } \nonumber\\ & \times\exp\left ( -{\sum_{l}\frac{\eta_{l}^{2}}{2}}\right ) j_{n}\left ( x\right ) \frac{\left ( -1\right ) ^{n_{1}+n_{2}}\eta_{1}^{m_{1}+n_{1}}\eta_{2}^{m_{2}+n_{2}}}{m_{1}!n_{1}!m_{2}!n_{2}!}. \label{eq : jnab}\ ] ] equation ( [ eq : hint ] ) describes the classical - field - assisted two - mode multiphoton processes as in trapped ions @xcite . the magnitude of @xmath63 depends on @xmath64 , @xmath65 , and @xmath66 . we find @xmath67 where the properties of @xmath68 have been studied in ref . @xcite . the specific expression * * of @xmath69 is given by @xmath70 t+in\phi\right\ } .\nonumber\end{aligned}\ ] ] similarly to eq . ( [ eq : jnab_abs ] ) , the magnitude of @xmath69 can be rewritten as @xmath71 it is clear that @xmath72 is independent of the reduced driving strength @xmath65 . from eqs . ( [ eq : jnab_abs])-([eq : jn_abs ] ) , we know that both @xmath73 and @xmath74 can be changed by adjusting @xmath65 and @xmath66 in a similar way . by introducing new variables @xmath75 , we expand eq . ( [ eq : hint ] ) in the fock state basis , and then have @xmath76 with @xmath77 and @xmath78 . here , @xmath79 denotes the ladder operator of the @xmath20th cavity field . the time - dependent transition element @xmath80 is given by @xmath81 with @xmath82 , @xmath83 replaced by @xmath84 , @xmath85 respectively . the complex transition amplitude @xmath86 and detuning @xmath87 are respectively @xmath88 the parameter @xmath89 is given by @xmath90 with @xmath91{cc}1 , & k<0\\ 0 , & k\geq0 \end{array } \right . , \\ l_{n}^{\left ( k\right ) } \left ( z\right ) & = \sum\limits_{l=0}^{n}\frac{\left ( n+k\right ) ! } { \left ( n - l\right ) ! } \frac{\left ( -1\right ) ^{l}}{\left ( l+k\right ) ! } \frac{z^{l}}{l!}.\end{aligned}\ ] ] here , @xmath92 is the generalized laguerre polynomials . it is clear that the classical - field - assisted multiphoton transitions can be derived from eq . ( [ eq : hint2 ] ) using different frequency - matching conditions . we now give detailed discussions on how to engineer two - mode multiphoton processes by tuning the driving field . let us assume that the driving field is tuned to satisfy the resonant condition @xmath93 then eq . ( [ eq : hint2 ] ) can be reduced to an effective hamiltonian @xmath94 when unwanted terms are neglected , that is , @xmath95 the time evolution operator governed by the hamiltonian in eq . ( [ eq : h_sideband1 ] ) is given by @xmath96 recall that @xmath97 and @xmath98 as defined previously . here the new parameters used in eq . ( [ eq : u_sideband1 ] ) are respectively @xmath99 as shown in eqs . ( [ eq : h_sideband1 ] ) and ( [ eq : u_sideband1 ] ) , @xmath100 photons in the @xmath20th resonator can be either created if @xmath101 or annihilated if @xmath102 while the qubit is flipped up . similarly , @xmath100 photons in the @xmath20th resonator can be either created if @xmath102 or annihilated if @xmath101 while the qubit is flipped down . thus different sideband excitations can be constructed , depending on the values of @xmath103 and @xmath104 . because the hamiltonian derived in eq . ( [ eq : hint ] ) is similar to that of the trapped ions @xcite , the algorithm using two - mode multiphonon processes in trapped ions can be directly applied into our model , and different superpositions of two - mode photons can be generated . as a special case , two - mode fock states of high photon numbers can in principle be more efficiently generated with just two steps as single - mode fock states of high photon numbers @xcite . however , we here design a new algorithm via different sideband transitions of low photon numbers by tuning the driving field with properly selecting the parameters @xmath105 , @xmath64 , @xmath21 , and @xmath66 . the detailed discussions of parameter selection will be given in sec . [ sec : parsel ] . let us first study a universal algorithm for generating arbitrary two - mode microwave photon states using sideband transitions with the following four hamiltonians @xmath106 , @xmath107 , @xmath108 , and @xmath109 . here , for the compact of notations , we have used @xmath110 to represent @xmath111 with @xmath112 . for instance , @xmath106 is actually @xmath94 with @xmath113 , @xmath114 , and @xmath115 . for different @xmath116 , @xmath103 , and @xmath104 , the interaction hamiltonian @xmath94 and its time evolution operator @xmath117 have already been given in sec . [ sec : eng_inter - mode ] . below , we will first study how to generate the target state by choosing pulse durations , frequencies , and phases of the driving fields at each generation step with different sideband excitations , and then we will apply our algorithm to the generation of noon states and discuss particular properties of the algorithm . we note that the state generation in our algorithm is studied in the displacement picture with the unitary transformation as shown in eq . ( [ eq : d ] ) . the arbitrary quantum states , we expect to be generated , is written as@xmath118 where @xmath119 means that the first and second cavities contain @xmath120 and @xmath83 photons , respectively , and @xmath121 means that the qubit is in the ground state . besides , @xmath122 and @xmath123 mean the maximum photon number and the probability amplitude on the state @xmath124 , respectively . we assume that the system is initially in the state @xmath125 we suppose the target state @xmath126 can be generated by alternately switching on and off the two - mode transitions @xmath106 , @xmath107 , @xmath108 , and @xmath109 . with the designed time evolution operators , the state generation procedure can be represented by,@xmath127 \left\vert \psi _ { 0}\right\rangle \nonumber\\ = & \left [ \bar{u}_{f}^{\dag}\left ( t_{f}\right ) u_{\bar{1}}^{p_{f}}\left ( t_{f}\right ) \bar{u}_{f}\left ( 0\right ) \right ] \cdots\nonumber\\ & \left [ \bar{u}_{1}^{\dag}\left ( t_{1}\right ) u_{\bar{1}}^{p_{1}}\left ( t_{1}\right ) \bar{u}_{1}\left ( 0\right ) \right ] \left\vert 0,0\right\rangle \left\vert g\right\rangle , \label{eq : evolution}\ ] ] where @xmath128 denotes the transition type for the @xmath129th step , and @xmath130 is the time duration for the @xmath129th step . the time evolution operator @xmath131 is given by @xmath132 \nonumber\\ & \times\exp\left [ ix\frac{\sigma_{z}}{2}\sin\left ( \tilde{\omega}_{\nu } t_{\nu}+\phi_{\nu}\right ) \right ] . \label{eq : ubar}\ ] ] as discussed above , the transitions of different types can be achieved by changing the frequency @xmath3 of the driving field , which is denoted by @xmath133 for the @xmath129th step . the phase of the driving field for the @xmath129th step is denoted by @xmath134 . we can express eq . ( [ eq : evolution ] ) in another equivalent form of iteration , @xmath135 with @xmath136 and @xmath137 given in eqs . ( [ eq : psi_0 ] ) and ( [ eq : psi_f ] ) , respectively . the ket @xmath138 is the state after the @xmath129th step . we note that the subscript @xmath139 of @xmath137 in eq . ( [ eq : evolution ] ) denotes the number of the final step . equation ( [ eq : iteration ] ) means that the initial state is restored from the target state by a composition of sideband transitions with proper time durations , frequencies and phases of driving fields . it is a recursion algorithm . . the @xmath82 and @xmath83-axis respectively denote the photon number of the first and second mode . the two - mode photon state is denoted by @xmath140 . the qubit state is represented by the @xmath141-axis with @xmath142 or @xmath143 respectively denoting the ground state @xmath144 or excited state @xmath145 . the state component @xmath146 is represented by a block at the location @xmath147 . if a state component is occupied , we color the corresponding block with red ; otherwise , the block is left uncolored . the arrows respectively represent the 10 , 01 , 11 , and 00 transitions with transition types labeled aside them . the solid arrow indicates a desired population transfer from the starting state to the end state , while the dashed arrow indicates the inevitable irrelevant oscillation when the desired population transfer is implemented . ( a ) schematic diagram for transferring the populations on states @xmath148 , @xmath149 , and @xmath150 to the state @xmath151 . this is achieved by consecutively using 01 , 10 , 01 , and 10 transitions . ( b ) schematic diagram for transferring populations on states @xmath152 and @xmath151 to the state @xmath153 to the state @xmath154 . this is achieved by using a 10 transition . ( d ) schematic diagram for transferring the population on the state @xmath155 without loss of generality , we use the maximum photon number @xmath156 as an example to show our algorithm . the more general case with arbitrary @xmath122 is given in appendix . [ append : univ_alg ] . the detailed steps for generating the target state@xmath157 with @xmath156 using our recursion algorithm are described as the following four procedures . * procedure ( i)*. as schematically shown in fig . [ fig : general](a ) , from the final state @xmath137 , we first transfer the populations in the state space spanned by @xmath158 to the state @xmath151 . this procedure consists of four steps as schematically shown in below @xmath159{0\bar{1}}\left\vert 0,1\right\rangle \left\vert e\right\rangle \xrightarrow[f-1]{\bar{1}0}\left\vert 1,1\right\rangle \left\vert g\right\rangle \xrightarrow[f-2]{0\bar{1}}\nonumber\\ & \left\vert 1,0\right\rangle \left\vert e\right\rangle \xleftarrow[f-3]{\bar{1}0}\left\vert 2,0\right\rangle \left\vert g\right\rangle .\label{eq : proc_i}\ ] ] in eq . ( [ eq : proc_i ] ) , the transition type and step number is labeled respectively above and below the arrow . the arrow points to the direction of the population transfer . in step @xmath129 , the population transfer is accomplished by properly tuning @xmath133 , @xmath130 , and @xmath134 . after this procedure , we obtain the state @xmath160 which only has popupaltions in the space @xmath161 . we note that two additional oscillations @xmath162 will also occur inevitably when the population transfer from the state @xmath163 to the state @xmath164 is implemented . but they do not cause population leakage outside the original space and no extra steps should be taken for them . we call these oscillations as irrelevant oscillations . for this procedure , these irrelevant oscillations are schematically shown by dashed arrows in fig . [ fig : general](a ) . the irrelevant oscillation can also occur in the following procedures and is shown by dashed arrows . * procedure ( ii)*. as schematically shown in fig . [ fig : general](b ) , starting from the state @xmath160 , we need to transfer the populations in the state space spanned by @xmath165 to the state @xmath166 . this procedure consists of following two steps @xmath167{1\bar{1}}\left\vert 1,0\right\rangle \left\vert e\right\rangle \xrightarrow[f-5]{00}\left\vert 1,0\right\rangle \left\vert g\right\rangle .\ ] ] after this procedure , we obtain the state @xmath168 , which only has popupaltions in the space @xmath169 . * procedure ( iii)*. this procedure is similar to procedure ( i ) . as schematically shown in fig . [ fig : general](c ) , starting from the state @xmath168 , here we need to transfer the population on the state @xmath166 to the state @xmath154 . this procedure consists of only one step as below @xmath170{\bar{1}0}\left\vert 0,0\right\rangle \left\vert e\right\rangle .\ ] ] after this procedure , we obtain the state @xmath171 , which only has popupaltions in the space @xmath172 . * procedure ( iv)*. this procedure is similar to the procedure ( ii ) . as schematically shown in fig . [ fig : general](d ) , starting from the state @xmath171 , we need to transfer the population on the state @xmath154 to the state @xmath173 . this procedure only consists of one step as below @xmath174{00}\left\vert 0,0\right\rangle \left\vert g\right\rangle .\ ] ] we thus obtain the state @xmath175 . therefore , the target @xmath126 can be generated from the initial state @xmath176 using inverse processes from the procedure ( iv ) to the procedure ( i ) . we note @xmath177 . therefore , the generation of the target state with @xmath178 needs @xmath179 steps . our algorithm takes a quadratic number of steps while an exponential one is required in ref . let us now analyze the reason . our algorithm employs four interaction hamiltonians , @xmath107 , @xmath108 , @xmath109 , and @xmath106 given in eq . ( [ eq : h_sideband1 ] ) . however , four interaction hamiltonians @xmath180 , @xmath181 , @xmath182 , and @xmath183are employed in ref . the former three interaction hamiltonians between our algorithm and those in ref . @xcite are qualitatively identical since they convert the same number of bosons for either mode when the two - level system is excited . however , the last ones show fundamental difference between our algorithm and that in ref . @xcite , because ours creates one boson ( photon ) of one mode but annihilate one boson ( photon ) of the other when the two - level system is excited . but in ref . @xcite , one boson for both modes can be simultaneously created when the two - level system is excited . this difference is critical for us to design an algorithm which can keep track of the populations with a constant total boson ( photon ) number . therefore , there is no population leakage outside the original space . however , the algorithm in ref . @xcite has population leakage . obviously , if the last interaction in ref . @xcite is changed to @xmath184 , a theoretically equivalent algorithm to ours can also be developed . in this sense , our algorithm can be regarded as the improved version of that in ref . @xcite . let us now study how to choose the pulse duration @xmath130 , the frequency @xmath133 and phase @xmath134 of the driving field to generate a target state in the @xmath129th step for different types of transitions . we suppose that the population transfer is taken as following @xmath185{p_{\nu}}\left\vert n_{1}+k_{1},n_{2}+k_{2}\right\rangle \left\vert e\right\rangle , \ ] ] in the @xmath129th step , where the transition type @xmath186 should be switched on based on the previous discussions . thus the driving frequency is taken as @xmath187 from the resonant condition in eq . ( [ eq : res_con ] ) . by introducing the notations@xmath188 then from eq . ( [ eq : iteration ] ) , we need to solve the equation,@xmath189 we thus have the explicit solution for the pulse duration @xmath130 as @xmath190 the phase of the driving field is determined by @xmath191 here , the notation @xmath192 is the value of @xmath193 for the @xmath129th step , which is given in eq . ( [ eq : phink ] ) and depends on @xmath134 . still recall @xmath194 with @xmath195 . similarly , if the population transfer is taken as @xmath196{p_{\nu}}\left\vert n_{1},n_{2}\right\rangle \left\vert g\right\rangle .\ ] ] in the @xmath129th step . the explicit solution for @xmath130 is then@xmath197 and the phase of the driving field is determined by @xmath198 according to the target state , the time duration , frequency and phase of the driving field for each step can be calculated using above equations . for example , if the @xmath199 " transition is used in the @xmath200rd step , then we use eq . ( [ eq : t_nu_e ] ) and eq . ( [ eq : phi_nu_e ] ) to obtain @xmath201 and @xmath202 by setting @xmath203 . as an example , we now apply our algorithm to the generation of the noon state , i.e. , the target state is @xmath204 the recursion algorithm restoring @xmath126 to the vacuum state @xmath173 is schematically shown in fig . [ fig : noon ] for the maximum photon number @xmath156 . in fig . [ fig : noon](a ) , we can find that all the populations in the hilbert space spanned by @xmath205 can be transferred to the state @xmath151 by consecutively using transitions 01 `` , 10 '' , 01 `` , and 10 '' , i.e. , @xmath159{0\bar{1}}\left\vert 0,1\right\rangle \left\vert e\right\rangle \xrightarrow[f-1]{\bar{1}0}\left\vert 1,1\right\rangle \left\vert g\right\rangle \xrightarrow[f-2]{0\bar{1}}\nonumber\\ & \left\vert 1,0\right\rangle \left\vert e\right\rangle \xleftarrow[f-3]{\bar{1}0}\left\vert 2,0\right\rangle \left\vert g\right\rangle .\end{aligned}\ ] ] after this procedure , as schematically shown in fig . [ fig : noon](b ) , all the populations on the state latexmath:[$|1,0\rangle @xmath173 by consecutively using transitions 00 `` , 10 '' , and 00 " , i.e. , @xmath207{00}\left\vert 1,0\right\rangle \left\vert g\right\rangle \xrightarrow[f-5]{\bar{1}0}\nonumber\\ & \left\vert 0,0\right\rangle \left\vert e\right\rangle \xrightarrow[f-6]{00}\left\vert 0,0\right\rangle \left\vert g\right\rangle .\end{aligned}\ ] ] the total step number is thus @xmath208 for generating the noon state @xmath209 . more generally , given an arbitrary @xmath122 , the total step number for generating the noon state in eq . ( [ eq : noon ] ) is @xmath210 the step number for generating noon state has been greatly reduced in comparison with that for generating an arbitrary state in eq . ( [ eq : f_general ] ) . obviously , the noon state can be generated without using the 11 transition . if we assume the lamb - dicke parameter @xmath211 , which is usually the case even in the ultrastrong regime in superconducting circuit qed systems @xcite . from eq . ( [ eq : rabi ] ) , we have the rabi frequencies @xmath212 , @xmath213 , @xmath214 , and @xmath215 . thus , the transition 11 generally takes more time among the four types of transitions employed by us . therefore , our algorithm may show a better efficiency for generating noon sates than generating arbitrary entangled states . this is especially true when the maximal photon number @xmath122 is higher and the lamb - dicke parameter @xmath66 is smaller . . ( a ) schematic diagram for transferring the population in the space latexmath:[$\{|0,2\rangle @xmath151 . this is achieved by consecutively using 01 , 10 , 01 , and 10 transitions . ( b ) schematic diagram for transferring the population on the state @xmath151 to the state @xmath173 . this is achieved by consecutively using 00 , 10 , and 00 transitions . , scaledwidth=48.0% ] in all of the above studies , we make an approximation that all unwanted terms have been neglected . however , these neglected terms will affect the fidelity of the prepared target state . let us now discuss how to minimize the effect of these unwanted terms in eq . ( [ eq : hint2 ] ) on the target state by choosing appropriate parameters . in principle , the effects of these unwanted terms can be perfectly removed by pulse calibration techniques . here , we study a method to minimize the effect of these unwanted terms by choosing the parameters when the pulse calibration can not be used . in our algorithm , we have used four interactions @xmath107 , @xmath108 , @xmath106 , and @xmath109 , all of them are constructed by the terms with the bessel function @xmath217 in eq . ( [ eq : hint2 ] ) . here , in the subscript of the bessel function , we also use @xmath218 to denote @xmath219 if @xmath220 . we hope to suppress all the terms with the bessel functions @xmath221 for @xmath222 . we focus on the case * * @xmath223 considering possible experimental conditions . in this case , only lower order bessel functions @xmath224 , @xmath225 , and @xmath226 play significant roles . thus , we need only to find proper parameters such that the effect of the terms with @xmath224 , @xmath227 , and @xmath226 are negligibly small . our idea is to make those terms nonresonant by properly choosing the parameters @xmath64 and @xmath105 of the qubit , and frequencies @xmath228 and @xmath0 of two microwave modes . that is , we assume that the frequency of the @xmath20th cavity mode satisfies @xmath229 where @xmath230 is a positive integer and @xmath231 is the greatest common divisor of @xmath228 and @xmath0 . assuming that the @xmath232transition is switched on , i.e. , the transition detuning @xmath233 , then from eq . ( [ eq : detuning ] ) , the frequency @xmath3 of the driving field must satisfy the condition @xmath234 from eqs . ( [ eq : detuning ] ) , ( [ eq : oil ] ) , and ( [ eq : omega_wave ] ) , the detuning of the term with @xmath235 , @xmath236 , @xmath237 is then given by @xmath238 thus the terms with the bessel function @xmath227 will have the detuning @xmath239 we expect that the terms with @xmath227 are nonresonant . thus , the relation that @xmath240 must hold . a simple but sufficient condition is @xmath241 where @xmath242 is an integer . similarly , for the terms with @xmath243 , @xmath244 , and @xmath245 , the sufficient conditions can be given by @xmath246 the conditions in eqs . ( [ eq : con1])-([eq : con_2 ] ) can be summarized as @xmath247 we can also assume that the longitudinal frequency @xmath105 of the qubit is @xmath248 where @xmath249 is the integer part , and @xmath250is the fraction part . to meet eq . ( [ eq : con_omegaz ] ) , there should be @xmath251 the nonresonant terms with @xmath252 , @xmath253 , and @xmath254 still have effect on the desired time evolution . these effects can be further eliminated by decreasing the stark shifts caused by the terms with @xmath255 in eq . ( [ eq : rabi ] ) . the ideal case is@xmath256 or equivalently,@xmath257 for @xmath258 , @xmath259 , @xmath260 , @xmath261 , and @xmath262 , where the constraint condition for @xmath120 and @xmath263 denotes the working space of our algorithm . equation ( [ eq : con_rabi0 ] ) means that the stark shifts should be negligibly smaller than the rabi frequencies for state generation . considering that @xmath122 is the maximum photon number of the target state , and using eq . ( [ eq : dnk1k2p ] ) and eq . ( [ eq : ozform ] ) , we can obtain * * @xmath264 here , @xmath265 means @xmath65 rounded down and @xmath266 means @xmath65 rounded up . we thus reduce eq . ( [ eq : con_rabi ] ) to @xmath267 a condition much stronger than eq . ( [ eq : con_rabi ] ) . if eqs . ( [ eq : con_rabi2 - 1])-([eq : con_rabi2 - 4 ] ) are fulfilled , the nonresonant terms can in principle be suppressed . we know from eq . ( [ eq : rabi ] ) that eqs . ( [ eq : con_rabi2 - 1])-([eq : con_rabi2 - 4 ] ) can be satisfied if , for example , the parameter @xmath64 of the qubit is tuned sufficiently small , assuming that the reduced driving frequency @xmath65 and lamb - dicke parameters @xmath66 have been appropriately chosen . beside the terms with bessel functions @xmath268 where @xmath269 , there are also unwanted terms with the bessel function @xmath217 , which , however , also satisfy the resonant condition @xmath270 here , we have used eqs . ( [ eq : omega_wave ] ) and ( [ eq : dnk1k2p ] ) to obtain eq . ( [ eq : unrester ] ) . the lamb - dicke parameters satisfy the condition @xmath271 for circuit qed systems even in the ultrastrong regime @xcite . from eq . ( [ eq : oil ] ) , we know that @xmath272 and @xmath273 are coprime numbers . we can further make @xmath272 ( or @xmath273 ) sufficiently large . thus the unwanted resonant terms will possess large @xmath274 ( or @xmath275 ) . in this way , the effects of these terms will be suppressed due to the exponential decrease via the term @xmath276 in eq . ( [ eq : rabi ] ) . the condition , that the term @xmath277 is negligibly small , can be summarized as that @xmath272 and @xmath273 should satisfy @xmath278 we now summarize the condition that minimizes the effects of unwanted terms . the parameter @xmath105 of the qubit should satisfy eq . ( [ eq : ozform ] ) and eq . ( [ eq : con_r ] ) . however , the parameter @xmath64 of the qubit is mainly constrained by current experiments . for example , typical values of @xmath279 are in the range @xmath280 ghz . the frequencies of the cavity modes @xmath21 should satisfy eq . ( [ eq : oil ] ) and eq . ( [ eq : con_li ] ) . the values of the reduced driving frequency @xmath55 and lamb - dick parameter @xmath53 should satisfy eq . ( [ eq : con_rabi ] ) or stronger conditions eqs . ( [ eq : con_rabi2 - 1])-([eq : con_rabi2 - 4 ] ) . appropriate values of @xmath65 and @xmath66 can be obtained via numerical simulations , which will be discussed below in sec . [ sec : unter ] . we now further numerically simulate the effect of the unwanted terms on the generation of target states by using examples of generating the following two target states @xmath281 for some given parameters . it is obvious that @xmath282 is an entangled state where every state component is evenly occupied . we thus call @xmath282 the evenly - populated state . the state @xmath283 is a two - photon noon state @xcite . both @xmath282 and @xmath283 possess a maximum photon number @xmath156 . the fidelities for generating these two states @xmath282 and @xmath283 are defined as @xmath284 here , @xmath285 and @xmath286 are respectively the actually generated states via the total hamiltonian in eq . ( [ eq : h ] ) . we now determine the detailed experimental parameters . from eq . ( [ eq : ozform ] ) and eq . ( [ eq : con_r ] ) , we set @xmath287 @xmath288 , @xmath289 , and @xmath290 ghz , which corresponds to @xmath291 ghz . from eq . ( [ eq : con_rabi ] ) or eqs . ( [ eq : con_rabi2 - 1])-([eq : con_rabi2 - 4 ] ) , the parameter @xmath279 should be made smaller , e.g. , we set @xmath292 ghz . the frequency of the @xmath20th cavity , i.e. , @xmath21 , is determined by eq . ( [ eq : oil ] ) and ( [ eq : con_li ] ) . since the microwave fields are usually of several gigahertz , here we set @xmath293 , and @xmath294 , thus yielding @xmath295 ghz and @xmath296 ghz . the lamb - dicke parameters for the first and second cavity modes are set to be identical , i.e. , @xmath297 we vary the lamb - dick parameter @xmath298 and the reduced driving frequency @xmath55 to simulate the effect of the unwanted terms on the fidelity of the expected target states in eqs . ( [ eq : psig ] ) and ( [ eq : psinoon ] ) . the pulses are taken according to the calculation of chapter iii b. that is , in the @xmath129th step , we use a sinusoidal driving with the driving frequency @xmath133 . since the sinusoidal driving lasts for a duration @xmath130 , the driving field can be considered as square - windowed sinusoidal signal and thus , strictly speaking , is not delta - shaped in the spectrum . the simulation results for generating target states in eq . ( [ eq : psig ] ) and eq . ( [ eq : psinoon ] ) are listed in table [ tab : psig ] and table [ tab : psinoon ] , respectively . we can easily find that larger reduced driving strengths @xmath65 and lamb - dick parameters @xmath298 can usually make the fidelity higher . for the evenly - populated state @xmath299 , the largest fidelity 0.939 can be obtained at @xmath300 and @xmath301 . however , for the noon state @xmath302 , the largest fidelity 0.92 can be obtained at @xmath303 and @xmath304 . [ c]c|c|cccccc & + & 0.2 & 0.3714 & 0.4571 & 0.5429 & 0.6286 & 0.7143 + & 0.3 & 0.115 & 0.393 & 0.519 & 0.608 & 0.739 & 0.675 + & 0.7857 & 0.589 & 0.817 & 0.872 & 0.851 & 0.911 & 0.852 + @xmath65 & 1.0286 & 0.615 & 0.85 & 0.876 & 0.906 & 0.886 & 0.857 + & 1.2714 & 0.724 & 0.872 & 0.886 & 0.906 & 0.852 & 0.878 + & 1.7571 & 0.821 & 0.939 & 0.899 & 0.859 & 0.825 & 0.837 + & 2 & 0.867 & 0.915 & 0.838 & 0.876 & 0.887 & 0.859 + [ tab : psig ] [ c]c|c|cccccc & + & 0.2 & 0.3714 & 0.4571 & 0.5429 & 0.6286 & 0.7143 + & 0.3 & 0.108 & 0.34 & 0.403 & 0.395 & 0.461 & 0.687 + & 0.7857 & 0.675 & 0.815 & 0.833 & 0.862 & 0.829 & 0.868 + @xmath65 & 1.0286 & 0.78 & 0.857 & 0.846 & 0.867 & 0.883 & 0.813 + & 1.5143 & 0.871 & 0.877 & 0.873 & 0.877 & 0.787 & 0.806 + & 1.7571 & 0.844 & 0.918 & 0.889 & 0.902 & 0.862 & 0.819 + & 2 & 0.876 & 0.909 & 0.832 & 0.92 & 0.853 & 0.806 + [ tab : psinoon ] in * * the above , we only discuss the effect of unwanted terms on the generation of target states . we now study the effect of dissipation on the fidelities of target states by numerical simulation for given parameters . when the environmental effect is included , the dynamical evolution of the sqc can be described by the master equation @xmath305 + \mathcal{d}\left [ \sqrt{\gamma_{eg}}\tilde{\sigma}_{ge}\right ] \rho+\mathcal{d}\left [ \sqrt{\gamma_{ee}}\tilde{\sigma}_{ee}\right ] \rho\nonumber\\ & + \mathcal{d}\left [ \sqrt{\gamma_{gg}}\tilde{\sigma}_{gg}\right ] \rho+\mathcal{d}\left [ \sqrt{\kappa_{1}}a_{1}\right ] \rho+\mathcal{d}\left [ \sqrt{\kappa_{2}}a_{2}\right ] \rho , \label{eq : me_schr}\ ] ] where @xmath306 and @xmath307 are the reduced density operator and the hamiltonian of the whole system , respectively . the total hamiltonian has been given in eq . ( [ eq : h ] ) . the compact notation @xmath308 \rho=\left ( 2c\rho c^{\dag}-c^{\dag}c\rho-\rho c^{\dag}c\right ) /2 $ ] represents the lindblad - type dissipation . we have noted that @xmath309 is the basis of @xmath310 , but the qubit dissipation is determined by the qubit basis @xmath311 . the ground ( @xmath312 ) and excited ( @xmath313 ) states of the qubit are given by the eigenstates of eq . ( [ eq : hq ] ) . if we define @xmath314 with @xmath315 and @xmath316 , and also define @xmath317 with @xmath318 and @xmath319 . we can easily verify @xmath320 where @xmath321 , and @xmath322 . in eq . ( [ eq : me_schr ] ) , @xmath323 is the pure - relaxation rate from the qubit excited state to the ground state . besides , @xmath324 and @xmath325 are the pure - dephasing rates originating from disturbed qubit eigenstates . the decay rates of the first and the second cavity fields are denoted by @xmath326 and @xmath327 , respectively . using parameters in sec . [ sec : unter ] and taking the reduced driving * * strength @xmath328 and lamb - dicke parameter @xmath301 from table [ tab : psig ] and table [ tab : psinoon ] , we find that the highest fidelity @xmath329 is achieved for generating the evenly - populated state @xmath299 in eq . ( [ eq : psig ] ) , and a high fidelity @xmath330 is also reached for generating the noon state @xmath331 in eq . ( [ eq : psinoon ] ) . we now assume that the decay rates in eq . ( [ eq : me_schr ] ) are taken as @xmath332 , @xmath333 mhz , and @xmath334 mhz . we assume that the density operators @xmath335 and @xmath336 are the actually generated states for the target states @xmath299 and @xmath331 . then the fidelities can be redefined as @xmath337 we perform numerical simulations using the above parameters and obtain @xmath338 @xmath339 and @xmath340 . the total time for generating @xmath299 is @xmath341 and that for generating @xmath331 is @xmath342 . both @xmath343 and @xmath344 are too small to induce significant decoherence at the decay rates specified by us . thus , the fidelity losses induced by dissipation are fairly small , which are @xmath345 for the evenly - populated state @xmath346 and @xmath347 for the noon state @xmath331 . we now discuss the advantages and disadvantages between our methods and the previous ones @xcite for generating arbitrarily entangled states of two microwave fields or two vibrational modes . [ c]c|c|c|c|c & * pop . leak . * & * no . at . lev . * & * st . eff . * & * mult . proc . * + ref . @xcite & yes & 2 & no & l. pn . @xcite & no & 3 & no & l. pn . @xcite & no & 3 & no & h. pn . @xcite & no & 2 & yes & none + ref . @xcite & no & 2 & no & h. pn . @xcite & no & 2 & yes & none + our proposal & no & 2 & no & l. + [ tab : compareuniversal ] the brief comparison between these methods is listed in table . [ tab : compareuniversal ] . in detail , @xcite provided an algorithm to generate arbitrarily entangled states of two vibrational modes . but due to population leakage outside the original space , it takes an exponential complexity of the number of steps . the succeeding proposals @xcite overcome the exponential drawback in several ways : ( 1 ) a third atomic level is used to shield oscillations that cause population leakage @xcite . but the disadvantage is that higher energy levels of systems usually have larger decay rates , which inevitably reduce the fidelities of the target states . ( 2 ) boson - number - dependent stark effects are used to realize independent operations of particular states @xcite . but the disadvantage is that the detunings of nonresonant terms are usually less by one order of the coupling strengths between the two - level system and boson modes . this means that the rabi frequencies are smaller , and the longer generation time is required . ( 3 ) multiphoton processes of high photon number are used to shield oscillations that cause population leakage or reduce the number of steps @xcite . but the disadvantage is that if the coupling strengths between the atom and cavity fields are not high enough , then the rabi frequencies become small , especially for states with high photon numbers , which obviously indicates longer generation time . besides the advantage that there is no population leakage , our method has also the following advantages compared with previous ones . @xcite : ( 1 ) it only uses the two energy levels of the qubit . thus , the fidelities of the target states should be higher because there is no other auxiliary energy levels . ( 2 ) the detunings of the nonresonant terms are in the order of the resonator frequencies . they are usually bigger than the coupling strengths between the qubit and resonator modes . thus the rabi frequency can be made bigger than those using boson - number - dependent stark effects . ( 3 ) we use multiphoton processes of low photon number , i.e. , one photon at most is converted for either mode . thus the rabi frequency can be bigger than those using multiphoton processes of higher photon number , especially when the coupling strengths between the qubit and cavity modes are not very big . of course , stronger couplings will further enhance the rabi frequencies and hence reduce the generation time . we point out that the real supercoducting qubit circuits are mutilevel systems , the information leakage to higher levels is not avoidable . however , the leakage can be neglected when the transition frequency between the first excited state and the second excited state is much larger than the qubit frequency . for example , in the flux qubit circuits , due to its large anharmonicity of energy levels , the information leakage is negligibly small . however , for the transmon and phase qubit , the anharmonicity is very weak . thus , the pulse should be carefully calibrated to avoid information leakage to higher levels . the pulse calibration can be done as in ref . @xcite . we now compare the differences between our algorithm and other ones for generating noon states . ref . @xcite uses mutliphoton processes to generate noon states . in superconducting systems , this means a low generation efficiency if the lamb - dicke parameter is not sufficiently big . @xcite uses synchronization technology to generate noon states , but the time duration for synchronization between two steps can be quite long and there exists inevitably information leakage . ref . @xcite and its experimental realization @xcite use two phase qubits with three active energy levels to generate noon states of two cavity modes . the experimental setup is complex and the high energy levels of qubits will reduce the decoherence time . * * ref . @xcite uses photon number - dependent stark effects to achieve independent operations . thus the rabi frequency is smaller than the qubit - cavity coupling strengths . @xcite requires that two qubits be initially prepared in a bell state and finally get decoupled from the qubits and cavity fields . @xcite uses one qubit but still needs one additional level to shield unwanted resonances . more recently , ref . @xcite uses one qubit of four levels which resonantly interacts with two resonators simultaneously to speed up the generation process of noon states . when applied to generating noon states , our algorithm has new features besides the common advantages for generating arbitrary two - mode photon states : ( 1 ) only carrier processes @xcite and one - photon processes are used . in this case , even though the coupling strengths between the qubit and cavity modes are small , large rabi frequencies can still be obtained . ( 2 ) the number of steps is reduced to linear dependence on the maximum photon number . these advantages indicate less generation time and thus guarantee a higher efficiency than preceding methods . now we discuss the experimental feasibility of our scheme . table [ tab : psig ] and table [ tab : psinoon ] show that without pulse calibration , higher fidelities can be achieved at bigger * * lamb - dicke parameters @xmath298 and reduced driving frequencies @xmath65 . these values are already in the ultrastrong regime . ref . @xcite has reported ultrastrong couplings between three resonator modes and a flux qubit , where the lamb - dicke parameter @xmath298 can reach as high as @xmath348 . in the ultrastrong regime , rabi frequencies can be made to approach the magnitude of @xmath64 , which usually ranges from @xmath259 to @xmath349 ghz . the decay rates of the qubit and cavity fields are usually in the magnitude of megahertz . thus the dissipation has small effect on the fidelities of target states . for singe - mode microwave fields , fock states with up to six photons @xcite and fock state superpositions @xcite have been experimentally demonstrated using phase qubits . the noon state up to @xmath200 photons has also been experimentally reported @xcite . we thus hope that our proposal is also experimentally feasible in the near future . in summary , we have proposed an approach to generate arbitrary superpositions of photon states of two microwave fields in two separated cavities . our method mainly depends on the coexistence of transverse and longitudinal couplings between the qubit and cavity fields . employing the longitudinal couplings , we derive a hamiltonian which is similar to that of trapped ions interacting with two vibrational modes @xcite . using four simple interaction hamiltonians derived from the longitudinal coupling , we design the state generation algorithm . our algorithm can be regarded as the improved version of that in @xcite when the transverse and longitudinal couplings coexist in circuit qed systems . but it has remedied the drawback that the number of steps exponentially depends on the maximal photon number , which is replaced by a quadratic dependence . compared with previous ones with quadratic complexity , our algorithm does not require atomic energy levels higher than two @xcite , boson - number - dependent stark effects @xcite , or multiboson processes of high boson numbers @xcite . when applied to the generation of noon states , whose engineering has been extensively studied @xcite , our algorithm needs only carrier and one - photon sideband transitions . meanwhile , the number of steps only linearly depends on the maximum photon numbers . in fact , these properties for generating noon states can be generalized to any states with a constant total photon number of both modes . we have also discussed how to avoid the effect of unwanted terms on the generation of target state . our numerical results show that fidelities above @xmath350 can be reached in the ultrastrong regime for the two - photon evenly - populated state and noon state when the environmental effect is neglected . the generation time can be very short , in which case , the environment has small effect on fidelities of the target states . we here note that due to the similarity of two - mode interaction hamiltonians , the algorithm using two - mode multi - phonon processes in ref . @xcite can be directly applied into our model . thus , two - mode fock states with high photon numbers can be generated with just two steps as one - mode fock states in @xcite . we have noted that our method for generating noon states is similar to a recent algorithm simplified from the one which employs stark effects to generate arbitrary entangled states @xcite . yxl is supported by the national basic research program of china grant no . 2014cb921401 , the nsfc grants no . 61025022 , and no . the basic principle of our algorithm is state space reduction . the state space of the target state in eq . ( [ eq : psi_f ] ) can be denoted as @xmath351 we will implement @xmath352 procedures , with each procedure containing some steps . in the 1st procedure , we aim to clear the populations in the subspace @xmath353 of @xmath354 . this can be achieved via alternatively switching @xmath122 10 transitions and @xmath122 01 transitions , i.e. , @xmath355{0\bar{1}}\left\vert 0,n_{\max}-1\right\rangle \left\vert e\right\rangle \xrightarrow[f-1]{\bar{1}0}\left\vert 1,n_{\max}-1\right\rangle \left\vert g\right\rangle \cdots\nonumber\\ & \xrightarrow[f-2n_{\mathrm{max}}+2]{0\bar{1}}\left\vert n_{\max } -1,0\right\rangle \left\vert e\right\rangle \xleftarrow[f-2n_{\mathrm{max}}+1]{\bar{1}0}\left\vert n_{\max},0\right\rangle \left\vert g\right\rangle .\end{aligned}\ ] ] to transfer the populations in the subspace @xmath356 of @xmath354 to the state @xmath357 . thus , @xmath354 is reduced to the state space @xmath358 where@xmath359 in the @xmath361th procedure , we aim to clear the populations in the subspace @xmath362 of @xmath363 . this can be achieved via alternatively switching @xmath364 11 transitions and @xmath364 00 transitions , i.e.,@xmath365{1\bar{1}}\left\vert 1,n_{\max}-\mu-1\right\rangle \left\vert e\right\rangle \nonumber\\ & \xrightarrow[f - n_{2\mu}-1]{00}\left\vert 1,n_{\max}-\mu-1\right\rangle \left\vert g\right\rangle \cdots\nonumber\\ & \xrightarrow[f - n_{2\mu}-2(n_{\mathrm{max}}-\mu)+2]{1\bar{1}}\left\vert n_{\max}-\mu,0\right\rangle \left\vert e\right\rangle \nonumber\\ & \xrightarrow[f - n_{2\mu}-2(n_{\mathrm{max}}-\mu)+1]{00}\left\vert n_{\max } -\mu,0\right\rangle \left\vert g\right\rangle .\end{aligned}\ ] ] with @xmath366 , to transfer the populations in the subspace @xmath367 of @xmath363 to the state @xmath368 . here thus , @xmath363 is reduced to the state space @xmath369 where @xmath370 in the @xmath371th procedure , we aim to clear the populations in the subspace @xmath372 of @xmath373 . this can be achieved via alternatively switching @xmath364 10 transitions and @xmath374 01 transitions , i.e. , @xmath375{\bar{1}0}\left\vert 1,n_{\max}-\mu-1\right\rangle \left\vert g\right\rangle \nonumber\\ & \xrightarrow[f - n_{2\mu+1}-1]{0\bar{1}}\left\vert 1,n_{\max}-\mu -2\right\rangle \left\vert e\right\rangle \cdots\nonumber\\ & \xrightarrow[f - n_{2\mu+1}-2(n_{\mathrm{max}}-\mu)+3]{\bar{1}0}\left\vert n_{\max}-\mu-1,0\right\rangle \left\vert e\right\rangle \nonumber\\ & \xleftarrow[f - n_{2\mu+1}-2(n_{\mathrm{max}}-\mu)+2]{0\bar{1}}\left\vert n_{\max}-\mu,0\right\rangle \left\vert g\right\rangle .\end{aligned}\ ] ] with @xmath376 , to transfer the populations in the subspace @xmath377 of @xmath373 to the state @xmath378 . thus , @xmath373 is reduced to the state space @xmath379 where@xmath380 in the ( @xmath352)th procedure , we aim to clear the populations in the subspace @xmath381 of @xmath382 . this can be achieved via switching one 00 transitions , i.e.,@xmath383{00}\left\vert 0,0\right\rangle \left\vert g\right\rangle .\ ] ] to transfer the populations in the subspace @xmath384 of @xmath382 to the state @xmath385 . thus , @xmath382 is reduced to the state space @xmath386 where @xmath387 is namely the intial state space for @xmath388 in eq . ( [ eq : psi_0 ] ) . base on the above discussion , we now calculate the number of steps to generate the target state in eq . ( [ eq : psi_f ] ) with an arbitrary maximum photon number @xmath122 . for alternatively switching 11 `` and 00 '' transitions , we need @xmath389 steps , given by @xmath390 for alternatively switching 10 `` and 01 '' transitions , we need @xmath391 steps , given by @xmath392 therefore , the total step number @xmath139 is @xmath393 for generating the target state in eq . ( [ eq : psi_f ] ) . h. wang , m. mariantoni , r. c. bialczak , m. lenander , e. lucero , m. neeley , a. d. oconnell , d. sank , m. weides , j. wenner , t. yamamoto , y. yin , j. zhao , j. m. martinis , and a. n. cleland , phys . 106 * , 060401 ( 2011 ) .

the quark structure of the nucleons , suggested by quantum chromodynamics , opens the possibility of a hadron - quark phase transition at high densities and/or temperatures , as suggested by @xcite . in the theories of strong interaction quark bag models suppose that breaking of physical vacuum takes place inside hadrons . if the hypothesis of the quark matter is true , then some compact objects identified with neutron stars could actually be strange stars , built entirely of strange matter @xcite . for a review of strange star properties , see @xcite . most of the investigations of quark star properties have been done within the framework of the so - called mit bag model . assuming that interactions of quarks and gluons are sufficiently small , neglecting quark masses and supposing that quarks are confined to the bag volume ( in the case of a bare strange star , the boundary of the bag coincides with stellar surface ) , the energy density @xmath0 and pressure @xmath1 of a quark - gluon plasma are related , in the mit bag model , by the equation of state ( eos ) @xcite @xmath2 where @xmath3 is the difference between the energy density of the perturbative and non - perturbative qcd vacuum ( the bag constant ) . equation ( [ 1 ] ) is essentially the equation of state of a gas of massless particles with corrections due to the qcd trace anomaly and perturbative interactions . more sophisticated investigations of quark - gluon interactions have shown that eq . ( [ 1 ] ) represents a limiting case of more general equations of state . for example mit bag models with massive strange quarks and lowest order qcd interactions lead to some corrections terms in the equation of state of quark matter . models incorporating restoration of chiral quark masses at high densities and giving absolutely stable strange matter can no longer be accurately described by using eq . ( [ 1 ] ) . on the other hand in models in which quark interaction is described by an interquark potential originating from gluon exchange and by a density dependent scalar potential which restores the chiral symmetry at high densities @xcite , the equation of state @xmath4 can be well approximated by a linear function in the energy density @xmath5 @xcite . @xcite has studied the linear approximation of the equation of state , obtaining all parameters of the eos as polynomial functions of strange quark mass , qcd coupling constant and bag constant . a complete description of static strange stars has been obtained based on numerical integration of mass continuity and tov ( hydrostatic equilibrium ) equations for different values of the bag constant @xcite . using numerical methods the maximum gravitational mass @xmath6 and the maximum radius @xmath7 of the strange star , have been obtained , as a function of the bag constant , in the form @xcite @xmath8 where @xmath9 mev @xmath10 . there are several proposed mechanisms for the formation of quark stars . quark stars are expected to form during the collapse of the core of a massive star after the supernova explosion as a result of a first or second order phase transition , resulting in deconfined quark matter @xcite . the proto - neutron star core or the neutron star core is a favorable environment for the conversion of ordinary matter to strange quark matter @xcite . another possibility is that some neutron stars in low - mass x - ray binaries can accrete sufficient mass to undergo a phase transition to become strange stars @xcite . this mechanism has also been proposed as a source of radiation emission for cosmological @xmath11-ray bursts @xcite . strange quark matter is filled with electromagnetic waves in thermodynamic equilibrium with quarks . photon emissivity is the basic parameter for determining macroscopic properties of objects made of strange matter . @xcite have shown that , because of very high plasma frequency near the strange matter edge , photon emissivity of strange matter is very low . for temperatures @xmath12 , where @xmath13 mev is the characteristic transverse plasmon cuttof energy , the equilibrium photon emissivity of strange matter is negligible small , compared to the black body one . the dispersion relation of the electromagnetic waves in quark matter can be written as @xmath14 , where @xmath15 is the frequency of the waves , @xmath16 is their wave number , @xmath17 is the plasma frequency , with @xmath18 the baryon number density of strange quark matter and @xmath19 is the chemical potential of the quarks ( usov 1998 ) . propagating modes exist only for @xmath20 . hence the lower limit of the energy @xmath21 of the electromagnetic quanta , which are in equilibrium with quarks is given by @xmath22 mev , where @xmath23 @xmath24 is normal nuclear matter density . at the quark star surface @xmath25 and the spectrum of equilibrium photons is very hard , with @xmath26 mev @xcite . the bremsstrahlung emissivity of quark matter has been estimated by @xcite . they have found that the surface radiation is about four orders of magnitude weaker than the equilibrium black body radiation , even they both have the same temperature dependence . the structure of a realistic strange star is very complicated , but its basic properties can be described as follows @xcite . beta - equilibrated strange quark - star matter consists of an approximately equal mixture of up @xmath27 , down @xmath28 and strange @xmath29 quarks , with a slight deficit of the latter . the fermi gas of @xmath30 quarks constitutes a single color - singlet baryon , with baryon number @xmath31 . this structure of the quarks leads to a net positive charge inside the star . since stars in their lowest energy state are supposed to be charge neutral , electrons must balance the net positive quark charge in strange matter stars . the electrons , being bounded to the quark matter by the electromagnetic interaction and not by the strong force , are able to move freely across the quark surface , but clearly can not move to infinity because of the electrostatic attraction of quarks . the electron distribution extends up to @xmath32 fm above the quark surface . the coulomb barrier at the quark surface of a hot strange star may also be a powerful source of @xmath33 pairs , which are created in the extremely strong electric field of the barrier . at surface temperatures of around @xmath34 k , the luminosity of the outflowing plasma may be of the order @xmath35 ergs@xmath36 @xcite . moreover , as shown by @xcite , the thermal luminosity from the star surface , due to both photon emission and @xmath33 pair production may be , for about one day for normal quark matter and for up to a hundred years for superconducting quark matter , orders of magnitude higher than the eddington limit . it is the purpose of the present paper to reconsider the problem of the photon emissivity , via bremsstrahlung radiation , of the quark stars . equilibrium radiation , transmitted through the surface , is dominant at temperatures @xmath37 k , while below this temperature the bremsstrahlung radiation from the surface layer prevails . hence bremsstrahlung is the main source of radiation for cold quark stars . this electromagnetic radiation is generated in the dense medium at the stellar surface in which extremely relativistic quarks move . in the 1950 s @xcite predicted that the cross section for bremsstrahlung from highly relativistic particles in dense media is suppressed , due to the interference between amplitudes of nearby interactions . the suppression has its roots in the uncertainty principle . the kinematics of the bremsstrahlung requires that the longitudinal momentum transfer between the fixed and the scattered particles must be small . on the other hand the uncertainty principle requires that the interaction must occur over a large longitudinal distance scale ( formation zone ) . if the charged particle coulomb scatters while traversing this zone , the bremsstrahlung amplitude from before and after the scattering can interfere , reducing the amplitude for photon emission @xcite . this effect has been recently confirmed experimentally for the bremsstrahlung emission of electrons in different materials at a 5% accuracy level @xcite . since the surface of the quark star consists of a very dense medium , with density of the order @xmath38 , multiple collisions of the electromagnetically radiating particle lead to a significant suppression of the quark - quark bremsstrahlung . by adopting a simple model for the elastic scattering of quarks , we derive the spectrum of the classical bremsstrahlung radiation and the emissivity of the quark matter , by considering the multiple scattering effects in the electromagnetic radiation of quark matter . for the range of densities existing at the surfaces of strange stars , which are higher than the nuclear density , the landau - pomeranchuk effect could reduce the bremsstrahlung emissivity of quark matter by an order of magnitude . photon emissivity of quark - gluon plasma ( qgp ) , which is conjectured to be formed in ultrarelativistic heavy ion collisions , has been extensively investigated recently ( for a recent review of direct photon emission from qgp , including comparisons of theoretical predictions with experiments see peitzmann & thoma ( 2002 ) ) . photons and dilepton pairs only interact electromagnetically and their mean free paths are much larger than the size of the qgp . hence these electromagnetic probes leave the hot and dense qgp without further scattering , thus providing important informations about the early stages of the collision and the structure of the qgp . as sources of direct photons one can consider quark annihilation , compton scattering and bremsstrahlung following the initial hard scattering of partons of the nuclei , as well as thermal photons from the qgp and from hadronic interactions in the hot hadronic gas after the hadronization of the plasma ( peitzmann & thoma 2002 ) . the first observation of direct photon production in ultrarelativistic heavy ion collisions has been reported by the wa98 collaboration in @xmath39 collisions at @xmath40 gev at the super proton synchrotron at cern ( aggarwal et al . 2000 ) . the results display a clear excess of direct photons above the expected background from hadronic decays in the range of transverse momentum @xmath41 gev@xmath42 in the most central collisions . these experimental findings provide a confirmation of the feasibility of direct photons as reliable probes in heavy ion collisions and may pave the way for the understanding of the formation and evolution of the qgp . the evaluation of the photo - emission rate from the qgp was initiated by kapusta et al . ( 1991 ) and by baier et al . . however , as shown by aurenche et al . ( 2000 ) , these initial estimations were incomplete , since the bremsstrahlung and inelastic pair annihilation processes contains collinear enhancements which cause them to contribute at the same parametric order in the coupling as the two to two processes even for large photon energies . but these calculations are also incomplete , as they did not incorporate the suppression of photon emission due to multiple scatterings during the photon emission process , which limits the coherence length of the emitted radiation ( the landau - pomeranchuk - migdal effect ) . the rate of photo - production by bremsstrahlung and inelastic pair annihilation in a hot , equilibrated plasma at zero chemical potential , fully including the lpm effect , was calculated , to leading orders in both the electromagnetic and strong coupling constants , by ( arnold , moore & jaffe 2001 , arnold , moore & jaffe 2002 ) . the emission of hard photons from the qgp plasma , using a model based on the thermodynamics of qcd , and the determination of the initial temperature of the expanding fireball has been considered recently by renk ( 2003 ) . the radiation emitted by the quark layer at the surface of the quark star must also cross the thin electron layer extending beyond the star s surface . we also consider the effect of this layer on the outgoing radiation and show that it can significantly reduce the intensity of the radiation coming from the quark star . the present paper is organized as follows . in section ii we discuss the landau - pomeranchuk - migdal effect for quark matter . the bremsstrahlung emissivity of quark matter is considered in section iii . the effect of the electron layer on the radiation outgoing from the quark star surface is analyzed in section iv . in section v we discuss and conclude our results . throughout this paper we use units so that @xmath43 , with @xmath44 the boltzmann constant . the dispersion relation shows that only photons with frequency @xmath45 can propagate inside strange matter . propagation of electromagnetic waves of frequencies lower than @xmath46 is exponentially damped , with the damping coefficient @xmath47 depending on @xmath15 and @xmath46 . for @xmath48 the value of @xmath47 is given by @xmath49 @xcite . this means that photons with @xmath50 emitted in various processes inside strange matter are absorbed . the mean free path of such a photon is of the order of @xmath51 fm @xcite . therefore , only photons produced just below the surface with momenta pointing outwards can leave strange matter . in the following we shall consider the photon emissivity of quark matter , by also taking into account the effect of multiple scatterings on the emitted radiation . this effect is called the landau - pomeranchuk - migdal ( lpm ) effect , and was considered initially by @xcite , who used classical electrodynamics to demonstrate bremsstrahlung suppression due to multiple scattering : the suppression comes from the interference between photons emitted by different elements of the charged particle trajectory . to study the astrophysical implications of this effect on quark matter we shall take into account the bremsstrahlung photons produced in the quark - quark interactions , that is in the process @xmath52 . the lpm effect can be discussed in a simple qualitative way as follows . let s consider a bremsstrahlung photon of energy @xmath15 produced by an ultrarelativistic quark of energy @xmath53 and mass @xmath54 , where @xmath55 . in this kinematic regime the average angle @xmath56 between the incident quark and the produced photon is small , @xmath57 . we assume that the angle between the incident and scattered quark is also small . moreover , we shall assume that all quarks have the same density dependent effective mass @xmath58 , where @xmath59 is the qcd coupling constant and @xmath60 , with @xmath61 the quark particle number densities . neglecting the scattering and quark - photon angles , it follows that the longitudinal momentum transfer @xmath62 to the static quark is @xmath63 , where @xmath64 @xcite . the uncertainty principle then requires that the spatial position of the bremsstrahlung process has a longitudinal uncertainty of @xmath65 . in alternate language , the quark and photon slowly split apart over the distance @xmath66 . in a sufficiently dense medium the quark mean free path is much less than @xmath66 ; in the lpm effect the relevant interaction is multiple scattering . bremsstrahlung is suppressed when the mean square multiple scattering angle over the distance @xmath66 , @xmath67 @xmath68 , where @xmath69 and @xmath70 ^{-1}$ ] , with @xmath71 , is the radiation length , is greater than or equal to @xmath72 , @xmath73 @xcite . therefore the bremsstrahlung differential cross section for the production of a photon is suppressed when @xmath74 , where @xmath75 @xcite . by assuming that the temperature of the star @xmath76 , with @xmath77 the fermi momentum , the fermi distribution factors force all the quarks to have @xmath78 . for an extreme relativistic quark gas @xmath79 , where @xmath80 is the statistical weight for quark matter . therefore the critical lpm frequency for bremsstrahlung photon emission in quark matter is given by @xmath81 where @xmath82 . for strange quark matter @xmath83 . emission of the bremsstrahlung radiation with frequency smaller than @xmath84 is suppressed , due to the effect of multiple scattering . as expected , the lpm frequency increases with the density of the strange matter . the variation of @xmath85 as a function of the particle number density is represented , for different values of the strong coupling constant @xmath59 , in fig . 1 . the most important parameter characterizing the lpm effect is the average time between two collisions , denoted @xmath86 . it can be calculated from @xmath87 , where @xmath88 is the density of the medium which depends on its composition and on temperature @xcite . @xmath89 denotes the cross section of a test particle having velocity @xmath90 and interacting with the particles of the medium . thus generally @xmath86 is a function of the temperature and density of the medium and of the momentum of the test particle . the value of @xmath86 is essential for the importance of the lpm effect : if @xmath91 is zero ( @xmath92 ) , the effect is absent @xcite . the starting point for the rigorous implementation of the lpm effect is the following textbook equation for the energy radiated per unit momentum by the charged particle @xcite , @xmath93 } \vec{n}\times \vec{v}_{l}dt\right| ^{2 } , \label{1a}\ ] ] where @xmath94 describes the trajectory of the charged particle , @xmath95 is its velocity , @xmath96 , @xmath97 is the fine structure constant and @xmath98 is the charge of @xmath99 quarks . if only one collision occurs and if the velocity is a constant ( @xmath100 ) before and after the collision ( @xmath101 ) , then the radiated energy is given by @xcite @xmath102 which is the standard textbook form for the calculation of the intensity . if , however , a series of collisions occurs , one has to change eqs . ( [ 1a ] ) and ( [ 2a ] ) to a sum over all pieces of the trajectory @xcite , @xmath103 } \vec{n}\times \vec{v}_{1l}dt+\int_{t_{1}}^{t_{2}}e^{i\left [ \omega t-\vec{k}\cdot \vec{r}_{l}\left ( t\right ) \right ] } \vec{n}\times \vec{v}_{2l}dt+ ... +\int_{t_{n-1}}^{t_{n}}e^{i\left [ \omega t-\vec{k}\cdot \vec{r}_{l}\left ( t\right ) \right ] } \vec{n}\times \vec{v}_{nl}dt+ .... \right| ^{2}\end{aligned}\ ] ] assuming that the velocity is a constant between two collisions leads to @xmath104 where @xmath105 is the total number of collisions . the calculation of the square gives @xmath106 + \nonumber \\ & & 2\frac{\alpha } { \left ( 2\pi \right ) ^{2}}\sum_{m = u , d , s}\left ( \frac{q_{m}}{e}\right ) ^{2}\textrm{re}\sum_{j > l}^{n}\frac{\vec{v}_{jm}\cdot \vec{v}_{lm}-\left ( \vec{n}\cdot \vec{v}_{jm}\right ) \left ( \vec{n}\cdot \vec{v}_{lm}\right ) } { \left ( \omega -\vec{k}\cdot \vec{v}_{jm}\right ) \left ( \omega -\vec{k}\cdot \vec{v}_{lm}\right ) } \exp \left [ i\sum_{i = l}^{j-1}\xi _ { i}\left ( \omega -\vec{k}\cdot \vec{v}_{im}\right ) \right]\nonumber \\ & & \left ( 1-e^{-i\left ( \omega -\vec{k}\cdot \vec{v}_{jm}\right ) \xi _ { j}}\right ) \times \left ( 1-e^{-i\left ( \omega -\vec{k}\cdot \vec{v}_{lm}\right ) \xi _ { l}}\right ) , \end{aligned}\ ] ] where we denoted @xmath107 . as the quark gas is taken to be in thermal equilibrium , many possible velocities can result in between two collisions . for this reason the non - diagonal terms in the above equation will give zero contribution after one averages over all velocities , providing of course that no correlation exists between the velocity before and after the collision . only the diagonal terms remain in this case . the time between two successive collisions is given by @xmath108 . taking an average over the time between two collisions can be obtained using the following distribution @xcite @xmath109 then the electromagnetic energy radiated away by the quarks is given by @xmath110 d\xi , \ ] ] where we have introduced the average velocities by means of the definitions @xmath111 , @xmath112 and we also denoted @xmath113 . the averaging over the time between two collisions can be done with the use of the mathematical identity @xmath114 . therefore we obtain for the intensity of the radiation emitted by a quark moving in a dense medium the expression @xmath115 since @xmath116 , the intensity can also be written as @xmath117 the angular integration is standard , and finally we obtain for the intensity of the radiation emitted by the quark moving in a dense medium the expression @xcite @xmath118 where the function @xmath119 is given by @xmath120 \nonumber \\ & & + \omega \ln \frac{1+\omega ^{2}\tau ^{2}\left ( 1+v_{l}\right ) ^{2}}{1+\omega ^{2}\tau ^{2}\left ( 1-v_{l}\right ) ^{2}}-2\omega v_{l}.\end{aligned}\ ] ] in order to compute the numerical values of the intensity we need to know the mean velocities of the quarks randomly moving in a dense medium . the multiple scattering of the quarks change the quark path and affect the emission of the radiation . if @xmath121 is the initial velocity of the quark at the moment @xmath122 , then at the moment @xmath123 its velocity will be @xmath124 , where @xmath125 is a unit vector perpendicular to the initial direction of motion and @xmath126 is the electron multiple scattering in the time interval @xmath127 to @xmath123 @xcite . @xcite took @xmath128 . therefore , in this approximation we obtain @xmath129 . over a distance @xmath130 the multiple scattering is given by @xmath131 . for @xmath28 we shall assume that it is the suppression distance , that is , the distance over the interaction actually spread : @xmath132 . in the most general case , with quantum corrections taken into account , @xmath133 is given as a root of a quadratic equation : @xmath134 , where @xmath135 is the classical suppression length @xcite . in the case of the standard approach to the bremsstrahlung radiation of charged particles , one assumes that the energy is emitted in a single collision . then , from eq . ( [ single ] ) , by taking @xmath136 and @xmath137 , it immediately follows that the single collision spectral distribution of the bremsstrahlung radiation @xmath138 is given by @xcite @xmath139 the total intensity is proportional to the frequency of the emitted radiation and is independent on the medium in which the motion of the particle takes place . the same result can be derived by taking the limit @xmath140 in eq . ( [ int ] ) : @xmath141 in the opposite limit of very high density , corresponding to @xmath142 , from eq . ( [ int ] ) we obtain @xmath143 and @xmath144 . therefore the electromagnetic radiation from a quark moving in a very dense medium is completely suppressed . to calculate the bremsstrahlung emissivity a sensible model for quark interactions has to be used . we adopt the basic assumption that the details of the interaction should be relatively unimportant , as long as the overall strength is correct and the relevant symmetries are respected . hence we use the well - known nambu - jona - lasinio ( njl ) model in its @xmath145 version in our calculations @xcite . it has the same symmetries as qcd and describes an effective pointlike interaction with a constant coupling strength . usually the njl model and its extensions are used in mean field calculations . as a consequence of adopting this model all the aspects related to color superconductivity are neglected . let @xmath146 be the cross - section for a given process of scattering of charged particles , which may be accompanied by the emission of a certain number of photons . for example , @xmath146 could refer to the scattering of a quark by an other quark , with the possible emission of hard photons . together with this process one could consider another process which differs from it only in that one extra photon is emitted . in this case the total cross - section @xmath147 can be represented as a product of two independent factors , the cross - section @xmath146 and the probability @xmath148 of emission of a single photon in the collision @xcite . the emission of a soft photon is a quasi - classical process . the probability of emission is the same as the classically calculated number of quanta emitted in the collision , that is the same as the classical intensity ( total energy ) of emission @xmath149 , divided by the frequency of the radiation @xmath15 @xcite . hence as a first step in obtaining the bremsstrahlung emissivity of quark matter we have to calculate the classical radiation intensity emitted by a quark moving in a dense medium in which many inter - particle collisions occur . the probability @xmath148 of emitting a photon of energy @xmath15 by any of the scattered quarks is @xmath150 . therefore the total cross section for the emission of soft bremsstrahlung photons is given by @xcite @xmath151 to calculate the total cross section for a photon emission we need to know @xmath146 , describing the scattering of quarks . within the two - flavor njl model the elementary quark - quark cross sections have been calculated by @xcite.the calculations have been done nonperturbatively in the coupling constant , using the so - called @xmath152 expansion , where @xmath153 is the number of colors . in the first order of the @xmath152 expansion and for temperature @xmath154 , where @xmath155 gev , the quark - quark scattering cross - sections have a remarkably simple dependence on the collision energy @xmath29 , given by @xmath156 @xcite . the cross section decreases with increasing energy . for a more general parametrization of the quark - quark scattering cross section , with the inclusion of the double scattering terms see also @xcite . we assume that the quarks form a degenerate fermi gas , with the particle number density @xmath88 given by @xmath157 p^{2}dp=\frac{g}{2\pi ^{2}}\int_{0}^{\infty } \frac{p^{2}dp}{\exp \left [ \frac{e(p)-\mu } { t}\right ] + 1},\ ] ] where @xmath158 is the quark kinetic energy and @xmath159 = 1/\left ( \exp \left [ \frac{e(p)-\mu } { t}\right ] + 1\right ) $ ] is the fermi - dirac distribution function @xcite . in order to estimate the importance of the lpm effect on the bremsstrahlung emissivity of quark matter we have to estimate @xmath86 , the average time between two collisions . in the case of the interaction of two quarks , denoted 1 and 2 , the average time can be obtained from @xcite @xmath160 , \ ] ] where @xmath161 the kinematic invariant @xmath29 is given by @xmath162 @xcite . by assuming that quarks are ultrarelativistic , with @xmath163 , the average collision time can be approximated by the following analytic representation : @xmath164 , \ ] ] where @xmath165 is the polylogarithm function defined as @xmath166 . the variation of the collision time as a function of the quark number density is presented , for different values of the temperature , in fig . 2 . in the framework of our model and for the range of temperatures considered , the collision time is almost independent of @xmath167 . as expected , @xmath86 decreases with increasing density . to calculate @xmath86 we have assumed again that the test quarks , moving in the dense medium , have energies of the same order as the fermi energy , @xmath168 . a rough estimate of the mean collision time can be also obtained from the relation @xmath169 . by assuming that the cross section of quark - quark interactions is around @xmath170 mb @xcite and the quarks are travelling with the speed of light , we obtain @xmath171 fm @xmath172 for @xmath173 @xmath24 and @xmath174 fm @xmath175 for @xmath176 @xmath24 . the knowledge of @xmath86 allows the calculation of the spectrum of the photons emitted via the bremsstrahlung mechanism from the quark star surface . the applicability of the formulas for the spectral distribution and intensity of the bremsstrahlung radiation we have derived in the previous section is limited by the quantum condition that @xmath15 be small as compared with the total kinetic energy of the quark , which is of the order of @xmath177 : @xmath178 . since at the star s surface the kinetic energy of the quarks is of the order of @xmath179 mev , the radiation formulae are valid only for frequencies @xmath180 mev . in order to find the maximum frequency @xmath181 of the radiation in the ultrarelativistic case we assume that it is of the order @xmath182 , that is it is inversely proportional to the duration of the collision and to the square of the energy of the radiating particle @xcite . if we assume that quarks have energies near the fermi energy , then the corresponding velocities are the fermi velocities , @xmath183 which are of the order of @xmath184 for @xmath185 and @xmath186 for @xmath187 . hence the maximum allowable frequency up to which the formalism can be applied is given by @xmath188 for a quark - quark interaction cross sections of the order of @xmath189 mb and for @xmath185 we obtain @xmath190 mev . in the following we shall restrict our study only to quark bremsstrahlung frequencies satisfying the condition @xmath191 mev . the frequency distribution of the radiation emitted by a single quark moving in the dense matter at the surface of the quark star is presented in fig . the total intensity @xmath192 of the bremsstrahlung radiation emitted by a single quark , @xmath193 obtained by integrating eq . ( [ int ] ) over all possible frequencies is represented , as a function of the collision time @xmath86 , in fig . 4 . for the same range of frequencies the calculation of the bremsstrahlung intensity by using eq . ( [ bremst ] ) , which does not take into account the effects of the multiple collisions , gives @xmath194 mev . by taking into account that in the high density quark - gluon plasma the collision time is finite and of the order of @xmath195 fm leads to an intensity of the order @xmath196 mev . hence the lpm effect could reduce the intensity of the photon bremsstrahlung of quarks at the surface of the star by a factor of @xmath197 to @xmath198 for @xmath199 @xmath24 . for collision times of the order of @xmath200 fm , which , due to the incertitudes in the quark - quark cross - sections , are also possible , the decrease in the intensity of the bremsstrahlung is of an order of magnitude . the variation of the intensity of the electromagnetic radiation is represented , as a function of the quark density , and for different values of the temperature , in fig . 5 the intensity of the radiation emitted by a single particle is decreasing with increasing density , due to the effects of the multiple scatterings of the quarks . the number of photons is proportional to the number of collisions , but if these happen too often , due to the high density of the medium , the photons emitted at different points of the trajectory start to interfere with each other and the intensity of the radiation is accordingly reduced . on the other hand @xmath192 increases with the temperature of the dense quark matter . if the quarks are extremely relativistic , with @xmath201 , the intensity of the total radiation emitted by a single quark can be expressed in the following closed form @xmath202 + \nonumber \\ & & \frac{1}{\tau } \left [ \arctan 2\left ( 1 + 3\pi /4\alpha _ { s}\right ) -\arctan 2\tau an_{b}^{1/3}\right ] - \nonumber\\ & & 4\left ( \frac{1 + 3\pi /4\alpha _ { s}}{\tau } -an_{b}^{1/3}\right ) + \frac{1}{2\tau } \left [ d\left ( 1 + 3\pi /4\alpha _ { s}\right ) -d\left ( \tau an_{b}^{1/3}\right ) \right ] , \end{aligned}\ ] ] where the function @xmath203 is given by @xmath204 $ ] . @xmath203 is a real function for all @xmath205 , and its power series expansion is given by @xmath206 . the approximation @xmath207 is appropriate for quarks having energies near the fermi energy and moving with velocities @xmath208 , which are of the order of @xmath189 . the emission of a photon in the quark matter is the result of the scattering of a quark from a state @xmath209 to a state @xmath210 by collision with another particle of momentum @xmath211 . the rate of collisions in the quark - gluon plasma in which a soft photon of energy @xmath55 is emitted can be defined as @xcite @xmath212 \left [ 1-f_{2}\left ( \vec{p}_{2}^{\prime } \right ) \right ] \prod_{i=1,2,1^{\prime } , 2^{\prime } } d^{3}\vec{p}_{i}d\omega \left ( \theta , \phi \right ) , \ ] ] where @xmath213 is the solid angle element in the direction @xmath214 , @xmath215 are the statistical weights and @xmath216 , @xmath217 are the velocities of the interacting particles . the bremsstrahlung emissivity per unit volume of the quark matter is given by @xmath218 withe the use of the explicit form of the cross - section for the photon emission , given by eq . ( [ cross ] ) , the photon emissivity can be written in the following form : @xmath219 where @xmath220 is a function of the quark density and of the temperature . the bremsstrahlung energy flux from the star , @xmath221 is given by the emissivity of a thin surface layer , with thickness of the order of the photon mean free path @xmath222 @xcite . taking into account only photons emitted outwards and produced not deeper than @xmath222 , the energy flux from the surface of the quark star is given by @xmath223 where @xmath224 . since the temperature dependence of @xmath225 is not significant , it follows that the quark bremsstrahlung spectrum has the same temperature dependence as the black body radiation . for arbitrary temperatures the coefficient of the bremsstrahlung emissivity can be represented , in an approximate analytical form , by @xmath226 i\left ( n_{b},t\right ) \lambda .\ ] ] the calculation of @xmath227 requires , in the general case , the estimation of integrals of the form @xmath228 $ ] , which can be expressed usually in terms of the polylogarithm function . in the limit of small temperatures , with @xmath229 , the integral can be evaluated by means of the substitution @xmath230 . for small @xmath167 one can extend the lower limit of integration to @xmath231 , so that the integral becomes @xmath232 = \left ( \pi ^{2}/6\right ) t^{2}$ ] . therefore in the limit of small temperatures @xmath233 , but there is no explicit temperature dependence of the coefficient @xmath225 : @xmath234 in this case the only @xmath167-dependence of @xmath225 is via the intensity of the radiation . the variation of the ratio of the bremsstrahlung emissivity coefficient and the stefan - boltzmann constant @xmath235 is represented , as a function of the mean collision time @xmath86 , and in the limit of small temperatures , in fig . we have assumed that the thickness of the emitting layer is @xmath236 fm . in the limit of large collisions times @xmath237 , and thus we recover the single - collision results of @xcite . for collision times specific to the quark - gluon plasma at the surface of the quark stars , of the order of @xmath238 fm , the ratio @xmath239 . therefore , taking into account multiple collisions in the high density plasma , reduces with an order of magnitude the value of the bremsstrahlung coefficient @xmath225 , making the soft photon radiation from the surface of the quark star about @xmath198 orders of magnitude weaker than the equilibrium black body radiation . the radiation emitted by the thin quark surface layer of the strange star must propagate through the electron layer formed due to the electrostatic attraction of the quarks . in the followings we shall consider the effect of the electron layer on the electromagnetic radiation emitted , via the bremsstrahlung mechanism , by the quarks in the layer of thickness @xmath222 . as a first step it is necessary to obtain a quantitative analytical description of the properties of the electrons outside the quark star . to find the electron distribution near the quark star surface we follow the approach of @xcite , but also taking into account the finite temperature effects as discussed by @xcite . the chemical equilibrium implies that the electron chemical potential @xmath240 is constant , where @xmath241 is the electrostatic potential per unit charge and @xmath242 is the electron s chemical potential . since far outside the star both @xmath241 and @xmath242 tend to zero , it follows that @xmath243 and @xmath244 . for the number density @xmath245 of the electrons at the quark star surface we use the expression @xcite @xmath246 where @xmath167 is the temperature of the electron layer , which can be taken as a constant , since we assume the electrons are in thermodynamic equilibrium with the constant temperature quark matter . the poisson equation for the electrostatic potential @xmath247 generated by the finite temperature electron distribution reads @xcite @xmath248 , z\leq 0 , \label{2}\ ] ] @xmath249 in eqs . ( [ 2])-([3 ] ) , @xmath250 is the space coordinate measuring height above the quark surface , @xmath97 is the fine structure constant and @xmath251 is the quark charge density inside the quark matter . the boundary conditions for eqs . ( [ 2])-([3 ] ) are @xmath252 as @xmath253 and @xmath254 for @xmath255 . in the case of the zero temperature electron distribution at the boundary @xmath256 we have the condition @xmath257 @xcite . in the following we will be interested mainly in the properties of the electron gas extending outside the quark surface of the star , described by eq . ( [ 3 ] ) . in order to find the exact solution of eq . ( [ 3 ] ) , we introduce first a new variable @xmath258 . then eq . ( [ 3 ] ) can be written in the form @xmath259 immediately leading to the first integral @xmath260 where an arbitrary integration constant has been set to zero . the general solution of eq . ( [ eqv ] ) is given by @xmath261 } { \exp \left [ 4\sqrt{\frac{\alpha \pi } { 3}}t\left ( z+z_{0}\right ) \right ] -1},\ ] ] where @xmath262 is a constant of integration . its value can be obtained from the condition of the continuity of the potential across the star s surface , requiring @xmath263 , where @xmath264 is the value of the electrostatic potential in the region @xmath265 , described by eq . ( [ 2 ] ) . therefore @xmath266 .\ ] ] hence the electron number density @xmath267 is given , at finite temperatures , by @xmath268 } { \exp \left [ 4\sqrt{\frac{\alpha \pi } { 3}}t\left ( z+z_{0}\right ) \right ] -1}\left\ { 8\left ( \frac{\exp \left [ 2\sqrt{\frac{\alpha \pi } { 3}}t\left ( z+z_{0}\right ) \right ] } { \exp \left [ 4\sqrt{\frac{\alpha \pi } { 3}}t\left ( z+z_{0}\right ) \right ] -1}\right ) ^{2}+1\right\ } t^{3},t\neq 0.\ ] ] in the limit of zero temperature , @xmath269 , from eq . ( [ eqv ] ) we obtain @xmath270 where @xmath271 and @xmath272 is an integration constant . @xmath272 can be determined from the boundary condition @xmath257 , which gives @xmath273 . therefore , in this case for the electron particle number distribution , extending several thousands fermis above the quark matter surface , we find the expression : @xmath274 the variation of the electron number density @xmath275 is represented , for different values of the temperature , in fig . 7 . in the absence of a crust of the quark star , the electron layer can extend to several thousands fermis outside the star s surface . the intensity of the photon beam coming from the thin outer layer of the quark star decreases due to the energy absorption in the electron layer . the variation of the intensity of the beam can be obtained by solving the equation of the radiative transfer @xcite . for a steady state , the general solution of the transport equation is by given @xmath276 where the optical depth @xmath277 is defined as @xmath278 , with @xmath279 the rosseland mean opacity ( giving the energy absorbed per unit time per unit volume from the beam with intensity @xmath280 ) , @xmath281 is the intensity of the radiation incident on the electron layer and @xmath282 is the emissivity of the electron . since in the present paper we are interested to study only the effect of the outer electrons on the quark matter emission , we shall take @xmath283 , that is we shall neglect the emissivity of the electron layer . hence , the variation of the intensity , due to absorption in the electron layer , of the photon beam emitted by the quarks at the strange star surface can be described by the following equation : @xmath284 in the case of the scattering of photons by free non - relativistic electrons , the opacity is given by @xmath285 , where @xmath286 is the thomson cross - section @xcite . this formula is valid only in the case of a non - degenerate gas at low - temperatures . for systems at high densities and/or temperatures , the corrections to the thomson scattering cross - section become important . for high - temperature non - degenerate electrons the opacity has been calculated first in @xcite . these calculations have been extended to the semi - degenerate ( @xmath287 ) case by @xcite and @xcite ( see also @xcite and @xcite ) . for a degenerate electron gas at @xmath287 , the rosseland mean opacity is given by @xcite @xmath288 @xmath289 ( the inverse of the rosseland mean ) is a function of the temperature and of the degeneracy parameter @xmath290 , where @xmath291 is the fermi energy . @xmath292 can be represented in the form @xcite @xmath293 where @xmath294 . the variation of the function @xmath295 as a function of the parameters @xmath296 and @xmath167 is represented in fig . 8 . for a non - degenerate electron gas , obeying the maxwell - boltzmann distribution , the rosseland mean opacity is given by @xcite @xmath297 where @xmath298 and the temperature is expressed in units of @xmath299 . in order to obtain a simple estimate of the absorption of the surface radiation of quarks due to the presence of the electron layer , we consider first the case of the completely degenerate electron gas at a small temperature @xmath167 . complete degeneracy corresponds to @xmath300 and @xmath301 . we also neglect the temperature dependent terms in eq . ( [ g ] ) , which are not important for small temperatures . hence from eq . ( [ g ] ) we obtain , as a first approximation , @xmath302 . therefore the rosseland opacity of the electron layer at the quark star surface can be approximated , for low temperatures of the degenerate electron gas , by the expression @xmath303 consequently , for the optical depth of the electron distribution above the quark star surface we find the expression @xmath304 . \label{tau0}\ ] ] for large @xmath250 , corresponding to the limit @xmath255 , from eq . ( [ tau0 ] ) we obtain @xmath305 at zero temperature the optical depth is proportional to the square of the value of the quark charge density at the surface of the quark star @xmath306 . the absorption of the bremsstrahlung radiation by the electron layer essentially depends on the quark charge density @xmath307 inside the quark star , where @xmath27 , @xmath28 and @xmath29 denotes the different quark flavors @xcite . the condition of the charge neutrality requires @xmath308 , where @xmath309 is the charge density of the electrons inside the star . @xmath309 is related to the baryon number density @xmath310 by means of the relation @xmath311 , where for the electron abundance @xmath312 we adopt the range @xmath313 @xcite . near the surface of the quark star the density can be approximated by @xmath314 g/@xmath315 . for a strange quark mass of the order of @xmath316 mev , for the baryon number density near the surface of the star we find @xmath317 mev@xmath318 . hence the corresponding quark charge density is @xmath319 mev , leading to @xmath320 mev . the corresponding bremsstrahlung luminosity @xmath321 of the quark star is given by @xmath322 , where @xmath323 is the radius of the star . for a @xmath324 star and for a degenerate electron layer , @xmath325 .\ ] ] the variation of the coefficient @xmath326 is represented , as a function of @xmath306 , in fig . 9 . for an increasing surface quark charge density @xmath326 is a rapidly decreasing function . therefore in the case of a large @xmath306 the electron layer significantly decreases the intensity of the bremsstrahlung radiation emitted by the thin quark layer at the surface of the star . the spectrum of the @xmath287 quark star essentially depends on the charge density @xmath327 at the surface of the star , on the degeneracy parameter @xmath296 and on the temperature . the numerical values of @xmath328 must be obtained by solving eq . ( [ 2 ] ) in the region @xmath329 for different values of @xmath167 , together with the boundary condition @xmath252 for @xmath253 . therefore even for the non - zero temperature quark star , the spectrum of the bremsstrahlung radiation is determined by the quark charge density inside the star . however , in order to obtain a simple estimate of the temperature effects on the optical length of the radiation in the electron layer , we shall assume that for a suitable range of temperatures the charge density at the surface can be approximated by @xmath330 mev @xcite . for this value of the charge density the variation of @xmath326 as a function of @xmath296 and @xmath167 is represented in fig . 10 . since the increase in the temperature also increases the electron number density of the electron layer , the absorption of the bremsstrahlung radiation is much stronger at high temperatures of the semi - degenerate electron gas . since the proposal of the existence of strange stars , much effort has been devoted to find observational properties that could distinguish strange stars from neutron stars . one of the most common and simplest method could be to find some specific signatures in the photon emissivity of strange stars . in the first estimate of the surface photon emissivity of quark matter @xcite it has been concluded that the most important photon radiation mechanism is the bremsstrahlung of quarks from a thin layer at the star s surface , and that the intensity of the radiation is about four orders of magnitude weaker than the equilibrium black body radiation , although both have the same temperature dependence . in the present paper we have reconsidered the electromagnetic emission from the surface of bare quark stars , by also taking into account the important effect of the multiple collisions of quarks in the very high density quark - gluon plasma existing at the surface of the quark stars . to obtain the bremsstrahlung spectrum of the photon emission of the quarks we have used several assumptions . the most important is the adoption of a specific model for the description of quark - quark elastic scattering , which is essential in obtaining the bremsstrahlung emissivity . the details of the interactions between quarks are generally poorly known , and therefore there are a lot of uncertainties in the description of the strong interactions in a quark - gluon plasma . in our calculations we have used a classical soft photon approximation , neglecting quantum effects . we have generally assumed that the quarks are semi - degenerate , with temperature @xmath331 and @xmath332 , but with the kinetic energies of the interacting particles of the order of the fermi energy . in order to give a complete description of the quark - quark bremsstrahlung we have considered , in a systematic way , the effects of the dense medium on the radiation spectrum . a dense medium in which many collisions occur will reduce the bremsstrahlung radiation , and as high the density is , as lower is the bremsstrahlung emission . on the other hand the emission of photons with frequency smaller than the critical lpm frequency is completely suppressed . @xmath85 depends on the qcd coupling constant and on the density at the quark star s surface . for strange stars @xmath85 is of the order of @xmath333 mev , so photons with @xmath334 can not be emitted from the surface of the star . hence only low frequencies are suppressed due to the lpm effect . on the other hand the multiple collision effect reduces with one order of magnitude the bremsstrahlung coefficient @xmath225 , making the soft photon radiation from the quark star @xmath198 orders of magnitude weaker than the black body radiation . a second effect , which essentially influences the radiation of photons by quark stars , is the absorption of the radiation in the electron layer formed outside the quark star . the electron layer could extend to several hundreds or thousands fermis , the electron number density being an increasing function of the temperature . the high - density semi - degenerate electron layer at the surface of a strange star plays an important role in the propagation of the electromagnetic radiation from the star s surface . the optical depth of the degenerate or semi - degenerate electron layer depends on the electron number density , temperature , the degeneracy parameter and the charge density at the surface of the quark star . for high surface charge densities the intensity of the quark bremsstrahlung radiation is significantly reduced by absorption in the electron layer . for the low - temperature star , with @xmath335 , for a surface potential @xmath336 mev , the electron layer reduces the intensity of the radiation by an order of magnitude . in this case the combined effects of the multiple collisions and absorption in the electron layer give @xmath337 . hence the energy radiated away by the quark star could be @xmath338 orders of magnitude smaller than the equilibrium black body radiation . for the sake of comparison we also present the expression for the photon emission rate of an equilibrated , hot qcd plasma , at zero chemical potential , with the lpm effect fully included . the spontaneous emission rate @xmath339 of photons with a given momentum @xmath340 can be represented , for two flavor qcd , as ( arnold et al . 2001 , arnold et al . 2002 ) @xmath341 , \label{ar1}\ ] ] where @xmath342 is the fermi - dirac distribution function , @xmath343 ^{-1}$ ] , @xmath344 is the thermal quark mass , given by @xmath345 , and @xmath346 with @xmath347 and @xmath348 @xmath349 $ ] . for @xmath350 , bremsstrahlung becomes the most important process , while for @xmath351 , pair annihilation dominates . the lpm effect suppresses both bremsstrahlung and pair annihilation processes , but the suppression is not severe ( @xmath352 or less ) , except for photons with @xmath353 or very hard pair annihilation . in the limit of small temperatures , @xmath269 and @xmath354 . hence @xmath355 and @xmath356 . by neglecting the logarithmically divergent terms , the integration over all possible ranges of frequencies in eq . ( [ ar1 ] ) gives @xmath357 where @xmath358 /9 $ ] , with @xmath359 is the riemann zeta function @xmath360 . therefore the perturbative qcd approach gives , in the low temperature limit , the same temperature dependence of the bremsstrahlung spectrum as in eq . ( [ arbr ] ) . however , we must point out that eqs.([ar1 ] ) and ( [ ar2 ] ) have been derived for a temperature of the quark - gluon plasma higher than @xmath361 mev , since the temperature of the fireball formed in heavy ion collisions is of the order of @xmath362 mev ( renk 2003 ) . recently , it has been argued that the strange quark matter in the color - flavor locked(cfl ) phase of qcd , which occurs for large gaps @xmath363 , is rigorously electrically neutral , despite the unequal quark masses , and even in the presence of electron chemical potential @xcite . however , @xcite pointed out that for sufficiently large @xmath364 the low density regime is rather expected to be in the 2-color - flavor superconductor phase in which only @xmath27 and @xmath28 quarks of two color are paired in single condensates , while the ones of the third color and @xmath29 quarks of all three colors are unpaired . in this phase , electrons are present . in other words , electrons may be absent in the core of strange stars but present , at least , near the surface where the density is lowest . nevertheless , the presence of cfl effect can reduce the electron density at the surface and hence increases the bremsstrahlung emissivity . in conclusion , in the present paper we have pointed out two effects that could significantly reduce or even fully suppress the bremsstrahlung radiation from the surface of quark stars : the significant decrease of the intensity due to the multiple collisions in a dense medium and the absorption by the outer electron layer of the stars . the possible observational significance for the detection of quark stars of these effects will be considered in a future publication . the authors would like to thank to the anonymous referee , whose comments helped to improve the manuscript . we are grateful for the useful comments of prof . v. v. usov . this work is supported by a rgc grant of hong kong government . aggarwal , m. m. et al . 2000 , phys . lett . , 85 , 3595 alcock , c. , farhi , e. & olinto , a. 1986 , apj , 310 , 261 alford , m. , rajagopal , k. & wilczek , f. 1998 , phys . b , 422 , 247 alford , m. , rajagopal , k. & wilczek , f. 1999 , nucl . phys . b , 537 , 443 anthony , p. l. , becker - szendy , r. , cavalli - sforza , m. , keller , l. p. , kelley , l. a. , klein , s. r. , niemi , g. , perl , m. l. , rochester , l. s. & white , j. l. 1995 , phys . lett . , 75 , 1949 arnold , p. , moore , g. d. & yaffe , l. g. 2001 , jhep , 0112 , 009 arnold , p. , moore , g. d. & yaffe , l. g. 2002 , jhep , 0206 , 030 aurenche , p. , gelis , f. & zaraket , h. 2000 , phys . d , 62 , 096012 baier , r. , nakkagawa , h. , niegawa , a. & redlich , k. 1992 z. phys . c , 53 , 433 berestetskii , v. b. , lifshitz , e. m. & pitaevskii , l. p. 1982 , quantum electrodynamics ( oxford : pergamon press ) boercker , d. b. 1987 , apj , 316 , l95 buchler , j. r. & yueh , w. r. 1976 , apj , 210 , 440 cheng , k. s. & dai , z. g. 1996 , phys . rev . , 77 , 1210 cheng , k. s. & dai , z. g. 1998 , phys . rev . lett . , 80 , 18 cheng , k. s. , dai , z. g. , wei , d. m. & lu , t. 1998a , science , 280 , 407 cheng , k. s. , dai , z. g. & lu , t. 1998 , int . j. mod . d , 7 , 139 cheng , k. s. & harko , t. 2000 , phys . d , 62 , 083001 chin , c. w. 1965 , apj , 142 , 1481 chiu , h .- y . 1968 , stellar physics ( waltham , massachussets : blaisdell publishing company ) chmaj , t. , haensel , p. & slominski , w. 1991 , nucl . b ( proc . suppl . ) , 24 , 40 cleymans , j. , goloviznin , v. v. & redlich , k. 1993 , phys . rev . d , 47 , 173 dai , z. g. , peng , q. h. & lu , t. 1995 , apj , 440 , 815 dey , m. , bombacci , i. , dey , j. , ray , s. & samanta , b. c. 1998 , phys . b , 438 , 123 gondek - rosinska , d. , bulik , t. , zdunik , l. , gourgoulhon , e. , ray , s. , dey , j. & dey , m. 2000 , a&a , 363 , 1005 gourgoulhon , e. , haensel , p. , livine , r. , paluch , e. , bonazzola , s. & marck , j .- a . 1999 , a&a , 349 , 851 haensel , p. , zdunik , j. l. & schaeffer , r. 1986 , a&a , 160 , 121 haglin , k. & pratt , s. 1994 , phys . b , 328 , 255 harko , t. & cheng , k. s. 2002 , a&a , 385 , 947 hatsuda , t. & kunihiro , t. 1994 , phys . rept . , 247 , 221 jackson , j. d. 1999 , classical electrodynamics ( new york : john wiley and sons ) kapusta , j. , lichard , p. & seibert , d. 1991 phys . d , 44 , 2774 kettner , ch . , weber , f. , weigel , m. k. & glendenning , n. k. 1995 , phys . d , 51 , 1440 klein , s. 1999 , rev . phys . , 71 , 1501 landau , l. d. & lifshitz , e. m. 1975 , the classical theory of fields ( oxford : pergamon press ) landau , l. d. & pomeranchuk , i. j. 1953 , dokl . sssr , 92 , 735 manley , d. m. 1997 , phys . d , 56 , 7325 page , d. & usov , v. v. 2002 , phys . lett . , 89 , 131101 peitzmann , t. & thoma , m. h. 2002 , phys . rept . , 364 , 175 rajagopal , k. & wilczek , f. 2001 , phys . rev . , 86 , 3492 renk , t. 2003 , hep - ph/0301133 , to appear in phys . c rose , s. j. 1995 , apj , 453 , l45 rybicki , g. b. & lightman , a. p. 1979 , radiative processes in astrophysics ( new york : john wiley and sons ) sampson , d. h. 1959 , apj , 129 , 734 shapiro , s. l. & teukolsky , s. a. 1983 , black holes , white dwarfs , and neutron stars ( new york : john wiley and sons ) usov , v. v. 1998 , phys . , 80 , 230 witten , e. 1984 , phys . d , 30 , 272 zdunik , j. l. 2000 , a&a , 359 , 311 zhuang , p. , hufner , j. , klevansky , s. p. & neise , l. 1995 , phys . d , 51 , 3728

the lupus star forming region , recently reviewed by @xcite , lies about 150 pc from the earth @xcite , in the gould belt . it is immediately visible by inspection of optical photographs of the sky , comprising a set of largely filamentary dark clouds . based on optical extinction maps @xcite , the ` cores to disks ' spitzer legacy programme ( c2d , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , has produced and analyzed infrared images of three areas : lupus i , iii , and iv . the c2d data products @xcite include : shorter - wavelength infrared array camera ( irac ) maps , used to identify and classify young stellar objects ( ysos , * ? ? ? * ) ; far - ir mips maps @xcite tracing the thermal emission of the dust component of the molecular clouds ; and ir extinction maps , based on the two micron all sky survey ( 2mass ; * ? ? ? * ) and on a combination of 2mass and irac data @xcite . in this paper , we present maps of the c2d fields in lupus , tracing the molecular hydrogen by the rotational emission of carbon monoxide ( co ) and isotopically substituted @xmath0co . the lupus complex has been extensively surveyed in the various 10 transitions of co : @xcite first mapped the lupus molecular clouds using the 1.2 m columbia telescope with 0.5 resolution to show the extent of the molecular gas in lupus and suggest a total mass of a few @xmath2 @xmath3 . improved maps in several transitions were obtained by nanten ( 2.7 resolution , but routinely undersampled ) : @xcite published @xmath4co maps of the whole complex , mostly sampled at 8 spacing , with some areas at 4 spacing ; @xmath0co 10 maps @xcite cover lupus i and iii with 8 spacing , while c@xmath5o 10 maps @xcite cover all the clouds in the complex at 2 spacing . @xcite mapped about 200 arcmin@xmath6 of lupus iv at sub - arcminute resolution and full - beam spacing in the 10 transitions of co , @xmath0co , and c@xmath5o . the @xmath0co 21 and co 43 maps presented here are the first large - scale fully sampled low-@xmath7 co maps of lupus , and the first mid-@xmath7 co maps of any kind . the large - scale structure of the lupus clouds can be traced by near - ir extinction @xcite and co 10 emission @xcite . the complex covers some 20 of galactic latitude ( about 50 pc ) . at low latitudes ( @xmath8 ) , a large mass of diffuse gas contains denser filamentary clouds ; at higher latitudes , the dense clouds ( lupus i and ii ) are more clearly separated . the evolution of the lupus clouds may have been driven by the influence of nearby ob stars @xcite . the lupus and ophiuchus molecular clouds face one another across the upper - scorpius subgroup of the scorpius - centaurus ob association ; the upper - centaurus - lupus subgroup lies on the opposite side of the lupus clouds from upper - sco . the shell around upper - sco , blown by stellar winds and a presumed supernova 1.5myr ago @xcite , borders the ne side of the lupus clouds ; on the plane of the sky , the ridge which dominates lupus i ( b228 , see below ) lies just on the trailing edge of the shell . the much older upper - cen - lup subgroup has driven an shell far beyond the lupus clouds @xcite ; it should have passed through the lupus complex some 47 myr ago , roughly consistent with the ages of t tauri stars in lupus iii and iv @xcite . lupus i ( figure [ fig - lui - opt ] ) is dominated by b228 @xcite or gf19 @xcite , a long ridge running nw se , extending over about 2 ( 5 pc ) parallel to the edge of the upper - sco shell . the molecular material appears to fall off steeply toward the center of the shell ( at the ne edge ) , but there is extensive material on the other side of the ridge . lupus iii ( figure [ fig - luiii - opt ] ) is a long ( about @xmath9 pc ) e w cloud ( gf21 , * ? ? ? * ) at the edge of the low - latitude cloud mass : at its eastern end , it curves up to the ne and breaks up into clumps ; toward the west lie two embedded ( but optically visible ) herbig ae / be stars , hr5999 ( a57 ) and the b6 hr6000 @xcite . @xcite consider lupus iii to be a cluster - forming structure with a dense head and spread - out tail , analogous to the @xmath10 oph and cha i clouds . a couple of pc north of lupus iii , there is another smaller cloud , which appears to be a separate condensation in the low - latitude gas ; we henceforth refer to this northern structure as lupus iiin . lupus iv ( figure [ fig - luiv - opt ] ) is the head of a filamentary structure running approximately e w ( gf17 , * ? ? ? * ; * ? ? ? * ) , comprising a small core , about 0.5 pc long ( e w ) , lying within a more diffuse extended cloud , about a pc across . the filament connects lupus iv to the low - latitude gas mass to the east ; @xcite describe it as a `` chain of small faint globules '' . individual dark clouds or cloud peaks have been identified by inspection of optical surveys on scales ranging from degrees ( e.g. , b228 ) to arcminutes @xcite ; these clouds are listed in table [ tab - dcs ] . near - ir extinction maps of lupus iii with a resolution of 30 @xcite have been used to identify dark cores in more detail , which were then fitted with bonnor - ebert profiles . single - pointing spectra in several molecular lines have been observed toward extinction - selected clumps : @xcite observed h@xmath11co along four lines of sight at 6 cm ( 6.6 beamwidth ) ; @xcite searched for ammonia , but did not detect it , toward two globules ( bhr120 & 140 ) ; @xcite observed @xmath0co and c@xmath5o 10 , but their 0.8 beamwidth is poorly - matched to ours ; @xcite observed a few of their lupus i cores @xcite in cs 32 and dco@xmath12 21 ( 0.7 beamwidth ) . previously identified co cores , both those observed by @xcite , and those identified from their c@xmath5o maps by @xcite , are listed in table [ tab - mcs ] . the positions of these extinction cores and c@xmath5o cores with respect to optical and @xmath0co 21 emission are shown in figures [ fig - lui - opt][fig - luiv - opt ] . of the members of the c2d small dark cloud sample mapped by ast / ro in @xmath0co 21 and co 43 @xcite , three lie within the lupus region , but outside our maps . lupus i contains a few ysos , concentrated to the b228 ridge @xcite . lupus iii contains a very dense cluster of ysos near the herbig ae / be stars , composed mainly of t tauri stars @xcite , but including the class 0 object lupus3mms @xcite . @xcite identified a population of cool stars and brown dwarfs toward lupus i and iii , which are not strongly concentrated toward the molecular clouds . the lupus iv region contains about as many ysos as the lupus i region @xcite , but they are older ( class ii / iii ) and largely found outside the molecular cloud , with the exception of one flat - spectrum yso candidate close to the cloud center . even the dispersed population is absent @xcite . lupus iv thus appears to have less ( or no ) ongoing star formation , compared to lupus i and iii . @xcite suggest that the class ii and iii ysos surrounding it represent an earlier generation of star formation ( possibly associated with the passage of the upper - cen - lup shell mentioned above ) , and that the very dense lupus iv core is poised to form new stars . the younger ysos ( class 0 , i , and flat - spectrum sources from the c2d samples , * ? ? ? * ) within the boundaries of our @xmath0co 21 maps are listed in table [ tab - yso ] , and plotted in figures [ fig - lui - opt][fig - luiv - opt ] . lupus i , iii , and iv were observed with the antarctic submillimeter telescope and remote observatory ( ast / ro , * ? ? ? * ) , during the period 2005 march november , in the 461ghz co 43 and 220ghz @xmath0co 21 transitions . the telescope , receiver , and spectrometer systems are described by @xcite : @xmath0co 21 observations used the 230ghz sis receiver , whose local oscillator system has been upgraded to use a synthesizer , frequency multiplied ( @xmath13 ) by a millitech multiplier . co 43 observations used the 450495ghz sis waveguide receiver @xcite . low - resolution acousto - optical spectrometers ( aoss , 1ghz bandwidth , 0.7mhz channel width ) were used for all observations . the large - scale maps presented in this paper were built up from multiple smaller maps ( ` submaps ' ) , mosaicked together as follows : each submap was centered on a point of a large - scale grid superimposed on the molecular cloud , the grid points separated by 24 and 12 for the 21 and 43 observations respectively . the submaps measure @xmath14 ( @xmath15 ) and @xmath16 respectively , giving substantial overlap between neighbouring 21 maps and some overlap ( one row or column of spectra ) between neighbouring 43 maps . the submaps use cell sizes of 1 and 0.5 , significantly oversampling the telescope beam ( 3.3 at 220ghz , 1.7 at 461ghz ) . the spectra are position switched , taken in batches of 4 or 5 , all of which share a reference measurement , about a degree away . the 4 or 5 source spectra are all at the same declination ( and hence at the same elevation , since ast / ro lies at the south geographic pole ) , but the reference position is the same for all spectra in a submap , and may thus differ in elevation from the source spectra by as much as a few arcminutes . the use of one reference position for each submap , however , allows the reference position to be checked for emission . each row of each submap is built up from a number of contiguous blocks of reference - sharing spectra . reference sharing increases the observing efficiency almost to that of on - the - fly mapping , but also shares one of its weaknesses : the sharing of a reference spectrum produces correlated noise in neighboring spectra , which can show up as artifacts in the map . rather than the long stripes characteristic of on - the - fly mapping , correlated noise is manifested in the ast / ro maps as short ( 4- or 5-cell ) horizontal blocks . the ast / ro observing system produces spectra calibrated onto the @xmath17 scale @xcite ; because of the unobstructed off - axis optical design of the telescope , this is essentially equivalent to @xmath18 . further reduction and processing were carried out ( using comb , the standard ast / ro data reduction programmpound / comb ] , and pdl ) . the spectra have velocity resolutions of 0.9kms@xmath1 ( @xmath0co 21 ) and 0.4kms@xmath1 ( co 43 ) ; they were checked for frequency shifts , and corrections were applied ( corrections ranged from 0 to almost 5kms@xmath1 see appendix [ app - data ] ) . linear baselines were subtracted from all spectra , along with polynomial baselines if necessary : a few percent of the 43 spectra required non - linear baseline subtraction , and almost none of the 21 spectra ( table [ tab - spectra ] ) . long - integration spectra were taken toward all reference positions , generally showing upper limits of @xmath19k ( @xmath0co 21 ) and @xmath20 k ( co 43 ) . some 21 reference positions ( for one submap in lupus i and one in lupus iii ) contained substantial emission ( @xmath21k ) , and these reference spectra were added back into all the spectra in the relevant submap . because of the overlap between submaps , there are over a thousand pointings on the sky toward which two or more spectra were measured at different times . these duplicate pointings can be used to estimate the internal consistency of the flux calibration at the telescope . distributions of the ratios of peak @xmath18 between spectra with the same pointings are shown in figure [ fig - peakratio ] for both 43 and 21 data . the typical inconsistencies between these spectra ( estimated by the standard deviation of the distributions of @xmath18 ratios ) are about 30% for co 43 data , and about 20% for @xmath0co 21 . initial map making was carried out by nearest - neighbor sampling the submap data onto one large grid , with overlapping observations co - added . each pixel of these maps corresponds directly to the spectrum observed by the telescope ; without smoothing , inconsistencies between neighbouring submaps show up as clear edges , which are much more obvious in the 43 than in the 21 data . the overlap between submaps allowed these inconsistencies in the 43 data to be checked and corrected . individual submaps with significantly different flux scales to their surroundings were identified and rescaled . these rescalings were checked visually : a submap was only rescaled if it was obviously higher or lower than its neighbours , if the measured scaling was consistent with the visual one , and if the rescaling improved the appearance of the map when it was applied . rescaling corrections of @xmath22% did not significantly improve the appearance of the map , and were not implemented . fewer than half the submaps needed correction , mostly by factors of 0.71.4 ( in line with the estimate of internal consistency given above ) , and the largest correction factor was @xmath23 . no corrections needed to be applied to the 21 data . the need to rescale probably arises from imperfect estimation of the atmospheric transparency in the calibration procedure , which works better when the atmosphere is fairly optically thin ( e.g. at 220 ghz ) than at higher opacities ( e.g. , at 461 ghz ) . the corrected 43 data ( and uncorrected 21 data ) were then gridded ( as above ) to produce final unsmoothed data cubes , which were used to generate the maps in this paper . all the quantities mapped are calculated directly from the data , without profile - fitting . for example , the line width is estimated by the ratio of integrated intensity to peak @xmath18 ( which overestimates the fwhm by about 10% if applied to a perfect gaussian line ) . sample spectra from the data cubes are shown in figures [ fig - spectra - lui][fig - spectra - luiv ] . integrated intensity maps ( figures [ fig - lui - int][fig - luiv - int ] ) , channel maps ( integrated over 2 kms@xmath1 channels , figures [ fig - lui - chan][fig - luiv - chan ] ) , and peak @xmath18 maps ( figures [ fig - lui - tmax][fig - luiv - tmax ] ) were all produced by bessel smoothing ( e.g. * ? ? ? * ) , which gives the ` most fair ' representation of the sky as observed by a single - dish telescope . however , if significant emission extends to the edge of the map , this technique tends to produce a roll - off of emission in the outermost pixels of the map , and this effect is visible in some of the lupus maps . the effective telescope diameter was picked to balance resolution against noise : 1.5 m for the 21 maps , and 1.2 m for the 43 , yielding overall beam sizes of 3.8 and 2.3 , respectively . maps of the velocity centroid ( figures [ fig - lui - vcen][fig - luiv - vcen ] ) and line width ( figures [ fig - lui - lwid][fig - luiv - lwid ] ) use both unsmoothed data ( in grayscale ) and gaussian - smoothed data ( contours ) , with gaussian fwhms of 5 ( 21 ) and 5.75 ( 43 ) yielding overall beams of 6 in both transitions . the nominal pointing accuracy of ast / ro is about 1 , but comparison of the @xmath0co 21 map of lupus iv with the optical extinction structure ( see figure [ fig - luiv - opt ] ) suggests a systematic offset in declination . discrepancies of similar magnitude can be seen in other ast / ro data from 2005 @xcite . the lupus iv maps have therefore all been shifted southwards by 1.5 . the submillimeter map is well correlated with the millimeter - wave one , and so it has been shifted as well . there is no evidence of similar systematic shifts in the lupus i or lupus iii data . the publicly available data cubes _ do not _ have this shift applied to them . @xcite estimated c@xmath5o 10 optical depths in the range 0.10.5 toward extinction peaks , implying that @xmath0co 21 will no longer be optically thin in dense parts of the lupus clouds . maps of @xmath0co 21 peak @xmath18 follow the extinction maps of @xcite better than maps of integrated intensity do , suggesting widespread breakdown of the optically thin approximation . the maximum peak @xmath18 toward lupus i and iv is about 3.5k in each map , implying a minimum excitation temperature , @xmath24 , of 8k ; the peak @xmath18 in lupus iii reaches 5k near the bright stars , so @xmath25k in this area . @xcite adopted 13k for their analysis of @xmath0co 10 toward lupus i. a comparison of the lower @xmath0co contours to the most recent c2d extinction maps suggests that excitation temperatures of 810 k will reproduce the extinction - traced column density quite well . while higher excitation temperatures are not ruled out by the molecular line data , they would lead to lower column density estimates which would be less compatible with estimates from extinction , which is probably the least biased estimator of the true column density @xcite . gas masses are estimated from the @xmath0co 21 data as follows : the ratio of the peak @xmath18 to the maximum possible brightness temperature ( for assumed @xmath24 ) yields an estimate of the peak optical depth @xmath26 ; this is multiplied by the line width to estimate @xmath27 , which is proportional to the column density of @xmath0co ( @xmath28 , appendix [ app - co ] ) . @xmath28 is converted to a total gas column density by assuming conversions between @xmath28 and extinction @xmath29 ( see below ) , and between @xmath29 and the column density of hydrogen ( @xmath30@xmath31 , * ? ? ? assuming a distance of 150pc , the total gas mass enclosed in a 1arcmin@xmath6 pixel is then calculated . the relationship between @xmath0co column density and extinction can be expressed as @xmath32 @xmath33 , with the extinction threshold @xmath34 reflecting the minimum column density required for the presence of @xmath0co . the most commonly used parameters come from a study of @xmath10 oph by ( * ? ? ? * flw ) : @xmath35@xmath31 @xmath36 , @xmath37mag ; other studies ( e.g. * ? ? ? * ; * ? ? ? * ) have found values of @xmath38 in the range 2.22.7 , and extinction thresholds of 01.6 mag . recently , a comprehensive study of the perseus complex @xcite found @xmath39@xmath31 @xmath36 @xmath40mag overall , and @xmath41@xmath31 @xmath36 , @xmath42mag toward the west end of the complex ( pwe ) , comprising several dark clouds , which is likely to be the best analog to the lupus clouds . estimated masses of the lupus clouds above @xmath29 thresholds of 2 and 3 mag are given in table [ tab - mass ] , along with the mean and dispersion of the line widths of material with @xmath43 . ( the line width distribution for @xmath44 is very similar , varying only by about 0.1kms@xmath1 . ) masses have been calculated for @xmath45k and for three @xmath0co - to-@xmath29 relations : pwe , perseus , and flw . the quoted mass is the estimate using @xmath46k and pwe , and the positive and negative errors give the full range of mass estimates . other estimates of mass and line width from the literature are also tabulated . previous isotopically substituted co results relied , directly or indirectly , on the older ratio of @xmath47 to @xmath29 ( @xmath48@xmath31 , * ? ? ? * ) , and have been rescaled to use the same ratio as in this work . the extinction mass of lupus iii has also been rescaled to assume a distance of 150pc instead of the 200pc used by @xcite and @xcite . the masses of lupus i , iii ( including lupus iiin ) and iv are estimated to be 280 , 150 , and 80 @xmath49 respectively ( for @xmath43 ) . the extinction mass of lupus i ( 251 @xmath49 ) is similar to the @xmath0co 21 estimate ; those of lupus iii and iv are significantly greater ( 250 and 120 @xmath49 ) , but still consistent with the @xmath0co 21 error range . this suggests that the lupus complex has spatial variations in the @xmath0co - to-@xmath29 ratio , as perseus does @xcite . by contrast , the @xmath0co 10 mass estimate toward lupus i is 880 @xmath49 @xcite . the @xmath0co 10 map covers a much larger area than ours ( almost 10deg@xmath6 ) and includes significant emission to the se and sw of our map boundaries . within the confines of the @xmath0co21 map , we estimate that the @xmath0co 10 mass should be 600700 @xmath49 , still about 50% more than the upper limit of our mass range . a similar discrepancy was reported by @xcite , whose mass estimate for lupus ii is about a factor of 3 lower than that of @xcite ; but their mass estimate for lupus iv is quite similar to ours . possible reasons for the discrepancies include the undersampling of the nanten maps ( which will tend to overestimate the area of structures ) and differences in the calibration of the two telescopes . @xcite measured only 240@xmath49 toward lupus i in c@xmath5o 10 , which is much closer to our estimate . the additional mass in the @xmath0co 10 map might comprise tenuous gas in which the 21 transition is not fully excited , and whose column density is low enough for it not to be detectable in c@xmath5o 10 . the @xmath0co 10 mass estimate toward lupus iii ( 220 @xmath49 ) is consistent with our data and close to the extinction result , while the c@xmath5o 10 estimates toward lupus iii and iiin ( 80 and 5 @xmath49 ) are much lower . the average line widths for the lupus clouds ( table [ tab - mass ] and * ? ? ? * ) have been deconvolved by subtracting the aos channel width in quadrature , and , at 1.52.7 kms@xmath1 , are broader than the @xmath0co 10 widths reported at emission peaks by @xcite , much broader than the 0.9 kms@xmath1 given by @xcite as the typical line width , and also broader than the c@xmath5o 10 line widths toward individual cores ( table [ tab - mcs ] ) . c@xmath5o 10 is expected to display narrower lines than @xmath0co 21 , since the more optically thin transition arises mainly from the denser gas in the center of the cores , which generally has a lower velocity dispersion . it is harder to see why there should be a discrepancy between @xmath0co 10 and 21 line widths . however , the 10 measurements are taken from two spectra extracted from the map , so the data are insufficient to investigate the disagreement . despite these differences , the various line width measurements follow the same pattern : lupus i and iii have quite similar line widths , lupus iiin has broader lines and lupus iv narrower . based on the line widths of @xmath0co 21 and the velocity gradients observed toward the clouds , it is clear that they are gravitationally unbound ( although their general filamentary shape makes the standard formula for virial mass hard to apply ) . lupus iv , however , may be marginally bound due to its high column density , narrow lines and low velocity gradients : the lower limit to its virial mass is estimated to be @xmath50140@xmath49 , while the upper limit to the mass derived from the @xmath0co 21 map is @xmath50150@xmath49 . @xcite found almost all of the dense cores that they identified in c@xmath5o 10 to be gravitationally unbound ; the greater line widths in @xmath0co 21 reinforce this conclusion . because of their small areal coverage , concentrating on @xmath0co 21 emission peaks , the co 43 maps do not provide the same extensive overview of the clouds . the 43 map of lupus i , however , covers most of the b228 ridge . throughout most of the ridge , the peak @xmath18 lies between 1 and 2k , corresponding to excitation temperatures of 79k ( assuming that the co emission is optically thick throughout , with a filling factor of unity ) . the ends of the ridge may be warmer , with peak @xmath18 of 34k implying excitation temperatures of 1012k . the 43 maps of lupus iii and iiin show the majority of the gas to be similar to the ridge in lupus i , with @xmath51k . within the main lupus iii cloud there are two maxima : one with peak @xmath18 of 4k ( @xmath52k ) , while the other peaks at 9k . this latter one appears very compact , and so its excitation temperature may be significantly higher than the 18k minimum implied by its peak @xmath18 . the 43 map of lupus iv covers only the very center of the cloud , with peak @xmath18 of 34k ( @xmath531012k ) . towards the b228 ridge , maps of extinction @xcite , peak @xmath18 ( figure [ fig - lui - tmax ] ) , and integrated intensity ( figure [ fig - lui - int ] ) are not well correlated . the extinction map is the most robust estimator of column density @xcite : it shows a very strong peak in the middle of the ridge ( dc 338.8 + 16.5 ) , slightly less to the se ( 339.2 + 16.1 ) , and a broad low plateau to the nw ( 338.7 + 17.5 ) . the @xmath0co 21 peak @xmath18 map has similar peaks toward all three dark clouds , while the co 43 peak @xmath18 map shows no significant peak toward 338.8 + 16.5 in the center . the integrated intensity map is very different from the extinction map , with the strongest emission in the nw followed by 339.2 + 16.1 in the se , and 338.8 + 16.5 comparatively weak in the center ( hardly appearing in the co 43 integrated intensity map ) . previous molecular line studies have found a velocity gradient of about 0.3 kms@xmath1pc@xmath1 along the b228 ridge @xcite , but the channel and velocity maps ( figures [ fig - lui - chan ] and [ fig - lui - vcen ] ) show very little evidence of a smooth velocity gradient along the ridge . rather , the two large clumps at the nw and se end of the ridge have velocities that differ by about 3 kms@xmath1 , while the ridge between them lies at an intermediate velocity . on the other hand , a velocity gradient across the center of the ridge shows up very clearly in both 21 and 43 maps : the gradient of about 1 kms@xmath1pc@xmath1 is quite strong , compared not only to previously reported gradients in this cloud , but also to the gradients of 0.10.4 kms@xmath1pc@xmath1 found in areas of taurus @xcite . the gradient is smooth and coherent over the whole of this region an extent of more than 1 pc and appears quite linear , which would be consistent with solid - body rotation around the long axis of the filament @xcite . a coherent pattern in this region may also be seen in the line width maps ( figure [ fig - lui - lwid ] ) , which show an increase in line width toward the leading ( ne ) edge of the ridge . there is no sign of a velocity gradient across the nw half of b228 , which suggests that the nw part of the ridge may be distinct from the center . one of the peaks in the center of the ridge , dc 3388 + 165 - 5 , is associated with iras 15398 - 3359 : this class 0 yso @xcite has a compact ( @xmath501 ) molecular outflow with a dynamical timescale of about 2kyr @xcite . it lies about 0.5 pc behind the steep edge of the ridge , in the e w elongated core that forms the southern half of dc 338.8 + 16.5 , and coincides with local maxima in integrated co 43 emission ( figure [ fig - lui - int ] ) , in peak @xmath18 of both transitions ( fig [ fig - lui - tmax ] ) , and in co 43 line width ( figure [ fig - lui - lwid ] ) . at the se end of the ridge , the ne spur of dc 339.2 + 16.1 is blueshifted from the ridge , while the rest of the clump is at about the same velocity , suggesting that the spur might be a separate cloud superposed on the ridge . @xcite identified four optical extinction peaks in this clump : dc 3392 + 161 - 1 to -4 . they associated peak 1 with iras 154203408/ht lup , a ctts @xcite , which is optically visible as a nebulous patch between peaks 1 , 2 and 3 in fig [ fig - lui - opt ] . the other 3 ` starless ' peaks were observed in cs 32 and dco@xmath12 21 @xcite : all three have @xmath17(cs)@xmath54 k ; dco@xmath12 was only detected ( with @xmath17(dco@xmath12)@xmath55 k ) toward peaks 3 and 4 . dco@xmath12 is a high - density tracer , and its presence toward peaks 3 and 4 along the ridgeline , and absence from peak 2 at the start of the ne extension , suggest that , even in this complex structure , dense gas is concentrated to the ridge . the co 43 data , with better spatial and velocity resolution , show additional structure in this clump : a patch of strong emission close to dc 3392 + 161 - 4 , about 10 across , is blueshifted by about 1 kms@xmath1 with respect to the main ridge structure , leaving a cavity in the emission at ambient velocity . molecular outflow is a possible explanation , but since there is no obvious red lobe , this would require the yso to lie behind the cloud . blue asymmetries in molecular lines can be caused by infall , but @xcite searched unsuccessfully for signs of infall toward dc 3392 + 161 - 4 . the parsec - scale ring - shaped structure to the sw of b228 is not covered by the co 43 map however , its @xmath0co 21 emission looks quite similar to that seen in the bulk of the ridge , and the detection of dco@xmath12 toward the dense core embedded in the se edge of the ring suggests a significant amount of dense gas . the structure may also have a strong velocity gradient : the channel maps ( figure [ fig - lui - chan ] ) show two complementary semicircular structures in adjacent channels , i.e. separated by about 2kms@xmath1 , equivalent to a velocity gradient of at least 1 kms@xmath1pc@xmath1 . in the line width map ( figure [ fig - lui - lwid ] ) , the redder nw side of the structure has somewhat broader lines . @xcite observed an extinction core at the se edge of the ring : they found no signs of infall , but both cs 32 and dco@xmath12 21 are stronger in this core than in any of the cores observed in b228 . @xcite found a ring - shaped structure in extinction in lupus iii , with a diameter of about 5 , which they interpret as the remnants of the molecular cloud that formed the nearby cluster of young stars . by contrast , this structure is larger ( 20 ) and is not associated with a known cluster . co emission toward lupus iii in both transitions is dominated by a compact source in the west of the cloud , close to core f identified by @xcite . proceeding from e to w are : core f ; the compact co source ( slightly blueshifted from the bulk of the cloud ) ; the hae / be stars hr5999 and 6000 ; and further @xmath0co 21 emission extending e w . this latter emission shows no particular sign of the ring structure found in extinction by @xcite , probably because the central void is smaller than the 220ghz beam . there is a significant velocity gradient across this area , of about 2 kms@xmath1pc@xmath1 , getting redder to the west , which is not simply due to the blueshifted emission of the compact source to the east . to the east of the compact source , extinction cores a e @xcite blend together at this resolution : the integrated intensity map ( fig [ fig - luiii - int ] ) shows two e w spurs off the emission peak , the northern containing cores c and d , and the southern containing e. further east still , the northern spur continues to core b , while a lies just ne of dc 3397 + 92 - 3 . the strongest h@xmath0co@xmath12 emission in this area ( @xmath21 kkms@xmath1 ) is found toward core e ; core d , with about half the h@xmath0co@xmath12 emission , contains the class 0 yso lupus 3 mms @xcite . lupus iiin , the small ( @xmath56 pc ) cloud core lying a few pc n of lupus iii , shows surprisingly complex structure : @xmath0co 21 channel and velocity maps ( figures [ fig - luiii - chan ] and [ fig - luiii - vcen ] ) show the core to be lying in a larger ( @xmath57 pc ) structure , elongated se nw , with a significant velocity change over the long axis . the velocity gradient is dominated by a sudden change just se of the cloud core , also seen as a sharp edge in the reddest 21 channel map . the e w elongated core of lupus iv has 3 peaks along its length : the central and western ones correspond to the extinction peak which reaches @xmath58 mag . the peaks are most obvious in the @xmath0co 21 channel maps ( figure [ fig - luiv - chan ] ) ; the peak @xmath18 map ( figure [ fig - luiv - tmax ] ) shows them with much lower contrast than the extinction maps as the @xmath0co transition is optically thick ( @xmath58 mag corresponds to @xmath59810 ) . @xcite found the @xmath0co 10 line to be optically thick as well , saturating with respect to 100 ir emission . the eastern peak has broader lines yielding similar integrated intensity ( figure [ fig - luiv - int ] ) ; at finer velocity resolution , these lines are found to be double peaked @xcite . although there is quite strong co 43 emission throughout the core , and the velocity structure of the 43 emission is similar to that of @xmath0co 21 , the integrated 43 emission is concentrated toward the e and w condensations only . two elongated structures running ne of the core are also visible in the co 43 peak @xmath18 map . the strong velocity gradients seen in lupus i and iii are absent from lupus iv . some velocity structure is evident : the central condensation in the core and the two elongated ne structures are blueshifted ( by 12 kms@xmath1 ) compared to the e and w condensations . @xcite reported significant velocity structure in the e w direction , but our velocity maps show that this e w structure is not found throughout the cloud . co 43 emission from the lupus clouds can be compared to that of @xmath0co 21 by gaussian smoothing the co 43 map and plotting the integrated intensities ( figure [ fig-4321-int ] ) and peak @xmath18 ( figure [ fig-4321-peak ] ) against one another , point - by - point . the peak @xmath18 comparison is noisier than that of integrated intensity , and is only shown for lupus i and iii , which have stronger emission . the large 43 map of lupus i provides an overview of the bulk of the molecular gas . clusters of points close to the origins of the integrated intensity and peak @xmath18 plots reflect the noise levels in the maps . the majority of the data are somewhat correlated , occupying regions of the plots bounded by ( 43)/(21)@xmath60 ( both plots ) , @xmath61kkms@xmath1(integrated intensity ) , and @xmath62k ( peak ) . the 43 emission is effectively saturated , while the @xmath0co 21 emission takes a wide range of values . this suggests that even regions of quite low column density contain sufficient dense gas to emit optically - thick co 43 radiation ( the critical density of co 43 is of order @xmath2@xmath63 , but see sec . [ sec - disc - bulk ] ) . another population of molecular gas can be identified toward lupus i : its @xmath0co 21 peak @xmath18 is similar to that of the bulk population , while its co 43 peak @xmath18 is greater ( @xmath64k ) . this suggests that the gas is warmer , but not denser than the bulk of the cloud . the integrated intensities of the transitions , however , are well correlated ; this probably arises from increased line widths in both transitions , and thus from increased velocity dispersion in the gas . inspection of the maps of integrated intensity , peak @xmath18 and line width toward lupus i ( figures [ fig - lui - int ] , [ fig - lui - tmax ] , [ fig - lui - lwid ] ) shows that this component , with enhanced integrated intensity , peak co 43 @xmath18 and line width , is found in the nw end of b228 . the much smaller co 43 maps toward lupus iii , iiin , and iv provide fewer data . however , two components can still be identified in lupus iii : one component consistent with the lupus i bulk population , and the other with stronger emission . in contrast to its counterpart in lupus i , both integrated intensity and peak @xmath18 of the brighter component are well correlated . this brighter component constitutes the compact peak close to the hae / be stars in lupus iii . heating by the nearby stars could increase the peak @xmath18 and hence the integrated intensities of both transitions , without a significant change in the line width . the molecular gas in lupus iiin and iv occupies a similar region of the integrated intensity plot to the bulk population of lupus i , suggesting broadly similar physical conditions . some positions show significant co 43 emission without @xmath0co 21 : the spectra toward one of these positions are shown in figure [ fig - spectra - lui ] , and the integrated intensity and peak @xmath18plots show that this occurs in all the clouds . the combination of strong co 43 emission and weak or absent @xmath0co 21 implies warm gas with low column density ; since the co 43 map areas were chosen to cover @xmath0co 21 peaks , there may be more of this population outside our 43 maps . spectra toward specific cores have been measured by nanten and the swedish - eso submillimeter telescope ( sest ) : @xcite list the c@xmath5o 10 line parameters for the spectrum at the peak of each dense core identified from their map , and @xcite list parameters for both @xmath0co and c@xmath5o transitions toward extinction - selected cores . because the ast / ro maps are fully sampled , it is possible to obtain equivalent spectra ( albeit with a beamwidth of 3.3 , compared to the 2.8 nanten beam , and sest s 0.8 beam ) and compare the peak @xmath18 ( figure [ fig - cotrans - peak ] ) and integrated intensities ( figure [ fig - cotrans - int ] ) . the peak @xmath18 in co 43 and @xmath0co 21 toward most of the cores lie in the same region of the plot as the bulk component of molecular gas described above . four cores are clearly brighter than the rest : the two identified in lupus iii are both associated with the bright compact source , the one in lupus i lies at the se end of b228 , and the one in lupus iv is in the middle of the central core . the majority of the extinction cores have @xmath0co 21/10 peak @xmath18 ratios in the range 0.30.7 , inconsistent with the 0.72 expected from lte ( appendix [ app - co ] ) . this could be caused by beam - dilution compared to the sub - arcminute sest beam ( but see below ) . the exception ( one of the lupus iii cores associated with the bright compact source ) has a @xmath18 ratio of almost 3 : it may be large enough not to suffer beam dilution , and warm enough to be close to the high - temperature line ratio limit . while the lupus i and iii cores are well mixed in the plot , the lupus iv cores have rather low line ratios . there is no particular reason for them to be more beam diluted , so the lower line ratio may reflect a lower temperature . the plot of @xmath0co 21 peak @xmath18 against that of c@xmath5o 10 does not show any clear separation between the extinction and c@xmath5o core samples . this is surprising , since the c@xmath5o cores observed by nanten should suffer almost as much beam dilution as the ast / ro data , compared to the extinction - selected cores observed by sest . the line ratios of the cores lie mainly between unity and @xmath65 ; correcting for the factor - of-2 @xmath18 discrepancy estimated above , this suggests line ratios of about 26 , in line with lte estimates for c@xmath5o optical depths of 0.10.5 @xcite . the co 43 and @xmath0co 21 integrated intensities of the cores are more scattered , but the high-@xmath18 cores in lupus iii also have the highest integrated intensities . in contrast , the high-@xmath18 cores in lupus i and iv have more average integrated intensity , suggesting that the increased @xmath18 ( and hence temperature and/or density ) is not accompanied by increased line width . the two more cores in lupus i with large @xmath0co 21 integrated intensity also lie in the se of b228 . the ratios of the two @xmath0co transitions ( 21/10 , for the extinction - selected cores only ) lie between about 0.7 and 1.5 ( with a few between 1.5 and 3 ) , which is consistent with our expectations from lte , but not with the peak @xmath18ratios above . line parameters were estimated from the sest data by fitting gaussian line profiles @xcite , in contrast to the ast / ro and nanten data , for which integrated intensity and peak @xmath18 were measured directly , and the line width estimated by the ratio . this difference in analysis could cause a systematic discrepancy , with the sest analysis yielding lower @xmath18 and broader line widths . the cores with higher ratios ( 1.53 ) are all in lupus i or iii , but are scattered throughout the clouds . ratios of @xmath0co 21 to c@xmath5o 10 integrated intensity are around 36 for the c@xmath5o - selected cores observed with nanten ( consistent with the peak @xmath18 ratios above ) , but the same ratios toward extinction cores are generally higher . these high ratios generally arise from rather low c@xmath5o 10 integrated intensities ( @xmath66 kkms@xmath1 ) . the yso population of lupus is dominated by class ii and iii objects @xcite , but younger objects are also found there . the class 0/i / f sources identified by @xcite lying within the @xmath0co 21 maps are listed in table [ tab - yso ] , together with the 21 line parameters toward them and , where applicable , co 43 parameters ; two objects that have been identified as background galaxies @xcite are excluded . spectra toward the ysos are shown in figure [ fig - spectra - yso ] : all 21 spectra are pointed within 0.6 of the yso position , and all 43 spectra are within 0.25 . spectra toward some ysos do not show any significant emission : values of @xmath18 below @xmath67k ( in either transition ) should be treated as noise . of the 4 ysos in lupus iii with very low or non - existent 21 emission , @xcite estimate 3 to have ages of order 100myr . the rest of the ysos have line widths similar to those of the surrounding molecular gas . this does not rule out the presence of line wings due to outflow : the class 0 source lupus 3 mms has quite average line widths in both transitions , but the spectra themselves clearly show wings . the flat - spectrum source j154506.3341738 ( in lupus i ) has broader - than - average lines ; nebulosity prevented it from being measured in the optical by @xcite . the estimates of column density , and hence mass , derived above assumed a blanket temperature of @xmath68k rather than the 13k assumed by @xcite for lupus i and the 1217k estimated toward lupus iv by @xcite , both based on optically thick co 10 emission . 8k is the minimum excitation temperature consistent with the peak @xmath0co 21 @xmath18 seen toward the majority of the gas in lupus i , iii , and iv , and with the peak co 43 @xmath18 over most of the b228 ridge . adopting excitation temperatures close to the minimum implies assumptions of optical thickness , lte , and a filling factor close to unity , but is required for consistency with the extinction maps @xcite : the @xmath69 and @xmath18(@xmath0co 21)@xmath70k contours are quite similar . this implies @xmath71k for both pwe and flw @xmath0co - to-@xmath29 relations , although it does allow a higher @xmath24 toward lupus i for the perseus @xmath0co - to-@xmath29 parameters . the gas temperature estimate is consistent with estimates for many dark clouds . @xcite , using @xmath0co and c@xmath5o 10 transitions , estimated the excitation temperatures of their sample of dense cores in lupus to lie in the range 715k , with the majority of good estimates being colder than 10k ; @xcite estimated gas temperatures toward a large sample of small dark clouds from co 21 , and found that the majority were colder than 10k . models of cold dark clouds yield dust temperatures @xmath7210k at the center @xcite . while the dust temperature is higher toward the cloud surface , gas temperatures are lower than dust temperatures for low column and volume densities ( e.g. , * ? ? ? * ) , so the @xmath0co 21 and co 43 transitions need not be dominated by the cloud centers . the co 10 estimates mentioned above are likely to be dominated by the cloud surfaces , which may be warmer than those of the majority of dark clouds due to external heating by the nearby ob association . the peak co 43 @xmath18 toward lupus iv implies a @xmath24 of 1012k , compared to @xmath73k from @xmath0co 21 . this may reflect a combination of external heating , as suggested by @xcite , with high enough density to couple the gas and dust temperatures more effectively than in lupus i or iii . the column density of h@xmath11 throughout most of the b228 ridge is within a factor of 2 of @xmath74 @xmath31@xcite . the width of the ridge on the sky is about 10 , or 0.4 pc ; if b228 is assumed to be a filamentary cloud , with a similar depth , the average volume density @xmath75 is a few @xmath76 @xmath63 . this is close to the critical density @xmath77 of the 21 transitions of co and its isotopologues , supporting the assumption of lte used throughout this work , but an order of magnitude lower than the critical density of co 43 ( a few @xmath2@xmath63 , appendix [ app - co ] ) . @xcite found that significant emission in many transitions could arise from gas with volume density more than an order of magnitude lower than @xmath77 ( although co was not included in that study ) , but the similarity between the excitation temperatures of @xmath0co 21 and co 43 suggests thermalized emission , and thus a volume density close to @xmath77 . the line emission is pervasive , which argues against its arising from cores or a high - density center of the ridge . however , if the bulk of the ridge material were taken to be close to the critical density of co 43 , the depth of the ridge along the line of sight would be an order of magnitude lower than its width in the plane of the sky , which seems implausible . it is more likely that the co 43 emission arises from a small fraction of dense gas found in clumps throughout the molecular cloud ; or from a thin warm shell around the outside of the cloud ; or that subthermal emission from the bulk of the molecular gas can account for the co 43 lines seen throughout the cloud . the latter possibility , in particular , may be checked by modelling the emission . @xcite estimated the optical depth of c@xmath5o 10 toward their dark cloud sample to lie in the range 0.10.5 by comparing the @xmath0co and c@xmath5o lines , and assuming an abundance ratio of 5.5 . a large fraction of their cores had line ratios in excess of 5.5 , inconsistent with their assumed isotopologue ratio ; however , ratios up to 8 are possible ( appendix [ app - co ] ) . @xcite estimated somewhat lower optical depths toward their c@xmath5o cores , assuming @xmath78k ; at @xmath46k , their estimated optical depth range becomes quite consistent with that of @xcite . the @xmath0co 10 optical depth toward the cloud cores is likely to range from 0.5 up to 24 , depending on the abundance ratio . @xcite estimated volume densities of a few @xmath2 @xmath63 for the dense cores they identified in lupus iii . if such densities are widespread among the dense cores in lupus , the co 43 transition is close to lte , and it becomes possible to estimate the expected value of @xmath79 toward the cores . at @xmath46k , the high - column - density limit of the ratio ( both transitions optically thick ) is @xmath500.4 ; as the optical depth of @xmath0co becomes moderate , the ratio increases toward unity . however , at low optical depths , co 43 may no longer be thermalized , making this estimate invalid . most of the cloud cores have 43/21 ratios clustered around 0.5 ( figure [ fig - cotrans - peak ] ) , with a few more at unity or above , and one over 1.5 . it is difficult to achieve a line ratio significantly above unity under lte : even at low optical depth , the temperature would need to be at least 20k . the central condensation in the b228 ridge ( 338.8 + 16.5 ) has high extinction , moderate peak @xmath18 and integrated @xmath18 in @xmath0co 21 , and co 43 emission quite similar to the bulk of the cloud . the line width is also quite similar to that of the bulk cloud , so the enhancement in @xmath0co 21 is largely due to the increased peak @xmath18 . elevated gas temperature would likely show up in the co 43 emission , so the increased @xmath18 is largely due to the increased column density as the @xmath0co 21 transition becomes optically thick . this dense cloud seems to have more in common with the bulk material around it than it does with the emission peak to the nw , having similar temperature and line width . it is also associated with two embedded ysos , including a known outflow source . the strong , coherent velocity gradient seen north and east of 338.8 + 16.5 runs perpendicular to the leading edge of the ridge ( i.e. , in the direction of the shell s expansion ) ; the leading edge ( i.e. , toward the center of the shell ) has a greater line width than the gas behind it . it is difficult to rule out temperature and optical - depth effects combining to mimic a velocity gradient , but similar patterns are seen in both @xmath0co 21 and co 43 maps : a truly optically thin tracer is required to confirm the gradient . if the apparent gradient truly reflects the kinematics of the gas , the total change in velocity across the ridge is about 1kms@xmath1 , or half the line width . however , the change in velocity across a jeans length ( around 0.1 pc for these clouds ) is only about 0.1kms@xmath1 , which is unlikely to add significant support against collapse . indeed , 338.8 + 16.5 seems to be part of this velocity field , and contains a known class 0 yso . the integrated intensity maximum at the nw end of b228 is due to the combination of enhanced peak @xmath18 and broader lines . throughout this area , the peak co 43 @xmath18 is @xmath80k , implying a gas temperature of at least 10k . although the 43 emission probably does not sample the whole of the gas column , there is supporting evidence for a warmer temperature : the _ spitzer _ ir maps @xcite show this end of the filament to be bluer than the central part , with strong 24 emission ( see also * ? ? ? * ) , which could be due to a higher dust temperature . no class 0/i / f ysos are found in the northern part of the ridge , and only one class iii @xcite , so there are no obvious internal heating sources . the nearby upper - sco ob subgroup lies to the ne of lupus i , but is unlikely to be causing the heating , since no similar effect is to be found in the center of the ridge . at the se end , the picture is complex : the extinction and peak @xmath18 for both transitions are high at the se end of 339.2 + 16.1 , while line width and integrated intensity peak further nw . the strong enhancement of peak co 43 @xmath18 at the se end suggests increased temperature as well as column density . the associated ysos all lie to the nw in this area , so internal heating seems unlikely . the emission from the upper - sco shell lies at a similar velocity ( 39kms@xmath1 , * ? ? ? * ) to the lupus clouds , consistent with their being in contact with one another , and lupus i being dynamically affected by the shell . if such an interaction is going on , the combination of continuity and conservation of momentum require that the sum of pressure and momentum flux is conserved by the interaction . the limitations of both molecular and atomic data make it impossible to look for clear diagnostics of interaction between the shell and ridge , but some rough estimates can be made . for both phases , @xmath81 where @xmath82 for , and 340 for h@xmath11 . based on the data @xcite , this quantity may be estimated to be of order @xmath83@xmath63k , about 90% of which is the momentum flux component due to the expansion velocity of the shell ( 10kms@xmath1 , * ? ? ? @xmath84 for the molecular gas in lupus i is also around @xmath83@xmath63k , but is approximately evenly divided between pressure and momentum flux ( due to velocity gradients ) . the rough similarity in this sum between the and molecular gas is consistent with interaction between them . if the shell is indeed affecting lupus i , it is likely doing so through the momentum flux of its expansion . this transfer of momentum could be causing the velocity gradient across the b228 ridge ( in the direction of the shell s expansion ) ; this process could be analogous to the ` streamers ' in the ophiuchus complex @xcite . the enhanced line width at the nw end of b228 and along the leading edge of the ridge ( implying increased pressure in the molecular gas ) is also consistent with the effect of the shell . more detailed studies of both and molecular gas may support the idea of lupus i being affected by the upper - sco shell : better estimates of the volume density of h@xmath11 and higher - resolution maps of are required . the brightest part of lupus iii near hr5999 and 6000 was mapped by @xcite in the millimeter - wave continuum , c@xmath5o 10 and h@xmath0co@xmath12 10 , all with higher resolution than the ast / ro data . the c@xmath5o emission peak coincides with the co peak in our maps , and extends to the east , with another e w elongation a few arcminutes to the north . the continuum map , together with a near - ir extinction map @xcite , shows the same southern e w structure extending further east , while the northern structure breaks up into two clumps , the denser western one containing a class 0 protostar ( lupus 3 mms ) . in the h@xmath0co@xmath12 map , there is no emission at the co peak ; the emission peaks strongly to the east in the southern e w structure , while the clumps in the northern structure show up as smaller peaks . @xcite suggest that the lack of h@xmath0co@xmath12 emission at the co peak can be accounted for by a long path length through gas with volume density significantly lower than the critical density ( @xmath85@xmath63 for h@xmath0co@xmath12 ) . they estimate a column density of @xmath86@xmath31 , equivalent to @xmath29 of about 13 mag , which is consistent with the extinction maps @xcite . the peak @xmath18 in co 43 ( about 9k ) implies @xmath87 k , while the 6k peak @xmath18 of @xmath0co 21 only requires @xmath88 k. compared to the rest of the lupus clouds , where the co 43 and @xmath0co 21 excitation temperatures are quite similar , this is a significant discrepancy . it could be explained by the co peak being compact , as suggested by the co 43 peak @xmath18 map , so that the @xmath0co 21 measurement is beam - diluted . alternatively , the co 43 could be tracing an outer shell heated by the hae / be stars . however , the @xmath0co 21 emission is also optically thick ( for @xmath89k , @xmath29 of 13 mag implies an optical depth of about 8) , and so will trace similar material . these temperature estimates tend to support the argument that the gas is too warm for depletion of h@xmath0co@xmath12 to explain the lack of emission toward the co peak @xcite , although the estimates are unlikely to apply to the center of the clump . the strong co 43 emission implies that the transition is thermalized , so a significant amount of the gas in the clump must have volume density @xmath90@xmath63 . while @xcite derived an average volume density of @xmath2@xmath63 , there must be enough significantly denser gas to thermalize the co 43 line , but not enough to excite the h@xmath0co@xmath12 10 transition . in projection , lupus iii lies far away from any part of sco - cen @xcite , in contrast to lupus i and lupus iv ( discussed below ) , so the hae / be stars probably influence it far more than the ob associations . however , there is evidence for lupus iii being further away than lupus i and iv , possibly as much as 50 pc ( e.g. * ? ? ? * ) . if this is the case , then lupus iii could lie behind sco - cen , and could be affected by either or both of upper - sco and upper - cen - lup . the large line widths seen toward lupus iiin could be caused by external influence in the same way as the broader lines seen toward lupus i. the peak @xmath18 in co 43 toward lupus iv of about 4k , implying @xmath91 k , occurs around the outside of the extinction peaks , which have peak @xmath18 around 3k ( @xmath25k ) . this suggests that the outside of the clump is significantly externally heated . the @xmath0co 21 peak @xmath18 implies @xmath92k , but even this measure is unlikely to sample the extinction maxima properly , since it will be dominated by the outer layers of the structure . the co 43 temperature estimates are in line with those seen at the nw end of b228 , where lupus i seems to be strongly affected by upper - sco . lupus iv is on the opposite side of the lupus complex to lupus i , upper - sco and its shell @xcite , but faces the upper - cen - lup subgroup , which lies to the w and sw . @xcite suggested that lupus iv was shaped by the influences of both subgroups , and noted that some of the velocity gradients they saw in lupus iv were along the vector toward upper - cen - lup . much of the enhancement in the peak co 43 @xmath18 lies on the s and w sides of the extinction peaks , which is consistent with a picture of external heating by the radiation field from upper - cen - lup . there is , however , no nearby shell , the upper - cen - lup shell having passed the lupus clouds long ago @xcite . if interaction with the shell causes the enhanced line width at the nw end of b228 , the lack of any such interaction in lupus iv would be consistent with its rather low line widths . fully sampled degree - scale maps of the @xmath0co 21 emission toward the lupus i , iii , and iv clouds trace the column density and temperature of the gas , the transition becoming optically thick in the cloud cores . the peak @xmath18 is well correlated with the near - ir extinction @xcite , and a comparison of the two suggests that the bulk of the molecular gas in lupus has a temperature of 810k , rather than the 1017k generally adopted elsewhere ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? this estimate is fairly robust to changes in the relationship between @xmath0co column density and @xmath29 . estimates of the cloud masses from the @xmath0co maps are reasonably consistent with those derived from extinction mapping . the differences between these estimates vary greatly from cloud to cloud , and suggest that there may be significant spatial variation in the @xmath0co - to-@xmath29 relationship , as found in the perseus complex @xcite . the line widths of @xmath0co 21 toward the clouds are higher than previous estimates @xcite , around 2kms@xmath1 , with lupus iiin rather broad and lupus iv rather narrow . fully sampled co 43 maps covering most of lupus i and small regions of lupus iii and iv trace dense gas : the peak @xmath18 generally indicates excitation temperatures quite close to those of @xmath0co 21 , and hence that the transition is largely thermalized . this suggests that the volume density @xmath93@xmath63 , although modelling will be required to ascertain the required density . co 43 emission is pervasive toward lupus i ( the map of which covers a large area ) , implying that this dense gas is found either throughout or all around the outside of the cloud , although it may comprise a fairly small fraction of the cloud mass . the physical conditions of the molecular gas vary along the b228 ridge in lupus i. at the nw end , the gas has broader lines and probably higher temperature than in the bulk of the cloud ; the column density is not particularly high and there is only one class iii yso . in the center of the ridge , the dark cloud 338.8 + 16.5 is associated with recent star formation @xcite ; in this area a coherent velocity gradient of about 1kms@xmath1pc@xmath1 runs across the ridge . the se end of the ridge is complex , with ysos , enhanced line width and integrated intensity on the nw side , and column density ( and possibly temperature ) peaking to the se . the enhanced line widths and velocity gradient in b228 are consistent with a dynamical interaction between lupus i and the shell around the upper - sco subgroup of sco - cen @xcite . to the north of lupus iii , the small cloud lupus iiin has similar characteristics to the bulk of the other clouds , albeit with broader lines and significant velocity structure . lupus iii itself contains a compact co peak which is probably heated by the nearby hae / be stars hr5999 and 6000 . the gas in this clump contains sufficiently dense gas to thermalize the co 43 transition ( @xmath94@xmath63 ) , but not to thermalize the h@xmath0co@xmath12 transition mapped by @xcite , which would require @xmath95@xmath63 . the rest of lupus iii seems to have quite similar physical conditions to those in the rest of the clouds , and shows no particular sign of being affected by the nearby ob subgroups . lupus iv contains peaks of very high column density @xcite associated with slightly warmer gas temperatures ( 1012k ) . these temperatures are estimated from the optically - thick co 43 transition , which is strongest around the extinction cores , suggesting significant external heating . the average line width of @xmath0co 21 , however , is significantly lower than those of lupus i and iii . lupus iv faces the upper - cen - lup subgroup of sco - cen and @xcite suggested that it is influenced by the ob stars ; the upper - cen - lup shell passed by the lupus clouds a few myr ago , so lupus iv is more likely to be affected by the radiation field from the ob stars . despite the basic similarities in their physical conditions , the three clouds have significant differences : lupus i appears to be strongly affected by external thermal and dynamical influences from the nearby upper - sco ob association , and does not display widespread star formation . lupus iii shows no sign of external influence parts of the cloud are heated internally by its own young stars . lupus iiin seems entirely quiescent , yet has a large average line width . lupus iv has the greatest column density and the narrowest average line width , has almost no star formation as yet , and may be heated externally by the upper - cen - lup ob association . a detailed spatial comparison of co and extinction maps will yield more accurate estimates of the physical conditions of the lupus clouds , as well as mapping the variation in the @xmath0co - to-@xmath29 ratio . mapping of additional co transitions is crucial to the understanding of the clouds : more optically - thin lines ( c@xmath5o and even @xmath96 ) are particularly important . maps of the upper - sco shell with comparable resolution to the lupus maps are required to look for more definitive signs of interaction between the shell and the molecular clouds . more sophisticated radiative transfer calculations are beyond the scope of this work , but are required to use the co 43 emission as a proper constraint on the physical conditions of the gas , and hence to understand the structure of the lupus clouds and how they are affected by the local environment . we thank neal evans , fernando comern , bruno mern , eric mamajek , and tracy huard for valuable discussions , and the many people who helped get ast / ro ready for the 2005 observing season , particularly jacob kooi and craig kulesa . christina hammock kept the liquid helium flowing through the winter . we thank her , and all the south pole station staff , for their work . we thank the anonymous referee , whose comments have improved this work . ast / ro was supported by the national science foundation , under nsf opp ant-0441756 . nfht was also supported by the university of exeter dvc ( resources ) discretionary fund and by the european commission ( grant mirg - ct-2006 - 044961 ) . this work has made use of : nasa s astrophysics data system ; the simbad database operated at cds , strasbourg ; and the _ skyview _ facility at nasa s goddard space flight center . _ facilities : _ . because the aos backends are analog devices , laser mode hopping causes shifts in their frequency calibration . a number of such shifts occurred over the nine - month period during which these data were taken . the fundamental frequency calibration for each spectrometer was obtained once , by connecting a frequency synthesizer to the if input of the aos to obtain the aos channel width and the channel number of a fiducial frequency . the frequency scale thus defined was used for all data in this paper , but some corrections had to be made . the change in channel width caused by a mode - hop is negligible , but the entire spectrum is offset by a few channels . the frequency shifts were tracked with a number of fiducials . each aos includes a comb generator which is usually used to obtain several frequency calibration scans per hour , giving excellent frequency tracking . however , the comb generator failed during 2005 , so other frequency standards had to be used . for the co 43 observations , the mesospheric co 43 line is so strong that it can be picked up without a switched measurement . so the ` sky ' spectra , used to estimate the sky temperature for calibration , show the line at an antenna velocity close to zero . the mesospheric co abundance has a strong seasonal variation , and became so weak at the end of the austral winter ( around september ) that it could no longer be seen in the sky spectra . finally , repeated spectra were taken toward the compact region ngc 3576 . this source has velocity structure on the scale of the ast / ro beam , so pointing uncertainties translate into velocity uncertainties . these three fiducials were combined to track the frequency scale . the majority of @xmath0co 21 spectra were corrected by 13kms@xmath1 , and some were corrected by up to 5kms@xmath1 . about 90% of the co 43 spectra were corrected by @xmath97kms@xmath1 , with the remainder corrected by 3.9kms@xmath1 . frequency shifts also show up as inconsistencies in the maps . the 43 observations showed no obvious inconsistencies , but some shifts had to be applied to the @xmath0co 21 data : channel maps of lupus i showed that data taken in 2005 november had a significant uncorrected frequency shift compared to earlier , better - calibrated data . in addition , the long - integration spectrum toward one of the reference positions with significant 21 emission was shifted by about half a channel with respect to the map spectra into which it was added . this was not corrected , because the facility for combining spectra in comb only handles integer - channel shifts , and the effect of the shift is negligible , even in the channel maps . the co and @xmath0co data in this work are analyzed under the assumption of lte ( i.e. , the excitation temperature , @xmath24 , and gas kinetic temperature , @xmath98 , are the same ) . for the two transitions considered in this work , equation ( 14.46 ) of @xcite yields @xmath99 for co 43 ( neglecting the cosmic microwave background term ) , and @xmath100 for @xmath0co 21 . these equations are correct if the transitions are optically thick , and completely fill the beam ; otherwise they underestimate @xmath24 . in lupus , the gas is cold enough ( @xmath101 ) that @xmath18 is significantly different from @xmath24 . the column density in the lower level of the @xmath0co 21 transition is given by @xmath102 and this can be converted to the total column density of @xmath0co by correcting for the partition function : @xmath103 ranges from 2.1 to 2.3 for @xmath24 of 710k . in lte , the ratio of @xmath0co 21 to 10 emission depends only on temperature . if both transitions are optically thick , the ratio is simply the ratio of brightness temperatures at different frequencies for a given @xmath24 , and will range from 0.7 ( at 7k ) to unity ( at high temperature ) . in the optically thin case , this ratio is multiplied by the ratio of optical depths , which also depends on @xmath24 via the partition function : this ratio ranges from 1 to 2 between 6k and 15k , with a high - temperature limit of 3 . the 21/10 line ratio should therefore range from 0.7 to about 2 in lupus . the ratio of @xmath0co 21 optical depth to c@xmath5o 10 optical depth is just the ratio of @xmath0co 21/10 optical depths , multiplied by the @xmath104 abundance ratio : the @xmath105 isotope ratio is @xmath106 in the local ism @xcite , but the double ratio @xmath104 is not as well known . combining the solar @xmath107 ratio of 500 @xcite with the local ism @xmath105 ratio yields @xmath108 , but @xcite point out that @xmath105 and @xmath107 should track one another , being similarly dependent on star formation history , and so a solar ratio ( @xmath109 , * ? ? ? * ) may be more appropriate . thus the ratio of optical depths should fall in the range of 4 to about 16 . the ratio of @xmath0co 21 emission to c@xmath5o 10 , however , is complicated by the fact that the c@xmath5o transition is likely to be fairly optically thin , while the @xmath0co transition will have moderate to high optical depth . this will tend to reduce the ratio : conditions in lupus are likely to yield emission ratios as low as 24 . below a critical volume density @xmath77 , lte fails ( @xmath110 ) . the critical density itself depends on physical conditions : the effective spontaneous emission rate @xmath111 is reduced at high optical depth , yielding a lower @xmath77 , and collisional transition rates are temperature - dependent . in the optically thin limit , @xmath77(co43 ) varies from @xmath112 @xmath63 at 40 k to @xmath113 @xmath63 at 10 k @xcite ; the critical densities of co 10 and 21 transitions are a few hundred and a few thousand @xmath63 @xcite . llllccccl 337.9 + 16.4 & 3379 + 164 & lu1 & & 15 39 37 & 34 46 30 & lm & y & fs342 + 337.6 + 16.4 & & & & 15 38 21 & 34 59 17 & hmstg & & fs341 + 338.2 + 16.4 & 3382 + 164 & lu4 & & 15 40 35 & 34 40 19 & lm & n & + 338.8 + 16.5 & 3388 + 165 - 2 & lu6 & & 15 42 19 & 33 50 59 & lm & n & b228 + & 3388 + 165 - 3 & lu8 & & 15 42 40 & 33 52 01 & lm & n & b228 + & 3388 + 165 - 4 & lu7 & sl 12 & 15 42 43 & 34 09 15 & lm & y & b228 + & 3388 + 165 - 5 & b228 & & 15 43 01 & 34 08 48 & lm & y & b228 + & 3388 + 165 - 6 & lu9 & & 15 43 18 & 34 13 30 & lm & n & b228 + 338.7 + 17.5 & & & & 15 39 01 & 33 27 38 & hmstg & & b228 + 339.0 + 15.8 & & lu12 & & 15 45 34 & 34 40 53 & vmf & & + 339.2 + 16.1 & 3392 + 161 - 1 & lu10 & & 15 44 54 & 34 17 33 & lm & y & b228 + & 3392 + 161 - 2 & & & 15 45 12 & 34 13 21 & lm & n & b228 + & 3392 + 161 - 3 & & & 15 45 15 & 34 20 43 & lm & n & b228 + & 3392 + 161 - 4 & & sl 13 & 15 45 29 & 34 24 40 & lm & n & fs349 , b228 + & & lu2 & & 15 39 56 & 34 42 50 & vmf & & + & & lu3 & & 15 40 10 & 33 40 07 & vmf & & b228 + & & lu5 & & 15 42 03 & 33 46 33 & vmf & & b228 + 340.7 + 9.7 & & & & 16 11 53 & 38 03 48 & hmstg & & lu iiin + 340.9 + 9.2 & & & & 16 13 57 & 38 16 40 & hmstg & & bhr134 + 340.6 + 9.0 & & & & 16 14 03 & 38 39 04 & hmstg & & + 340.2 + 9.0 & 3402 + 90 - 1 & lu34 & & 16 11 23 & 39 01 33 & lm & pm & + & 3402 + 90 - 2 & lu36 & & 16 11 37 & 38 58 21 & lm & pm & + & 3402 + 90 - 3 & lu35 & & 16 11 45 & 39 01 39 & lm & pm & + 339.7 + 9.2 & 3397 + 92 - 1 & & sl 14 & 16 09 42 & 39 09 28 & lm & y & + & 3397 + 92 - 2 & & & 16 10 07 & 39 03 47 & lm & y & + & 3397 + 92 - 3 & lu32 & & 16 10 23 & 39 10 48 & lm & y & + 339.4 + 9.5 & 3394 + 95 & lu30 & & 16 07 49 & 39 12 04 & lm & n & + & & lu31 & & 16 09 08 & 39 03 55 & vmf & & + & & lu33 & & 16 10 27 & 39 05 18 & vmf & & + 336.4 + 8.2 & 3364 + 82 - 1 & lu23 & & 16 00 53 & 42 04 08 & lm & n & + 336.6 + 7.8 & & & & 16 02 49 & 42 13 41 & hmstg & & bhr120 + 336.7 + 8.2 & 3364 + 82 - 2 & & & 16 01 26 & 41 53 06 & lm & n & + 336.7 + 7.8 & & & & 16 03 15 & 42 06 27 & hmstg & & + 336.9 + 8.3 & 3369 + 83 & lu25 & sl 7 & 16 02 31 & 41 39 48 & lm & n & + 336.9 + 7.8 & & lu26 & & 16 04 10 & 42 00 40 & vmf & & + & & lu24 & & 16 00 18 & 42 03 47 & vmf & & + llcccccccccccccccccl 337.6 + 16.4 & & 15 38 29 & 35 02 14 & & & & 2.1 & 4.4 & 2.1 & 0.4 & 0.7 & 1.3 & & & & & & & + 338.7 + 17.5 & & 15 39 17 & 33 30 06 & 2.3 & 6.2 & 2.8 & 2.4 & 6.2 & 2.6 & 0.6 & 0.8 & 1.6 & & & & & & & + 337.9 + 16.5 & & 15 39 24 & 34 45 39 & & & & 1.9 & 3.4 & 1.8 & 2.5 & 1.5 & 0.6 & & & & & & & + 338.1 + 16.7 & & 15 39 25 & 34 27 33 & & & & 1.9 & 2.7 & 1.4 & 1.0 & 0.9 & 0.9 & & & & & & & + & lu1 & 15 39 28 & 34 46 22 & & & & 1.9 & 3.4 & 1.8 & & & & 4.7 & 3.7 & 0.8 & 2.1 & 1.2 & 0.5 & + & lu2 & 15 39 56 & 34 42 50 & & & & 2.0 & 3.0 & 1.5 & & & & 4.6 & 3.3 & 0.7 & 1.9 & 1.0 & 0.5 & + & lu3 & 15 40 10 & 33 40 07 & 1.8 & 4.5 & 2.5 & 1.8 & 3.0 & 1.6 & & & & 5.5 & 9.9 & 1.7 & 0.5 & 0.8 & 1.6 & + 338.8 + 17.2 & & 15 40 14 & 33 38 41 & 2.0 & 5.0 & 2.5 & 2.4 & 5.7 & 2.4 & 0.5 & 0.8 & 1.3 & & & & & & & + & lu4 & 15 40 32 & 34 39 40 & & & & 2.3 & 5.0 & 2.2 & & & & 3.3 & 2.9 & 0.8 & 1.2 & 0.6 & 0.5 & + & lu5 & 15 42 03 & 33 46 33 & 1.9 & 4.3 & 2.2 & 1.5 & 4.0 & 2.7 & & & & 3.0 & 4.7 & 1.5 & 0.7 & 0.5 & 0.7 & + & lu6 & 15 42 04 & 33 50 36 & 1.4 & 2.5 & 1.8 & 2.1 & 5.2 & 2.4 & & & & 1.9 & 1.9 & 0.9 & 0.5 & 0.4 & 0.8 & + & lu7 & 15 42 24 & 34 09 02 & 1.8 & 3.3 & 1.8 & 2.9 & 5.3 & 1.9 & & & & 4.0 & 3.6 & 0.9 & 1.3 & 1.1 & 0.8 & + & lu8 & 15 42 35 & 33 52 50 & 1.7 & 4.6 & 2.7 & 2.5 & 6.0 & 2.4 & & & & 4.5 & 6.2 & 1.3 & 1.1 & 0.8 & 0.7 & + 338.8 + 16.5 & & 15 42 35 & 34 08 58 & 1.6 & 2.8 & 1.8 & 2.8 & 7.2 & 2.6 & 1.9 & 2.2 & 1.2 & & & & & & & + 339.0 + 16.7 & & 15 42 48 & 33 53 56 & 1.5 & 4.0 & 2.7 & 2.3 & 6.6 & 2.8 & 1.4 & 1.6 & 1.1 & & & & & & & + & b228 & 15 43 02 & 34 09 06 & 1.8 & 3.3 & 1.9 & 3.0 & 5.2 & 1.7 & & & & 5.2 & 5.9 & 1.1 & 0.8 & 0.8 & 1.0 & + & lu9 & 15 43 10 & 34 13 50 & 1.6 & 2.3 & 1.5 & 2.4 & 4.4 & 1.8 & & & & 4.2 & 3.6 & 0.8 & 0.9 & 0.4 & 0.4 & + 339.1 + 16.1 & & 15 44 59 & 34 18 08 & 2.2 & 5.8 & 2.6 & 3.0 & 9.1 & 3.1 & 1.0 & 1.5 & 1.1 & & & & & & & + & lu10 & 15 45 06 & 34 17 39 & 2.2 & 6.3 & 2.8 & 2.9 & 8.6 & 3.0 & & & & 3.4 & 3.3 & 0.9 & 0.8 & 0.5 & 0.6 & + 339.1 + 15.9 & & 15 45 30 & 34 25 51 & 3.2 & 4.9 & 1.5 & 2.9 & 5.9 & 2.0 & 2.3 & 1.5 & 0.7 & & & & & & & + & lu30 & 16 07 51 & 39 11 12 & & & & 2.4 & 4.9 & 2.1 & & & & 3.2 & 2.2 & 0.7 & 0.3 & 0.2 & 0.7 & + 339.6 + 9.3 & & 16 08 53 & 39 06 26 & 9.0 & 15.5 & 1.7 & 5.6 & 13.1 & 2.3 & 1.9 & 2.4 & 1.2 & & & & & & & + & lu31 & 16 09 08 & 39 03 55 & 4.2 & 8.0 & 1.9 & 3.3 & 8.0 & 2.4 & & & & 1.5 & 1.8 & 1.1 & 0.4 & 0.2 & 0.4 & + & lu32 & 16 10 19 & 39 12 16 & 2.3 & 3.0 & 1.3 & 2.2 & 5.2 & 2.3 & & & & 4.3 & 3.9 & 0.9 & 0.3 & 0.2 & 0.7 & + & lu33 & 16 10 27 & 39 05 18 & 2.5 & 3.7 & 1.5 & 2.1 & 2.9 & 1.4 & & & & 5.1 & 4.1 & 0.8 & 1.2 & 0.7 & 0.6 & + & lu34 & 16 11 16 & 39 02 38 & & & & 0.9 & 1.9 & 2.1 & & & & 1.0 & 1.0 & 1.0 & 0.5 & 0.2 & 0.4 & + & lu36 & 16 11 28 & 39 00 21 & & & & 1.3 & 3.2 & 2.5 & & & & 4.6 & 3.7 & 0.8 & 1.2 & 0.7 & 0.5 & + & lu35 & 16 11 36 & 39 04 18 & & & & 1.4 & 2.7 & 2.0 & & & & 4.4 & 2.8 & 0.6 & 0.6 & 0.5 & 0.7 & + 340.7 + 9.7 & & 16 11 58 & 38 04 23 & 1.5 & 6.0 & 4.1 & 2.3 & 5.4 & 2.4 & 0.5 & 0.9 & 1.8 & & & & & & & lu iii n + & lu24 & 16 00 18 & 42 03 47 & & & & 1.4 & 2.6 & 1.8 & & & & 4.1 & 2.4 & 0.6 & 0.8 & 0.3 & 0.4 & + & lu23 & 16 00 57 & 42 04 55 & & & & 1.6 & 2.9 & 1.8 & & & & 4.2 & 3.8 & 0.9 & 1.2 & 0.7 & 0.6 & + 336.4 + 8.2 & & 16 00 57 & 42 03 16 & & & & 2.1 & 3.7 & 1.7 & 2.1 & 1.4 & 0.5 & & & & & & & + 336.7 + 8.2 & & 16 01 46 & 41 52 34 & 3.6 & 5.4 & 1.5 & 3.4 & 6.5 & 1.9 & 2.0 & 2.0 & 0.8 & & & & & & & + 336.9 + 8.2 & & 16 02 34 & 41 41 54 & 2.2 & 2.9 & 1.3 & 2.0 & 3.7 & 1.8 & 1.5 & 1.3 & 0.8 & & & & & & & + & lu25 & 16 02 36 & 41 42 26 & 1.2 & 1.2 & 1.0 & 1.9 & 3.3 & 1.7 & & & & 4.9 & 2.5 & 0.5 & 0.6 & 0.2 & 0.4 & + 336.7 + 7.8 & & 16 03 12 & 42 07 43 & & & & 1.8 & 3.7 & 2.0 & 0.7 & 0.6 & 0.9 & & & & & & & + 336.8 + 7.9 & & 16 03 37 & 42 00 55 & & & & 1.5 & 3.1 & 2.1 & 0.7 & 0.7 & 0.8 & & & & & & & + & lu26 & 16 04 10 & 42 00 40 & & & & 1.4 & 2.7 & 1.8 & & & & 5.2 & 3.8 & 0.7 & 1.1 & 0.6 & 0.5 & + llcccccccccc j154214.6341026 & & i & low & 15 42 14.6 & 34 10 26 & 1.8 & 2.2 & 1.2 & 2.5 & 4.7 & 1.9 + j154301.3340915 & iras 153983359 & 0 & & 15 43 01.3 & 34 09 15 & 4.3 & 8.2 & 1.9 & 2.8 & 4.8 & 1.7 + j154506.3341738 & & f & neb & 15 45 06.3 & 34 17 38 & 2.1 & 5.9 & 2.8 & 2.9 & 8.6 & 3.0 + j160703.9391112 & & f & ir , old & 16 07 03.9 & 39 11 12 & & & & 0.4 & 0.9 & 2.4 + j160708.6391407 & & f & v , old & 16 07 08.6 & 39 14 08 & & & & 0.6 & 1.8 & 3.2 + j160829.7390311 & sz102 & i & note & 16 08 29.7 & 39 03 11 & 2.6 & 3.4 & 1.3 & 2.7 & 6.1 & 2.2 + j160831.1385600 & & f & & 16 08 31.1 & 38 56 00 & 0.6 & 1.1 & 1.7 & 0.3 & 0.2 & 0.8 + j160846.8390207 & & i & & 16 08 46.8 & 39 02 07 & 4.2 & 7.8 & 1.9 & 2.7 & 6.0 & 2.3 + j160918.1390453 & lupus3mms & 0 & neb & 16 09 18.1 & 39 04 53 & 3.5 & 6.2 & 1.8 & 3.3 & 8.0 & 2.4 + j160934.2391513 & & f & & 16 09 34.2 & 39 15 13 & 2.3 & 4.1 & 1.8 & 1.7 & 3.8 & 2.2 + j161013.1384617 & & f & ir , old & 16 10 13.1 & 38 46 17 & & & & 0.4 & 0.6 & 1.4 + j161027.4390230 & & f & low & 16 10 27.4 & 39 02 30 & 2.6 & 4.4 & 1.7 & 1.9 & 4.5 & 2.4 + j161204.5380959 & & f & v , low & 16 12 04.5 & 38 09 59 & & & & 1.2 & 4.3 & 3.7 + j160115.6415235 & & f & neb & 16 01 15.6 & 41 52 35 & 4.5 & 6.4 & 1.4 & 3.6 & 6.2 & 1.7 + lccccc lupus i & 2.57 & 9258 & 11856 & 0 & 0 + lupus iii & 1.87 & 6740 & 8412 & 12 & 0.1 + lupus iv & 0.92 & 3300 & 3300 & 0 & 0 + lupus & 5.36 & 19298 & 23568 & 12 & 0.05 + lupus i & 0.76 & 10964 & 12067 & 939 & 8 + lupus iii n & 0.04 & 625 & 650 & 10 & 2 + lupus iii & 0.17 & 2401 & 3240 & 150 & 5 + lupus iv & 0.09 & 1225 & 1425 & 15 & 1 + lupus & 1.06 & 15215 & 17382 & 1114 & 6 + lccccc @xmath114(@xmath43 ) & @xmath115 & @xmath116 & @xmath117 & @xmath118 & @xmath119 + @xmath114(@xmath44 ) & @xmath120 & @xmath121 & @xmath122 & @xmath123 & @xmath124 + @xmath125 ( kms@xmath1 ) & 2.3 & 2.7 & 2.2 & 2.4 & 1.5 + rms @xmath126 ( kms@xmath1 ) & 0.7 & 0.9 & 0.9 & 0.9 & 0.4 + @xmath114(@xmath0co 10 ) & 880 & & & 220 & + @xmath127kms@xmath1(@xmath0co 10 ) & 1.9 & & & 1.7 & + @xmath114(@xmath0co 10 ) & & & & & @xmath128 + @xmath114(c@xmath5o 10 ) & 240 & 5 & & 80 & 160 + @xmath127kms@xmath1(c@xmath5o 10 ) & 0.61.7 & 1.8 & 1.2 & & 0.50.9 + @xmath114(@xmath43)&251 & & & 250 & 120 + co 21 emission from this work ( contours at 1 , 3 , 5 , 7 , 9kkms@xmath1 ) . annotations : dark cloud positions ( table [ tab - dcs ] ) from ( * ? ? ? * open triangles ) , ( * ? ? ? * filled triangles ) , and ( * ? ? ? * squares ) ; embedded ysos from ( * ? ? ? * star - shapes ) ; bright stars ( by name ) ; c@xmath5o 10 peaks from ( * ? ? ? * diamonds ) . filled squares denote cloud cores observed in molecular lines by @xcite . only objects within the @xmath0co map are plotted . the @xmath129 arrows are 15 long . ] co 21 and ( right ) co 43 , integrated from @xmath131 to @xmath132kms@xmath1 . @xmath0co 21 contours at 2 , 4kkms@xmath1 ( black ) and 6 , 8kkms@xmath1 ( white ) ; co 43 contours at 3 , 5 ( black ) and 7 , 9 , 11kkms@xmath1 ( white ) . the broken outline denotes the extent of the co 43 map . ] co 21 , integrated from @xmath131 to @xmath133kms@xmath1 , and ( right ) co 43 , integrated from @xmath131 to @xmath134kms@xmath1 toward lupus iii and from @xmath135 to @xmath133kms@xmath1 toward lupus iiin . contours at 2 , 4kkms@xmath1 ( black ) and 6 , 8 16kkms@xmath1 ( white ) . ] co 21 and ( right ) co 43 , integrated from @xmath136 to @xmath137kms@xmath1 . @xmath0co 21 contours at 1 , 2 , 3kkms@xmath1 ( black ) and 4 , 5 , 6 , 7kkms@xmath1 ( white ) ; co 43 contours at 3kkms@xmath1 ( black ) and 5 , 7kkms@xmath1 ( white ) . ] of ( left ) @xmath0co 21 and ( right ) co 43 toward lupus i. left : contours at 0.5 , 1k ( black ) and 1.5 , 2.5 , 3.5k ( white ) . right : contours at 1.5 , 2k ( black ) and 2.5 , 3 , 3.5 , 4k ( white ) . the broken outline denotes the extent of the co 43 map . ] co 21 and ( right ) co 43 emission toward lupus i. contours are at 5 , 5.5 , 6kms@xmath1 ( black ) and 6.5 , 7 , 7.5kms@xmath1 ( white ) . the gray scale is unsmoothed , while the contours have been smoothed with a gaussian kernel . only spectra with s / n@xmath13010 are shown . solid outlines indicate the extent of the maps , while the broken outline indicates the extent of the co 43 map . ] : @xmath0co 21 contours are at 5 , 5.5 , 6kms@xmath1 ( black ) and 7.5 , 8 , 8.5 , 9kms@xmath1 ( white ) ; co 43 contours are at 4 , 4.5,kms@xmath1 ( black ) and 7.5 , 8 , 8.5 , 9kms@xmath1 ( white ) . only spectra with s / n@xmath1305 are shown . ] co 21 and ( right ) co 43 toward lupus i. contours are at 2 , 2.4, 3.6kms@xmath1 . gray scale is unsmoothed , while contours have been smoothed with a gaussian kernel . only spectra with s / n@xmath13010 are shown . solid outlines indicate the extent of the maps , while the broken outline indicates the extent of the co 43 map . ] of co transitions for cores seen in c@xmath5o 10 . the dashed line indicates a line ratio of unity , while dotted lines trace ratios of 0.67 , 1.5 ( left ) , 0.33 , 0.67 , 1.5 , 3 ( center ) , and 0.67 , 1.5 , 3 ( right ) . symbols denote cores in lupus i ( squares ) , lupus iii / iii n ( triangles ) , and lupus iv ( diamonds ) ; filled symbols are c@xmath5o - selected cores from @xcite , and open symbols are extinction - selected cores from @xcite . ]

complex systems are usually described as consisting of mutually interacting objects , often exhibiting complex global behavior resulting from the interactions between these objects . this behavior is typically characterized as `` emergent '' or `` self - organizing '' as the system s constituting parts do not usually obey a central controller @xcite . analytic treatment generally does not yield the complete theory of a complex system . therefore , modeling and simulation techniques play a major role in our understanding of how these systems work @xcite . methodologies such as agent - based modeling ( abm ) , system dynamics , discrete event simulation , among others , are frequently employed for this purpose @xcite . of these , abm provides an instinctive approach for describing many complex systems , as agents are regularly a suitable match to the individual and heterogeneous objects composing these systems . the local interactions of these objects , as well as their individual and adaptive behavior , are often critical for understanding global system response @xcite . abms are commonly implemented as a stochastic process , and thus require multiple runs ( observations ) with distinct pseudo - random number generator ( prng ) seeds in order to have appropriate sample sizes for testing hypotheses and differentiating multiple scenarios under distinct parameterizations @xcite . computational models of complex systems in general , and abms in particular , are usually very sensitive to implementation details , and the influence that seemingly negligible aspects such as data structures , discrete time representation and sequences of events can have on simulation results is notable @xcite . furthermore , most model implementations are considerably elaborate , making them prone to programming errors @xcite . this can seriously affect model validation when data from the system being modeled can not be obtained easily , cheaply or at all . model verification can also be compromised , to the point that wrong conclusions may be drawn from simulation results . a possible answer to this problem is the independent replication of such models @xcite . replication consists in the reimplementation of an existing model and the replication of its results @xcite . replicating a model in a new context will sidestep the biases associated with the language or toolkit used to develop the original model , bringing to light dissimilarities between the conceptual and implemented models , as well as inconsistencies in the conceptual model specification @xcite . additionally , replication promotes model verification , model validation @xcite , and model credibility @xcite . more specifically , model verification is promoted because if two or more distinct implementations of a conceptual model yield statistically equivalent results , it is more likely that the implemented models correctly describe the conceptual model @xcite . thus , it is reasonable to assume that a computational model is untrustworthy until it has been successfully replicated @xcite . model parallelization is a an illustrative example of the importance of replication . parallelization is often required for simulating large models in practical time frames , as in the case of abms reflecting systems with large number of individual entities @xcite . by definition , model parallelization implies a number of changes , or even full reimplementation , of the original model . extra care should be taken in order to make sure a parallelized model faithfully reproduces the behavior of the original serial model . there are inclusively reports of failure in converting a serial model into a parallel one @xcite . although replication is considered the scientific gold standard against which scientific claims are evaluated @xcite , most conceptual models have only been implemented by the original developer , and thus , have never been replicated @xcite . several reasons for this problem have been identified , namely : a ) lack of incentive @xcite ; b ) below par communication of original models @xcite ; c ) insufficient knowledge and uncertainty of how to validate results of a reimplemented model @xcite ; and , d ) the inherent difficulty in reimplementing a model @xcite . this work targets c ) , with positive influence on d ) . replication is evaluated by comparing the output of the reimplementation against the output of the original model @xcite , and this process , as will be discussed , is empirically driven and model - dependent ( or even parameter - dependent ) . furthermore , it is sometimes unclear as to what output features best describe model behavior . a robust and ready to use output comparison method would thus reduce or eliminate uncertainty of how to validate reimplementation results ( reason c ) , eliminating this obstacle in the overall process of model replication ( reason d ) . we present a model comparison technique , which uses principal component analysis ( pca ) to convert simulation output into a set of linearly uncorrelated statistical measures , analyzable in a consistent , model - independent fashion . it is appropriate for ascertaining statistical equivalence of a model replication with its original implementation . besides model - independence , this technique has three additional desirable features : a ) it automatically selects output features that best explain implementation differences ; b ) it does not depend on the distributional properties of simulation output ; and , c ) it simplifies the modelers work , as it can be used directly on model output , avoiding manual selection of specific points or summary statistics . we evaluate this technique against classic model comparison methods using the pphpc abm @xcite as a test case . this model is thoroughly studied in terms of simulation output for a number of parameter configurations , providing a solid base for this discussion . the paper is organized as follows . first , in section [ sec : background ] , we review commonly used methods for comparing the output of simulation models , as well as previous work on model replication using these methods . section [ sec : simmod ] introduces pphpc , the simulation model used as a test case for the proposed model - independent comparison methodology , which is described in section [ sec : micomp ] . next , section [ sec : expsetup ] delineates the experimental setup for assessing this methodology . in section [ sec : results ] , results of the classic and model - independent comparison approaches are presented . a discussion and an evaluation of these results is undertaken in section [ sec : discussion ] . section [ sec : conclusions ] summarizes what was accomplished in this paper . the typical process of model comparison is described by wilensky and rand @xcite , and can be roughly divided into three steps : 1 ) choice of replication standard ( rs ) ; 2 ) selection of output statistical summaries , i.e. focal measures ( fms ) , to compare ; and , 3 ) comparison of fms using statistical techniques . [ [ choice - of - replication - standard ] ] choice of replication standard + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + axtell et al . @xcite defined three rss for the level of similarity between fms ( carley @xcite calls the rs the emphasis of demonstration ) : _ numerical identity _ , _ distributional equivalence _ and _ relational alignment_. the first , _ numerical identity _ , implies exact numerical output , but it is difficult to demonstrate for stochastic models and not critical for showing that two such models have the same dynamic behavior . to achieve this goal , _ distributional equivalence _ is a more appropriate choice , as it aims to reveal the statistical similarity between two outputs . finally , _ relational alignment _ between two outputs exists if they show qualitatively similar dependencies with input data , which is frequently the only way to compare a model with another which is inaccessible ( e.g. , implementation has not been made available by the original author ) , or with a non - controllable `` real '' system ( such as a model of the human immune system @xcite ) . for the remainder of this text we assume the _ distributional equivalence _ rs when discussing model replication . [ [ selection - of - focal - measures ] ] selection of focal measures + + + + + + + + + + + + + + + + + + + + + + + + + + + after selecting the rs criteria , a set of statistical summaries representative of each output must also be defined . it is these statistical summaries , and not the complete model outputs , that will be compared in order to assert the similarity between the original model and the replication . as models may produce large amounts of data , the statistical summaries should be chosen as to be relevant to the actual modeling objective . consequently , the selection of output statistical summaries , i.e. fms , is an empirical exercise and is always dependent of the model under study @xcite . [ [ comparison - of - fms - using - statistical - techniques ] ] comparison of fms using statistical techniques + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + there are three major statistical approaches used to compare fms : 1 ) statistical hypothesis tests ; 2 ) confidence intervals ; and , 3 ) graphical methods @xcite . in statistical hypothesis testing , a null hypothesis is tested against an alternative hypothesis . the test can have two results : 1 ) fail to reject the null hypothesis ; or , 2 ) reject the null hypothesis in favor of the alternative hypothesis . in this case , the tests of interest are two - sample ( or multi - sample ) hypothesis tests which test for the null hypothesis that the observations in each sample are drawn from the same distribution , against the alternative that they are not . these observations are the statistical summaries obtained from the outputs of two ( or more ) model implementations , or from the outputs of a model and the system being modeled . although statistical procedures for comparing model and system outputs using hypothesis tests have been proposed @xcite , confidence intervals are usually preferred for such comparisons , as they provide an indication of the magnitude by which the statistic of interest differs from model to system . confidence intervals are also commonly used when evaluating different models that might represent competing system designs or alternative operating policies @xcite . graphical methods , such as q - q plots , can also be employed for comparing output data , though their interpretation is more subjective than the previous methods . in this paper we will focus on hypothesis tests , which are more common for comparing a model implementation and its replication @xcite . the choice of which statistical test(s ) to use depends on a number of factors , namely : 1 ) whether a specific fm follows a normal distribution ; 2 ) the number of observations ( i.e. , number of runs ) in each sample ( i.e. , model implementation ) ; 3 ) the dimensionality of the fms ( i.e. , unidimensional or multidimensional ) ; and , 4 ) the number of samples ( i.e. , model implementations ) to be compared . for example , in @xcite , the authors discuss the use of the hotelling s @xmath0 test , for assessing the validity of a multivariate response simulation model representing a real , observable system with multiple outputs . the study is focused towards relating the cost of sampling against the probability of type i and ii errors according to an acceptable similarity range between model and system output averages . this analysis was later extended to confidence intervals and joint confidence regions @xcite . in reference @xcite , axtell et . all compared two initially different models , with one iteratively modified in order to be aligned with the other . the authors evaluated how can distinct equivalence standards be statistically assessed using non - parametric statistical tests ( namely the kolmogorov - smirnov @xcite and mann - whitney u @xcite tests ) , and how minor variations in model design affect simulation outcomes . they concluded that comparing models developed by different researchers and with different tools ( i.e. , programming languages and/or modeling environments ) , can lead to exposing bugs , misinterpretations in model specification , and implicit assumptions in toolkit implementations . the concepts and methods of `` computational model alignment '' ( or `` docking '' ) were first discussed in this work . edmonds and hales @xcite performed two independent replications of a previously published model involving co - operation between self - interested agents . several shortcomings were found in the original model , leading the authors to conclude that unreplicated simulation models and their results can not be trusted . this work is one of the main references in model replication , describing in detail the process of running two model implementations with different parameters , selecting comparison measures and performing adequate statistical tests . in reference @xcite , the authors presented an abm replication case study , describing the difficulties that emerged from performing the replication and determining if the replication was successful . a standard @xmath1-test was used for comparing model outputs . the authors concluded that model replication influences model verification and validation and promotes shared comprehension concerning modeling decisions . miodownik et al . @xcite replicated an abm of social interaction @xcite , originally implemented in matlab , using the ps - i environment for abm simulations @xcite . a statistical comparison of the mean levels of `` civicness '' at the end of the simulation ( over 10 runs ) was performed using the mann - whitney u test . results showed that , while distributional equivalence was obtained in some cases , the two models were mostly only relationally aligned . the authors attribute this mainly to the fact that some aspects of the original model were not implementable with ps - i . a method for replicating insufficiently described abms was discussed in reference @xcite , consisting in modeling ambiguous assumptions as binary parameters and systematically applying statistical tests to all combinations for their equivalence to the original model . the approach was used to replicate epstein s demographic prisoner s dilemma model @xcite , with only partial success , suggesting the existence of some undefined assumptions concerning the original model . the authors also conducted a number of statistical tests regarding the influence of specific design choices , highlighting the importance that these be explicitly documented . alberts et al . @xcite implemented a cuda @xcite version of toy infection model @xcite , and compared it with the original version implemented in netlogo @xcite , as well as to an in - house optimized serial version . statistical validation was performed visually using q - q plots . multiple serial and parallel variants of the pphpc model were compared in reference @xcite . simultaneous comparison of the several variants was accomplished by applying the multi - group non - parametric kruskal - wallis test @xcite to predetermined fms of the several model outputs . results showed that all model variants could not be considered misaligned for a number of different parameters . this methodology has a number of disadvantages . first , it relies on fms which are model - dependent and , probably , user - dependent . in order to apply it to a different model , the process of selecting fms and the appropriate statistical tests must be redone . furthermore , for different model parameters , the originally selected fms may be of no use , as simulation output may change substantially ( e.g. , the warm - up period for the steady - state statistical summaries can be quite different ) . also , it might not be clear which fms best capture the behavior of a model . the predator - prey for high - performance computing ( pphpc ) model is a reference model for studying and evaluating spatial abm ( sabm ) implementation strategies , capturing important characteristics of sabms , such as agent movement and local agent interactions . it is used in this work as a test case for the proposed model comparison method . the model is thoroughly described in reference @xcite using the odd protocol @xcite . here we present a summarized description of the model . pphpc is a predator - prey model composed of three entity classes : _ agents _ , _ grid cells _ and _ environment_. _ agents _ can be of type prey or predator . while prey consume passive cell - bound food , predators consume prey . agents have an energy state variable , @xmath2 , which increases when feeding and decreases when moving and reproducing . when energy reaches zero , the agent is removed from the simulation . instances of the _ grid cell _ entity class are where agents act , namely where they try to feed and reproduce . grid cells have a fixed grid position and contain only one resource , cell - bound food ( _ grass _ ) , which can be consumed by prey , and is represented by the countdown state variable @xmath3 . the @xmath3 state variable specifies the number of iterations left for the cell - bound food to become available . food becomes available when @xmath4 , and when a prey consumes it , @xmath3 is set to @xmath5 ( given as an initial simulation parameter ) . the set of all grid cells forms the _ environment _ entity , a toroidal square grid where the simulation takes place . the environment is defined by its size and by the restart parameter , @xmath5 . the temporal scale is represented by a positive integer @xmath6 , which represents the number of iterations . a simulation starts with an initialization process , where a predetermined number of agents are randomly placed in the simulation environment . cell - bound food is also initialized at this stage . after initialization , and to get the simulation state at iteration zero , outputs are collected . the scheduler then enters the main simulation loop , where each iteration is sub - divided into four steps : 1 ) agent movement ; 2 ) food growth in grid cells ; 3 ) agent actions ; and , 4 ) gathering of simulation outputs . note that the following processes are explicitly random : a ) initialization of specific state variables ( e.g. , initial agent energy ) ; b ) agent movement ; c ) the order in which agents act ; and , d ) agent reproduction . for process c ) , this implies that the agent list should be explicitly shuffled before agents can act . six outputs are collected at each iteration @xmath7 : @xmath8 , @xmath9 , @xmath10 , @xmath11 , @xmath12 , and @xmath13 . @xmath14 and @xmath15 refer to the total prey ( _ sheep _ ) and predator ( _ wolf _ ) population counts , respectively , while @xmath16 holds the quantity of available cell - bound food . @xmath17 and @xmath18 contain the mean energy of prey and predator populations . finally , @xmath13 refers to the mean value of the @xmath3 state variable in all grid cells . reference parameters for the pphpc model are specified in reference @xcite . parameters are qualitatively separated into size - related and dynamics - related groups . although size - related parameters also influence model dynamics , this separation is useful for parameterizing simulations . concerning size - related parameters , a base grid size of @xmath19 is associated with @xmath20 prey and @xmath21 predators . different grid sizes should have proportionally assigned agent population sizes , such that the initial agent density and the initial ratio between prey and predators remains constant . we define _ model size _ as the association between grid size and initial agent population . for example , model size 200 corresponds to a grid size of @xmath22 with prey and predators at iteration 0 . for the dynamics - related parameters , two parameter sets , 1 and 2 , are proposed . the two parameter sets generate distinct dynamics , with the second set typically yielding more than twice the number of agents than the first . we will refer to a combination of model size and parameter set as `` size@xmath23set '' , e.g. @xmath24 for model size 400 , parameter set 1 . the reference number of iterations , @xmath6 , is , not counting with the initial simulation state at iteration 0 . in order to statistically compare output behaviors of different pphpc implementations , we should first select a set of fms , i.e. , of statistical summaries representative of individual outputs @xcite . we can then compare these fms for distributional equivalence instead of directly comparing outputs . typically , statistical summaries consist of long - term or steady - state means . however , being limited to analyze average system behavior can lead to incorrect conclusions @xcite . consequently , we extract and compare a broader range of statistical summaries from individual outputs , namely : 1 ) maximum value ( @xmath25 ) ; 2 ) iteration where maximum value occurs ( @xmath26 ) ; 3 ) minimum value ( @xmath27 ) ; 4 ) iteration where minimum value occurs ( @xmath28 ) ; 5 ) steady - state mean ( @xmath29 ) ; and , 6 ) steady - state sample standard deviation ( @xmath30 ) . thus , we compare a total of 36 fms ( six statistical summaries from six outputs ) . in order to discard the initial transient period and avoid initialization bias , steady - state measures , @xmath29 and @xmath30 , are taken for @xmath31 and @xmath32 for parameter sets 1 and 2 , respectively @xcite . assuming that individual model implementations are executed @xmath33 times , there will be @xmath33 observations for each associated output . as such , each of the 36 fms will be represented by a sample of @xmath33 observations per implementation . individual fms are tested for the null hypothesis that the points in each sample are drawn from the same distribution , against the alternative that they are not . as discussed in @xcite , all measures , except @xmath26 and @xmath28 , are amenable to be compared using parametric methods , as they follow ( or approximately follow ) normal distributions . thus , for these we use the two - sample @xmath1-test , while for @xmath26 and @xmath28 we use the non - parametric mann - whitney u test . if two implementations produce the same dynamic behavior it is reasonable to expect that any statistical summary collected over a number of independent runs also comes from the same distribution @xcite . thus , conclusions on the alignment of two configurations can be drawn from the @xmath34-values produced by the statistical tests , more specifically by observing if all or most of the 36 @xmath34-values ( one per fm ) are above the typical 1% or 5% significance levels . as this effectively constitutes a multiple hypothesis testing scenario , @xmath34-values can be adjusted with the bonferroni procedure or similar method @xcite . two implementations of the pphpc model are used for evaluating the model comparison technique proposed in the next section . the first is developed in netlogo @xcite , while the second is a java implementation with several parallel variants @xcite . for the results discussed in this paper , simulations performed with the java implementation were executed with the ex parallelization strategy using eight threads . in this strategy , each thread processes an equal part of the simulation environment , and reproducible simulations are guaranteed . individual threads use their own sub - sequence of a global random sequence , obtained using a random spacing approach @xcite . the mersenne twister pseudo - random number generator ( prng ) @xcite is used by both implementations for driving the model s random processes . complete details of both implementations are available in the provided references , and their source code is available at https://github.com / fakenmc / pphpc/. a model - independent comparison or alignment method should work directly from simulation output , automatically selecting the features that best explain potential differences between the implementations being compared , thus avoiding the disadvantages of the classical approach . additionally , such a method should not depend on the distributional properties of simulation output , and should be directly applicable by modelers . our proposal consists of automatically extracting the most relevant information from simulation output using pca @xcite . pca is a widely used technique @xcite which extracts the largest variance components from data by applying singular value decomposition to a mean - centered data matrix . in other words , pca is able to convert simulation output into a set of linearly uncorrelated measures which can be analyzed in a consistent , model - independent fashion . this technique is especially relevant for abms , as it considers not only equilibrium but also dynamics over time @xcite . procedure [ proc : micomp ] summarizes this process for a time - series type output , although the general idea is applicable for other types of multivariate output @xcite . 1 . perform @xmath33 replications for each model implementation , and collect the respective outputs . 2 . for each model output @xmath35 ( e.g. , predator population , @xmath36 ) : 1 . group the @xmath35 s from all replications row - wise in matrix @xmath37 , i.e. the @xmath38 output from the @xmath39 replication will be the @xmath39 row of matrix @xmath37 . if we are comparing @xmath40 implementations and performing @xmath33 replications of @xmath6 time steps for each , the size of this matrix will be @xmath41 . 2 . determine matrix @xmath42 , which is the column mean - centered version of @xmath37 . 3 . apply pca to matrix @xmath42 , considering that rows ( replications ) correspond to observations and columns ( iterations or time steps ) to variables . this yields : 1 ) matrix @xmath43 , containing the representation of the original data in the principal components ( pcs ) space ; and , 2 ) vector @xmath44 , containing the eigenvalues of the covariance matrix of @xmath42 in descending order , each eigenvalue corresponding to the variance of the columns of @xmath43 . the columns of @xmath43 correspond to pcs , and are orderer by decreasing variance , i.e. the first column corresponds to the first pc , and so on . rows of @xmath43 correspond to observations . the @xmath45 column of @xmath43 contains @xmath46 modelindependent observations for the @xmath45 pc , @xmath33 for each implementation . groups of @xmath33 observations associated with model implementations can be compared using statistical methods . more specifically , hypothesis tests can be used to check if the output projections on the pc space , grouped by model implementation , are drawn from populations with the same distribution . there are two possible lines of action : 1 . a multivariate approach , consisting of applying a manova test @xcite on the @xmath46 output projections , organized in @xmath40 groups ( one per model implementation ) of @xmath33 @xmath47-dimensional observations each , along the first @xmath47 pcs ( dimensions ) such that these explain a userdefined minimum amount of variance . 2 . a univariate approach , consisting of applying a hypothesis test to individual pcs on the @xmath46 output projections , organized in @xmath40 groups ( one per model implementation ) of @xmath33 observations each . possible tests include the @xmath1-test and the mann - whitney u test for comparing two groups ( @xmath48 ) , or anova @xcite and kruskal - wallis test , which are the respective parametric and non - parametric versions for comparing more than two groups ( @xmath49 ) . the use of the manova multivariate test has the advantage of yielding a single @xmath34-value from the simultaneous comparison of output projections along multiple dimensions . an equally succinct answer can be obtained with the univariate approach using the bonferroni correction or a similar method for handling @xmath34-values from multiple comparisons @xcite . however , both approaches will not prioritize dimensions , even though the first pcs are more important for characterizing model differences , as they explain more variance . in the univariate case one can subjectively attribute higher relevance to the first pcs , or objectively prioritize dimensions according to the explained variance using the weighted bonferroni procedure @xcite . parametric tests such as manova and @xmath1-test make several assumptions about the data being compared , namely that @xcite : * all samples are mutually independent . * each sample is drawn from a normally distributed population ( multivariate normality for manova ) . * samples are drawn from populations with equal variances ( for manova , the variance - covariance matrix should be the same for each population ) . the first assumption is guaranteed by the way the samples are obtained , as described in section [ sec : expsetup ] . the second and third assumptions can be checked using appropriate statistical tests . more specifically , group sample normality can be assessed using the shapiro - wilk test @xcite ( royston test @xcite in the multivariate approach ) , while equality of variances among groups can be verified with the bartlett test @xcite ( box s m test @xcite for homogeneity of variance - covariance matrices in the multivariate approach ) . however , box s m test is very sensitive and can lead to false negatives ( type ii errors ) . fortunately , manova is considerably robust to violations in the homogeneity of variance - covariance matrices when groups have equal sample sizes @xcite . if these assumptions are not met in the univariate approach , non - parametric tests ( e.g. , mann - whitney u test or the kolmogorov - smirnov test ) can be used instead . non - parametric alternatives , e.g @xcite , exist for the multivariate approach , but they are not as well established as the aforementioned methods . the eigenvalues vector @xmath44 is also important for this process , for two reasons : 1 ) to determine a minimum number of pcs for the manova test , such that these explain a prespecified percentage of variance ; 2 ) alignment or otherwise of the @xmath40 model implementations can be empirically assessed by analyzing how the explained variance is distributed along pcs . the percentage of variance explained by each pc can be obtained as shown in eq . [ eq : pcvar ] . @xmath50 where @xmath7 identifies the @xmath51 pc , @xmath52 is the eigenvalue associated with the @xmath51 pc , and @xmath53 is the sum of all eigenvalues . if the variance is well distributed along many pcs , it is an indication that the compared implementations are aligned , at least for the output being analyzed . on the other hand , if most of the variance is explained by the first pcs , it is a clear sign that at least one model implementation is misaligned . the rationale being that if all implementations show the same dynamical behavior , then the projection of their outputs in the pc space will be close together and have similar statistics , i.e. means , medians and variance . as such , pca will be unable to find components which explain large quantities of variance , and the variance will be well distributed along the pcs . if at least one model implementation is misaligned , the projection of its outputs in the pc space will be farther apart than the projections of the remaining implementations . as such , pca will yield at least one component which explains large part of the overall variance . the alignment of two or more implementations can be assessed by analyzing the following information : 1 ) the @xmath34-values produced by the univariate and multivariate statistical tests , which should be above the typical 1% or 5% significance levels in case of implementation alignment ; in the univariate case , it may be useful to adjust the @xmath34-values using the weighted bonferroni procedure to account for multiple comparisons ; 2 ) in case of misalignment , the total number of pcs required to explain a prespecified amount of variance should be lower than in case of alignment ; also , more variance should be explained by the first pcs of the former than by the same pcs of the latter ; and , 3 ) the scatter plot of the first two pc dimensions , which can offer visual , although subjective feedback on model alignment ; e.g. , in case of misalignment , points associated with runs from different implementations should form distinct groups . determining the alignment of models with multiple outputs may be less straightforward . if model implementations are clearly aligned or misaligned , conclusions can be drawn by analyzing the information provided by the proposed method for each individual output . otherwise , an objective approach may be required . we suggest two such approaches : a ) a bonferroni or similar multiple comparison @xmath34-value correction ; or , b ) the concatenation of all outputs , centered and scaled , followed by a model - independent examination of the concatenated output . the first method can be appropriate for the multivariate case , in which we have one @xmath34-value per output . it can be extended to the univariate case if we only consider the uncorrected @xmath34-value from the first pc . if we consider multiple pcs or their already corrected @xmath34-values ( weighted bonferroni ) , the analysis becomes more complex and it can become unclear how to interpret the results . additionally , multiple comparison correction methods often assume independency of test statistics , which might not be possible to assure when testing different outputs of the same simulation model , which are most likely correlated . in the second approach , we concatenate all outputs and examine the resulting concatenated output . this reduces a model with @xmath54 outputs to a model with one output . in order to perform output concatenation , we center and scale the outputs such that their domains are in the same order of magnitude this can be performed using range scaling on each output @xmath35 , for example , as shown in the following eqs . : @xmath55 @xmath56 \label{eq : scaledoutput}\ ] ] where @xmath57 is the value of output @xmath35 at iteration @xmath7 , @xmath58 is the mean of output @xmath35 , and @xmath59 is the range scaled version of @xmath35 . other centering and scaling methods , such as auto - scaling or level scaling @xcite , can be used as an alternative to range scaling in eq . [ eq : rangescale ] . for a model with @xmath54 outputs , @xmath60 , the resulting concatenated output is given by @xmath61 where @xmath62 is the concatenation operator , and @xmath63 is the concatenation of all model outputs . model implementations can thus be compared with the proposed method using the `` single '' model output @xmath63 . in order to test the model comparison methods , we define a base pphpc configuration using the netlogo implementation and compare it with three configurations using the java implementation . all configurations are tested for model sizes 100 , 200 , 400 and 800 , and parameter sets 1 and 2 , as described in section [ sec : simmod : params ] . the four configurations follow the conceptual model , except when stated otherwise : 1 . netlogo implementation . 2 . java implementation . 3 . java implementation : no agent list shuffling prior to agent actions java implementation : the restart parameter , @xmath5 , is set to one unit less than specified ( 9 instead of 10 for parameter set 1 , 14 instead of 15 for parameter set 2 ) . the goal is to assess how the two model comparison methods ( classic and model - independent ) expose the increasing differences of comparing configuration 1 with configurations 2 to 4 . more specifically , we are interested in checking if the proposed model - independent comparison method is able to expose these differences in the same way as the classic approach . as such , we define three comparison cases : case i : : compare configuration 1 with configuration 2 . these configurations should yield distributionally equivalent results . case ii : : compare configuration 1 with configuration 3 . a small misalignment is to be expected . case iii : : compare configuration 1 with configuration 4 . there should be a mismatch in the outputs . for each `` size@xmath23set '' combination , independent samples of the six model outputs are obtained from @xmath64 replications for each configuration , in a total of @xmath65 runs . each replication @xmath66 is performed with a prng seed obtained by taking the md5 checksum of @xmath67 and converting the resulting hexadecimal string to an integer ( limited to 32-bits for netlogo and 128-bits for the java implementation ) , guaranteeing independence between seeds , and consequently , between replications . the same samples were used for the evaluation of both classical and model - independent comparison approaches . in order to evaluate how the tested methodologies respond to larger sample sizes , we performed additional @xmath68 replications per configuration for the @xmath24 combination , in a total of @xmath69 runs . the prng seeds for individual replications were obtained in the same fashion . for the model - dependent comparisons , results were obtained with the simoututils software @xcite . the micompr r package @xcite provides an implementation of the proposed model - independent comparison approach , and was used to produce the corresponding results . the data generated by this computational experiment , as well as the scripts used to set up the experiment , are made available to other researchers at https://zenodo.org/record/46848 . in this section we mainly focus on the results for the @xmath24 combination . results for the remaining size / set combinations are provided as supplementary material , and are referred to when appropriate . .[tab : mdcomps]@xmath70-values for the classic model comparison method for model size 400 , parameter set 1 , and @xmath64 runs per configuration . @xmath70-values were obtained with the @xmath1-test for @xmath25 , @xmath27 , @xmath29 and @xmath30 statistics , and with the mann - whitney u test for @xmath26 and @xmath28 statistics . case i weights two similar configurations , case ii compares configurations with a small implementation difference , and case iii compares configurations with a different parameter . values lower than @xmath71 are underlined , while values lower than @xmath72 are double - underlined . [ cols="^ , < , > , > , > , > , > , > " , ] the shapiro - wilk test does not reject univariate normality in approximately 95% of samples used for the @xmath1-tests . of the remaining 5% , only 1% show significant rejection at the @xmath73 level . this percentage is even lower for pc1 samples , the most critical for evaluating configuration alignment using the univariate approach . samples also seem to generally have similar variance , as denoted by the 90% of non - significant results for the bartlett test . multivariate normality is also verified by the royston test in most instances . since @xmath34-values are the probability of obtaining a result at least as extreme than what was actually observed ( assuming that the null hypothesis is true ) , type i error rates are close to what would be expected , namely 95% for @xmath74 , 4% for @xmath75 and 1% for @xmath76 . as such , the assumptions tested by the shapiro - wilk , bartlett and royston tests seem to be generally verified . the same does not appear to hold for the manova assumption of homogeneity of variance - covariance matrices , evaluated with box s m test . however , as stated in section [ sec : micomp ] , this test is extremely sensitive and manova is resilient to violations in this assumption when samples have equal sizes , which is the case here . larger sample sizes make hypothesis test more powerful , i.e. , more likely to reject the null hypothesis when it is false , thus decreasing the probability of committing a type ii error . there are a number of methods to determine the adequate sample size ( number of runs in this case ) , required to obtain a specific statistical power @xcite . however , we are essentially interested in determining , within the discussed frameworks , if and how conclusions concerning configuration alignment change with sample size . for this purpose , the classic and the proposed model - independent comparison methods were applied to samples of @xmath77 and @xmath68 replicates for the @xmath24 combination . the comparison with samples of size 10 used the first 10 observations from the @xmath64 runs per configuration setup , while the comparison with samples of size 100 used a new set of @xmath68 runs per configuration . results from this analysis are provided in supplementary tables s5.1 ( classic approach , 10 runs ) , s5.2 ( model - independent approach , 10 runs ) , s6.1 ( classic approach , 100 runs ) and s6.2 ( model - independent approach , 100 runs ) . using samples of 10 observations , the classic model comparison method still manages to find configuration differences in cases ii and iii , although not as clearly as for @xmath64 . in case ii , the difference is found on the steady - state mean of the @xmath36 and @xmath78 outputs . with @xmath64 , differences are detected in several more fms . the proposed model - independent comparison approach is also able to spot differences in cases ii and iii , likewise detecting differences in the @xmath36 and @xmath78 outputs for case ii . for @xmath68 , both methods confirm what was observed for smaller sample sizes , although results are more pronounced , with very significant @xmath34-values for cases ii and iii . two details emerge , however . first , for case ii , the classic approach hardly recognizes configuration differences for the mean predator energy output , @xmath79 . such differences are exposed by the proposed model - independent method , both by the @xmath1-test and manova . second , the model - independent method perceives a difference where there should not be any , namely in output @xmath78 of case i , via the manova test . the pc1 @xmath1-test @xmath34-value is not significant , as is the case for the vast majority of remaining pcs . for the 72 pcs ( dimensions ) considered by manova , the @xmath1-test @xmath34-value is significant at @xmath73 in three cases : pc11 , pc22 and pc48 . after weighted bonferroni correction , only the pc11 @xmath34-value , associated with 2.4% of the variance , remains significant , with a value of @xmath80 . since these pcs , and pc11 in particular , account for a small amount of output variability , conclusions concerning output similarity ( and consequently , model alignment ) , do not necessarily change . however , this result raises some concerns about the sensitivity of manova to type i errors ( i.e. , rejecting a true null hypothesis ) in this context . overall , what seems clear for a sample of @xmath68 replicates , is that a @xmath1-test on the first pc of each individual output ( or even the concatenated output ) , is enough to pick up implementation differences . the classic comparison approach and the proposed model - independent method were able to identify the changes introduced in configurations 3 and 4 , even though these were of different nature . the change introduced in configuration 3 , case ii , confirms the sensitivity of this type of models : a simple and seemingly innocuous implementation discrepancy lead to statistically different behavior . however , the change in configuration 4 , case iii , was expected to be noticed , since it implied the alteration of a model parameter . nonetheless , in specific conditions , namely for parameter set 2 , configuration 3 presented more dissimilar behavior than configuration 4 , when compared to configuration 1 . since parameter set 2 generates more agents during simulations ( as discussed in section [ sec : simmod : params ] ) , the effect of not shuffling the agent list in configuration 3 , case ii , may be more pronounced . additionally , the @xmath5 parameter is larger in parameter set 2 than in parameter set 1 . as such , the smaller relative difference of decreasing @xmath5 by one unit in configuration 4 may lead to less significant discrepancies in case iii . in the context of a smaller sample size , namely for @xmath77 runs per configuration , both methods were capable of discerning differences in cases ii and iii . differentiation was better with increasing number of replicates , i.e. , when going from @xmath77 , then @xmath64 , and finally @xmath68 . in practice , however , more replicates did not yield different conclusions regarding output and configuration alignment . globally , results for the proposed model - independent method were in accordance with the classic approach on an output - per - output basis , with the advantage that the outputs are compared as a whole , instead of being summarized into a number of fms . the proposed method was also able to expose a few implementation differences not detected by the classic approach , and , in other instances , it identified these differences with additional detail . as such , the proposed method was shown to be a direct , and often more effective replacement for the classic model - dependent comparison approach . however , in spite of these advantages , interpretation of results should be done thoroughly . in other words , the different types of information produced by the method , detailed in section [ sec : micomp ] , should be considered and evaluated before concluding on model alignment . the @xmath34-values yielded by the @xmath1 and manova tests offer an objective assessment concerning the alignment of individual outputs . while in most instances the two tests provided the same response , there were situations in which results differed . in case of alignment , the @xmath1-test on the first pc never produced @xmath34-values significant at @xmath73 , even prior to weighted bonferroni correction . manova , however , signaled a few false positives for the @xmath78 output , the most worrying of which for the @xmath68 observations per sample case , which also affected the @xmath63 output . a careful analysis revealed the cause to be a misalignment in one or more pcs , which in themselves did not account for much output variance , but were sufficient to provoke a type i error in the manova test . in case of configuration misalignment , manova was generally more sensitive than the pc1 @xmath1-test , and with the exception of @xmath81 , flagged a few misalignments unnoticed by the @xmath1-test . overall , we recommend both tests are performed , and in case of disagreement , try to find the cause . namely , 1 ) if the pc1 @xmath1-test @xmath34-value is significant ( and the manova @xmath34-value is not ) , adjust it with the weighted bonferroni correction , and consider that instead ; 2 ) if the manova @xmath34-value is significant ( and the pc1 @xmath1-test @xmath34-value is not ) , perform the @xmath1-test on other pcs to find which ones are likely to be influencing the manova result . decide on alignment considering how much variance the culprit pcs explain . as an alternative to performing both tests , it is also possible to perform the @xmath1-test on all pcs , and consider the resulting @xmath34-values after weighted bonferroni correction . this also has the advantage of automatically privileging the most important pcs , i.e. , the ones that explain more variance . furthermore , if samples do not appear to be drawn from a normal distribution , a non - parametric test can be used instead of the @xmath1-test ( e.g. , mann - whitney or kolmogorov - smirnov ) . the eigenvalue structure , reflected in the number of pcs required to explain a prespecified percentage of variance , as well as in the variance explained by the first pc or pcs , is also an interesting metric . for aligned outputs , in contrast with misaligned ones , more pcs are required to explain the same amount of variance , and less variance is explained by the first pc or pcs . the information provided by this metric in general , and by the number of pcs in particular , is more useful when assessing multiple cases , for which it can be relatively compared . nonetheless , concerning the variance explained by the first pc , some notes can be provided . it seems to stay at about 5 - 20% for aligned outputs , and usually 25% or more , up to 95% , otherwise . there are exceptions , namely for model size 100 , so unless this value is extremely high , e.g. above 40% , it can be difficult to draw conclusions without having an aligned baseline to compare with ( e.g. , case i ) . this aligned baseline can be obtained with different sets of runs ( using distinct prng seeds ) from the same configuration . however , the specified variance is above all necessary for selecting the number of pcs for manova . while conclusions concerning configuration alignment do not appear to change with different percentages of variance , it seems preferable to use more variance than less , otherwise one may not be able to use manova . our selection of 90% seems appropriate . another possibility is to use a fixed number of pcs for manova instead of specifying a percentage of variance to explain . a potential problem with this approach is that manova could eventually consider very different percentages of variance from output to output and/or case to case . the scatter plot of the first two pc dimensions is also useful for quickly assessing possible output misalignments . if the outputs have a clear mismatch , points associated with runs from different implementations form visibly distinct groups . although this technique is basically a type of face validation , it provides a useful intuition on the alignment of two or more implementations for any given output . the concatenated output provides a single answer on whether two or more model implementations are aligned or not . in terms of false positives , i.e. , reporting a misalignment where none is expected , this was observed once by a manova test for the @xmath68 runs per configuration instance . the @xmath1-test , however , did not flag any configuration difference . false negatives , i.e. failure to detect output mismatch , occurred a few times for both manova and the @xmath1-test , and were essentially associated with smaller model sizes . additionally , neither tests were able to detect differences in case ii for the @xmath77 runs per configuration instance . overall , output concatenation does provide a practical and quick assessment of model alignment , but should not be fully relied upon to make a final decision . additionally , the analysis of individual outputs allows to diagnose which model aspects are most affected by possible mismatches , something which the concatenated output does not permit . altogether , no single metric of the proposed method provides a definitive answer , and decisions concerning model alignment should be taken based on the different types of information it produces . a technique for comparing simulation models in a model - independent fashion was presented in this paper . the method dispenses the need for selecting model - specific fms , and is appropriate for comparing multiple implementations of a conceptual model or replications of an existing model . the technique was validated against a classic model comparison approach using the pphpc abm on a number of model sizes and parameter sets . results showed that the proposed method offered similar conclusions concerning output and model alignment , while providing more detailed feedback in some situations . although the proposed framework is presented in the context of simulation output comparison , it is a generic statistical comparison methodology . as such , it can be employed in circumstances where observations yield a large number of variables @xcite . this work was supported by the fundao para a cincia e a tecnologia ( fct ) projects uid / eea/50009/2013 , uid / mat/04561/2013 , and partially funded with grant sfrh / bd/48310/2008 , also from fct . the author vitor v. lopes acknowledges the financial support from the prometeo project of senescyt ( ecuador ) . s. alberts , m. k. keenan , r. m. dsouza , and g. an . data - parallel techniques for simulating a mega - scale agent - based model of systemic inflammatory response syndrome on graphics processing units . , 88(8):895907 , 2011 . r. a. berg , h. c. hoefsloot , j. a. westerhuis , a. k. smilde , and m. j. werf . centering , scaling , and transformations : improving the biological information content of metabolomics data . , 7(1):142 , 2006 . n. fachada , m. a. t. figueiredo , v. v. lopes , r. c. martins , and a. c. rosa . spectrometric differentiation of yeast strains using minimum volume increase and minimum direction change clustering criteria . , 45:5561 , aug . 2014 . n. fachada , v. v. lopes , and a. rosa . simulating antigenic drift and shift in influenza a. in _ proc . of the 2009 acm symposium on applied computing _ , sac 09 , pages 20932100 , new york , ny , usa , 2009 . acm . lee , t. filatova , a. ligmann - zielinska , b. hassani - mahmooei , f. stonedahl , i. lorscheid , a. voinov , j. g. polhill , z. sun , and d. c. parker . the complexities of agent - based modeling output analysis . , 18(4):4 , 2015 . b. mller , s. balbi , c. m. buchmann , l. de sousa , g. dressler , j. groeneveld , c. j. klassert , q. b. le , j. d. millington , h. nolzen , d. c. parker , j. g. polhill , m. schlter , j. schulze , n. schwarz , z. sun , p. taillandier , and h. weise . standardised and transparent model descriptions for agent - based models : current status and prospects . , 55:156163 , may 2014 . h. r. parry and m. bithell . large scale agent - based modelling : a review and guidelines for model scaling . in a. j. heppenstall , a. t. crooks , l. m. see , and m. batty , editors , _ agent - based models of geographical systems _ , pages 271308 . springer netherlands , 2012 . r. g. sargent . a new statistical procedure for validation of simulation and stochastic models . technical report syr - eecs-2010 - 06 , department of electrical engineering and computer science , syracuse university , nov . 2010 . o. will and r. hegselmann . a replication that failed on the computational model in ` michael w. macy and yoshimichi sato : trust , cooperation and market formation in the u.s . and japan . proceedings of the national academy of sciences , may 2002 ' . , 11(3):3 , 2008 .

in recent years the spin - boson model @xcite @xcite has experienced a strong revival , as it is well suited to describe dissipative and decoherence effects on the dynamics of a two - level system ( tls ) or qubit coupled to a bath . crucial for the effects of the environment on the dynamics of the tls is the shape of the spectral density of the harmonic bath . it is common to assume an ohmic spectral density , which is linear in the continuous bath modes . in this work we concentrate on a so - called structured bath , for which the spectral density is ohmic at low frequencies but exhibits a lorenztian - shaped peak at a certain frequency @xmath0 . it has been shown in @xcite that a spin - boson model with such an effective spectral density can be exactly mapped on the model of a tls which is coupled to a single harmonic oscillator ( ho ) of frequency @xmath0 , where the latter feels the influence of an ohmic bath.due to its wide applicability the tls - oscillator system has been object of intense research along the years . so it reflects for example the physics of single atoms with a large electric dipole moment coupled to the microwave photons of a cavity @xcite , or quantum dots in photonic crystals @xcite . more recently the model has received quite some attention in the field of quantum computation , where two - level systems are used to implement the two logical states of a qubit . we will especially focus on the solid - state implementation of such systems . here , two prominent realizations of a qubit - oscillator system are the cooper - pair box ( cpb ) @xcite @xcite coupled to a transmission line resonantor @xcite @xcite and the josephson flux qubit @xcite read out by a dc - squid @xcite @xcite . inspired by experiments with real atoms interacting with a cavity mode , one speaks for the cpb case of circuit quantum electrodynamics , as now the cpb plays the role of an artificial atom and the waveguide acts as a cavity . from such a setup one expects a huge step towards the realization of a quantum computer , as the transmission line resonator can be used to couple qubits together @xcite , store the information of qubits or to provide non - demolition read - out schemes @xcite . concerning the flux qubit , the read - out usually happens through a damped dc - squid , which is inductively coupled to the qubit . however , through the squid enviromental noise is transferred to the qubit leading to decoherence and dissipation within its dynamics . the effect of this noise on the qubit depends very much on the strength @xmath1 of the coupling between qubit and squid and one faces a conflicting situation . on the one hand one wants a strong coupling for a good read - out resolution . on the other hand the coupling should be minimized to keep the negative effects of the environment as small as possible . in @xcite it has been shown that the qubit - squid system can be described by a spin - boson model with an effective spectral density @xmath2 exhibting a peak at the plasma frequency @xmath0 of the squid . applying the above mentioned mapping an equivalent point of view is to consider the squid as an lc - circuit coupled to the ohmic bath and model it as a harmonic oscillator . a detailed description of a nondestructive read - out scheme is e.g. given in @xcite . the spin - boson model can be formally solved using e.g. real - time path integral methods @xcite . however , in order to get closed - form analytical results , approximations must be invoked . a quite common one is the so - called weak coupling approximation ( wca ) , which is perturbative in the bath spectral density @xcite . however , it has been shown that for strong qubit - ho coupling @xmath1 and for small detuning @xmath3 , where @xmath4 is the qubit energy splitting , such an approximation breaks down @xcite , as coherent exchange processes between tls and oscillator are disregared . for an unbiased qubit the non - interacting blip approximation ( niba ) used in @xcite @xcite circumvents this problem as it is non - perturbative in the coupling @xmath1 and therefore takes correctly into account the influence of the oscillator on the tls . moreover , it allows an analytic treatment of the dynamics . however , the niba is known to break down for a biased qubit at low temperatures @xcite . another approach , which treats the system non - perturbatively in the bath is the flow - equation renormalization method @xcite , where the spin - boson hamiltonian is diagonalized using infinitesimal unitary transformations . however , whithin this approach analytical solutions are difficult to find . recently a polaron transformation was used by huang to obtain analytically the population dynamics and confirm the shiba s relation for an unbiased tls @xcite.in the case in which the qubit and the ho are considered as the central quantum system being coupled to an ohmic bath , the numerical , ab - initio quasiadiabatic propagator path - integral ( quapi ) method @xcite is a nice tool as it enables to cover both the resonant regime , where the oscillator frequency is close to the qubit energy splitting , and the dispersive regime with the oscillator being far detuned from the qubit @xcite . moreover , it can be applied to a biased as well as to an unbiased tls and therefore be used as a testbed for analytical results . for qubits being operated at the degeneracy point , which means an unbiased tls , very often a rotating wave approximation ( rwa ) is applied @xcite , which is expected to be valid for small detuning and yields as starting point the jaynes - cummings hamiltonian @xcite . this model was first used to study a two - state atom interacting with a single , close to resonance cavity mode of the electromagnetic field and predicts e.g. the repeated revival and collapse of rabi oscillations within the atomic excitation probability . by condsidering the tls - ho system in the representation of displaced ho states , brito were able to truncate the infinite hilbert space of this system without loosing the effects of the ho on the tls dynamics @xcite . however , so far none of these works could provide an analytical expression for the dynamics of the dissipative qubit being valid for zero as well as non - zero detuning and for both a biased and unbiased tls . in this work an analytic expression for the dissipative qubit s dynamics which includes the effects of a finite detuning and of a static bias is derived . specifically , starting from the qubit - ho perspective , the eigenvalues and eigenfunctions of the non - dissipative tls - ho system are found approximately using van - vleck perturbation theory up to second order in the coupling @xmath1 . notice that no rotating wave approximation is required . dissipation effects are then evaluated by solving a born - markov master equation for the reduced density matrix in the system s eigenbasis . the structure of the work is as follows . the dissipative tls - ho hamiltonian and the main dynamical quantities are introduced in . inspired by the work of goorden @xcite , we demonstrate in how the eigenstates and eigenenergies of the non - dissipative hamiltonian can be found approximately using van - vleck perturbation theory @xcite . in this way we can provide an analytical formula for the non - dissipative dynamics whichs takes into account the full hilbert space of the qubit - ho system . after that , we show how for low temperatures ( @xmath5 ) this infinite hilbert space can be truncated and discuss the relevant contributions of the ho to the dynamics . in the influence of the environment is investigated , by looking at solutions of the bloch - redfield equations . specifically , analytical expressions fo the tls dynamics are obtained and compared with numerical solutions . the main physical features of the coupled tls - ho system are discussed in . to illustrate the effects of counter - rotating terms in the hamiltonian of the qubit - ho system , which are neglected performing a rwa , we compare in our calculations to results obtained from the jaynes - cummings model . in this section we introduce the hamiltonian for a qubit coupled through a harmonic oscillator to a thermal bath . further , a formula for the population difference between the qubit s two logical states is derived . to set up the model we consider the hamiltonian of a qubit - ho system , @xmath6 , which is coupled to an environmental bath , @xmath7 , by the interaction hamiltonian @xmath8 , so that the total hamiltonian becomes @xmath9 the hamiltonian , @xmath10 consists of @xmath11 the hamiltonian of the tls / qubit and the harmonic oscillator , and the interaction term @xmath12 the hamiltonian of the tls is given in the subspace @xmath13 , corresponding to a clockwise or counterclockwise current in the superconducting loop of a three - junction josephson qubit or more generally to the qubit s two logical states . in the case of a superconducting flux - qubit , the energy bias @xmath14 can be tuned by an applied external flux , @xmath15 , and is zero at the so - called degeneracy point . the tunnelling amplitude is described by @xmath16 . for @xmath17 the states @xmath18 and @xmath19 are eigenstates of @xmath20 , whereas at the degeneracy point those eigenstates are a symmetric and antisymmetric superposition of the two logical states . further , @xmath21 and @xmath22 are the annihilation and creation operator for the ho with frequency @xmath0 , and @xmath1 characterizes the coupling strength . we also introduce the energy splitting @xmath23 between the groundstate @xmath24 and the excited state @xmath25 of the tls . using the transformation @xmath26 with @xmath27 and @xmath28 , we obtain the hamiltonian of the tls in the this basis : @xmath29 . the states @xmath19 and @xmath18 become in the energy basis @xmath30 the hamiltonian @xmath31 is diagonal in the eigenbasis @xmath32 with @xmath33 being the occupation number : @xmath34 . for the eigenbasis of the combined hamiltonian @xmath35 we write @xmath36 following caldeira and leggett @xcite , we model the environmental influences originating from the circuitry surrounding the qubit and the oscillator as a bath of harmonic oscillators being coupled bilinearly to the ho . thus , the environment is described by @xmath37 and the interaction hamiltonian is @xmath38 the operators @xmath39 and @xmath40 are the creation and destruction operator , respectively , for the @xmath41 bath oscillator , @xmath42 is its frequency and @xmath43 gives the coupling strength . the whole bath can be described by its spectral density , which we consider to be ohmic : @xmath44 in @xcite it is shown that the above model is equivalent to that of a tls being coupled directly to a harmonic bath including the single oscillator of frequency @xmath0 ; i.e. , a spin - boson model @xcite with a peaked effective spectral density , @xmath45 the relation between @xmath46 and the coupling parameter @xmath1 between the qubit and the ho is @xmath47 @xcite . this second perspective is suitable for calculating the dynamics of the qubit using a path - integral approach , as it was done for example in @xcite for the case of an unbiased qubit ( @xmath48 ) . the approach in @xcite , however , being based on the niba @xcite , is not suitable to investigate the low temperature dynamics of a biased tls . thus , in this manuscript we will consider the tls and the single oscillator as central quantum system and solve the bloch - redfield master equations for the density matrix of this system , which are valid also for the case of a biased tls . the main goal of this work is to determine the dynamics @xmath49 of the qubit . that means , we wish to calculate the population difference @xmath50 between the @xmath51 and @xmath52 states of the qubit . the reduced density matrix of the tls , @xmath53 is found after tracing out the oscillator and bath degrees of freedom from the total density matrix @xmath54 . in turn @xmath55 is the reduced density matrix of the qubit - ho system . how to calculate this density matrix will be shown later . after some algebra , illustrated in more detail in [ apppt ] , we arrive at an expression for @xmath49 , given in terms of diagonal and off - diagonal elements of @xmath56 in the tls - ho eigenbasis @xmath57 . it reads @xmath58 where @xmath59 + 2 \sin \theta { \langle j { \rm g}|n\rangle } { \langle j { \rm e}|n\rangle } \right\ } \rho_{n n}(t ) \label{p0},\\ \fl p_{nm}(t ) = 2 \sum_j \biggl\ { \cos \theta \biggl[{\langle j { \rm g}|n\rangle } { \langle m|j { \rm g}\rangle } - { \langle j { \rm e}|n\rangle } { \langle m|j { \rm e}\rangle } \biggr ] \nonumber\\ + \sin \theta \biggl[{\langle j { \rm e}|n\rangle } { \langle m|j { \rm g}\rangle } + { \langle j { \rm e}|m\rangle } { \langle n|j { \rm g}\rangle } \biggr ] \biggr\ } \re \ { \rho_{nm}(t ) \ } \label{pnm}\end{aligned}\ ] ] with @xmath60 . how to determine the eigenstates of @xmath6 is described in the next section . in this section we show how to find the eigenvalues of the unperturbed qubit - ho hamiltonian @xmath6 approximately by using van - vleck perturbation theory @xcite . the idea is to take advantage of the degenerate or doublet structure of the energy spectrum of the uncoupled ( @xmath61 ) tls - ho system near resonance , e.g. at @xmath62 . then , as long as the perturbation is small compared to the energy separation of the different doublets , the full hamiltonian will exhibit a similar spectrum of bundled energy levels . the eigenenergies of the uncoupled tls - ho system are immediately found by applying the hamiltonian @xmath63 on the eigenstates in : @xmath64 the dashed lines in show the energy spectrum corresponding to vs. the oscillator frequency @xmath0 for the five lowest eigenstates . except for the groundstate , @xmath65 , the states @xmath66 and @xmath67 are degenerate in the resonant case @xmath68 . close to resonance the spectrum exhibits a doublet structure . with the coupling being switched on , the full hamiltonian @xmath69 reads @xmath70 in the basis @xmath71 . in order to diagonalize the hamiltonian @xmath72 we consider @xmath73 as a small perturbation , which is resonable as long as @xmath74 . applying van - vleck perturbation theory we construct an effective hamiltonian , @xmath75 having the same eigenvalues as @xmath76 but no matrix elements connecting states which are far off from degeneracy . thus , @xmath77 will be block - diagonal with all quasi - degenerate energy levels being in one common block . as in our case always two states are nearly degenerate , each block of @xmath78 builds a two - by - two matrix . this can be easily diagonalized in order to determine the eigenstates . following @xcite we calculate the transformation matrix @xmath79 up to second order in @xmath1 . the general formulas for both an arbitrary hamiltonian and @xmath76 are given in [ appvanvleck ] . the only surviving matrix elements of the effective hamiltonian , apart from the ones being of zeroth order in @xmath1 , are @xmath80 and @xmath81 . \label{heff2}\end{aligned}\ ] ] thus , @xmath82 has the matrix structure @xmath83 where the section shown corresponds to the basis states @xmath84 and @xmath85 . from this form it is easy to calculate the eigenstates and eigenenergies . the groundstate @xmath86 , which is an eigenstate of @xmath87 , has the eigenenergy @xmath88 the other eigenstates of @xmath87 are , @xmath89 , @xmath90 corresponding to the eigenenergies @xmath91\\ = \hbar \left [ ( j+\frac{1}{2 } ) \omega + w_1 + w_0 \mp \frac{1}{2 } \sqrt{\delta_j^2 + 4(j+1 ) |\delta|^2 } \right ] , } \ ] ] @xmath92 by construction these are also eigenenergies of @xmath93 . using the transformation ( [ vanvlecktrafo ] ) we get the eigenvectors of @xmath93 as @xmath94 the energy spectrum of @xmath93 is shown in for the case of an unbiased tls ( @xmath95 ) . we want to emphasize that our findings are also valid for the more general case @xmath96 . at resonance , where the spectrum for the uncoupled case is degenerated , avoided crossings can be seen . the gap between two formerly degenerated levels for @xmath97 is @xmath98 which is as predicted by the jaynes - cumming model @xcite . as we will show in , the second order correction @xmath99 in , whichleads to a shift in the resonance frequency , is a result of the counter - rotating terms in @xmath93 . as such it can be interpreted as a bloch - siegert shift @xcite . with the coupling to the bath being turned off , the time evolution of the density matrix of the qubit - ho system is given by @xmath100 and consequently @xmath101 with @xmath102 . with that becomes @xmath103 where we defined @xmath104 . from we notice that the dynamics of the qubit is characterized by an infinite number of oscillation frequencies rather than showing rabi oscillations with a single distinct frequency . this is clearly a consequence of the coupling of the ho to the tls . further we assume that at @xmath105 the qubit starts in the state @xmath51 and that the occupation numbers of the ho are boltzmann distributed , so that @xmath106 where @xmath107 is the partition function of the oscillator and @xmath108 denotes the inverse temperature of the system . in the tls - ho eigenbasis this becomes @xmath109\\ \times \left [ \cos \left ( \frac{\theta}{2 } \right ) { \langle j{\rm g}|m\rangle } + \sin \left ( \frac{\theta}{2 } \right ) { \langle j { \rm e}|m\rangle } \right].}\ ] ] with we found a formula which describes using the approximate eigenenergies and eigenstates in and the non - dissipative dynamics up to second order in @xmath1 , thereby taking into account all oscillator levels . thus , we still have to deal with an infinite hilbert space . typically experiments , see e.g. in @xcite , run in a temperature regime for which @xmath110 . considering the exponential function in we assume the higher oscillator levels to be only sparsely populated and the maximum value of the sum in is truncated to @xmath111 . nevertheless , states @xmath112 with @xmath113 still play a role in the dynamics . in fact , due to the van - vleck transformation @xmath114 , for example the state @xmath115\ ] ] yields nonvanishing contributions to the matrix elements @xmath116 and @xmath117 occurring in due to the fact that the energy eigenstates of the coupled tls - ho system are made of linear combinations which involve also these states . using together with one finds that coefficients @xmath118 with @xmath119 are of higher than second order in @xmath1 . the same is valid for @xmath120 , @xmath121 , @xmath122 and @xmath123 . thus , those terms play no role in our calculation of @xmath49 . furthermore , @xmath124 neglecting also these contributions we find that @xmath125 for @xmath126 . in the end it will be sufficient to concentrate on eigenstates of @xmath93 up to @xmath127 . this trunctation leaves us with ten possible oscillation frequencies @xmath128 , where @xmath129 and @xmath130 . as an example we calculate the dynamics of an unbiased tls ( @xmath95 ) . here the coefficients @xmath131 , @xmath132 , @xmath133 , @xmath134 and @xmath135 vanish due to symmetry , so that @xmath136 additionally as a benchmark we consider the mostly studied resonant case , where @xmath137 . in this case we find with the transitions frequencies @xmath138 the dynamical quantity @xmath49 and its fourier transform are shown in . one clearly sees the influence of the coupling to the ho on the dynamics of the tls . instead of rabi oscillations with a single frequency , @xmath49 oscillates with six different frequencies , which are in the fourier spectrum symmetrically located around the point @xmath139 . among those frequencies @xmath140 and @xmath141 are dominating . they correspond to transitions between the first or second energy level of the qubit - ho system and its groundstate and their weight is almost equal . to summarize , one notices that due to the coupling with the oscillator additional frequencies are induced into the qubit dynamics . theoretically , the number of those frequencies is infinite . at low temperatures , however , transitions between the lower energy levels of the system are clearly dominating . again , for simplicity we have shown here the case of an unbiased tls . for @xmath96 the behaviour is similar only that in the fourier spectrum the weight difference of the two dominating peaks will be more pronounced . in the preceding section we neglected the influence of the bath on the qubit - ho system . yet , in order to model a realistic situation , we have to pay attention to environmental influences , as they lead to decoherence and dissipation in the dynamics of the qubit , which is harmful for quantum computing application . thus , we will now consider the full hamiltonian @xmath142 . as shown in , we need for the calculation of the population difference @xmath49 the density matrix @xmath56 of the qubit - ho system . starting from the liouville equation of motion for the full density matrix @xmath143 of @xmath142 , @xmath144,\ ] ] where the index stands for the interaction picture and following @xcite , we can provide a born - markov master equation for @xmath56 being in the schrdinger picture and expressed in the basis of the eigenstates of @xmath93 : @xmath145 the free dynamics of the system is given by the first term of the right - hand side in the above equation . the rate coefficients are defined as @xmath146 x_{nk } x_{lm } \\ -\delta_{ml } \sum_{l^\prime } g(\omega_{l^\prime k } ) n_{l^\prime k } x_{n l^\prime } x_{l^\prime k } + \delta_{nk } \sum_{k^\prime } g(\omega_{l k^\prime } ) n_{k^\prime l } x_{l k^\prime } x_{k^\prime m}}\ ] ] with @xmath147 $ ] and @xmath148 . for the derivation of the master equation , besides the born and markov approximations , some more assumptions have been made , which we briefly mention . first , we consider our system and the bath to be initially ( at @xmath105 ) uncorrelated ; i.e. , @xmath149 with @xmath150 and @xmath151 the partition function of the bath . further , with the bath consisting of infinite degrees of freedom , we assume the effects of the interaction with the qubit - ho system to dissipate away quickly , such that the bath remains in thermal equilibrium for all times @xmath152 : @xmath153 besides , an initial slip term which occurs due to the sudden coupling of the system to the bath is neglected @xcite . and as last approximation the lamb - shift of the oscillation frequencies @xmath128 was not taken into account @xcite . in @xmath154 describes matrix elements of the operator @xmath155 in the qubit - ho eigenbasis . by use of , and those states were expressed in the basis @xmath156 , and we will also calculate the oscillator matrix elements in this basis . for that purpose the operator @xmath157 is defined . four different situations can be distinguished . there are matrix elements were neither the oscillator nor the qubit state are changed , namely @xmath158 and @xmath159 with @xmath160 . we see that those elements are independent of @xmath161 , the occupation number of the oscillator . next , we look at the case where a single quantum is emitted or absorbed from the oscillator and get @xmath162 and @xmath163 with @xmath164 for a transition within the qubit we have @xmath165 . and finally , if the qubit and the oscillator state are changed simultaneously , one obtains @xmath166 and @xmath167 , @xmath168 comparing the magnitude of the transition terms , we notice that those consisting in changes of the oscillator occupation only are the dominant ones , as they have a part which is of zeroth order in @xmath1 . further , for the case in which the qubit is operated at the degeneracy point @xmath169 and @xmath170 vanish . with those results we can calculate the matrix elements @xmath154 . they are given in [ apposcillatormatrixelements ] . like in we assume the system to be operated at low temperatures and thus take as highest qubit - ho state the eigenstate @xmath127 . for determination of @xmath49 the formulas of can be used . unlike in the non - dissipative case @xmath56 is not given anymore by the simple expression . rather we have to solve a system of coupled differential equations , namely . to do this analytically we will follow three different approaches and compare them finally to the numerical solution of . we start by introducing @xmath171 which yields the set of differential equations for @xmath172 @xmath173 as a first approach we make the full secular approximation ; i.e. , we neglect fast rotating terms in and keep only contributions where @xmath174 vanishes . in this way the off - diagonal elements of @xmath175 are decoupled from the diagonal ones so that @xmath176 the equation for the off - diagonal elements is then @xmath177 which becomes with @xmath178 as through the fsa the oscillatory motion of the dynamics is separated from the relaxation one we can divide into two parts , @xmath179 where @xmath180 describes the relaxation and @xmath181 the dephasing parts of the dynamics . with the latter takes the form @xmath182 with the dephasing rates @xmath183 . expressions for the dephasing coefficients @xmath184 can be found in [ ratecoeff ] and the initial conditions @xmath185 are given by . the diagonal elements are more difficult to obtain , as one has to solve a system of coupled differential equations , . calculating the corresponding rate coefficients of this system for the five lowest eigenstates , we find that there are only eight independent ones , namely @xmath186 , @xmath187 , @xmath188 , @xmath189 , @xmath190 , @xmath191 , @xmath192 and @xmath193 . they are given by @xmath194 where @xmath161 and @xmath195 adopt the above values . furthermore , @xmath196 , @xmath197 , @xmath198 and @xmath199 vanish . the remaining rate coefficients are combinations of the above . we find that @xmath200 and @xmath201 however , the system is still too complicated to be solved analytically . thus , we invoke a further approximation : we consider the factor @xmath202 $ ] with @xmath203 in and use that @xmath204 . it depends strongly on the temperature @xmath205 for which value of @xmath128 this limit is reached approximately . for the parameters we are working with one usually is in the region where @xmath206 . thus , we will neglect in the following terms containing the factor @xmath207 . furthermore , one sees from that @xmath208 and @xmath209 . with that @xmath210 and @xmath211 can be neglected . using the matrix of the system ( [ diagonalfsa ] ) becomes @xmath212 the eigenvalues and eigenvectors of this matrix and the associated time evolution of the elements @xmath213 are given in of [ appdiagonalelements ] . unlike for the dephasing part , we can not extract a simple analytical expression for the relaxation rate as @xmath214 now consists of a sum of several exponential functions , cf . together with . but still we are able to provide an analytical formula for @xmath49 using . in order to obtain a simple expression for the relaxation part , we consider the long - time dynamics of the system . in other words , rather than looking at the many relaxation contributions to the populations @xmath213 , we focus on the smallest eigenvalue of the relaxation coefficients , as it will dominate at long times . further , we consider only the rate matrix associated to the three lowest qubit - ho eigenstates , @xmath215 , @xmath216 and @xmath217 in and obtain with - that @xmath218 here , we have not neglected the rate coefficients containing the term @xmath207 and further took @xmath188 into account despite of being of third order in @xmath1 as such contribution removes the degeneracy between the two lowest eigenvalues at resonance , cf . the smallest eigenvalue reads @xmath219^{1/2 } \biggr\}.\end{aligned}\ ] ] with the system being detuned this expression can be simplified further , namely @xmath220 in the relaxation rate @xmath221 as obtained from is shown as a function of the oscillator frequency @xmath0 . clearly , it is maximal at resonance ( @xmath222 ) , whereas it decays with @xmath0 being detuned from the resonance . this effect has already been predicted by blais @xcite . as the qubit is not directly coupled to the bath but rather through the oscillator , the latter being detuned filters out the environmental noise at the qubit transition frequency . additionally , we show the second smallest eigenvalue of . we notice that close to the resonant point ( @xmath222 ) there is an avoided crossing . finally , we find that @xmath223 where like in @xmath104 . for getting @xmath224 we have in principle to find the steady - state solution of . here , we just assume for @xmath225 a boltzmann distribution for the qubit - ho system , so that @xmath226 with @xmath227 . thus , @xmath228 + 2 \sin \theta { \langle j { \rm g}|n\rangle } { \langle j { \rm e}|n\rangle } \right\ } \rho_{n n}(\infty ) \label{pinf}.\ ] ] the formula for the long - time dynamics is obtained as @xmath229 to get further insight on the dominant frequencies we evaluate the fourier transform of according to @xmath230 yielding @xmath231.\end{aligned}\ ] ] an improvement to the fsa is to take into account certain non - vanishing contributions of @xmath174 . we have to keep in mind , that there are quasi - degenerate levels close to resonance . in our case the first with second energy level and the third with fourth one build a doublet , meaning that they are close together in energy space . the level spacing is approximately proportional to @xmath1 for the former and @xmath232 for the latter . because of that and as the transitions from level three and four are less probable , we will in the following only consider the first and second level as being almost degenerate . taking this into account in we arrive for the diagonal elements at @xmath233 a numerical analysis shows that the effect of the last two terms on the right - hand side of the above equation will in the worst case lead to very small wiggles in @xmath213 and play no role in @xmath49 . thus , we finally write @xmath234 which is the same equation as we got in the fsa approach . however , the off - diagonal contributions @xmath235 , @xmath236 , @xmath237 , @xmath238 , @xmath239 and @xmath240 have to be examined more carefully . from ( [ masterwithansatz ] ) we find that one has to solve the equations @xmath241 with @xmath242,@xmath243 or @xmath244 as solution one gets @xmath245 here , the oscillation frequencies and the decay of the off - diagonal elements are given by @xmath246\ ] ] with @xmath247 ^ 2 + 4 \pi^2 \mathcal{l}_{nm , jk } \mathcal{l}_{jk , nm}}.\ ] ] the amplitudes of the oscillations are given through the coefficients @xmath248}{2 r_{nm , jk}}\ ] ] and @xmath249.\ ] ] thus , we have again all ingredients to calculate analytically the relaxation and dephasing part of . for the psa we can not provide a simple expression for the dephasing rates as in the fsa , where we had @xmath250 . as one can see from and , @xmath251 and @xmath252 are a mixture of contributions decaying with @xmath253 and @xmath254 . similar to our findings for the relaxation rate , also here the smallest eigenvalue will dominate the dephasing behaviour . from the right graph in we find that this is @xmath255 . comparing it with the dephasing rates we got using the fsa , left graph in , we see that for negative detuning ( @xmath256 ) the rate @xmath257 approximates @xmath258 , whereas for positive detuning ( @xmath259 ) this is done by @xmath260 . in the fsa @xmath261 and @xmath262 correspond to the frequencies @xmath140 and @xmath141 , respectively . in the psa the frequency @xmath263 is given by @xmath264 for @xmath256 and @xmath265 for @xmath259 . hence , for negative detuning oscillations with frequency @xmath141 will dominate the dynamics , while those with @xmath140 will almost vanish . for positive detuning it is the other way round . in and this behaviour is reflected by the coefficients @xmath266 and @xmath267 . around resonance ( @xmath268 ) the psa tells us by and that the dephasing rates and frequencies are a mixture of @xmath262 and @xmath261 or @xmath140 and @xmath141 , respectively . from the left graph in one notices further that the fsa rate @xmath261 grows linearly with @xmath0 for positive detuning . however , as the weight of the corresponding frequency @xmath141 will be almost zero , @xmath261 will give no relevant contribution to @xmath269 in this regime but the dephasing will rather be associated to the fsa rate @xmath262 . hence , out of resonance the fsa will still fairly well describe the dynamics of @xmath49 . comparing the expressions for @xmath270 and @xmath271 given in [ ratecoeff ] by and with the approximative expressions for the relaxation rate at positive and negative detuning , we see that for zero bias ( @xmath272 ) the psa dephasing rate is equal to @xmath273 . for a biased system an additional term is added depending on the spectral density of the bath at @xmath274 . in we compare the three analytical solutions described above to the numerical solution of the master equation for the case of an unbiased tls being at resonance with the oscillator . concerning both the dynamics of @xmath49 and its fourier spectrum we see a good agreement between the different solutions . the one being closest to the numerical solution is the psa solution . we also want to mention that going to stronger damping @xmath275 , the fsa results start to show deviations from the numerical solution . here , one should use the psa only . however , for the parameter regime used in the following , we will mainly apply due to its simple , analytical form . having solved the master equation analytically and numerically we can examine the dynamics of the system and its fourier transform for different situations . first , we will look at a qubit operated at the degeneracy point ( @xmath48 ) being in and out of resonance with the oscillator . then , we will concentrate on the biased qubit in the same regime of parameters . for unbiased qubits we can compare our predictions with the analytical results obtained in @xcite by starting from a spin - boson model with the effective spectral density . in @xcite a so - called weak damping approximation ( wda ) based on the non - interacting blip approximation ( niba ) is applied . the wda allows a non - perturbative treatment of the coupling between the tls and ho and hence can reproduce the occurence of two dominating frequencies as expected e.g. from exact quapi calculations @xcite . the niba , and hence the wda , however , become not reliable for a biased tls . we find that the overall agreement between our approach and the wda is very good . however , in the wda solution the frequencies are slightly shifted compared to the ones obtained from our master equation . this may result from the perturbative expansion we have performed with respect to @xmath1 by applying the van - vleck perturbation theory . first , we look at the resonant case shown in . in agreement with previous works @xcite , we find that the dynamics is dominated by two frequencies corresponding to @xmath140 and @xmath141 with separation being approximately @xmath276 . the weight of the latter is a bit larger . the reason for the bigger weight is that at resonance ( @xmath222 ) the qubit - ho eigenstate @xmath277 is not a symmetric superposition of the states @xmath278 and @xmath279 unlike it is predicted by the janyes - cummings model ( cf e.g. @xcite ) . we notice that the two unequal peaks have indeed been experimentally observed in @xcite ( see fig . 4b therein ) . considering the states @xmath280 and @xmath281 in and , one already sees that for a symmetric superposition of these states we need that @xmath282 vanishes or that @xmath283^{-\frac{1}{2}}$ ] ( cf . ) . besides , in order to get the qubit - ho eigenstates one still has to perform the van - vleck transformation , which adds contributions to @xmath216 and @xmath217 from states corresponding to oscillator levels higher than @xmath111 . thus , our system behaves for @xmath284 as being negatively detuned , which means that the peak belonging to the higher frequency dominates , as we will show below . slightly increasing @xmath0 will give a stronger weight to the peak at @xmath140 . this effect is not very pronounced for the non - dissipative dynamics of the unbiased qubit ( ) , as there the two frequencies are still almost equally weighted . looking however at the fourier transform of the dissipative dynamics , one notices that the relaxation rate also contributes to the weight of the peaks with a prefactor @xmath285 . as for a negative detuned system @xmath262 is slightly bigger than @xmath261 , the difference between the two peaks becomes more clear in the dissipative case . for @xmath96 the effect can already be noticed in the non - dissipative case.next , we consider in the case of negative detuning , where @xmath286 . no matter which approach one is looking at , clearly the frequency @xmath141 is dominating . furthermore , paying attention to the timescale of the dynamics , one notices that the relaxation time is _ enhanced _ compared to the one we found for the resonant system . this behaviour was already explained by the formula for the relaxation rate . again , the numerical and the solution obtained by using the long - time ansatz in agree quite well with each other , whereas the amplitude of the oscillation with frequency @xmath141 is stronger in the wda approach . also remarkable is the fact that looking at the fourier transform in one sees in the inset already small contributions of the higher oscillator levels . the transitions corresponding to @xmath287 and @xmath288 give raise to small additional peaks , while the contributions of @xmath289 and @xmath290 are negatively weighted and cause dips . the wda approach does not show this additional contributions . they are , however , confirmed by the numerical quapi calculations in @xcite ( see figure 2 therein ) . in the case of positive detuning ( @xmath291 ) shown in we find a quite good agreement between all three approaches . also for postive detuning the relaxation time is enhanced compared to the resonant case . contrary to the negatively detuned situation the additional peaks have vanished . besides , now the frequency @xmath140 is dominating the dynamics . this behaviour , namely that for negative detuning @xmath141 and for positive detuning @xmath140 is dominating , was already found in @xcite . we will briefly explain how one can explain this observation physically . for this we look at . for a detuned system ( @xmath292 ) the qubit - ho eigenstates are not symmetric superpositions of the states @xmath293 and @xmath294 . they rather asymptotically approach the eigenstates of the uncoupled qubit - ho hamiltonian . in we see that for a negatively detuned system ( line a ) the qubit - ho eigenstate @xmath295 approaches the state @xmath296 , whereas the main contribution to the state @xmath297 will come from the state @xmath298 . from the left diagram in we see that the state @xmath217 is energetically higher than the state @xmath216 . however , due to the boltzmann distributed occupation of the oscillator , the state @xmath299 will be more populated than the state @xmath300 and consequently also @xmath217 will exhibit a larger population than @xmath216 , as the latter only feels a small contribution from the state @xmath299 . thus , transitions from @xmath217 to the groundstate are more likely to occur than those from @xmath216 to the groundstate . this explains the dominance of @xmath141 in and . in this case the frequency @xmath301 and @xmath302 . as far as not excluded by selection rules , minor peaks from transitions to the levels lying in between can be also seen.for positive detuning ( line c in ) @xmath295 approaches @xmath298 , while @xmath297 is close to @xmath296 . from the right graph in we see that the state @xmath216 , being lowest in energy apart from the groundstate , is now also more probable to be occupied than @xmath217 . therefore , as confirmed by and , the frequency @xmath140 is dominating whereas @xmath141 is represented only by a small peak in the fourier spectrum . furthermore , as there are no additional energy levels between the state @xmath216 , which is most probably to be populated , and the ground level , other transitions than those corresponding to @xmath140 or @xmath141 are very unlikely to occur . in the dip corresponding to @xmath303 appears only very faintly . we will now examine a qubit being operated at finite bias . we consider the case @xmath304 . for negative bias - offset the behaviour is analogous . again three different situations are taken into account : the qubit being in resonance with the oscillator ( @xmath305 ) , negative ( @xmath306 ) and positive ( @xmath307 ) detuning.for the resonant case ( @xmath222 ) depicted in we see a similiar behaviour as for the unbiased qubit . again two frequencies , @xmath140 and @xmath141 , are dominating the dynamics . left to the peak at @xmath140 a small dip can be found in the fourier spectrum . this corresponds to the transition @xmath303 . for infinite time the dynamics relaxes to an equilibrium value which is nonzero in contrast to the unbiased case . this can be seen in the fourier spectrum through a relaxation peak at @xmath308 . the peak arises because of the term @xmath309 in . the first part of this sum gives rise to the negative shift of this peak . the reason that for the analytical solution the peak is not as strongly shifted as for the numerical one is technical : in order to plot the delta function in we gave it a finite width , which surpresses the negative contribution of the first term in . like for the unbiased qubit the highest energy level playing a role for the dynamics is @xmath310 ; i.e. , only the ground and first excited level of the oscillator are of importance . in the dynamics and its fourier transform for a negatively detuned qubit - oscillator system with @xmath96 are shown . like for the unbiased case detuning gives raise to longer relaxation times for the qubit . also in agreement with the unbiased case is the dominance of the frequency @xmath141 . we see that for small @xmath152 the long - time solution slightly overestimates the maxima of the oscillations and underestimats its minima . furthermore , we get here the unphysical situation that the maximum of the third oscillation in @xmath49 exceeds the value of one . the reason for that behaviour is that , by construction , we underestimate with the relaxation at short times . as for certain paramteres the term @xmath311 in can become negative , it increases too fast towards the equilibrium and gives thus raise to the observed deviations in the short time behaviour . on a longer timescale both graphs agree quite well . for the case of positive detuning ( @xmath259 ) , which is presented in , the upward shift of the dynamics obtained from and compared to the numerical graph of @xmath49 at small times is even stronger . to visualize that it is not a failure of the fsa approach we show in and additionally the analytical fsa solution of and calculated in . the latter agrees very well with the numerical solution . at long time - scales and for the fourier spectrum all three approaches match with each other very well.to conclude this paragraph we want to mention that all the results found both for the unbiased and the biased qubit confirm the numerical quapi results in @xcite . so far we have always considered the qubit for certain values of @xmath14 and finite or zero detuning . in this section , we fix the oscillator frequency at @xmath284 . that means that an unbiased qubit will be at resonance with the oscillator . changing the bias to positive or negative values will always lead to negative detuning , as @xmath312 . shows a density plot of the fourier transform of the symmetrized correlation function against the bias of the qubit and the fourier frequency @xmath313 . we consider this correlation function rather than @xmath49 , as it is symmetric in the bias @xmath14 . the symmetrized correlation function is defined as follows @xcite : @xmath314 where @xmath315 . expressed in terms of the population difference @xmath49 this becomes , @xmath316 with @xmath317 and @xmath318 being symmetric and antisymmetric in @xmath14 and @xmath319 . the fourier transform of @xmath320 is defined as @xmath321 considering now we see that for any bias the spectrum is dominated by two frequencies , namely @xmath140 and @xmath141 . detuning the system @xmath141 gets more and more important , as we could already observe in the two previous sections for the positively detuned systems . furthermore , the peaks are shifted to higher frequency values and at @xmath322 the relaxation peak occurs . we want to compare these results to a circuit qed experiment performed by wallraff @xcite . there the qubit is realized by a cooper pair box , which is coupled to a superconducting transmission line resonator . the properties of the system are determined by probing the resonator spectroscopically . the amplitude of a microwave probe beam transmitted through the resonator is measured versus the probe frequency and the gate charge of the cooper pair box ( see figure 4 in @xcite ) . via the gate charge the qubit can be detuned in situ from the degeneracy point . the frequency of the resonator is chosen in such a way that it is in resonance with a qubit being operated at the degeneracy point . for the resonant case two dominating frequencies , being almost equally weighted and symmetrically positioned around the cavity frequency , are observed . going away from the degeneracy point the system becomes detuned and the frequency of the cavity dominates . the behaviour we observe in is similar . however , as we are looking at the dynamics of the qubit , it corresponds to a spectroscopic measurement on the tls rather than on the oscillator . as explained above the two lowest excited states of the coupled tls - ho system , namely @xmath216 and @xmath217 , evolve from an almost symmetric superposition of basis states @xmath323 at resonance @xmath324 to the states @xmath325 and @xmath326 ( cf the left graph in ) . for @xmath327 the two peaks of the rabi splitting are observed . for @xmath96 , which means negative detuning in this case , the peak with the lower frequency corresponding to @xmath140 approaches more and more the frequency @xmath0 of the oscillator , as the state @xmath216 becomes @xmath325 for large detuning and then @xmath328 . furthermore , the transition peak at @xmath140 gets weaker as also the occupation probability of @xmath216 decreases . at @xmath329 the symmetrized correlation function vanishes at @xmath140 and increases again for higher values of @xmath330 . here , the amplitude @xmath331 in @xmath49 changes its sign . in contrast the peak at @xmath141 becomes stronger with the detuning and approaches more and more the qubit splitting energy @xmath332 , as @xmath217 approaches @xmath299 and then @xmath333 . additionally , looking at the logarithmic plot one sees around @xmath334 a peak appearing , which corresponds to the frequency @xmath303 and is forbidden at @xmath95 . for large detuning it arises from transitions from @xmath299 to @xmath325 and therefore has the value @xmath335 . the amplitude of this peak is very small compared to the peaks at @xmath140 and @xmath141 and is not resolved in the experiment of wallraff . van - vleck perturbation theory enabled us to find approximately the eigenstates and eigenenergies of the full hamiltonian of the tls - ho system without performing a rotating - wave approximation . using those eigenstates and eigenenergies in a born - markov master equation we could calculate the dynamics of such a system under the influence of an environmental bath . in the following we will show how the results change if we neglect counter - rotating terms in the tls - ho hamiltonian for @xmath95 . for this we rewrite the interaction part in as @xmath336 where we identified with @xmath337 and @xmath338 a rotating and counter - rotating part of @xmath339 , respectively , and introduced the two - level transition operators @xmath340 . neglecting the counter - rotating part @xmath338 in @xmath93 leads to the jaynes - cummings hamiltonian @xmath341 this hamiltonian can be diagonalized exactly and its eigenstate and eigenvalues can for example be found in @xcite . in order to see the effect of not taking into account the counter - rotating terms we diagonalize @xmath342 using van - vleck perturbation theory . looking at the formula fo the effective hamiltonian in [ appvanvleck ] and keeping in mind that we set @xmath14 to zero , we see that the second order contributions in @xmath1 vanish neglecting @xmath338 ; i.e. , @xmath343 , @xmath344 are zero in . considering further the transformation matrix @xmath79 , equations and in [ appvanvleck ] show that @xmath345 for @xmath346 . thus , with we find that _ for the jaynes - cummings model @xmath347 is identical to @xmath342 _ and therefore the eigenstates of @xmath347 are simultaneously eigenstates of @xmath342 . consequently , one can determine from the eigenstates and eigenenergies of @xmath342 . the energy of the groundstate @xmath348 is @xmath349 . for the higher states we get @xmath350 corresponding to the eigenenergies @xmath351 = \hbar \left [ ( j+\frac{1}{2 } ) \omega \mp \frac{1}{2 } \sqrt{\delta_{\rm jc}^2 + 4(j+1 ) |\delta|^2 } \right],\ ] ] with @xmath352 and @xmath353 . comparing these eigenstates and eigenenergies to the ones found for @xmath93 , , and , we see that the counter - rotating terms yield second order corrections in @xmath1 not present in the jaynes - cummings hamiltonian . these corrections give rise to a very prominent effect concerning the resonance condition between tls and ho . from @xmath354 we find the tls being in resonance with the oscillator for @xmath97 . considering @xmath355 in this resonance condition is shifted to @xmath356.\ ] ] this second order correction to the resonance frequency due to counter - rotating terms is known as bloch - siegert shift @xcite . the eigenstates and are always a superposition of two basis states of the unperturbed system . this is like for the eigenstates and of the effective hamiltonian . however , in order to find the eigenstates @xmath357 of @xmath358 we had to apply the transformation @xmath359 on the effective eigenstates so that @xmath357 is in the end a superpostion of _ several _ states of the basis @xmath360.to calculate the reduced density matrix of the qubit - ho system described by @xmath342 and taking into account the influence of an environmental oscillator bath we can again use the born - markov master equation . we just have to use and as basis states and the corresponding eigenenergies . for the population difference @xmath49 we apply , which becomes for @xmath48 @xmath361 \re \ { \rho_{nm}(t ) \}.\end{aligned}\ ] ] before doing a qualitative comparison between the results obtained from the original hamiltonian @xmath93 and the simplified form @xmath342 , we want to analyse which transitions between the different eigenstates of @xmath342 yield contributions to @xmath49 . for this it is helpful to rewrite and . in the following we neglect the upper index jc denoting an eigenstate of @xmath342 . for the state @xmath357 we have three possibilities : the first one corresponds to @xmath362 . in this case the only non - vanishing component of @xmath215 is @xmath363 . second , @xmath364 can be an even number , which means , expressed in terms of oscillator quanta , that @xmath365 . then @xmath366 and third for an odd state @xmath367 we find @xmath368 it is quite easy to see that the part @xmath369 in vanishes for any @xmath364 . that means that the diagonal elements of the reduced density matrix yield no contributions to @xmath49 for @xmath48 and the equilibrium value of the dynamics will be also zero . not so easy to see are the combinations of @xmath364 and @xmath370 yielding contributions to the off - diagonal part @xmath371 . considering transitions to the groundstate ( @xmath372 ) , we find using that @xmath373 that @xmath374 with ( [ evenstate ] ) and ( [ oddstate ] ) these elements are non - zero only if @xmath375 or @xmath376 . for the more general case with @xmath377 and both being different from zero we have to investigate products of components like @xmath378 . for @xmath379 we find that @xmath380 and that @xmath381 so that @xmath382 for the case of transitions between odd states , one gets that @xmath383 and similarly under exchange of @xmath384 and @xmath385 , so that @xmath386 for transitions from an even to an odd state and vice versa we have to pay attention to the fact that @xmath387 which yields the selection rule @xmath388 . further , @xmath389 yielding @xmath390 . to sum up : @xmath371 is non - zero if one of the three following cases is valid : @xmath391 , @xmath392 , @xmath388 or @xmath390 . the allowed transitions are shown in figure [ figselecrules ] . one sees that transitions between almost degenerate levels are forbidden . this behaviour we have also found in for the non - dissipative dynamics resulting from @xmath93 . in the following we compare the numerical solution of the born - markov master equation originating from the jaynes - cummings hamiltonian with the solution using the eigenstates of the full hamiltonian @xmath93 , which we found by applying van - vleck perturbation theory . the fixed parameters we use are in units of @xmath16 : @xmath95 , @xmath393 , @xmath394 and @xmath395 . the oscillator frequency @xmath0 is varied . for all the three possible cases ( positive , negative and zero detuning ) one notices from figures [ positivedetuning ] [ nodetuning ] that the jaynes - cummings approach _ underestimates _ the dephasing rate @xmath396 ( means a larger peak at frequency @xmath140 ) and _ overestimates _ the rate @xmath397 ( smaller peak at @xmath141 ) compared to the approach with the full hamiltonian.the case of positive detuning ( @xmath398 ) is shown in figure [ positivedetuning ] . as here the dynamics is dominated by the frequency @xmath140 , the equilibrium value is reached on a too long time scale using the jaynes - cummings hamiltonian . on the contrary , for negative detuning ( @xmath286 ) , which is shown in figure [ negativedetuning ] and where @xmath141 dominates , the equilibrium value is reached too fast within the jaynes - cummings approach . furthermore , considering the graph of the fourier transform we find that small contributions which come from higher level transitions , and which have already been discussed in , are not caught be the jaynes - cummings approach . for the resonant case ( @xmath284 ) we find that the jaynes - cummings method predicts @xmath140 to be slightly dominating whereas the approach starting from @xmath6 results in @xmath141 being dominating . the reason for this discrepancy is that due to the counter - rotating terms we have for @xmath6 no symmetric or antisymmetric superposition of the unperturbed eigenstates at @xmath284 in contrast to the jaynes - cummings model . to conclude this section we can say that for an unbiased tls - ho system the jaynes - cummings model gives a good insight in the qualitative behaviour of @xmath49 both for a slightly detuned and a non - detuned system . however , it under- or overestimates dephasing times for the system . furthermore , we find taking into account counter - rotating terms in @xmath358 that at @xmath222 the dressed eigenstates are not a symmetric or antisymmetric superposition of the uncoupled states . moreover , the effects of transitions between states of different manifolds are neglected . in conclusion , we discussed the dynamics of a biased and unbiased tls coupled through a harmonic oscillator to an environmental bath described by an ohmic spectral density . in particular , we examined the regime of weak damping and moderate coupling between oscillator and tls . an equivalent description of our system is provided by the spin - boson model with a structured spectral density . in contrast to many other works in this field , our starting point was not the jaynes - cummings hamiltonian , but a more general one given in , where no initial rotating wave approximation has been applied . in we provided with a formal expression for the population difference and showed in how the hamiltonian of the coupled qubit - ho system can be diagonalized approximately using van - vleck perturbation theory . this approach is valid both for the oscillator being in resonance with the qubit ( @xmath222 ) and for finite detuning ( @xmath399 ) . in an analytical expression for the non - dissipative dynamics was provided up to second order in the qubit - ho coupling @xmath1 taking into account the infinite hilbert space of the system . for low temperatures ( @xmath400 ) we truncated the hilbert space and found that transition processes between the groundstate and the two first excited energy levels of the qubit - ho system dominate the dynamics ( ) . in the influence of the bath was taken into account by solving the born - markov master equation for the density matrix of the qubit - ho system . to do this analytically we considered two variants of the secular approximation : first , in , the full secular approximation ( fsa ) , where all fast oscillating terms are neglected , and second , in , the partial secular approximation , where attention was paid to the fact that the first two excited energy levels are almost degenerate . using an ansatz for the long time dynamics in , we could provide a general expression for the relaxation and dephasing rates of the qubit , showing that the relaxation time can be enhanced by detuning the oscillator into the off - resonant regime . it was found that all three approaches agree quite well with the numerical solution of the master equation . the dynamics of both a biased and an unbiased qubit were intesively studied for zero and finite detuning in . the results agree qualitatively with the numerical findings within the ab - initio quapi approach @xcite . furthermore , in a good agreement with the results of the weak damping approximation performed in @xcite for a symmetric spin - boson model was found . besides , we saw that at resonance ( @xmath222 ) the first two excited qubit oscillator states are not a symmetric or antisymmetric superposition of the states @xmath401 and @xmath325 , as predicted by the jaynes - cummings model , and thus could give an explanation for the differently weighted peaks of @xmath140 and @xmath141 in the fourier spectrum of the dynamics . we further could explain the dominance of frequency @xmath141 in the case of negative detuning @xmath402 and of frequency @xmath140 for positive detuning , respectively . moreover , we showed that for large negative detuning @xmath141 approaches the energy splitting @xmath403 of the qubit , whereas @xmath140 approximates the oscillator frequency @xmath0 . this behaviour agrees nicely with spectroscopic experiments performed on a circuit qed architecture @xcite . in we compared our results for an unbiased system to the ones obtained starting with the jaynes - cummings model . we visualize the effects of the counter - rotating terms in the bloch - siegert shift of the resonance frequency and in contributions of states with larger oscillator number to the tls dynamics . apart from this the jaynes - cummings model and our approach agree quite well for the non - dissipative case . also for the dissipative case at resonance an initial rwa represents a good approximation to our starting hamiltonian and seems to be favourable as it is analytically exactly diagonalizable . for detuned systems , however , we find discrepancies concerning relaxation and dephasing times and a rwa becomes less appropriate to give precise results . thus , we think that our approach represents an improvement as it is valid in a wider parameter range avoiding an initial rotating wave approximation . to our knowledge it provides for the first time analytical results for the dynamics of an unbiased _ and _ biased qubit coupled to a structured environment being valid both in the resonant _ and _ off - resonant regime . furthermore , due to the generality of the qubit - oscillator model , we expect our results to be of interest for a wide range of experimental applications . we acknowledge financial support under dfg program sfb631 . further we would like to thank g. begemann for his help with some of the figures . here , we show how to derive given in . in order to trace out the oscillator degrees of freedom we transform @xmath404 into the basis @xmath71 and by using @xmath405 we find @xmath406 performing the trace over the oscillator @xmath407 similarily , we find for the off - diagonal elements of the reduced density matrix @xmath408 , \end{aligned}\ ] ] @xmath409.\end{aligned}\ ] ] using and in we can express @xmath49 in the energy basis , yielding @xmath410 + \sin \theta \left [ \rho_{\rm red;ge}(t)+\rho_{\rm red;eg}(t ) \right]\\ \fl = \cos(\theta ) \biggl ( \sum_{j=0}^\infty \sum_n \left [ { \langle j{\rm g}|n\rangle}^2 - { \langle j { \rm e}|n\rangle}^2 \right ] \rho_{nn } ( t ) \\ + \sum_{j=0}^\infty \sum_{\stackrel{n , m}{n \neq m } } \left [ { \langle j { \rm g}|n\rangle } { \langle m|j { \rm g}\rangle } - { \langle j { \rm g}|n\rangle } { \langle m|j { \rm g}\rangle } \right ] \re \{\rho_{nm}(t ) \ } \biggr ) \\ \fl \phantom{\mathrel{= } } + \sin ( \theta ) \biggl ( 2 \sum_{j=0}^\infty \sum_n { \langle j { \rm e}|n\rangle } \rho_{nn}(t ) { \langle n|j { \rm g}\rangle } \\ + \sum_{j=0}^\infty \sum_{\stackrel{n , m}{n \neq m } } \left [ { \langle j { \rm e}|n\rangle}{\langle m|j { \rm g}\rangle } + { \langle j { \rm e}|m\rangle}{\langle j { \rm g}|n\rangle}\right ] \re \{\rho_{nm}(t)\}\biggr)}.\ ] ] with we can write @xmath411 where @xmath412 + 2 \sin \theta { \langle j { \rm g}|n\rangle } { \langle j { \rm e}|n\rangle } \right\ } \rho_{n n}(t ) , \\ \fl p_{nm}(t ) = 2 \sum_j \biggl\ { \cos \theta \biggl[{\langle j { \rm g}|n\rangle } { \langle m|j { \rm g}\rangle } - { \langle j { \rm e}|n\rangle } { \langle m|j { \rm e}\rangle } \biggr ] \nonumber\\ + \sin \theta \biggl[{\langle j { \rm e}|n\rangle } { \langle m|j { \rm g}\rangle } + { \langle j { \rm e}|m\rangle } { \langle n|j { \rm g}\rangle } \biggr ] \biggr\ } \re \{\rho_{nm}(t ) \}. \end{aligned}\ ] ] let us consider the hamiltonian @xmath413 consisting of the free hamiltonian @xmath414 and a small perturbation @xmath415 , which is proportional to the coupling constant @xmath1 . additionally we assume , that the energy levels @xmath416 of @xmath414 are grouped into manifolds , with @xmath46 being the index of the manifold and @xmath417 is used to distinguish between different energy levels within the same manifold . the energy levels @xmath416 are eigenenergies of @xmath414 : @xmath418 through the perturbation @xmath419 different manifolds are coupled together . as long as the coupling @xmath1 is small , namely that @xmath420 for @xmath421 , also the energy levels of the total hamiltonian @xmath142 are clustered into manifolds . using the transformation @xmath422 , we construct an effective hamiltonian @xmath423 , which acts only within the individual manifolds ; i.e. , @xmath424 for @xmath421 , and has the same eigenvalues as @xmath142 within the manifolds . we expand @xmath79 and @xmath423 in terms of the small parameter @xmath1 up to second order : @xmath425 for calculating @xmath426 and @xmath427 we use that @xmath428 for @xmath421 and furthermore choose that @xmath79 has no matrix elements within a manifold , namely @xmath429 . now , one can iteratively calculate @xmath79 and @xmath423 order by order . for details see @xcite . here , we give only the results . for the transformation one has @xmath430 and @xmath431 \nonumber \\ + \sum_k \frac{1}{e_{j,\beta } - e_{j,\alpha } } \frac{{\langle j,\alpha| } v { |k,\beta\rangle } { \langle k,\beta| } v { |j,\beta\rangle}}{e_{k , \beta}-e_{j , \alpha } } \nonumber \\ + \sum_k \frac{1}{e_{j,\beta } - e_{j,\alpha } } \frac{{\langle j,\alpha| } v { |k,\alpha\rangle } { \langle k,\alpha| } v { |j,\beta\rangle}}{e_{k , \alpha}-e_{j , \beta } } , \textrm{\quad for \quad } \alpha \neq \beta.\end{aligned}\ ] ] the effective hamiltonian is up to second order @xmath432 + \or(g^3).\nonumber\end{aligned}\ ] ] in the case of the hamiltonian @xmath76 the first order matrix elements are @xmath433 and for the second order contributions @xmath434 where e. g. @xmath435 these matrix elements change sign under index transposition and all other matrix elements vanish . finally , we get the transformation up to second order in @xmath1 : @xmath436 here , we give the matrix elements @xmath154 specified in . @xmath437 matrix elements including the groundstate are given separately because of the special shape of @xmath215 : @xmath438 all other matrix elements are zero . here , we give the rate coefficients of the master equation for the reduced density matrix elements . they are : @xmath439 @xmath440 @xmath441 @xmath442 @xmath443 @xmath444 @xmath445 @xmath446 @xmath447 @xmath448 @xmath449 @xmath450 @xmath451 @xmath452 @xmath453 @xmath454 the solutions of the fsa master equation for the diagonal elements reads : @xmath455 @xmath456 @xmath457 @xmath458 @xmath459 99 leggett a j 1987 1 weiss u 1999 _ quantum dissipative systems _ 2nd edn ( singapore : world scientific ) grifoni m and hnggi p 1998 _ phys . rep . _ * 304 * 229 garg a , onunchic j n and ambegaokar v 1985 _ j. chem . phys . _ * 83 * 4491 raimond j m , brune m and haroche s 2001 565 yoshie t , scherer a , hendrickson j , khitrova g , gibbs h m , rupper g , ell c , shcheklin o b and deppe d g 2004 _ nature _ * 432 * 200 reithmaier j p 2004 _ nature _ * 432 * 197 nakamura y , pashkin yu a and tsai j s 1999 _ nature _ * 398 * 786 makhlin y , schn g and shnirman a 2001 357 vion d , aassime a , cottet a , joyez p , pothier h , urbina c , esteve d and devoret m h 2002 _ science _ * 296 * 886 collin e , ithier g , aassime a , joyez p , vion d and esteve d 2004 157005 blais a , huang r - s , wallraff a , girvin s m and schoelkopf r j 2004 a * 69 * 062320 wallraff a , schuster d i , blais a , frunzio l , huang r - s , majer j , kumar s , girvin s m and schoelkopf r j 2004 _ nature _ * 431 * 162 schuster d i , wallraff a , blais a , frunzio l , huang r - s , majer j , girvin s m and schoelkopf r j 2005 123602 wallraff a , schuster d i , blais a , frunzio l , majer j , devoret m h , girvin s m and schoelkopf r j 2005 060501 schuster d i 2007 _ nature _ * 445 * 515 wallraff a , schuster d i , blais a , gambetta j m , schreier j , frunzio l , devoret m h , girvin s m and schoelkopf r j 2007 050501 mooij j e , orlando t p , levitov l , tian l , van der wal c h and lloyd s 1999 _ science _ * 285 * 1036 van der wal c h , ter haar a c j , wilhelm f k , schouten r n , harmans c j p m , orlando t p , lloyd s and mooij j e 2000 _ science _ * 290 * 773 chiorescu i , nakamura y , harmans c j p m and mooij j e 2003 _ science _ * 299 * 1869 chiorescu i , bertet p , semba k , nakamura y , harmans c j p m and mooij j e 2004 _ nature _ * 431 * 159 johansson j , saito s , meno t , nakano h , ueda m , semba k and takayanagi h 2006 127006 blais a , gambetta j , wallraff a , schuster d i , girvin s m , devoret m h and schoelkopf r j 2007 a * 75 * 032329 tian l , lloyd s and orlando t p 2002 b * 65 * 144516 van der wal c h , wilhelm f k , harmans c j p m and mooij j e 2003 _ eur . phys . j. _ b * 31 * 2003 lupacu a , verwijs c j m , schouten r n , harmans c j p m and mooij j e 2004 177006 thorwart m , paladino e and grifoni m 2004 _ chem . * 296 * 333 wilhelm f k , kleff s and von delft j 2004 _ chem . phys . _ * 296 * 345 goorden m c , thorwart m and grifoni m 2004 267005 goorden m c , thorwart m and grifoni m 2005 _ eur . phys . j. _ b * 45 * 405 nesi f , grifoni m and paladino e 2007 316 kleff s , kehrein s and von delft j 2003 _ physica e _ * 18 * 343 kleff s , kehrein s and von delft j 2004 b * 70 * 014516 huang p and zheng h 2008 _ preprint _ cond - mat/0707.0725v3 makri n and makarov d e 1995 _ j. chem . phys . _ * 102 * 4600 makri n and makarov d e 1995 _ j. chem . phys . _ * 102 * 4611 jaynes e t and cummings f w 1963 _ ieee proc . _ * 51 * 90 shore b w and knight p l 1993 _ j. mod . _ * 40 * 1195 brito f and caldeira a o 2008 _ preprint _ quant - ph/08060992v1 shavitt i and redmon l t 1980 _ j. chem . * 73 * 5711 cohen - tannoudji d , dupont - roc j and grynberg g 1992 _ atom - photon interactions : basic processes and applications _ ( new york / chichester / brisbane / toronto / singapore : john wiley & sons ) caldeira a o and leggett a j 1983 374 blum k 1996 _ density matrix theory and applications _ 2nd edn ( new york and london : plenum press ) louisell w h 1973 _ quantum statistical properties of radiation _ ( new york / london / sydney/ toronto : john wiley & sons ) bloch f and siegert a 1940 522 vogel w , welsch d g and wallentowitz s 2001 _ quantum optics : an introduction _ 2nd edn ( berlin : wiley - vch )

the eddington limit corresponds to a situation where the radiative acceleration outwards equals the gravitational acceleration inwards . eddington ( 1926 ) originally only considered electron scattering for the opacity ( the classical eddington limit ) , but in an ionization zone of an important element the true opacity can be much higher . therefore , the opacity - modified eddington limit in stellar atmospheres can occur at significantly greater gravities than the classical eddington limit , as is evident for example from the calculations of lamers & fitzpatrick ( 1988 ) , gustafsson & plez ( 1992 ) and asplund & gustafsson ( 1996 ) . if hydrostatic equilibrium is required , a stellar luminosity which locally exceeds the eddington luminosity automatically forces the development of a gas pressure ( @xmath2 ) inversion , as seen from the equation of hydrostatic equilibrium : @xmath3 from rearranging the above expression one obtains @xmath4,\ ] ] with the radiative acceleration defined by : @xmath5 ( e.g mihalas 1978 ) . here @xmath6 denotes the physical flux and @xmath7 the total mass extinction coefficient ( with dimension @xmath8g@xmath1 ) . obviously a positive @xmath2-gradient must occur when the eddington limit is encountered if hydrostatic equilibrium is valid . it should be emphasized that a density inversion does not necessarily imply that the eddington limit is exceeded . a density inversion predominantly occurs due to a changing molecular weight in the ionization zone of a dominant species , while a @xmath2-inversion reflects a super - eddington luminosity due to high opacity ( asplund et al . since both are related to ionization they may often occur in the same atmospheric layers , which has probably caused the confusion found in the literature ( e.g. maeder 1989 ) . it should be noted that @xmath2-inversions only seem possible to develop in optically thick conditions . if the radiative acceleration exceeds gravity when lines dominate the opacity , the radiative force is highly unstable to perturbations induced by velocity gradients , since the spectral lines may then absorb unattenuated continuum flux . therefore , rather than the development of a @xmath2-inversion , a stellar wind will be initiated , which is the case for radiatively driven winds of hot stars . the aim of the present study is to investigate whether @xmath2-inversions in late - type stars , where the opacity instead is dominated by continuous opacity , are subject to a similar instability which would prevent their existence . before considering possible instabilities , in sect . [ s : eddlimit ] the derived eddington limit for h - deficient stars is presented and a possible connection between the limit and the declines of the r coronae borealis ( rcrb ) stars is discussed . the location of the eddington limit in a @xmath9-log@xmath10 diagram can be estimated by model atmosphere calculations ( lamers & fitzpatrick 1988 ; gustafsson & plez 1992 ) . it should be emphasized that these models consistently include the radiative acceleration in the equation of hydrostatic equilibrium . therefore , the location of the limit is not obtained by extrapolation towards lower gravities nor does it imply zero density , as has sometimes been claimed in the literature ( e.g. humphreys & davidson 1994 ; nieuwenhuijzen & de jager 1995 ) . an interesting conclusion from the above studies is that the theoretical opacity - modified eddington limit coincides with the upper luminosity limit for stars ( lamers & fitzpatrick 1988 ) . it is therefore possible that the eddington limit , possibly in connection with rotation ( langer 1997a , b ; owocki & gayley 1997 ) , prevents the evolution of stars into the super - eddington regime by the existence of radiative instabilities . these may give rise to the eruptions of the luminous blue variables ( lbvs ) ( cf . humphreys & davidson 1994 ) . here a similar investigation but for the h - deficient rcrb and related stars will be carried out ; a preliminary report can be found in asplund & gustafsson ( 1996 ) . the presence of @xmath2-inversions in such model atmospheres was noted by asplund et al . ( 1997a ) , who speculated that it may be the trigger mechanism for the enigmatic visual declines of the rcrb stars . the variability of these stars is likely due to the formation of an obscuring dust cloud close to the stellar surface in the line - of - sight , but how the dust condensation can proceed despite the relatively high equilibrium temperatures has long been an enigma ( cf . clayton 1996 ) . the presence of shocks seems to be a very promising mechanism for the gas to cool sufficiently and thereby trigger dust formation ( woitke et al . 1996 ) . the models used here are recently constructed line - blanketed model photospheres with typical abundances of rcrb stars ( asplund et al . 1997a ) . in fig . [ f : gamma ] the variation of @xmath11 with optical depth is displayed for a sequence of models corresponding to constant luminosity and mass ( e.g. @xmath12l@xmath13 and @xmath14m@xmath13 ) . the super - eddington luminosities only occur in the deep atmospheric layers where hei ionizes and , as expected , where @xmath2-inversions occur . since the rcrb stars presumably evolve towards higher @xmath9 ( kilkenny 1982 ) , the stars will move from the sub - eddington regime into having super - eddington luminosities , as is seen in fig . [ f : eddlimit ] where the theoretical eddington limit in the @xmath9-log@xmath10 diagram is shown . compared with solar abundances the limit is shifted towards higher @xmath9 due to the higher ionization potential of he compared with h , but also extends to slightly larger log@xmath10 . the location of the eddington limit for higher @xmath9 has not yet been explored , since a self - consistent hydrodynamical treatment is needed for @xmath15k when @xmath16 occurs for @xmath17 and line opacity may determine @xmath18 . the exact location of the computed eddington limit will naturally depend on the detailed properties of the model atmospheres . for @xmath19k convection in the deeper layers of the hei ionization zone restricts the limit to lower log@xmath10 , since the radiative flux diminishes in the presence of a significant convective flux . also , the lower temperatures decrease @xmath7 and @xmath18 further . including turbulent pressure would therefore bring the limit towards greater gravities by typically 0.2dex , due to less efficient convection with lower densities . a c / he ratio of 1% has been assumed on the basis of the measured ratio in extreme helium ( ehe ) stars ( jeffery 1996 ) , which presumably the rcrb stars are related to ( lambert et al . a higher carbon abundance pushes the eddington limit towards higher gravities the same way as turbulent pressure does . a lower overall metallicity by 0.5dex would decrease the limit by about 0.2dex by lowering the temperatures in the inner layers with a diminished line - blanketing . in fig . [ f : eddlimit ] the estimated parameters of the rcrb stars ( lambert et al . 1997 ; asplund et al . 1997c ) are also shown . obviously , rather than being randomly located in the diagram as one a priori might have guessed , the rcrb stars fall close to the limit , which suggests that a connection between the declines of rcrb stars and the eddington limit exists . thus , there is an alluring resemblance between the rcrb stars and the luminous blue variables in their proximity to the eddington limits and their eruptive behaviour ( asplund & gustafsson 1996 ) . if the rcrb stars are evolving towards higher @xmath9 ( kilkenny 1982 ) and the ehe domain at constant luminosity , their immediate progenitors should be the h - deficient carbon ( hdc ) stars ( lambert et al . the ehe and hdc stars experience no visual fadings despite being otherwise similar to the rcrb stars . it is therefore plausible that the reason for the non - variability is that the hdc stars have not yet reached the eddington limit while the ehe stars are located on the stable side of the limit at higher @xmath9 . if the declines of the rcrb stars are indeed triggered by the super - eddington luminosities , one could expect a correspondence between the maximum value of @xmath11 in the atmosphere and frequency of declines . in the pulsation - induced dust condensation model for the rcrb stars ( woitke et al . 1996 ) , in which the dust formation is triggered by propagating shock waves due to ordinary pulsations , one would sooner expect a correlation with pulsation strength . intriguingly , v854cen , which is the rcrb star most frequent in decline , has very weak photospheric pulsations ( lawson & cottrell 1997 ) but by a wide margin the largest @xmath11 found of the 18 rcrb stars analysed sofar . rysgr , on the other hand , has by far the strongest photospheric pulsations but only average amount of declines , which might reflect the fact that @xmath11 is not unusually high . pulsation may , however , push some stars across the limit by periodically shifting them towards higher @xmath9 , e.g. @xmath20k for rcrb itself ( rao & lambert 1997 ) . the proposed connection between the eddington limit and the declines of the rcrb stars is certainly not without problems , which deserve to be addressed further . a few stars seem not to behave as expected if the proposal is correct , in particular this is true for some of the coolest ehe stars and the three hot rcrb stars . as seen above , a @xmath2-inversion must develop at the eddington limit in hydrostatic equilibrium . one should ask , however , whether strong convection may take place to prevent it , since the ionization zone causing the high opacity and thereby the super - eddington luminosities also favours convection . in fact , a super - eddington luminosity in the diffusion regime automatically requires the presence of convection as shown by langer ( 1997a ) . generalizing langer s derivation for a non - negligible radiation pressure to also include the effects of ionization , the schwarzschild criterion @xmath21 for the onset of convection is equivalent to @xmath22 % \right\ } } % { 1 } \frac{4\alpha \left\ { 2 + 8\alpha + x\left(1-x\right ) \left[5/2 + \epsilon / kt + 4\alpha\right ] \right\ } } { 5 + 40\alpha + 32\alpha^2 + x\left(1-x\right ) \left[5/2 + \epsilon / kt + 4\alpha\right]^2}\ ] ] ( e.g. mihalas & weibel mihalas 1984 ) . here , while @xmath23 denotes the ionization fraction of the most abundant element with an ionization potential @xmath24 . convection must be initiated as @xmath11(r ) approaches unity , since the right hand side is always less than 1 . thereby the radiative flux and hence @xmath11 diminishes when convection is efficient . the eddington limit can therefore not be exceeded in adiabatic convection zones in the stellar interior and hence @xmath2-inversions will not develop . thus , inversions can only be present in non - adiabatic convection zones close to the stellar surface . it should be remembered that the location of the eddington limit is dependent on the convection treatment . a @xmath2-inversion will for example disappear if the density scale height is applied instead of the normal pressure scale height for the calculation of the convective flux through the mixing length theory . the mixing length theory is rarely a good description of stellar convection , and in particular for these extreme situations it may be very misleading . ultimately , hydrodynamical simulations will be necessary to investigate the effects of convection on @xmath2-inversions . using the momentum equation one can investigate under which conditions a @xmath2-inversion can exist in the presence of a stellar mass loss . or equivalently : which mass loss rates can one expect from a star if the super - eddington luminosities would instead drive a stellar wind ? in the presence of a stellar outflow , the equation of hydrostatic equilibrium must be replaced with the momentum equation : @xmath25 where we have restricted ourselves to the one - dimensional , time - independent case . also other external forces , such as turbulent forces ( @xmath26 ) and centrifugal forces ( @xmath27 ) , can be contained in @xmath28 besides radiation . it should therefore be kept in mind that the possibility of an instability may be underestimated when restricting the following discussion to only radiative forces , as other forces also tend to be destabilizing ( e.g. nieuwenhuijzen & de jager 1995 ; langer 1997a , b ; owocki & gayley 1997 ) . together with a constant , spherically symmetric mass loss rate @xmath29 , and the equation of state , it is possible to rewrite the momentum equation into the so called wind equation " : @xmath30 where the isothermal sound speed defined by @xmath31 has been introduced . by an inspection of eq . [ e : windeq ] , it is possible to determine under which circumstances @xmath32- or @xmath2-inversions are allowed . in normal cases , d@xmath33/dr@xmath34 in the surface layers and therefore the parenthesis on the right hand side of eq . [ e : windeq ] is positive as long as a temperature inversion is not present , which is not possible in optically thick layers with a radiative flux . therefore , if @xmath35 but @xmath36 , the velocity gradient @xmath37 must be negative . since a constant @xmath38 has been assumed , a density inversion will be present if @xmath39 and therefore a @xmath2-inversion is possible ( though not required ) . if , on the other hand , @xmath40 while @xmath35 , then @xmath37 must be positive , and the density gradient can not be positive : a density inversion can not exist . since a @xmath2-inversion requires a density inversion , neither inversion can occur in the stellar atmosphere despite the super - eddington luminosities . again the possibility of a temperature inversion has been neglected . a sufficient criterion for the disappearance of a @xmath2-inversion is therefore that the outflow velocity due to mass loss is high enough ( @xmath41 ) . the critical mass loss rate can be estimated by comparing @xmath42 with @xmath43 . the values for @xmath43 and @xmath32 will be taken from static models , which is justified for these rough estimates , since the atmosphere below the sonic point is normally little affected by mass loss . a model atmosphere with @xmath44k , log@xmath45 [ cgs ] and solar abundances requires a critical mass loss rate @xmath46 . hence @xmath2-inversions can exist in stars despite high mass loss rates . achmad & lamers ( 1997 ) have arrived at similar conclusions and estimates . they also verify the finding numerically with dynamical steady state atmospheres , though not altogether self - consistently . an even higher critical mass loss rate is found for the rcrb stars . the lack of hydrogen makes the continuous opacity significantly lower than for solar abundances and hence the density at a given optical depth will be correspondingly greater , typically by a factor of 50 ( asplund et al . since the density is higher , the outflow velocity will be lower in order to carry the same @xmath38 . a similar model as above but with abundances typical for rcrb stars , will therefore require @xmath47 . the mass loss rates of rcrb stars are poorly known , in particular since it may be episodic , but rough estimates suggest ( e.g. feast 1986 ) , orders of magnitude smaller than @xmath48 . it is concluded that @xmath2-inversions are not an artifact of hydrostatic equilibrium but may also exist in dynamical steady state atmospheres , even with a high mass loss rate . dynamical stability has been extensively studied , in particular in connection with the termination of the agb ( e.g. paczyski & zikowski 1968 ; wagenhuber & weiss 1994 ) . radial adiabatic oscillations will grow exponentially in time for dynamical instability , which is formally encountered if the first generalized adiabatic exponent @xmath49 is less than @xmath50 in a region effectively isolated from the rest of the star . an ideal gas has @xmath51 while pure radiation has @xmath52 , but the combined effect of non - negligible radiation pressure and ionization may push @xmath53 below the limiting value . for dynamical instability to occur , the pressure weighted volumetric mean value of @xmath53 between the depth @xmath54 and the stellar surface @xmath55 @xmath56 must be reduced below @xmath50 in a significant fraction of the star . stars with a pronounced core - envelope structure and high luminosity - to - mass ratios , like the supergiants investigated here , may show dynamical instabilities in ionization zones ( e.g. stothers & chin 1993 ; wagenhuber & weiss 1994 ) . in fig . [ f : gamma1 ] the run of @xmath53 and @xmath57 as functions of optical depth are shown for two model atmospheres corresponding to @xmath58 and log@xmath59 for a solar and a h - deficient composition . in particular , @xmath57 is reduced much below @xmath50 for the model with solar abundances in the hi and hei ionization zones . at the surface @xmath60 dominates the total pressure and hence @xmath53 is very close to @xmath50 . a more complete study for h - rich model atmospheres has been presented by lobel et al . ( 1992 ) , who arrive at the same conclusions . in the rcrb model only the hei ionization zone exists , which explains the quite different depth variation of @xmath53 . the higher temperatures at depth in the h - deficient model make hei ionization occur at smaller @xmath61 . also seen is the minor effect of the ci ionization zone at @xmath62 . these models only extend down to @xmath63 and hence it is not possibly to tell how @xmath57 varies further in , though the heii ionization zone will also reduce @xmath53 . it seems that atmospheres of late - type supergiants are close to being dynamically unstable , or may even be so for sufficiently high luminosity - to - mass ratios . violent instabilities due to this might lead to ejection of material , as found for the termination of the agb ( wagenhuber & weiss 1994 ) , as well as for the lbvs ( stothers & chin 1993 ) . also rcrb stellar models seem to suffer from a similar dynamical instability as the stars on the tip of the agb ( a. weiss , private communication ) . it should be remembered though that the above - mentioned evolutionary studies lack a proper hydrodynamical treatment , and therefore interpreting the instabilities found in the models as actually occurring in the stars must still be made with some caution . it is clear from the discussion in sect . [ s : pgasinversion ] and [ s : dynamical ] that @xmath2-inversions may exist in stars with high luminosity - to - mass ratios and that such stars are close to being dynamically unstable . thus , these stars may be vulnerable to radiative instabilities , since the additional destabilization from radiative forces may render them unstable . the existence of radiatively driven sound waves in hot stars due to high @xmath18 has been suggested by hearn ( 1972 , 1973 ) , and further elaborated by berthomieu et al . ( 1976 ) , spiegel ( 1976 ) and carlberg ( 1980 ) . the instability is based on the increase in opacity upon compression , since the radiation force is in phase with the velocity perturbation and hence causes amplification . for these high temperatures it requires essentially isothermal perturbations , since @xmath64 here @xmath65 , while @xmath66 is the flux weighted mean opacity . thus only short wavenumbers will be amplified . the growth rate is especially high as @xmath11 approaches unity ( berthomieu et al . 1976 ) but the super - eddington case has not been investigated . carlberg ( 1980 ) found that the instability may provide rapid transfer of momentum from the radiation to the gas with typical growth rates of @xmath67s@xmath1 for @xmath68k . it is possible that a similar instability is exhibited also in late - type supergiants , with the important differences being the super - eddington luminosities in the @xmath2-inversions and that isothermal perturbations are not necessary , since both ( @xmath69)@xmath70 and ( @xmath71)@xmath72 in the upper layers of the @xmath2-inversions ( fig . [ f : opacity ] ) . to study the stability of a @xmath2-inversion a local , linear , non - adiabatic stability analysis has been carried out . small , sinusoidal perturbations around the equilibrium values are assumed for the temperature , density and vertical velocity : @xmath73 where @xmath74 and @xmath75 are small . the atmosphere is assumed to be static ( @xmath76 ) , since the outflow velocity must anyway be less than the isothermal sound speed for a @xmath2-inversion to exist , as demonstrated above . the analysis will be restricted to vertical motion and therefore multi - dimensional instabilities of rayleigh - taylor - type will not be found . carlberg ( 1980 ) investigated also such a gradient instability for hot stars but found slow growth rates compared to the radiatively driven sound waves . such a gradient instability would resemble convective motion which , however , is already present in the layers of interest in the unperturbed model and is thus not specifically searched for here . the equations to be linearized are the material equations for conservation of mass , momentum and energy : @xmath77 ( e.g. mihalas & weibel mihalas 1984 ) , where d / dt , @xmath78 , @xmath79 and @xmath80 denote the lagrangean co - moving derivative , the specific internal energy of the gas , the wavelength integrated planck function ( @xmath81 ) and mean intensity ( @xmath82 ) , respectively . the absorption and planck mean opacities have been replaced by the rosseland mean opacity for absorption @xmath83 , which is a good approximation for the relevant optical depths . the scattering contribution to the total extinction is here denoted by @xmath84 . furthermore lte ( @xmath85 ) has been assumed . the material equations must be coupled to the equations for the radiation field to account for radiative transfer effects through exchanges of energy and momentum between the photons and the gas . in this exploratory investigation the effects of radiation are included with a time - independent treatment , i.e. the radiation field is assumed to be quasi - static . the radiation momentum and energy equations will then look like : @xmath86 here @xmath87 and @xmath88 denote the wavelength integrated eddington flux ( @xmath89 ) and radiation pressure ( @xmath90 ) respectively . in the diffusion regime for the unperturbed atmosphere , @xmath91 is valid . the equilibrium radiation quantities are related by : @xmath92 for the first relation radiative equilibrium is implicitly imposed , which is not strictly true in the cases which will be investigated below . however , the analysis will be restricted only to atmospheric depths where the convective flux carries @xmath93 of the total flux ( i.e. in the upper part of the convection zone ) and radiative equilibrium can therefore be largely justified . to close the relations the eddington approximation ( @xmath94 ) will be made , which is , like lte , well justified in the atmospheric layers of interest . in the following , the radiation energy and momentum equations will be combined to a single equation , which means that there is one more variable to perturb , for which @xmath80 is chosen : @xmath95 together with the definitions of the isothermal sound speed @xmath43 , the density scale height @xmath96^{-1}$ ] and an ideal gas for @xmath2 , linearization to first order yields : @xmath97 + v_1\left[ik - h_\rho^{-1}\right ] & = & 0 \\ % linearized conservation of momentum % \frac{\rho_1}{\rho_0}\left[g_{\rm eff } - g_\rho + \frac{\rho_1}{\rho_0}\left[g + c_{\rm t}^2\left(ik + \frac{1}{t_0}\frac{{\rm d}t}{{\rm dr } } - \frac{1}{\mu_0}\frac{{\rm d}\mu}{{\rm dr } } \right)\right ] & + & \nonumber \\ + \frac{\rho_1}{\rho_0}\left[- c_{\rm t}^2 \mu_\rho % \frac{\rho_0}{\mu_0 } % \frac{\partial \mu}{\partial \rho } \left(ik - h_\rho^{-1 } + \frac{1}{t_0}\frac{{\rm d}t}{{\rm dr } } - \frac{2}{\mu_0}\frac{{\rm d}\mu}{{\rm dr}}\right)\right ] & + & \nonumber \\ % + \frac{t_1}{t_0}\left[- g_{\rm t } + + \frac{t_1}{t_0}\left [ c_{\rm t}^2\left(ik - h_\rho^{-1 } - \frac{1}{\mu_0}\frac{{\rm d}\mu}{{\rm dr } } \right)\right ] & + & \nonumber \\ + \frac{t_1}{t_0}\left[- c_{\rm t}^2 \mu_t % \frac{t_0}{\mu_0 } % \frac{\partial \mu}{\partial t } \left(ik - h_\rho^{-1 } + \frac{1}{t_0}\frac{{\rm d}t}{{\rm dr } } - \frac{2}{\mu_0}\frac{{\rm d}\mu}{{\rm dr}}\right)\right ] & + & \nonumber \\ + v_1\left [ - i\omega\right ] % & + & \\ + \frac{j_1}{j_0}\left[- ik\frac{4\sigma_{\rm r } t_0 ^ 4}{3c\rho_0 } \right ] & = & 0 \\ % linearized conservation of energy \frac{\rho_1}{\rho_0}\left[-i\omega \left(\rho_0 \frac{\partial e}{\partial \rho } % \frac{1}{t_0}\frac{\partial l}{\partial \rho } - c_{\rm t}^2 \right)\right ] & + & \nonumber \\ + \frac{t_1}{t_0}\left[-i\omega t_0 \frac{\partial e}{\partial t } + % \frac{1}{\rho_0}\frac{\partial l}{\partial t } 16\left(\kappa_0 - \sigma_0\right)\sigma_{\rm r } t_0 ^ 4\right ] & + & \nonumber \\ % + v_1\left[ik\gamma + \frac{1}{c_{\rm v}t_0}g_{\rm eff } - % \gamma h_\rho^{-1 } + % \gamma \frac{1}{t_0}\frac{{\rm d}t}{{\rm dr } } \right ] & = & 0 . \\ % & + & \\ + v_1\left[g_{\rm rad } % + e_0\left ( ik - h_\rho^{-1}\right ) % \frac{h}{\rho_0}\frac{{\rm d}\rho}{{\rm dr } } + % - \rho_o\frac{\partial e}{\partial \rho } h_\rho^{-1 } + % \frac{\partial e}{\partial t}\frac{{\rm d}t}{{\rm dr } } + + \frac{{\rm d}e}{{\rm dr } } + c_{\rm t}^2 h_\rho^{-1 } \right ] & + & \nonumber \\ + \frac{j_1}{j_0}\left[-4\left ( \kappa_0 - \sigma_0\right)\sigma_{\rm r } t_0 ^ 4\right ] & = & 0 \\ % linearized radiation energy + momentum \frac{\rho_1}{\rho_0}\left[-\frac{ik}{\rho_0(\kappa_0 - \sigma_0 ) } \frac{1}{4}\left(\frac{t_{\rm eff}}{t_0}\right)^4\left(1 + \frac{\partial { \rm ln } \kappa}{\partial { \rm ln } \rho}\right)\right ] & + & \nonumber \\ + \frac{t_1}{t_0}\left[-4 - \frac{ik}{\rho_0(\kappa_0 - \sigma_0 ) } % \frac{h_0}{b_0 } \frac{1}{4}\left(\frac{t_{\rm eff}}{t_0}\right)^4 \frac{\partial { \rm ln } \kappa}{\partial { \rm ln } t}\right ] & + & \nonumber \\ + \frac{j_1}{j_0}\left[1 + \frac{k^2}{3\kappa_0(\kappa_0 - \sigma_0)\rho_0 ^ 2}\right ] & = & 0 . \end{aligned}\ ] ] the notation has been abbreviated according to and . derivatives with respect to @xmath98 and @xmath80 have been ignored , which is justified in optically thick layers where line opacity is unimportant . true enough , photoionization edges introduce some velocity dependence on @xmath18 but this minor effect is neglected . lccc + + variable & units & solar & rcrb + + log@xmath61 & & 0.5 & 1.2 + @xmath18 & [ cms@xmath99 & @xmath100 & @xmath101 + @xmath102 & [ k ] & 9610 & 14950 + @xmath32 & [ g@xmath103 & @xmath104 & @xmath105 + @xmath2 & [ dyncm@xmath99 & @xmath107 & @xmath108 + @xmath109 & [ cms@xmath1 ] & @xmath111 & @xmath112 + @xmath113d@xmath102/dr & [ cm@xmath1 ] & @xmath115 & @xmath116 + @xmath117d@xmath32/dr & [ cm@xmath1 ] & @xmath118 & @xmath119 + @xmath120d@xmath33/dr & [ cm@xmath1 ] & @xmath121 & @xmath122 + @xmath123 & [ @xmath124g@xmath1 ] & @xmath125 & @xmath126 + @xmath84 & [ @xmath124g@xmath1 ] & @xmath127 & @xmath128 + @xmath129 & & @xmath130 & @xmath131 + @xmath132 & & @xmath133 & @xmath134 + d@xmath78/dr & [ @xmath135s@xmath99 & @xmath136 & @xmath137 + @xmath138 & [ @xmath8s@xmath99 & @xmath139 & @xmath140 + @xmath141 & [ @xmath8s@xmath99 & @xmath142 & @xmath143 + + in order for the above set of equations to have a non - trivial solution for the perturbations @xmath74 , @xmath75 and @xmath144 , the determinant of the equations must be equal to 0 . the three complex eigenvalues @xmath145 for a given real wavenumber @xmath146 then correspond to the allowed perturbations . two of the modes will represent radiation - modified gravity - acoustic waves while the third is a non - propagating , heavily damped , thermal wave . here two different cases will be investigated numerically , one corresponding to a @xmath2-inversion in a supergiant with solar abundances and the second the same but for an rcrb star . as before , both model atmospheres have @xmath147k and log@xmath148 . values for some of the physical parameters at some atmospheric layer where the instability may occur are found in table [ t : values ] . the calculated growth rates are shown in fig . [ f : growthrate ] , which reveals amplification for @xmath149cm@xmath1 ( solar composition ) and @xmath151cm@xmath1 ( rcrb composition ) . for large @xmath146 , the perturbations are as expected damped due to the negligible optical depths of the disturbances ( @xmath152 ) , which make the sound waves little affected by the radiative forces . instead , the smoothing effects from radiation energy exchange due to large gradients in the perturbed variables dominate . the maximum growth rates are slightly smaller for a solar composition than for a h - deficient composition : @xmath153 respectively @xmath154s@xmath1 , which corresponds to an e - folding time on the order of the sound crossing time of the atmosphere . for the largest wavenumbers with such growth rates , the real parts are significantly higher , which implies that the sound waves are only slightly amplified per wavelength . for smaller @xmath146 , however , the growth rate is comparable with the oscillation rate . in fact , for the model with solar composition , the wave is essentially non - propagating for the smallest @xmath146 . however , the use of a local analysis is not justified when the wavelengths of the perturbations are comparable with the scale heights of the variations for the variables . for @xmath155cm@xmath1 , the propagation speed is slightly less than the isothermal sound speed in both cases . thus , radiation - modified sound waves can be amplified in late - type stars with significant radiative forces . however , the perturbations are not very strongly excited with minimum growth times @xmath156s , contrary to the findings by carlberg ( 1980 ) for hot stars , mainly as a result of the much smaller radiative fluxes ( @xmath157 ) . for the instability to generate large amplification , the sound waves must be reflected at the surface and pass through the region with high radiative forces several times . however , the inwards propagating modes are damped for the same reason as the outgoing are amplified , with the magnitude of the damping rate similar to the growth rate , which suggests that the instability will still not be very efficient . in order to estimate the behaviour of the instability in the global , non - linear regime , a more sophisticated study than presented here must be carried out . in order to gain further theoretical insight into which atmospheric properties contribute to the instability presented above , the equations can be simplified by assuming the unperturbed atmosphere to be isothermal and neglecting ionization and scattering . furthermore , only the limits of optically thin and optically thick disturbances will be studied and hence the combined radiative energy and momentum equation will not be needed . the gas energy equation will thus instead be written as : @xmath158 where @xmath159 denotes the net radiative cooling rate of the gas . depending on the optical thickness of the disturbance , @xmath159 will be described by either newtonian cooling or equilibrium diffusion ( e.g. mihalas & weibel mihalas 1984 ) . the rate of work done by @xmath18 has been ignored . the linearized continuity equation remains the same but the linearized momentum equation reads as : @xmath160 & + & \nonumber \\ + \frac{t_1}{t_0}\left [ c_{\rm t}^2\left(ik - h_\rho^{-1}\right ) - g_{\rm t}\right ] + v_1\left[-i\omega\right ] & = & 0,\end{aligned}\ ] ] where @xmath161 and @xmath162 , while the energy equation looks like : @xmath163 + \frac{t_1}{t_0}\left[\frac{i\omega}{\gamma -1 } - \tilde{{\cal l}}\right ] + v_1\left[-h_\rho^{-1}\right ] = 0.\ ] ] here @xmath164 represents the term arising from the net cooling rate @xmath159 and @xmath165 the ratio of specific heats . for newtonian cooling @xmath166 and in the diffusion regime @xmath167 @xmath164 is the inverse of the radiative cooling time - scale ( e.g. mihalas & weibel mihalas 1984 ) , where the factor @xmath168 takes into account the optical depth effects for optically thick conditions . the simplified dispersion relation then reads : @xmath169 & + & \nonumber \\ - \omega k\left[g_\rho + \left(\gamma-1\right)g_{\rm t } - g_{\rm eff } + \gamma c_{\rm t}^2 h_\rho^{-1}\right ] & + & \nonumber \\ -\tilde{{\cal l}}\left(\gamma-1\right)\left [ k^2c_{\rm t}^2 + h_\rho^{-1}\left(g_{\rm eff } - g_\rho\right)\right ] & + & \nonumber \\ - ik\tilde{{\cal l}}\left(\gamma-1\right)\left [ g_\rho - g_{\rm eff } + c_{\rm t}^2 h_\rho^{-1}\right ] & = & 0.\end{aligned}\ ] ] for @xmath155cm@xmath1 , @xmath170 is much smaller than @xmath171 for the two investigated models and can be ignored . in situations of rapid radiative cooling ( @xmath172 ) , the dispersion relation can be simplified to yield the solutions @xmath173,\ ] ] i.e. isothermally propagating , radiation - modified sound waves . in the limit of slow cooling ( @xmath174 ) , as expected adiabatic acoustic waves are recovered : @xmath175,\ ] ] which travel with the adiabatic sound speed . for @xmath176cm@xmath1 , the propagation speed will be modified by the neglected terms @xmath177 , though the growth / damping rates will remain the same . for both isothermal and adiabatic perturbations , the sound waves are amplified by the action of the radiative acceleration , as long as the opacity increases with temperature and density ( assuming constant radiative flux ) : @xmath178 furthermore , in a @xmath2-inversion @xmath28 is negative and hence tends to excite an initial disturbance . however , the normal momentum conservation counteracts this due to propagation in a density stratified medium ; the density inversion will tend to damp the amplitude for outwards propagating waves . in fact , with this simplified atmosphere and the values in table [ t : values ] , in both models the last term dominates . hence , the amplitude of sound waves will decay as they propagate outwards with these assumptions , despite the super - eddington luminosities and the increase in opacity upon compression . thus , since the numerical results in sect . [ s : radinstab ] show amplification for the case without these simplifying assumptions , the effects of e.g. ionization and a temperature gradient tend to work together with the radiative force and its derivatives to overcome the damping contribution from the density gradient . it has been shown that h - deficient model atmospheres may have radiative forces which exceed gravity , similar to the case of some luminous h - rich stars ( lamers & fitzpatrick 1988 ; gustafsson & plez 1992 ) . such super - eddington luminosities manifest themselves as @xmath2-inversions in hydrostatic atmospheres . the inversions are , however , not an artifact of this assumption but can also be present in dynamical atmospheres . the inversions are not removed unless a very high mass loss rate is present , and is neither necessarily removed by convection if the large radiative forces occur close to the surface . thus , super - eddington luminosities does _ not _ automatically initiate a stellar wind . the location of the h - deficient rcrb stars in the immediate proximity of the computed opacity - modified eddington limit ( fig . [ f : eddlimit ] ) , is certainly suggestive of a connection between the enigmatic visual declines of the stars and the eddington limit . it is therefore proposed that instabilities as the rcrb stars encounter the eddington limit during their evolution towards higher @xmath9 are the unknown trigger mechanism for their famous variability . the eddington limit may thus be the `` smoking gun '' for the declines : gas is ejected due to the high radiative forces in the atmospheres , which then cools rapidly first radiatively and later from adiabatic expansion to reach sub - equilibrium temperatures and thus possibly enable dust condensation ( cf . woitke et al . 1996 for the similar case of shocks instead induced by pulsations ) . such gas ejections may be observable as absorption components in strong lines , as observed by rao & lambert ( 1997 ) in rcrb at maximum light . if a connection indeed exists , it could explain why the ehe and hdc stars do not show rcrb - like variability as a result of being located on the stable side of the eddington limit . it also points to an interesting similarity between the rcrb stars and the lbvs . thus , the two types of stars known to be situated at the eddington limit both show eruptive behaviour , which suggests a similar underlying physical explanation for their variability . a search for radiative and dynamical instabilities in the atmospheres has been carried out for both h - rich and h - deficient late - type supergiants , but only partly successfully . sound waves are found to be amplified by the large radiative forces but the linear stability analysis reveal only rather slow growth rates compared to for early - type stars despite the super - eddington luminosities . such radiation - modified sound waves may thus be partly responsible for the semi - regular pulsational variations of such supergiants , but it is doubtful whether the instability is efficient enough to eject gas clouds from the atmospheres as speculated above . the atmospheres are also found to be close to dynamically unstable , which might give rise to increased mass loss , as previously suggested for lbvs ( stothers & chin 1993 ) and for the termination of the agb phase ( e.g. paczyski & zikowski 1968 ; wagenhuber & weiss 1994 ) . this does not , however , explain in what way the rcrb stars are special compared to other similar stars . in fact , the h - rich models seem to be more vulnerable to such instabilities , which suggests that dynamical instabilities are not the answer to the declines of the rcrb stars . another possibility is of course that the same dynamical instabilities occur in both types of stars but that the h - deficient and c - rich environment of the rcrb stars greatly favours dust condensation , and thus the observational manifestation , compared to for h - rich compositions . one can also imagine that the eddington limit is still responsible for the behaviour of the rcrb stars though in a more indirect way . ( 1997a , b ) has shown that the effect of rotation coupled with strong radiative forces may lead to increased mass loss in the equatorial plane , and it is possible that such a mechanism is at work in these supergiants , despite the presumably relatively small rotational velocities . the difference in observed variability between the rcrb stars and the hdc and ehe stars could then be the result of viewing angle . there are some observational evidence for such bipolarity ( e.g. rao & lambert 1993 ; clayton et al . 1997 ) violent instabilities due to strange mode pulsations leading to dramatically increased mass loss is another possibility , which is partly related to the eddington limit , since both have their origin in ionization zones , as do the dynamical instabilities discussed above . whether any of these instabilities , or a combination thereof , could help explain the variability of the rcrb stars and the lbvs certainly deserves further investigation . finally , it should be remembered that the stability analyses presented here assume the unperturbed model to be a realistic description of the stellar atmosphere . at least for the rcrb stars , there are strong indications that this is in fact not the case ( gustafsson & asplund 1996 ; asplund et al . 1997a , b , c ; lambert et al . 1997 ) . if the supergiant atmospheres are indeed distinctly different from the predictions of standard models , the efficiency of radiative instabilities may have been underestimated . one can suspect that models based on the normal assumptions , such as the mixing length theory for convection , near the eddington limit may be very unsuitable . clearly , radiative hydrodynamical simulations of such atmospheres would be valuable , also since they would shed further light on possible instabilities . the constructive suggestions by f.p . pijpers and a. weiss have significantly improved the paper . discussions with w. glatzel , b. gustafsson , l .- a . willson and p. woitke have been very helpful . the author is grateful to l. achmad , h.j.g.l.m . lamers , n. langer and s. owocki for communicating their work prior to publication . achmad l. , lamers h.j.g.l.m . , 1997 , a&a 320 , 196 asplund m. , gustafsson b. , 1996 , in : hydrogen - deficient stars , jeffery c.s . , heber u. ( eds . ) . series vol . asplund m. , gustafsson b. , kiselman d. , eriksson k. , 1997a , a&a 318 , 521 asplund m. , gustafsson b. , lambert d.l . , rao n.k . , 1997b , a&a 321 , l17 asplund m. , gustafsson b. , rao n.k . , lambert d.l . , 1997c , a&a , in press berthomieu g. , provost j. , rocca a. , 1976 , a&a 47 , 413 carlberg r.g . , 1980 , apj 241 , 1131 clayton g.c . , 1996 , pasp 108 , 225 clayton g.c . , bjorkman k.s . , nordsieck k.h . , zellner n.e.b . , schulte - ladbeck r.e . , 1997 , apj 476 , 870 eddington a.s . , 1926 , the internal constitution of the stars , cambridge university press , cambridge feast m.w . , 1986 , in : hydrogen - deficient stars and related objects , hunger k. , schnberner d. , rao n.k . ( eds . ) . reidel , dordrecht , p. 151 gustafsson b. , asplund m. , 1996 , in : hydrogen - deficient stars , jeffery c.s . , heber u. ( eds . ) . series vol . gustafsson b. , plez b. , 1992 , in : instabilities in evolved super- and hypergiants , de jager c. , nieuwenhuijzen h. ( eds . ) . north holland , amsterdam , p. 86 hearn a.g . , 1972 , a&a 19 , 417 hearn a.g . , 1973 , a&a 23 , 97 humphreys r.m . , davidson k. , 1994 , pasp 106 , 1025 jeffery c.s . , 1996 , in : hydrogen - deficient stars , jeffery c.s . , heber u. ( eds . ) . asp conf series vol . 96 , p. 152 kilkenny d. , 1982 , mnras 200 , 1019 lambert d.l . , , gustafsson b. , asplund m. , 1997 , submitted to a&a lamers h.j.g.l.m . , fitzpatrick e.l . , 1988 , apj 324 , 279 langer n. , 1997a , in : luminous blue variables : massive stars in transition , nota a. , lamers h.j.g.l.m . ( eds . ) . series , in press langer n. , 1997b , a&a , in press lawson w.a . , cottrell p.l . , 1997 , mnras 285 , 266 lobel a. , achmad l. , de jager c. , nieuwenhuijzen h. , 1992 , a&a 264 , 147 maeder a. , 1989 , in : physics of luminous blue variables , davidson k. , moffat a.f.j . , lamers h.j.g.l.m . ( eds . ) . kluwer , dordrecht , p. 15 mihalas d. , 1978 , stellar atmospheres , w.h . freeman and company , san fransisco mihalas d. , weibel mihalas b. , 1984 , foundations of radiation hydrodynamics , oxford university press , oxford nieuwenhuijzen h. , de jager c. , 1995 , a&a 302 , 811 owocki s.p . , gayley k.g . , 1997 , in : luminous blue variables : massive stars in transition , nota a. , lamers h.j.g.l.m . ( eds . ) . series , in press paczyski b. , zikowski j. , 1968 , acta astr . 18 , 225 rao n.k . , lambert d.l . , 1993 , aj 105 , 1915 rao n.k . , lambert d.l . , 1997 , mnras 284 , 489 seaton m.j . , yan y. , mihalas d. , pradhan a.k . , 1994 , mnras 266 , 805 spiegel e.a . , 1976 , in : physique des mouvements dans les atmosphres stellaires , cayrel r. , steinberg m. ( eds ) , cnrs , paris , p. 19 stothers r.b . , chin c .- w . , 1993 , apj 408 , l85 wagenhuber j. , weiss a. , 1994 , a&a 290 , 807 woitke p. , goeres a. , sedlmayr e. , 1996 , a&a 313 , 217

shock waves occur in many physical situations in nature . for example , shock waves can be produced by solar winds ( bow shocks ) , supersonic or near sonic aircrafts ( transonic shocks around the body ) , explosions ( blast waves ) , and various natural processes . when such a shock hits an obstacle ( steady or flying ) , shock reflection - diffraction phenomena occur . one of the most important problems in mathematical fluid dynamics is the problem of shock reflection - diffraction by a wedge . when the plane shock hits a wedge head on , it experiences a reflection - diffraction process , and then a fundamental question is what types of wave patterns of reflection - diffraction configurations it may form around the wedge . the complexity of reflection - diffraction configurations was first reported by ernst mach @xcite in 1878 , who first observed two patterns of reflection - diffraction configurations : regular reflection ( two - shock configuration ) and mach reflection ( three - shock configuration ) ; also see @xcite . the problem remained dormant until the 1940 s when von neumann , friedrichs , bethe , as well as many experimental scientists , among others , began extensive research into all aspects of shock reflection - diffraction phenomena , due to its importance in applications . see von neumann @xcite and ben - dor @xcite ; also see @xcite and the references cited therein . it has been found that the situation is much more complicated than what mach originally observed : the mach reflection can be further divided into more specific sub - patterns , and various other patterns of shock reflection - diffraction may occur such as the von neumann reflection and the guderley reflection ; see @xcite and the references cited therein . then the fundamental scientific issues include : \(i ) _ structure of the shock reflection - diffraction configurations _ ; \(ii ) _ transition criteria between the different patterns of shock reflection - diffraction configurations _ ; \(iii ) _ dependence of the patterns upon the physical parameters such as the wedge angle @xmath0 , the incident - shock - wave mach number @xmath1 , and the adiabatic exponent @xmath2_. careful asymptotic analysis has been made for various reflection - diffraction configurations in lighthill @xcite , keller - blank @xcite , hunter - keller @xcite , morawetz @xcite , @xcite , and the references cited therein ; also see glimm - majda @xcite . large or small scale numerical simulations have been also made ; see , e.g. @xcite , @xcite , @xcite , and the references cited therein . however , most of the fundamental issues for shock reflection - diffraction phenomena have not been understood , especially the global structure and transition of different patterns of shock reflection - diffraction configurations . this is partially because physical and numerical experiments are hampered by various difficulties and have not been able to select the correct transition criteria between different patterns . in particular , numerical dissipation or physical viscosity smear the shocks and cause boundary layers that interact with the reflection - diffraction patterns and can cause _ spurious mach steams _ ; cf . woodward - colella @xcite . furthermore , some difference between two different patterns are only fractions of a degree apart ( e.g. , see fig . 5 below ) , a resolution even by sophisticated modern experiments ( e.g. @xcite ) has been unable to reach , as pointed out by ben - dor in @xcite : _ for this reason it is almost impossible to distinguish experimentally between the sonic and detachment criteria _ " ( cf . section 5 below ) . in this regard , it seems that the ideal approach to understand fully the shock reflection - diffraction phenomena , especially the transition criteria , is still via rigorous mathematical analysis . to achieve this , it is essential to establish first the global existence , regularity , and structural stability of solutions of the shock reflection - diffraction problem . on the other hand , shock reflection - diffraction configurations are the core configurations in the structure of global solutions of the two - dimensional riemann problem for hyperbolic conservation laws ; while the riemann solutions are building blocks and local structure of general solutions and determine global attractors and asymptotic states of entropy solutions , as time tends to infinity , for multidimensional hyperbolic systems of conservation laws . see @xcite and the references cited therein . in this sense , we have to understand the shock reflection - diffraction phenomena , in order to understand fully entropy solutions to multidimensional hyperbolic systems of conservation laws . in this paper , we first formulate the shock reflection - diffraction problem into an initial - boundary value problem in section 2 . then we employ the essential feature of self - similarity of the initial - boundary value problem to reformulate the problem into a boundary value problem in the unbounded domain in section 3 . in section 4 , we present the unique solution of normal reflection for this problem when the wedge angle is @xmath3 . in section 5 , we exhibit the local theory of regular reflection - diffraction , introduce a stability criterion to determine state ( 2 ) at the reflection point on the wedge , and present the von neumann s detachment and sonic conjectures . then we discuss the role of the potential flow equation in the shock reflection - diffraction problem even in the level of the full euler equations in section 6 . based on the local theory , we reduce the boundary value problem into a free boundary problem in the context of potential flow in section 7 . in section 8 , we describe a global theory for regular reflection - diffraction for potential flow , established in chen - feldman @xcite and bae - chen - feldman @xcite . in section 9 , we discuss some open problems and new mathematics required for further developments , which are also essential for solving multidimensional problems in conservation laws and other areas in nonlinear partial differential equations . the full euler equations for compressible fluids in @xmath4 , are of the following form : @xmath5 where @xmath6 is the density , @xmath7 the fluid velocity , @xmath8 the pressure , and @xmath9 the internal energy . two other important thermodynamic variables are the temperature @xmath10 and the energy @xmath11 . the notation @xmath12 denotes the tensor product of the vectors @xmath13 and @xmath14 . choosing @xmath15 as the independent thermodynamical variables , then the constitutive relations can be written as @xmath16 governed by @xmath17 for a polytropic gas , @xmath18 or equivalently , @xmath19 where @xmath20 may be taken to be the universal gas constant divided by the effective molecular weight of the particular gas , @xmath21 is the specific heat at constant volume , @xmath22 is the adiabatic exponent , and @xmath23 is any constant under scaling . when a flow is potential , that is , there is a velocity potential @xmath24 such that @xmath25 then the euler equations for the flow take the form : @xmath26 where @xmath27 especially , @xmath28 when @xmath29 , by scaling and @xmath30 is the bernoulli constant , which is usually determined by the boundary conditions if such conditions are prescribed . from the second equation in , we have @xmath31 then system can be rewritten as the following time - dependent potential flow equation of second order : @xmath32 with . for a steady solution @xmath33 , i.e. , @xmath34 , we obtain the celebrated steady potential flow equation in aerodynamics : @xmath35 in applications in aerodynamics , or is used for discontinuous solutions , and the empirical evidence is that entropy solutions of or are fairly close to entropy solutions for , provided the shock strengths are small , the curvature of shock fronts is not too large , and the amount of vorticity is small in the region of interest . furthermore , we will show in section 6 that , for the shock reflection - diffraction problem , the euler equations for potential flow is actually _ exact _ in an important region of the solution ( see theorem 6.1 below ) . then the problem of shock reflection - diffraction by a wedge can be formulated as follows : . _ seek a solution of system satisfying the initial condition at @xmath36 : @xmath37 and the slip boundary condition along the wedge boundary : @xmath38 where @xmath39 is the exterior unit normal to the wedge boundary , and state @xmath40 and @xmath41 satisfy @xmath42 that is , given @xmath43 , and @xmath22 , the other variables @xmath44 and @xmath45 are determined by . in particular , the mach number @xmath46 is determined by @xmath47 where @xmath48 is the sonic speed of fluid state @xmath41 . _ notice that the initial - boundary value problem ( * problem 2.1 * ) is invariant under the self - similar scaling : @xmath49 therefore , we seek self - similar solutions : @xmath50 then the self - similar solutions are governed by the following system : @xmath51 where @xmath52 , and @xmath53 is the pseudo - velocity . the eigenvalues of system are @xmath54 where @xmath55 is the sonic speed . when the flow is pseudo - subsonic , i.e. , @xmath56 , the eigenvalues @xmath57 become complex and thus the system consists of two transport equations and two nonlinear equations of hyperbolic - elliptic mixed type . therefore , system is _ hyperbolic - elliptic composite - mixed _ in general . since the problem is symmetric with respect to the axis @xmath58 , it suffices to consider the problem in the half - plane @xmath59 outside the half - wedge : @xmath60 then the initial - boundary value problem ( * problem 2.1 * ) in the @xmath61coordinates can be formulated as the following boundary value problem in the self - similar coordinates @xmath62 : . _ seek a solution to system satisfying the slip boundary condition on the wedge boundary and the matching condition on the symmetry line @xmath58 : @xmath63 the asymptotic boundary condition as @xmath64 : @xmath65 _ , width=220,height=201 ] it is expected that the solutions of * problem 3.1 * contain all possible patterns of shock reflection - diffraction configurations as observed in physical and numerical experiments ; cf . @xcite and the references cited therein . the simplest case of the shock reflection - diffraction problem is when the wedge angle @xmath0 is @xmath66 . in this case , the reflection - diffraction problem simply becomes the normal reflection problem , for which the incident shock normally reflects , and the reflected shock is also a plane . it can be shown that there exist a _ unique _ state @xmath67 and a _ unique _ location of the reflected shock @xmath68 such that state @xmath69 is subsonic inside the sonic circle with center at the origin and radius @xmath70 , and is supersonic outside the sonic circle ( see fig . [ fig : nr ] ) . that is , in this case , the normal reflection solution is unique . in this case , @xmath71 and @xmath72 is the unique root of @xmath73 that is , @xmath74 in other words , given @xmath43 , and @xmath22 , state @xmath69 is uniquely determined through and . for a wedge angle @xmath75 , different reflection - diffraction patterns may occur . various criteria and conjectures have been proposed for the existence of configurations for the patterns ( cf . ben - dor @xcite ) . one of the most important conjectures made by von neumann @xcite in 1943 is the _ detachment conjecture _ , which states that the regular reflection - diffraction configuration may exist globally whenever the two shock configuration ( one is the incident shock and the other the reflected shock ) exists locally around the point @xmath76 ( see fig . the following theorem was rigorously shown in chang - chen @xcite ( also see sheng - yin @xcite , bleakney - taub @xcite , neumann @xcite ) . [ local ] there exists @xmath77 such that , when @xmath78 , there are two states @xmath79 and @xmath80 such that @xmath81 where @xmath82 is the sonic speed . then the conjecture can be stated as follows : : _ there exists a global regular reflection - diffraction configuration whenever the wedge angle @xmath0 is in @xmath83_. it is clear that the regular reflection - diffraction configuration is not possible without a local two - shock configuration at the reflection point on the wedge , so this is the weakest possible criterion . in this case , the local theory indicates that there are two possible states for state ( 2 ) . there had been a long debate to determine which one is more physical for the local theory ; see courant - friedrichs @xcite , ben - dor @xcite , and the references cited therein . since the reflection - diffraction problem is not a local problem , we take a different point of view that the selection of state ( 2 ) should be determined by the global features of the problem , more precisely , by the stability of the configuration with respect to the wedge angle @xmath0 , rather than the local features of the problem . : _ since the solution is unique when the wedge angle @xmath84 , it is required that our global regular reflection - diffraction configuration should be stable and converge to the unique normal reflection solution when @xmath85 , provided that such a global configuration can be constructed . _ we employ this stability criterion to conclude that our choice for state ( 2 ) must be @xmath86 . in general , @xmath86 may be supersonic or subsonic . if it is supersonic , the propagation speeds are finite and state ( 2 ) is completely determined by the local information : state ( 1 ) , state ( 0 ) , and the location of the point @xmath76 . this is , any information from the reflected region , especially the disturbance at the corner @xmath87 , can not travel towards the reflection point @xmath76 . however , if it is subsonic , the information can reach @xmath76 and interact with it , potentially altering the reflection - diffraction type . this argument motivated the second conjecture as follows : : _ there exists a regular reflection - diffraction configuration when @xmath88 for @xmath89 such that @xmath90 at @xmath76 . _ this sonic conjecture is based on the following fact : if state ( 2 ) is sonic when @xmath91 , then @xmath92 for any @xmath93 . this conjecture is stronger than the detachment one . in fact , the regime between the angles @xmath94 and @xmath95 is very narrow and is only fractions of a degree apart ; see fig . 5 from sheng - yin @xcite . when @xmath96.,width=211,height=192 ] in this section , we discuss the role of the potential flow equation in the shock reflection - diffraction problem for the full euler equations . under the hodge - helmoltz decomposition @xmath97 with @xmath98 , the euler equations become @xmath99 where @xmath100 is the vorticity of the fluid , @xmath101 is the entropy , and the gradient @xmath102 is with respect to the self - similar variables @xmath103 from now on . when @xmath104 , and @xmath105 on a curve @xmath106 transverse to the fluid direction , we first conclude from that , in the domain @xmath107 determined by the fluid trajectories : @xmath108 past @xmath106 , @xmath109 this implies that @xmath110 since @xmath98 . then we conclude that @xmath111 since @xmath112 , which yields that the right - hand side of equation vanishes . furthermore , from , @xmath113 which implies that @xmath114 by scaling , we finally conclude that the solutions of system in the domain @xmath107 is determined by the following system for self - similar solutions : @xmath115 or the potential flow equation for self - similar solutions : @xmath116 with @xmath117 then we have @xmath118 for our problem ( see fig . 4 ) , we note that , for state ( 2 ) , @xmath119 then , if our solution @xmath120 is @xmath121 and the gradient of the tangential component of the velocity is continuous across the sonic arc @xmath122 , we still have along @xmath122 on the side of @xmath123 . thus , we have [ p - dominate ] let @xmath120 be a solution of our * problem 3.1 * such that @xmath120 is @xmath121 in the open region @xmath124 and the gradient of the tangential component of @xmath125 is continuous across the sonic arc @xmath126 . let @xmath107 be the subregion of @xmath123 formed by the fluid trajectories past the sonic arc @xmath122 , then , in @xmath107 , the potential flow equation with coincides with the full euler equations , that is , equation with is exact in the domain @xmath107 for * problem 3.1*. the regions such as @xmath107 also exist in various mach reflection - diffraction configurations . theorem [ p - dominate ] applies to such regions whenever the solution @xmath120 is @xmath121 and the gradient of the tangential component of @xmath125 is continuous . in fact , theorem 8.3 indicates that , for the solutions @xmath127 of with , the @xmath128 regularity of @xmath127 and the continuity of the tangential component of the velocity field @xmath129 are optimal across the sonic arc @xmath122 . the importance of the potential flow equation with in the time - dependent euler flows was also observed by hadamard @xcite through a different argument . furthermore , when the wedge angle @xmath0 is close to @xmath66 , it is expected that the curvature of the reflected shock is small so that , in the other part @xmath130 of @xmath123 , the vorticity @xmath131 is small and the entropy is close to the constant . then , in the reflection - diffraction domain @xmath132 , the potential flow equation with dominates , provided that the exact state along the reflected shock is given . , width=230,height=163 ] equation with is a nonlinear equation of mixed elliptic - hyperbolic type . it is elliptic if and only if @xmath133 which is equivalent to @xmath134 the study of partial differential equations of mixed hyperbolic - elliptic type can date back 1940s ( cf . linear models of partial differential equations of mixed hyperbolic - elliptic type include the lavrentyev - betsadze equation : @xmath135 the tricomi equation : @xmath136 and the keldysh equation : @xmath137 nonlinear models of mixed - type equations for with include the transonic small disturbance equation : @xmath138 or , for @xmath139 , @xmath140 which has been studied in @xcite and the references cited therein . also see @xcite for the models for self - similar solutions from the pressure gradient system and nonlinear wave equations . for the potential equation with , shocks are discontinuities in the pseudo - velocity @xmath141 . that is , if @xmath142 and @xmath143 are two nonempty open subsets of @xmath144 and @xmath145 is a @xmath146curve where @xmath147 has a jump , then @xmath148 is a global weak solution of ( [ potential-1 ] ) with in @xmath149 if and only if @xmath127 is in @xmath150 and satisfies equation in @xmath151 and the rankine - hugoniot condition on @xmath11 : @xmath152_s=0,\ ] ] where the bracket @xmath153 $ ] denotes the difference of the values of the quantity along the two sides of @xmath11 . then the plane incident shock solution in the @xmath154coordinates with states @xmath155 and @xmath156 corresponds to a continuous weak solution @xmath127 of ( [ potential - eq ] ) in the self - similar coordinates @xmath62 with the following form : @xmath157 respectively , where @xmath158 are the velocity of state ( 1 ) and the location of the incident shock , uniquely determined by @xmath159 through ( [ fbconditionselfsim-0 ] ) . then @xmath160 in fig . 2 , and * problem 3.1 * in the context of the potential flow equation can be formulated as : [ bvp ] ( boundary value problem ) ( see fig . seek a solution @xmath127 of equation with in the self - similar domain @xmath161 with the boundary condition on @xmath162 : @xmath163 and the asymptotic boundary condition at infinity : @xmath164 where ( [ boundary - condition-2 ] ) holds in the sense that @xmath165 _ for our problem , since @xmath166 does not satisfy the slip boundary condition , the solution must differ from @xmath166 in @xmath167 , thus a shock diffraction - diffraction by the wedge vertex occurs . in chen - feldman @xcite , we first followed the von neumann criterion and the stability criterion introduced in section 5 to establish a local existence theory of regular shock reflection near the reflection point @xmath76 in the level of potential flow , when the wedge angle is large and close to @xmath66 . in this case , the vertical line is the incident shock @xmath168 that hits the wedge at the point @xmath160 , and state ( 0 ) and state ( 1 ) ahead of and behind @xmath11 are given by @xmath169 and @xmath166 defined in and , respectively . the solutions @xmath127 and @xmath166 differ only in the domain @xmath170 because of shock diffraction by the wedge vertex , where the curve @xmath171 is the reflected shock with the straight segment @xmath172 . state ( 2 ) behind @xmath172 can be computed explicitly with the form : @xmath173 which satisfies @xmath174 on @xmath175 ; the constant velocity @xmath176 and the angle @xmath94 between @xmath172 and the @xmath177axis are determined by @xmath178 from the two algebraic equations expressing ( [ fbconditionselfsim-0 ] ) and continuous matching of state ( 1 ) and state ( 2 ) across @xmath172 , whose existence is exactly guaranteed by the condition on @xmath178 under which regular shock reflection - diffraction is expected to occur as in theorem 5.1 . moreover , @xmath179 is the unique solution in the domain @xmath180 , as argued in @xcite . denote @xmath181 the sonic arc of state @xmath182 with center @xmath183 and radius @xmath184 . also we introduce the following notation for the other parts of @xmath185 : @xmath186 then * problem 7.1 * can be formulated as : . _ seek a solution @xmath127 in @xmath123 to equation with subject to the boundary condition on @xmath187 , the rankine - hugoniot conditions on the shock @xmath188 : @xmath189_{{\gamma_{shock}}}=0 , \label{rh-1}\\ & & [ \rho(\nabla\varphi,\varphi,\rho_0)\nabla\varphi\cdot\nu]_{{\gamma_{shock}}}=0,\label{rh-2}\end{aligned}\ ] ] and the dirichlet boundary condition on the sonic arc @xmath122 : @xmath190 _ it should be noted that , in order that the solution @xmath127 in the domain @xmath123 is a part of the global solution to * problem 7.1 * , that is , @xmath127 satisfies the equation in the sense of distributions in @xmath161 , especially across the sonic arc @xmath122 , it is requaired that @xmath191 that is , we have to match our solution with state ( 2 ) , which is the necessary condition for our solution in the domain @xmath123 to be a part of the global solution . to achieve this , we have to show that our solution is at least @xmath146 with @xmath192 across @xmath122 . then the problem can be reformulated as the following free boundary problem : . _ seek a solution @xmath127 and a free boundary @xmath193 such that _ * @xmath194 and @xmath195 * @xmath127 satisfies the free boundary condition along @xmath196 ; * @xmath197 solves in @xmath198 , is subsonic in @xmath198 , and satisfies @xmath199 the boundary condition on @xmath200 implies that @xmath201 and thus ensures the orthogonality of the free boundary with the @xmath177-axis . formulation implies that the free boundary is determined by the level set @xmath202 , which is a convenient formulation to apply useful free boundary techniques . the free boundary condition along @xmath196 is the conormal boundary condition on @xmath196 . condition ensures that the solution of the free boundary problem in @xmath123 is a part of the global solution as pointed out earlier . condition is the slip boundary condition . * problem 7.3 * involves two types of transonic flow : one is a continuous transition through the sonic arc @xmath126 as a fixed boundary from the pseudo - supersonic region ( 2 ) to the pseudo - subsonic region @xmath123 ; the other is a jump transition through the transonic shock as a free boundary from the supersonic region ( 1 ) to the subsonic region @xmath123 . in this section , we describe a global theory for regular reflection - diffraction established in chen - feldman @xcite and bae - chen - feldman @xcite . in chen - feldman @xcite , we have developed a rigorous mathematical approach to solve * problem 7.3 * and established a global theory for solutions of regular reflection - diffraction , which converge to the unique solution of the normal shock reflection when @xmath0 tends to @xmath66 . introduce the polar coordinates @xmath203 with respect to the center @xmath204 of the sonic arc @xmath126 of state ( 2 ) , that is , @xmath205 then , for @xmath206 , we denote by @xmath207 the @xmath208-neighborhood of the sonic arc @xmath209 within @xmath123 ; see fig . 4 . in @xmath210 , we introduce the coordinates : @xmath211 then @xmath212 and @xmath213 . there exist @xmath214 and @xmath215 such that , when @xmath216 , there exists a global self - similar solution @xmath217 with @xmath218 of * problem 7.1 * ( equivalently , * problem 7.2 * ) for shock reflection - diffraction by the wedge . the solution @xmath127 satisfies that , for @xmath219 , @xmath220 moreover , 1 . [ ellipticityinomega ] equation is elliptic in @xmath123 ; 2 . [ phi - ge - phi2 ] @xmath221 ; 3 . [ bdryreg ] the reflected shock @xmath171 is @xmath222 at @xmath223 and @xmath224 except @xmath223 ; 4 . [ c11normestimate ] there exists @xmath225 such that @xmath226 ; in particular , in the coordinates ( [ coordnearsonic ] ) , @xmath227 5 . [ psi - x - est ] there exists @xmath228 so that , in the coordinates ( [ coordnearsonic ] ) , @xmath229 6 . [ fbnearsonic ] there exist @xmath230 and a function @xmath231 such that , in the coordinates ( [ coordnearsonic ] ) , @xmath232 and @xmath233)}<\infty,\qquad \frac{d\hat{f}}{dx}\!\ge\omega>0\!\;\;\ ; \text{for}\;\;0<x<{\varepsilon}_0.\ ] ] furthermore , the solution @xmath127 is stable with respect to the wedge angle @xmath0 in @xmath234 and converges in @xmath234 to the unique solution of the normal reflection as @xmath235 . the existence of a solution @xmath127 of * problem 7.1 * ( equivalently , * problem 7.2 * ) , satisfying ( [ phi - states-0 - 1 - 2 ] ) and property ( [ c11normestimate ] ) , follows from ( * ? ? ? * main theorem ) . property ( [ ellipticityinomega ] ) follows from lemma 5.2 and proposition 7.1 in @xcite . property ( [ phi - ge - phi2 ] ) follows from proposition 7.1 and section 9 in @xcite which assert that @xmath236 , where the set @xmath237 is defined by ( 5.15 ) in @xcite . property ( [ psi - x - est ] ) follows from propositions 8.18.2 and section 9 in @xcite . property ( [ fbnearsonic ] ) follows from ( 5.7 ) and ( 5.25)(5.27 ) in @xcite and the fact that @xmath236 . we remark that estimate above confirms that our solutions satisfy the assumptions of theorem 6.1 for the velocity field @xmath129 . one of the main difficulties for the global existence is that the ellipticity condition for ( [ potential-1 ] ) with is hard to control , in comparison to our work on steady flow @xcite . the second difficulty is that the ellipticity degenerates along the sonic arc @xmath126 . the third difficulty is that , on @xmath126 , the solution in @xmath123 has to be matched with @xmath179 at least in @xmath146 , i.e. , the two conditions on the fixed boundary @xmath126 : the dirichlet and conormal conditions , which are generically overdetermined for an elliptic equation since the conditions on the other parts of boundary have been prescribed . thus , one needs to prove that , if @xmath127 satisfies ( [ potential-1 ] ) in @xmath123 , the dirichlet continuity condition on the sonic arc , and the appropriate conditions on the other parts of @xmath238 derived from problem 7.3 , then the normal derivative @xmath239 automatically matches with @xmath240 along @xmath126 . indeed , equation ( [ potential-1 ] ) , written in terms of the function @xmath241 in the @xmath242coordinates defined near @xmath126 such that @xmath126 becomes a segment on @xmath243 , has the form : @xmath244 plus the _ small _ terms that are controlled by @xmath245 in appropriate norms . equation is _ elliptic _ if @xmath246 . hence , it is required to obtain the @xmath128 estimates near @xmath126 to ensure @xmath247 which in turn implies both the ellipticity of the equation in @xmath123 and the match of normal derivatives @xmath248 along @xmath126 . taking into account the _ small _ terms to be added to equation , one needs to make the stronger estimate @xmath249 and assume that @xmath245 is suitably small to control these additional terms . another issue is the non - variational structure and nonlinearity of this problem which makes it hard to apply directly the approaches of caffarelli @xcite and alt - caffarelli - friedman @xcite . moreover , the elliptic degeneracy and geometry of the problem makes it difficult to apply the hodograph transform approach in chen - feldman @xcite and kinderlehrer - nirenberg @xcite to fix the free boundary . for these reasons , one of the new ingredients in our approach is to develop further the iteration scheme in @xcite to a partially modified equation . we modified equation ( [ potential-1 ] ) in @xmath123 by a proper shiffmanization ( i.e. a cutoff ) that depends on the distance to the sonic arc , so that the original and modified equations coincide when @xmath127 satisfies @xmath250 , and the modified equation @xmath251 is elliptic in @xmath123 with elliptic degeneracy on @xmath209 . then we solved a free boundary problem for this modified equation : the free boundary is the curve @xmath196 , and the free boundary conditions on @xmath196 are @xmath202 and the rankine - hugoniot condition ( [ fbconditionselfsim-0 ] ) . on each step , an _ iteration free boundary _ curve @xmath126 is given , and a solution of the modified equation @xmath252 is constructed in @xmath123 with the boundary condition ( [ fbconditionselfsim-0 ] ) on @xmath196 , the dirichlet condition @xmath253 on the degenerate arc @xmath126 , and @xmath254 on @xmath255 and @xmath256 . then we proved that @xmath127 is in fact @xmath128 up to the boundary @xmath126 , especially @xmath257 , by using the nonlinear structure of elliptic degeneracy near @xmath126 which is modeled by equation ( [ degenerate - equation ] ) and a scaling technique similar to daskalopoulos - hamilton @xcite and lin - wang @xcite . furthermore , we modified the _ iteration free boundary _ curve @xmath196 by using the dirichlet condition @xmath202 on @xmath196 . a fixed point @xmath127 of this _ iteration procedure _ is a solution of the free boundary problem for the modified equation . moreover , we proved the precise gradient estimate : @xmath258 for @xmath259 , which implies that @xmath127 satisfies the original equation ( [ potential - eq ] ) . this global theory for large - angle wedges has been extended in chen - feldman @xcite to the sonic angle @xmath260 , for which state ( 2 ) is sonic , such that , as long as @xmath261 $ ] , the global regular reflection - diffraction configuration exists . the global existence result in theorem 8.1 can be extended up to the sonic wedge - angle @xmath94 for any @xmath2 and @xmath262 . moreover , the solutions satisfy the properties ( i)(vi ) in theorem 8.1 . the condition @xmath262 depends explicitly only on the parameters @xmath22 and @xmath263 . for the case @xmath264 , we have been making substantial progress as well , and the final detailed results can be found in chen - feldman @xcite . by theorem 8.1([c11normestimate ] ) , the solution @xmath127 constructed there is at least @xmath128 near the sonic arc @xmath126 . the next question is to analyze the behavior of solutions @xmath265 to regular reflection - diffraction , especially the optimal regularity of the solutions . we first define the class of regular reflection - diffraction solutions . [ regreflsoldef ] _ let @xmath266 , @xmath263 , and @xmath267 be constants , let @xmath44 and @xmath268 be defined by ( [ state-1-velocity ] ) and ( [ shocklocation ] ) . let the incident shock @xmath168 hits the wedge at the point @xmath160 , and let state @xmath40 and state @xmath41 ahead of and behind @xmath196 be given by and , respectively . the function @xmath269 is a regular reflection - diffraction solution if @xmath270 is a solution to * problem 7.1 * such that _ \(a ) there exists state @xmath182 of form ( [ state2 ] ) with @xmath271 , satisfying the entropy condition @xmath272 and the rankine - hugoniot condition ( [ fbconditionselfsim-0 ] ) along the line @xmath273 which contains the points @xmath76 and @xmath223 , such that @xmath274 is on the sonic circle of state @xmath182 , and state @xmath182 is supersonic along @xmath172 ; \(b ) there exists an open , connected domain @xmath275 such that ( [ phi - states-0 - 1 - 2 ] ) holds and equation is elliptic in @xmath123 ; \(c ) @xmath276 on the part @xmath277 of the reflected shock . the global solution constructed in @xcite is a regular reflection - diffraction solution , which is a part of the assertions in theorems 8.18.2 . if state @xmath182 exists and is supersonic , then the line @xmath278 necessarily intersects the sonic circle of state @xmath182 ; see the argument in @xcite starting from ( 3.5 ) there . thus , the only assumption regarding the point @xmath223 is that @xmath279 intersects the sonic circle within @xmath161 . we note that , in the case @xmath280 , the regular reflection becomes the normal reflection , in which @xmath281 and the solution is smooth across the sonic line of state ( 2 ) ; see ( * ? ? ? * section 3.1 ) . condition @xmath282 in definition [ regreflsoldef ] rules out this case . moreover , for @xmath283 , the property @xmath271 in part ( a ) of definition [ regreflsoldef ] is always true for state @xmath182 of form ( [ state2 ] ) , satisfying the entropy condition @xmath272 and the rankine - hugoniot condition ( 7.1 ) along the line @xmath273 which contains the point @xmath76 . these are readily derived from the calculations in ( * ? ? ? * section 3.2 ) . there may exist a global regular reflection - diffraction configuration when state @xmath182 is subsonic which is a very narrow regime @xcite . such a case does not involve the difficulty of elliptic degeneracy , which we are facing for the configurations in the class of solutions in the sense of definition 8.3 . since @xmath202 on @xmath196 by , condition ( c ) in definition [ regreflsoldef ] is equivalent to @xmath284 that is , @xmath188 is below @xmath279 . in bae - chen - feldman @xcite , we have developed a mathematical approach to establish the regularity of solutions of the regular reflection - diffraction problem in the sense of definition 8.3 . first , we have shown that any regular reflection - diffraction solutions can not be @xmath222 across the sonic arc @xmath285 . [ noc2acrosssoniclinethm ] there does not exist a global regular reflection - diffraction solution in the sense of definition 8.3 such that @xmath270 is @xmath222 across the sonic arc @xmath126 . now we study the one - sided regularity up to @xmath126 from the elliptic side , i.e. , from @xmath123 . for simplicity of presentation , we now use a localized version of @xmath286 : for a given neighborhood @xmath287 of @xmath122 and @xmath288 , define @xmath289 since @xmath287 will be fixed in the following theorem , we do not specify the dependence of @xmath286 on @xmath287 . [ mainthm1 ] let @xmath270 be a regular reflection - diffraction solution in the sense of definition [ regreflsoldef ] and satisfy the following properties : there exists a neighborhood @xmath287 of @xmath122 such that 1 . [ c11normestimate - a ] @xmath127 is @xmath128 across the sonic arc @xmath126 : @xmath290 ; 2 . [ psi - x - est - a ] there exists @xmath228 so that , in the coordinates ( [ coordnearsonic ] ) , @xmath291 3 . [ fbnearsonic - a ] there exists @xmath292 , @xmath230 , and a function @xmath231 such that , in the coordinates ( [ coordnearsonic ] ) , @xmath293 and @xmath294)}<\infty,\;\;\;\ ; \frac{d\hat{f}}{dx}\!\ge\omega>0\!\;\;\ ; \text{for}\;\;0<x<{\varepsilon}_0.\ ] ] then we have 1 . [ c2alpsonic ] @xmath270 is @xmath295 up to @xmath126 away from @xmath296 for any @xmath297 . that is , for any @xmath298 and any given @xmath299 , there exists @xmath300 depending only on @xmath301 , and @xmath302 so that @xmath303 2 . [ limitsonic ] for any @xmath304 , @xmath305 3 . [ 2ndderjumpsonic ] @xmath306 has a jump across @xmath126 : for any @xmath307 , @xmath308 4 . [ nolimatp1 ] the limit @xmath309 does not exist . we remark that the solutions established in @xcite satisfy the assumptions of theorem 8.5 . in particular , we proved that the @xmath128-regularity is _ optimal _ for the solution across the open part @xmath310 of the sonic arc ( the degenerate elliptic curve ) and at the point @xmath296 where the sonic circle meets the reflected shock ( as a free boundary ) . to achieve the optimal regularity , one of the main difficulties is that the sonic arc @xmath126 is the transonic boundary separating the elliptic region from the hyperbolic region , near where the solution is governed by the nonlinear degenerate elliptic equation for @xmath241 . we carefully analyzed the features of equation and established the @xmath311 regularity of solutions in the elliptic region up to the open sonic arc @xmath310 . as a corollary , we showed that the @xmath128-regularity is actually optimal across the transonic boundary @xmath312 from the elliptic to hyperbolic region . since the reflected shock @xmath312 is regarded as a free boundary connecting the hyperbolic region ( 1 ) with the elliptic region @xmath123 for the nonlinear second - order equation of mixed type , another difficulty for the optimal regularity of the solution is that the point @xmath296 is exactly the point where the degenerate elliptic arc @xmath310 meets a transonic free boundary for the nonlinear partial differential equation of second order . as far as we know , this is the first optimal regularity result for solutions to a free boundary problem of nonlinear degenerate elliptic equations at the point where an elliptic degenerate curve meets the free boundary . to achieve this , we carefully constructed two sequences of points on where the corresponding sequences of values of @xmath313 have different limits at @xmath296 ; this has been done by employing the one - sided @xmath311 regularity of the solution up to the open arc @xmath310 and by studying detailed features of the free boundary conditions on the free boundary @xmath312 , i.e. , the rankine - hugoniot conditions . we remark that some efforts were also made mathematically for the reflection - diffraction problem via simplified models . one of these models , the unsteady transonic small - disturbance ( utsd ) equation , was derived and used in keller - blank @xcite , hunter - keller @xcite , hunter @xcite , and morawetz @xcite for asymptotic analysis of shock reflection - diffraction . also see zheng @xcite for the pressure gradient equation and canic - keyfitz - kim @xcite for the utsd equation and the nonlinear wave system . furthermore , in order to deal with the reflection - diffraction problem , some asymptotic methods have been also developed . lighthill @xcite studied shock reflection - diffraction under the assumption that the wedge angle is either very small or close to @xmath66 . keller - blank @xcite , hunter - keller @xcite , harabetian @xcite , and gamba - rosales - tabak @xcite considered the problem under the assumption that the shock is so weak that its motion can be approximated by an acoustic wave . for a weak incident shock and a wedge with small angle for potential flow , by taking the jump of the incident shock as a small parameter , the nature of the shock reflection - diffraction pattern was explored in morawetz @xcite by a number of different scalings , a study of mixed equations , and matching the asymptotics for the different scalings . also see chen @xcite for a linear approximation of shock reflection - diffraction when the wedge angle is close to @xmath66 and serre @xcite for an apriori analysis of solutions of shock reflection - diffraction and related discussions in the context of the euler equations for isentropic and adiabatic fluids . another related recent effort has been on various important physical problems in steady potential flow , as well as steady fully euler flow , for which great progress has been made . the problems for global subsonic flow past an obstacle and for local supersonic flow past an obstacle with sharp head are classical , due to the works of shiffman @xcite , bers @xcite , finn - gilbarg @xcite , dong @xcite , and others for the pure elliptic case and to the works of gu @xcite , shaeffer @xcite , li @xcite , and others for the pure hyperbolic case . recent progress has been made on transonic flow past nozzles ( e.g. @xcite ) , transonic flow past a wedge or conical body ( e.g. @xcite ) , and transonic flow past a smooth obstacle ( e.g. @xcite ) . also see @xcite for the existence of global subsonic - sonic flow , @xcite for global supersonic flow past an obstacle with sharp head , and the references cited therein . for some of other recent related developments , we refer the reader to chen @xcite for the local stability of mach configuration , elling - liu @xcite for physicality of weak prandtl - meyer reflection for supersonic potential flow around a ramp , serre @xcite for multidimensional shock interaction for a chaplygin gas , canic - keyfitz - kim @xcite for semi - global solutions for the shock reflection problem , glimm - ji - li - li - zhang - zhang - zheng @xcite for the formation of a transonic shock in a rarefaction riemann problem for polytropic gases , zheng @xcite for various solutions to some two - dimensional riemann problems , gues - mtivier - williams - zumbrun @xcite and benzoni - gavage and serre @xcite for revisits of the local stability of multidimensional shocks and phase boundaries , among many others . as we have seen from the previous discussion , the shock reflection - diffraction problem involves several core challenging difficulties that we have to face in the study of nonlinear partial differential equations . these nonlinear difficulties include _ free boundary problems , oblique derivative problems for nonsmooth domains , degenerate elliptic equations , degenerate hyperbolic equations , transport equations with rough coefficients , mixed and/or composite equations of hyperbolic - elliptic type , behavior of solutions when a free boundary meets an elliptic degenerate curve , and compressible vortex sheets_. more efficient numerical methods are also required for further understanding of shock reflection - diffraction phenomena . furthermore , the wave patterns of shock reflection - diffraction configurations are the core patterns and configurations for the global solutions of the two - dimensional riemann problem ; these solutions are building blocks and local structure of general entropy solutions and determine global attractors and asymptotic states of entropy solutions , as time goes infinity , for two - dimensional systems of hyperbolic conservation laws . therefore , a successful solution to the shock reflection - diffraction problem not only provides our understanding of shock reflection - diffraction phenomena and behavior of entropy solutions to multidimensional conservation laws , but also provides important new ideas , insights , techniques , and approaches for our developments of more efficient analytical techniques and methods to overcome the core challenging difficulties in multidimensional problems in conservation laws and other areas in nonlinear partial differential equations . the shock reflection - diffraction problem is also an excellent test problem to examine our capacity and ability to solve rigorously various challenging problems for nonlinear partial differential equations and related applications . bae , g .- q . chen , and m. feldman , _ regularity of solutions to regular shock reflection for potential flow _ , inventiones mathematicae , 2009 ( to appear ) ; doi 10.1007/s00222 - 008 - 0156 - 4 ( online ) ; preprint arxiv:0804.2500 , april 2008 . l. a. caffarelli , _ a harnack inequality approach to the regularity of free boundaries , i. lipschitz free boundaries are @xmath314 _ , rev . mat . iberoamericana , * 3 * ( 1987 ) , 139162 ; _ ii . flat free boundaries are lipschitz _ , comm . pure appl . math . * 42 * ( 1989 ) , 5578 ; _ iii . existence theory , compactness , and dependence on @xmath315 _ , ann . scuola norm . sci . ( 4 ) , * 15 * ( 1989 ) , 583602 . t. chang , g .- q . chen , and s. yang , _ @xmath316-d riemann problem in gas dynamics and formation of spiral _ , in : nonlinear problems in engineering and science numerical and analytical approach ( beijing , 1991 ) , 167179 , science press : beijing , 1992 . t. chang , g .- q . chen , and s. yang , _ on the riemann problem for two - dimensional euler equations i : interaction of shocks and rarefaction waves ; ii : interaction of contact discontinuities _ , discrete contin . systems , * 1 * ( 1995 ) , 555584 ; * 6 * ( 2000 ) , 419430 . chen , _ euler - poisson - darboux equations and hyperbolic conservation laws _ , in : nonlinear evolutionary partial differential equations , pp . 1125 , ams / ip studies in advanced mathematics , vol . 3 , international press : cambridge , massachusetts , * 1997*. g .- q . chen , _ euler equations and related hyperbolic conservation laws _ , chapter 1 , handbook of differential equations , evolutionary equations , vol . 2 , eds . c. m. dafermos and e. feireisl , elsevier : amsterdam , the netherlands , 2005 . chen , _ on nonlinear partial differential equations of mixed type _ , in : proceedings of the 4th international congress of chinese mathematicians , eds . l. jin , k .- f . liu , l. yang , and s .- t . yau , isbn : 7 - 04 - 023267 - 7 , higher education press : beijing , jan . 2008 . a. yu . demyanov and a. v. panasenko , _ numerical solution to the problem of the diffraction of a plane shock wave by a convex corner _ , fluid dynamics , * 16 * ( 1981 ) , 720725 ( translated from the original russian ) . i. m. gamba and c. s. morawetz , _ a viscous approximation for a @xmath316-d steady semiconductor or transonic gas dynamic flow : existence theorem for potential flow _ , comm . pure appl . math . * 49 * ( 1996 ) , 9991049 . h. m. glaz , p. colella , i. i. glass , and r. l. deschambault , _ a detailed numerical , graphical , and experimental study of oblique shcok wave reflection _ , lawrence berkeley laboratory report * lbl-20033 * ( 1985 ) . j. glimm , x .- ji , j .- q . li , x .- li , p. zhang , t. zhang , and y .- x . zheng , _ transonic shock formation in a rarefaction riemann problem for the 2-d compressible euler equations _ , siam j. appl . math . 2008 ( in press ) . m. s. ivanov , d. vandromme , v. m. formin , a. n. kudryavtsev , a. hadjadj , and d. v. khotyanovsky , _ transition between regular and mach reflection of shock waves : new numerical and experimental results _ , shock waves , * 11 * ( 2001 ) , 199207 . a. majda , _ one perspective on open problems in multi - dimensional conservation laws _ , in : multidimensional hyperbolic problems and computations ( minneapolis , mn , 1989 ) , 217238 , i m a vol . math . appl . * 29 * , springer : new york , 1991 . p. woodward , _ simulation of the kelvin - helmholtz instability of a supersonic slip surface with the piecewise - parabolic method ( ppm ) , _ in : numerical methods for the euler equations of fluid dynamics ( rocquencourt , 1983 ) , 493508 , siam : philadelphia , pa , 1985 .

the pion distribution amplitude ( da ) is an important element for applying the qcd factorization theory and qcd light - cone sum rules ( lcsr ) to exclusive processes involving pion . for examples , it is important for understanding the semi - leptonic decays @xmath5 and @xmath6 , the pion - photon transition form factor ( tff ) @xmath7 , the exclusive process @xmath8 , and etc .. inversely , one can make use of all those processes to obtain stringent constraints on the pion da @xcite . because of its simplest structure , the pion leading - twist da has attracted much attention after the pioneering works done by refs.@xcite . unfortunately , at present , there is no definite conclusion on the behavior of pion leading - twist da . for example , the babar measurement for the pion - photon tff @xcite supports the chernyak - zhitnitsky ( cz)-like @xcite behavior ; while the corresponding data by the belle collaboration @xcite supports the asymptotic - like behavior @xcite . the pion leading - twist da at the scale @xmath9 can be expanded into a gegenbauer polynomial as @xcite , @xmath10 , \label{dagegexp}\end{aligned}\ ] ] where the @xmath11 are gegenbauer polynomials and the @xmath12 are gegenbauer moments . theoretically , the gegenbauer moments have been calculated by qcd sum rules @xcite and the lattice gauge theory @xcite . phenomenologically , the gegenbauer moments have also been estimated from the pion - photon transition form factor @xcite , the pion electromagnetic form factor @xcite , the @xmath13 tffs @xcite , and etc . most of those estimations are consistent with each other within errors . it is noted that the central values of them are different from each other , especially , the lattice qcd estimations are bigger than those of qcd sum rules . due to the forthcoming more accurate data , it is helpful to provide a more accurate theoretical prediction on the pion da for a better comparison with the data . since its invention @xcite , the shifman - vainshtein - zakharov ( svz ) sum rules has been widely adopted for dealing with the hadron phenomenology . in addition , the background field theory provides a systematic description for its key components , i.e. the vacuum condensates , from the viewpoint of qcd field theory @xcite . it assumes that the quark and gluon fields are composed of the background fields and the quantum fluctuations around them . thus , to take the background field theory as the foundation for the qcd sum rules , it not only has distinct physical picture , but also greatly simplifies the calculation due to its capability of adopting different gauges for quantum fluctuations and background fields . the qcd sum rules within the background field theory has already been adopted to study the meson properties , cf.refs.@xcite . previous sum rule estimations for the pion da moments are usually done up to dimension - four condensates , or including part of the dimension - six condensates contributions . in the present paper , we shall adopt the svz sum rules under the background field theory to make a detailed study on the pion leading - twist da up to dimension - six operators . for the purpose , as the first time , we shall expand both the quark propagator and the vertex operator @xmath0 up to a more complex form with full dimension - six operators contributions so as to achieve an accurate qcd sum rule estimation . we shall show that those dimension - six operators shall result in new dimension - six condensates that do provide sizable contributions and should be taken into consideration for a sound estimation , i.e. they shall bring sizable changes in our previous predictions on the pion da behavior . the remaining parts of the paper are organized as follows . in section ii , we first present a brief introduction of the qcd sum rules and then provide the quark propagator and the vertex @xmath0 up to dimension - six operators under the background field theory . in section iii , we present the calculation technology for deriving the sum rules for pion leading - twist da moments . numerical results are given in section iv . section v is reserved for a summary . for convenience , formulas for the propagators and the vacuum matrix elements in @xmath14-dimension space are presented in the appendix . within the svz sum rules , the hadrons are represented by their interpolating quark currents taken at large virtualities . the correlator of those currents can be treated within the operator product expansion , whose short - distance coefficients are calculated using qcd perturbation theory , whereas its long - distance parts are written according to the non - perturbative but universal vacuum condensates , such as the quark condensate @xmath15 , the gluon condensate @xmath16 , and etc . furthermore , under the background field theory , the quark and gluon fields are composed of background fields and quantum fluctuations around them . because of the influence from background fields , the quark and gluon propagators shall include nonperturbative component inevitably , which however can be treated via a systematic way . within the background field theory , the gluon field @xmath17 and quark field @xmath18 in the qcd lagrangian or green function should be replaced by @xmath19 where in the right hand side of the arrow , @xmath17 with @xmath20 and @xmath18 indicate the background fields of the gluon and quark , respectively . @xmath21 and @xmath22 stand for the gluon and quark s quantum field , i.e. , the quantum fluctuation on the background fields . the lagrangian with the background field theory can be found in ref.@xcite . usually , the background quark and gluon fields satisfy the following equations of motion @xmath23 where @xmath24 and @xmath25 indicate the fundamental and adjoint representations of the gauge covariant derivatives respectively . as an advantage of the background field theory , one can take different gauges for dealing with the quantum fluctuations and background fields . more specifically , one can adopt the background gauge " @xcite @xmath26 for the gluon quantum field , the schwinger gauge or the fixed - point gauge " @xcite @xmath27 for the background field . within the framework of the background field theory , the quark propagator would be affected by the background quark and/or gluon fields , which satisfies the equation @xmath28 if taking @xmath29 eq.([quaprofun1 ] ) can be changed as @xmath30 where @xmath31 , and @xmath32 moreover , using the fixed - point gauge " as shown in eq.([fpgau ] ) , the gluon background field can be expressed by the gauge invariant @xmath33 as @xmath34 where @xmath33 is a notation for @xmath35 , where the indexes @xmath36 indicate the covariant derivative up to @xmath37-th order . @xmath38 with @xmath39 stands for the gluon field strength tensor and @xmath40 is the structure constant of @xmath41 group . substituting eq.([gluexp ] ) into eq.([quaprofun3 ] ) , we obtain the expressions for @xmath42 . by further using eq.([quaprofun2 ] ) , we can obtain the required quark propagator in the background field , which can be expressed as @xmath43 the quark propagators with various gauge invariant tensors @xmath33 that shall result in up to dimension - six operators are presented in appendix a. the feynman diagrams for the quark propagators ( [ appa1],@xmath44,[prop2 ] ) that with various gauge invariant tensors are shown in fig.([propagator ] ) , where thirteen figures , i.e. fig.([propagator]a ) , @xmath44 , fig.([propagator]m ) , correspond to @xmath45 , @xmath44 , @xmath46 , respectively . the cross ( @xmath47 ) attached to the gluon line indicates the tensor of the local gluon background field with @xmath37 " stands for @xmath37-th order covariant derivative . because the fixed - point gauge " , as indicated by eq.([fpgau ] ) , violates the translation invariance , the quark propagator from @xmath48 to @xmath49 , @xmath50 , can not be directly obtained by applying the replacement @xmath51 in eq.([prop4 ] ) . it can be related with @xmath52 via the relation @xcite @xmath53 where @xmath54 stands for the charge conjugation matrix and the symbol @xmath55 indicates transpose of both the dirac and the color matrices . when applying the svz sum rules to calculate the moments of meson , one would encounter the vertex operator , @xmath56 , where @xmath57 indicates some kind of dirac matrices , e.g. , @xmath58 for the pion leading - twist da , @xmath59 or @xmath60 for the pion twist-3 das , and etc .. we have @xmath61 where the @xmath44 stands for the higher - order terms which are irrelevant for our present analysis and @xmath62 we can expand the operator @xmath0 into series with operators @xmath63 and @xmath64 . for the purpose , we first expand @xmath0 as @xmath65 \nonumber\\ & & + 3 ( z\cdot \tensor{\partial } ) ( z\cdot \underline{b})^2 + ( z\cdot b)^3 , \nonumber\\ & & \cdots \cdots , \nonumber\end{aligned}\ ] ] where the underline " below @xmath66 ( or the latter @xmath48 ) indicates that the operation @xmath67 does not act on it . in deriving those equations , the following equation has been adopted , @xmath68.\ ] ] by keeping only those terms that shall leads to operators up to dimension - six , we obtain @xmath69 g_{\mu\nu;\rho\sigma } , \nonumber\\ ( z\cdot \tensor{d})^n_{5(1 ) } & = & -\frac{n(n-1)}{3 } ( z\cdot \tensor{\partial})^{n-2 } \nonumber\\ & & \times \underline{x}^\mu \underline{x}^\rho \underline{x}^\lambda z^\nu z^\sigma g_{\mu\nu } g_{\rho\sigma;\lambda } , \nonumber\\ ( z\cdot \tensor{d})^n_{5(2 ) } & = & -\frac{n(n-1)}{3 } ( z\cdot \tensor{\partial})^{n-2 } \nonumber\\ & & \times \underline{x}^\mu \underline{x}^\rho \underline{x}^\lambda z^\nu z^\sigma g_{\mu\nu;\lambda } g_{\rho\sigma } , \nonumber\\ ( z\cdot \tensor{d})^n_{5(3 ) } & = & \left [ -\frac{i}{15 } \times n ( z\cdot \tensor{\partial})^{n-1 } \underline{x}^\mu \underline{x}^\rho \underline{x}^\sigma \underline{x}^\lambda z^\nu \right . \nonumber\\ & & - \frac{i}{45 } n(n-1)(n-2 ) ( z\cdot \tensor{\partial})^{n-3 } \nonumber\\ & & \left . \times \underline{x}^\mu \left ( \underline{x}^\rho z^\sigma z^\lambda + z^\rho \underline{x}^\sigma z^\lambda + z^\rho z^\sigma \underline{x}^\lambda \right ) z^\nu \right ] \nonumber\\ & & \times g_{\mu\nu;\rho\sigma\lambda } , \nonumber\\ ( z\cdot \tensor{d})^n_{6(1 ) } & = & \frac{i}{6 } n(n-1)(n-2 ) ( z\cdot \tensor{\partial})^{n-3 } \nonumber\\ & & \times \underline{x}^\mu \underline{x}^\rho \underline{x}^\lambda z^\nu z^\sigma z^\tau g_{\mu\nu } g_{\rho\sigma } g_{\lambda\tau } , \nonumber\\ ( z\cdot \tensor{d})^n_{6(2 ) } & = & \left [ -\frac{n(n-1)}{8 } ( z\cdot \tensor{\partial})^{n-2 } \underline{x}^\mu \underline{x}^\rho \underline{x}^\lambda \underline{x}^\tau z^\nu z^\sigma \right . \nonumber\\ & & - \frac{n(n-1)(n-2)(n-3)}{12 } ( z\cdot \tensor{\partial})^{n-4 } \nonumber\\ & & \left . \times \underline{x}^\mu \underline{x}^\rho z^\lambda z^\tau z^\nu z^\sigma \right ] g_{\mu\nu } g_{\rho\sigma;\lambda\tau } , \nonumber\\ ( z\cdot \tensor{d})^n_{6(3 ) } & = & -\frac{2}{9 } n(n-1 ) ( z\cdot \tensor{\partial})^{n-2 } \nonumber\\ & & \times \underline{x}^\mu \underline{x}^\rho \underline{x}^\lambda \underline{x}^\tau z^\nu z^\sigma g_{\mu\nu;\lambda } g_{\rho\sigma;\tau } , \nonumber\\ ( z\cdot \tensor{d})^n_{6(4 ) } & = & -\frac{n(n-1)}{8 } ( z\cdot \tensor{\partial})^{n-2 } \nonumber\\ & & \times \underline{x}^\mu \underline{x}^\rho \underline{x}^\lambda \underline{x}^\tau z^\nu z^\sigma g_{\mu\nu;\lambda\tau } g_{\rho\sigma } , \nonumber\\ ( z\cdot \tensor{d})^n_{6(5 ) } & = & \left [ -\frac{i}{72 } \times n ( z\cdot \tensor{\partial})^{n-1 } \underline{x}^\mu \underline{x}^\rho \underline{x}^\sigma \underline{x}^\lambda \underline{x}^\tau z^\nu \right . \nonumber\\ & & - \frac{i}{216 } n(n-1)(n-2 ) ( z\cdot \tensor{\partial})^{n-3 } \nonumber\\ & & \times \underline{x}^\mu \left ( z^\rho z^\sigma \underline{x}^\lambda \underline{x}^\tau + z^\rho \underline{x}^\sigma z^\lambda \underline{x}^\tau + z^\rho \underline{x}^\sigma \underline{x}^\lambda z^\tau \right . \nonumber\\ & & \left . + \underline{x}^\rho z^\sigma z^\lambda \underline{x}^\tau + \underline{x}^\rho z^\sigma \underline{x}^\lambda z^\tau + \underline{x}^\rho \underline{x}^\sigma z^\lambda z^\tau \right ) z^\nu \nonumber\\ & & - \frac{i}{360 } n(n-1)(n-2)(n-3)(n-4 ) \nonumber\\ & & \left . \times ( z\cdot \tensor{\partial})^{n-5 } \underline{x}^\mu z^\rho z^\sigma z^\lambda z^\tau z^\nu \right ] g_{\mu\nu;\rho\sigma\lambda\tau } , \nonumber\end{aligned}\ ] ] where the subscript @xmath70 with @xmath71 stands for the dimension of the operator , in which @xmath72 stands for the @xmath73-th type of the operator with same dimension . for example , there is two type of dimension four operators , three type of dimension five operators and five type of dimension - six operators . fig.([vertex ] ) represents the feynman diagrams for the vertex operator @xmath74 under the background field theory , where thirteen figures , i.e. fig.([vertex]a ) , @xmath44 , fig.([vertex]m ) , correspond to @xmath75 , @xmath44 , @xmath76 , respectively . the cross ( @xmath47 ) attached to the gluon line indicates the tensor of the local gluon background field with @xmath37 " stands for the @xmath37-th order covariant derivative . stands for the background quark field . the left big dot and the right big dot stand for the vertex operators @xmath77 and @xmath78 in the correlator , respectively . the full feynman diagrams can be found in figs.([feyna],[feynb ] ) . , scaledwidth=45.0% ] considering the definition @xmath79 where @xmath80 is the pion decay constant and the @xmath37-th order moments of the pion da are defined as @xmath81 especially , the @xmath49-th moment satisfies the normalization condition @xmath82 to derive the sum rules for the pion leading - twist da moments @xmath83 , we introduce the following correlation function ( correlator ) , @xmath84 where @xmath85 , @xmath86 and the currents @xmath87 in physical region , the correlator ( [ correlator ] ) can be treated by inserting a complete set of intermediate hadronic states , which can be simplified with the help of eq.([mom ] ) as @xmath88 where the quark - hadron duality has been adopt and @xmath89 stands for the continuum threshold . on the other hand , we apply the ope for the correlator ( [ correlator ] ) in deep euclidean region . substituting the replacement rules ( [ bfrep0],[bfrep ] ) into the correlator ( [ correlator ] ) and applying the corresponding feynman rules , we obtain @xmath90 the feynman diagrams for the correlator ( [ ope ] ) are showed in fig.([feyn ] ) , whose details are further presented in figs.([feyna],[feynb ] ) . the solid and helical lines are for the quark and gluon propagators , respectively . the cross @xmath91 symbol stands for the background quark field . the left big dot and the right big dot are the vertex operators @xmath77 and @xmath92 , respectively . fig.([feyn]a ) corresponds to the first term in eq.([ope ] ) , fig.([feyn]b ) and its permutation - diagrams correspond to the second and third terms of eq.([ope ] ) , respectively . as for fig.([feyna ] ) , we observe * figs.([feyna]l,[feyna]p,[feyna]q,[feyna]s ) contribute zero because their quark loop has @xmath93=0 $ ] . the contribution from fig.([feyna]g ) is negligible due to strong suppression from small @xmath94-quark current masses . * fig.([feyna]a ) provides the perturbative contribution , figs.([feyna]b,[feyna]r ) provide the contribution proportional to the dimension - four condensate @xmath16 , figs.([feyna]c,[feyna]h,[feyna]m,[feyna]u ) provide the contribution proportional to the dimension - six condensate @xmath95 and figs.([feyna]e,[feyna]j,[feyna]o,[feyna]t ) provide the contribution proportional to the dimension - six condensate @xmath96 . here @xmath16 and @xmath95 are abbreviations for @xmath97 and @xmath98 . * the remaining diagrams involves the contributions proportional to the dimension - six condensate either @xmath95 or @xmath99 , where @xmath100 . as for fig.([feynb ] ) , we observe that * fig.([feynb]a ) provides the contribution proportional to the dimension three condensate @xmath15 and the dimension five condensate @xmath101 . fig.([feynb]b ) provides the contribution proportional to either the dimension five condensate @xmath101 or the dimension - six condensate @xmath102 . * the remaining three diagrams ( [ feynb]c,[feynb]d,[feynb]e ) provide the contribution proportional to the dimension - six condensate @xmath102 , where @xmath103 . we adopt the @xmath104-scheme to deal with the infrared divergences emerged in fig.([feyna ] ) , whose divergent terms shall be absorbed into the poin da following the way suggested by ref.@xcite . during the calculation , we shall meet with the following matrix elements : @xmath105 the formulas relating the first three vacuum matrix elements with the conventional condensates can be found in ref.@xcite , and the corresponding formulas for the remaining vacuum matrix elements under the @xmath14-dimensional space , @xmath106 , are presented in appendix b. as a combination the correlator within different regions , the sum rules for the pion da moments can be derived by the dispersion relation @xmath107 where @xmath108 is the borel parameter and the borel transformation operator @xmath109 our final result reads @xmath110 where @xmath111 , \nonumber\\ & & \mathbb{b } = \frac{8(165n^2 + 323n + 131)}{729(n+1 ) } + \frac{8(51n+44)}{243 } \ln \left(\frac{m^2}{\mu^2}\right ) + \theta(n-2 ) \left\ { -\frac{4(16n^3 - 89n^2 - 39n-30)}{729n(n+1 ) } + \frac{392n}{243 } \ln \left(\frac{m^2}{\mu^2}\right ) \right . \nonumber\\ & & \quad\quad \left . + \sum^{n-2}_{k=0 } ( -1)^k \left [ \frac{64n(k+1)}{243(n+1)(n - k ) } - \frac{8(10k^2 - 10nk+58k-7n+48)}{243(n - k)(n - k-1 ) } + \frac{8(8k-5n-2)}{243(k+1)(n - k+1 ) } \right ] \right\}. \nonumber\end{aligned}\ ] ] as a final remark , we can obtain the gegenbauer moments of the pion leading - twist da with the help of @xmath83 . that is , by substituting eq.([dagegexp ] ) into eq.([nmom ] ) , we have @xmath112 to do the numerical calculation , we adopt @xcite : @xmath113 and @xmath114 . as for the condensates , * we adopt @xmath115 which is derived from the current algebra @xcite and the partially conserved axial current @xcite . * we adopt @xcite , @xmath116 which is derived by using @xmath117 with @xmath118 @xcite . * as suggested in ref.@xcite , we adopt @xcite @xmath119 where @xmath120 @xcite . more definitely , we take @xcite @xmath121 and @xmath122 . and we take @xmath123 . * combining ( [ relgluqua ] ) with ( [ ratquaglu ] ) , together with the ratio @xmath124 @xcite , we obtain @xmath125 and @xmath126 the leading - order @xmath127 is fixed by @xmath128 @xcite and the renormalization scale is taken as @xmath129 . usually , the continuum threshold @xmath89 is taken as the square of the first exciting state of pion , i.e. , @xmath130 , which however may underestimate some continuum states contributions . at the present , we determine the value of @xmath89 from the sum rules of the @xmath49-th moment @xmath131 together with its normalization condition ( [ normalize0 ] ) , which leads to @xmath132 . .the moments @xmath83 of the pion leading - twist da at the scale @xmath133 and various condensates contributions over the total dispersion integration . @xmath134 and @xmath135 are abbreviations for @xmath136 and @xmath137 . [ cols="^,^,^,^",options="header " , ] [ daparameter ] using those moments ( [ mom21],[mom22],[mom23 ] ) , together with the formulas ( [ gegmom1],[gegmom2],[gegmom3 ] ) , we can obtain the gegenbauer moments @xmath138 , @xmath139 and @xmath140 . for example , we obtain @xmath141 and @xmath142 . we present a comparison of @xmath143 under the svz sum rules and the lattice qcd in tab.[taba2 ] . phenomenologically , a lcsr analysis on the pion electromagnetic form factor gives @xmath144 @xcite and @xmath145 @xcite . a lcsr analysis on the pion transition form factor gives @xmath144 @xcite , @xmath146 @xcite , @xmath147 @xcite , @xmath148 @xcite and @xmath149 @xcite . a lcsr analysis on the @xmath150 gives @xmath151 @xcite , @xmath152 @xcite and @xmath153 @xcite . it shows that our present estimation of @xmath154 agrees with most of the estimations within errors derived in the literature . especially , our central value is close to the lattice qcd estimations @xcite . following the idea of ref.@xcite , by including the correlation effect from the transverse distribution , we can construct a pion da model as @xmath155 - \textrm{erf } \left [ \sqrt{\frac{m_q^2}{8\beta_\pi^2 x(1-x ) } } \right ] \right\ } , \label{da_model}\end{aligned}\ ] ] where the light constitute quark @xmath156 gev , the error function @xmath157 , and the longitudinal part can be constructed as @xmath158.\ ] ] the parameters @xmath159 , @xmath160 , @xmath161 and @xmath162 can be fixed by the pion da normalization , the constraint from @xmath163 @xcite and the above determined gegenbauer moments @xmath154 and @xmath164 . several typical values for those parameters are presented in table [ daparameter ] . the background field theory provides an unambiguous physical picture for both the qcd physical vacuum and the svz sum rules . under this framework , the ope of the correlator can be calculated systematically . for the first time , we provide the quark propagator and the vertex up to dimension - six operators under the background field theory . it is noted that those newly dimension - six terms for the quark propagator are important for a sound estimation of a physics process up to dimension - six condensates . because the mass terms are kept explicitly in the quark propagator , it is applicable for both the massless and massive cases . in the present paper , we have studied the pion leading - twist da within the qcd sum rules under the background field theory . the resultant sum rules for the da moments up to dimension - six condensates have been presented . it has been found that the dimension - six condensates can provide sizable contributions to the da moments in comparison to contributions from the lower dimension condensates . within errors , our estimation of @xmath4 agrees with most of the estimations derived in the literature . its central value is larger than those determined by previous qcd sum rules , which is mainly caused by the contribution from the dimension - six condensate @xmath137 . by varying the second and four gegenbauer moments within their reasonable region , i.e. @xmath141 and @xmath142 . the solid line stands for the central values for @xmath165 and the shaded band stands for the theoretical uncertainties caused by the gegenbauer moments . , scaledwidth=50.0% ] as an application of the suggested pion da model ( [ da_model ] ) , we redraw the pion - photon tff @xmath7 in fig.([fig_pptff ] ) . all the formulas for @xmath7 can be found in our previous paper @xcite , only one need to change the pion da used there to be our present model . fig.([fig_pptff ] ) indicates that in the large @xmath166 region , the pion - photon tff @xmath7 lies in between the belle data and the babar data , which is consistent with the recent @xmath167 factorization estimation with the joint resummation @xcite . * acknowledgments * : this work was supported in part by the natural science foundation of china under grant no.11075225 , no.11235005 and no.11275280 . within the framework of the background field theory , the full quark propagator with various gauge invariant tensors that shall result in up to dimension - six operators can be written as @xmath168 in which the propagators with various gauge invariant tensors are @xmath169 \gamma^\nu g_{\mu\nu;\rho } \right\},\\ & & s_f^{4(1)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left\ { \left [ \frac{1}{4 } \left ( \frac{\gamma^\mu ( m - \not\ ! p)}{(m^2-p^2)^3 } - 2\frac{p^\mu}{(m^2-p^2)^3 } \right ) \gamma^\nu \gamma^\rho \gamma^\sigma \right.\right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . + \frac{1}{2 } \left ( \frac{(m + \not\ ! p ) \gamma^\mu}{(m^2-p^2)^3 } g^{\nu\sigma } + 4\frac{\gamma^\mu ( m - \not\ ! p)}{(m^2-p^2)^4 } p^\nu p^\sigma \right ) \gamma^\rho \right ] g_{\mu\nu}g_{\rho\sigma } \right\ } , \\ & & s_f^{4(2)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left\ { \frac{i}{4 } \left [ \frac{g^{\{\mu\rho}\gamma^{\sigma\ } } ( m - \not\ ! p)}{(m^2-p^2)^3 } - 2\frac{g^{\{\mu\rho}p^{\sigma\}}}{(m^2-p^2)^3 } + 4\frac{\gamma^{\{\mu}p^\rho p^{\sigma\ } } ( m - \not\ ! p)}{(m^2-p^2)^4 } \right ] \gamma^\nu g_{\mu\nu;\rho\sigma } \right\ } , \\ & & s_f^{5(1)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left\ { -\frac{i}{3 } \left [ \left ( 3\frac{\gamma^\mu ( m-\not\ ! p ) \gamma^\nu}{(m^2-p^2)^4 } ( p^\lambda \gamma^\rho + p^\rho \gamma^\lambda ) - \frac{\gamma^\nu ( g^{\mu\lambda } \gamma^\rho + g^{\mu\rho } \gamma^\lambda)}{(m^2-p^2)^3 } \right ) \gamma^\sigma \right.\right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . + 4 \left ( \frac{\gamma^\mu ( m-\not\ ! p)}{(m^2-p^2)^4 } g^{\{\nu\sigma } p^{\lambda\ } } + 2 \frac{p^\mu g^{\{\nu\sigma } p^{\lambda\}}}{(m^2-p^2)^4 } + 6\frac{\gamma^\mu ( m-\not\ ! p)}{(m^2-p^2)^5 } p^\sigma p^\lambda \right ) \gamma^\rho \right ] g_{\mu\nu}g_{\rho\sigma;\lambda } \right\ } , \\ & & s_f^{5(2)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left\ { \frac{2i}{3 } \left [ \left ( \frac{g^{\mu\lambda}}{(m^2-p^2)^3 } + \frac{6p^\mu p^\lambda}{(m^2-p^2)^4 } \right ) \gamma^\nu \gamma^\rho \gamma^\sigma \right.\right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . - 2 \left ( \frac{\gamma^\mu ( m-\not\ ! p)}{(m^2-p^2)^4 } g^{\{\nu\sigma } p^{\lambda\ } } + 2 \frac{p^\mu g^{\{\nu\sigma } p^{\lambda\}}}{(m^2-p^2)^4 } + 6\frac{\gamma^\mu ( m-\not\ ! p)}{(m^2-p^2)^5 } p^\sigma p^\lambda \right ) \gamma^\rho \right ] g_{\mu\nu;\lambda}g_{\rho\sigma } \right\ } , \\ & & s_f^{5(3)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left\ { \frac{4}{15 } \left [ \frac{g^{\{\rho\sigma } p^\lambda \gamma^{\mu\ } } ( m-\not\ ! p)}{(m^2-p^2)^4 } - 2 \frac{g^{\{\rho\sigma } p^\lambda p^{\mu\}}}{(m^2-p^2)^4 } + 6 \frac{\gamma^{\{\mu } p^\rho p^\sigma p^{\lambda\ } } ( m-\not\ ! p)}{(m^2-p^2)^5 } \right.\right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . - \frac{g^{(\mu\rho\sigma\lambda)}}{(m^2-p^2)^3 } \right ] \gamma^\nu g_{\mu\nu;\rho\sigma\lambda } \right\ } , \\ & & s_f^{6(1)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \frac{i}{8 } \left\ { \left [ \frac{\gamma^\mu ( m - \not\ ! p)}{(m^2-p^2)^4 } - 4\frac{p^\mu}{(m^2-p^2)^4 } \right ] \gamma^\nu \gamma^\rho \gamma^\sigma \gamma^\lambda \gamma^\tau \right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left . + 2\left [ 3\frac{\gamma^\mu ( m - \not\ ! p)}{(m^2-p^2)^4 } g^{\sigma\tau } + 16\frac{\gamma^\mu ( m - \not\ ! p)}{(m^2-p^2)^5 } p^\sigma p^\tau - 4\frac{g^{\mu\sigma}p^\tau + g^{\mu\tau}p^\sigma}{(m^2-p^2)^4 } \right ] \gamma^\nu \gamma^\rho \gamma^\lambda \right\ } g_{\mu\nu}g_{\rho\sigma}g_{\lambda\tau } , \\ & & s_f^{6(2)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left(-\frac{1}{8}\right ) \left\ { \left [ 3\frac{\gamma^\mu ( m - \not\ ! p ) \gamma^\nu}{(m^2-p^2)^4 } g^{\{\lambda\tau } \gamma^{\rho\ } } + 16\frac{\gamma^\mu ( m - \not\ ! p ) \gamma^\nu}{(m^2-p^2)^5 } \gamma^{\{\rho } p^\lambda p^{\tau\ } } \right.\right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left . - 4\frac{\gamma^\nu}{(m^2-p^2)^4 } g^{\mu\{\lambda } p^\tau \gamma^{\rho\ } } \right ] \gamma^\sigma + 4\left [ \frac{m + \not\ ! p}{(m^2-p^2)^4 } g^{(\nu\sigma\tau\lambda ) } + 6\frac{m+\not\ ! p}{(m^2-p^2)^5 } g^{\{\nu\sigma } p^{\tau } p^{\lambda\ } } \right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . + 48\frac{m + \not\ ! p}{(m^2-p^2)^6 } p^\nu p^\sigma p^\tau p^\lambda \right ] \gamma^\mu \gamma^\rho \right\ } g_{\mu\nu}g_{\rho\sigma;\lambda\tau } , \\ & & s_f^{6(3)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left ( -\frac{2}{9 } \right ) \left\ { 3\left ( \left [ 2\frac{\gamma^\mu ( m - \not\ ! p)}{(m^2-p^2)^5 } p^\lambda p^\tau - \frac{g^{\{\mu\lambda } p^{\tau\}}}{(m^2-p^2)^4 } - \frac{4p^\mu p^\lambda p^\tau}{(m^2-p^2)^5 } \right ] \gamma^\nu \gamma^\rho \right.\right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left . + ( \mu \leftrightarrow \lambda ) + ( \rho \leftrightarrow \tau ) + ( \mu \leftrightarrow \lambda , \rho \leftrightarrow \tau ) \right ) \gamma^\sigma + 4\left [ \frac{m + \not\ ! p}{(m^2-p^2)^4 } g^{(\nu\lambda\sigma\tau ) } \right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . + 6\frac{m+\not\ ! p}{(m^2-p^2)^5 } g^{\{\nu\lambda } p^{\sigma } p^{\tau\ } } + 48\frac{m + \not\ ! p}{(m^2-p^2)^6 } p^\nu p^\lambda p^\sigma p^\tau \right ] \gamma^\mu \gamma^\rho \right\ } g_{\mu\nu;\lambda}g_{\rho\sigma;\tau } , \\ & & s_f^{6(4)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left ( -\frac{1}{2 } \right ) \left\ { \left [ \frac{m + \not\ ! p}{(m^2-p^2)^4 } g^{(\nu\tau\lambda\sigma ) } + 6\frac{m+\not\ ! p}{(m^2-p^2)^5 } g^{\{\nu\tau } p^{\lambda } p^{\sigma\ } } \right.\right.\nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . + 48\frac{m + \not\ ! p}{(m^2-p^2)^6 } p^\nu p^\tau p^\lambda p^\sigma \right ] \gamma^\mu \gamma^\rho - 3\left [ \frac{g^{\{\mu\lambda } p^{\tau\}}}{(m^2-p^2)^4 } + \frac{8p^\mu p^\lambda p^\tau}{(m^2-p^2)^5 } \right ] \gamma^\nu \gamma^\rho \gamma^\sigma \right\ } g_{\mu\nu;\lambda\tau}g_{\rho\sigma } , \\ & & s_f^{6(5)}(x,0 ) = i \int \frac{d^4p}{(2\pi)^4 } e^{-ip\cdot x } \left\ { -\frac{i}{18 } \left [ \frac{g^{[\rho\sigma\lambda\tau } \gamma^{\mu ] } ( m-\not\ ! p)}{(m^2-p^2)^4 } - 4 \frac{g^{[\rho\sigma\lambda\tau } p^{\mu]}}{(m^2-p^2)^4 } + 6 \frac{g^{\{\rho\sigma } p^\lambda p^\tau \gamma^{\mu\ } } ( m-\not\ ! p)}{(m^2-p^2)^5 } \right.\right . \nonumber\\ & & \quad\quad\quad\quad\ \ \left.\left . - 12 \frac{g^{\{\rho\sigma } p^\lambda p^\tau p^{\mu\}}}{(m^2-p^2)^5 } + 48 \frac{p^{\{\rho } p^\sigma p^\lambda p^\tau \gamma^{\mu\ } } ( m-\not\!p)}{(m^2-p^2)^6 } \right ] \gamma^\nu g_{\mu\nu;\rho\sigma\lambda\tau } \right\ } , \label{prop2}\end{aligned}\ ] ] where @xmath170 } & = & g^{(\mu\nu\rho\sigma ) } p^\lambda + g^{(\lambda\nu\rho\sigma ) } p^\mu + g^{(\mu\lambda\rho\sigma ) } p^\nu + g^{(\mu\nu\lambda\sigma ) } p^\rho + g^{(\mu\nu\rho\lambda ) } p^\sigma , \nonumber\\ g^{\{\mu\nu } p^{\rho\ } } & = & g^{\mu\nu } p^\rho + g^{\mu\rho } p^\nu + g^{\nu\rho } p^\mu , \nonumber\\ g^{\{\mu\nu } p^\rho p^{\sigma\ } } & = & g^{\mu\nu } p^\rho p^\sigma + g^{\mu\rho } p^\nu p^\sigma + g^{\mu\sigma } p^\nu p^\rho + g^{\nu\rho } p^\mu p^\sigma + g^{\nu\sigma } p^\mu p^\rho + g^{\rho\sigma } p^\mu p^\nu , \nonumber\\ g^{\{\mu\nu } p^\rho p^\sigma p^{\lambda\ } } & = & g^{\mu\nu } p^\rho p^\sigma p^\lambda + g^{\mu\rho } p^\nu p^\sigma p^\lambda + g^{\mu\sigma } p^\rho p^\nu p^\lambda + g^{\mu\lambda } p^\rho p^\sigma p^\nu + g^{\nu\rho } p^\mu p^\sigma p^\lambda + g^{\nu\sigma } p^\rho p^\mu p^\lambda \nonumber\\ & & + g^{\nu\lambda } p^\rho p^\sigma p^\mu + g^{\rho\sigma } p^\mu p^\nu p^\lambda + g^{\rho\lambda } p^\mu p^\sigma p^\nu + g^{\sigma\lambda } p^\rho p^\mu p^\nu \nonumber\\ g^{\{\mu\rho } \gamma^{\sigma\ } } & = & g^{\mu\rho } \gamma^\sigma + g^{\mu\sigma } \gamma^\rho + g^{\rho\sigma } \gamma^{\mu } , \nonumber\\ g^{[\rho\sigma\lambda\tau } \gamma^{\mu ] } & = & \gamma^\mu g^{(\rho\sigma\lambda\tau ) } + ( \mu \leftrightarrow \rho ) + ( \mu \leftrightarrow \sigma ) + ( \mu \leftrightarrow \lambda ) + ( \mu \leftrightarrow \tau ) , \nonumber\\ g^{\{\rho\sigma } p^\lambda \gamma^{\mu\ } } & = & \gamma^\mu g^{\{\rho\sigma } p^{\lambda\ } } + ( \mu \leftrightarrow \rho ) + ( \mu \leftrightarrow \sigma ) + ( \mu \leftrightarrow \lambda ) , \nonumber\\ g^{\{\rho\sigma } p^\lambda p^\tau \gamma^{\mu\ } } & = & \gamma^\mu g^{\{\rho\sigma } p^\lambda p^{\tau\ } } + ( \mu \leftrightarrow \rho ) + ( \mu \leftrightarrow \sigma ) + ( \mu \leftrightarrow \lambda ) + ( \mu \leftrightarrow \tau ) , \nonumber\\ \gamma^{\{\mu } p^\rho p^{\sigma\ } } & = & \gamma^\mu p^\rho p^\sigma + \gamma^\rho p^\mu p^\sigma + \gamma^\sigma p^\mu p^\rho , \nonumber\\ \gamma^{\{\mu } p^\rho p^\sigma p^{\lambda\ } } & = & \gamma^\mu p^\rho p^\sigma p^\lambda + ( \mu \leftrightarrow \rho ) + ( \mu \leftrightarrow \sigma ) + ( \mu \leftrightarrow \lambda ) , \nonumber\\ \gamma^{\{\mu } p^\rho p^\sigma p^\lambda p^{\tau\ } } & = & \gamma^\mu p^\rho p^\sigma p^\lambda p^\tau + ( \mu \leftrightarrow \rho ) + ( \mu \leftrightarrow \sigma ) + ( \mu \leftrightarrow \lambda ) + ( \mu \leftrightarrow \tau ) , \nonumber\\ g^{\mu\{\lambda } p^\tau \gamma^{\rho\ } } & = & g^{\mu\lambda } p^\tau \gamma^\rho + g^{\mu\tau } p^\lambda \gamma^\rho + g^{\mu\rho } p^\tau \gamma^\lambda + g^{\mu\tau } p^\rho \gamma^\lambda + g^{\mu\rho } p^\lambda \gamma^\tau + g^{\mu\lambda } p^\rho \gamma^\tau . \nonumber\end{aligned}\ ] ] in those expressions , the mass terms are kept explicitly , which are appropriate not only for the light - quark case but also for the heavy - quark one . the vacuum matrix elements in the @xmath14-dimension space are @xmath171 , \nonumber\\ & & \textrm{tr}_c \left<0\left| g_{\mu\nu;\lambda } g_{\rho\sigma;\tau } \right|0\right > \nonumber\\ & & \quad\quad = -\frac{8}{27d^2(d-1 ) } g_s^2 \sum_{u , d , s } \left < g_s\bar{\psi}\psi\right>^2 \nonumber\\ & & \quad\quad \times \left [ 2 g_{\lambda\tau } ( g_{\mu\rho}g_{\nu\sigma } - g_{\mu\sigma}g_{\nu\rho } ) + g_{\lambda\rho } ( g_{\tau\mu}g_{\sigma\nu } - g_{\tau\nu}g_{\sigma\mu } ) \right . \nonumber\\ & & \quad\quad \left . - g_{\lambda\sigma } ( g_{\tau\mu}g_{\rho\nu } - g_{\tau\nu}g_{\rho\mu } ) \right ] , \nonumber\\ & & \left<0\left| g^a_{\mu\nu } g^b_{\rho\sigma } g^c_{\lambda\tau } \right|0\right > \nonumber\\ & & \quad\quad = \frac{\left < fg^3\right>}{24 d ( d-1 ) ( d-2 ) } f^{abc } \nonumber\\ & & \quad\quad \times \left\ { \left [ g_{\mu\rho } ( g_{\nu\lambda}g_{\sigma\tau } - g_{\nu\tau}g_{\sigma\lambda } ) - g_{\mu\sigma } ( g_{\nu\lambda}g_{\rho\tau } - g_{\nu\tau}g_{\rho\lambda } ) \right ] \right . \nonumber\\ & & \quad\quad \left . - \left [ g_{\nu\rho } ( g_{\mu\lambda}g_{\sigma\tau } - g_{\mu\tau}g_{\sigma\lambda } ) - g_{\nu\sigma } ( g_{\mu\lambda}g_{\rho\tau } - g_{\mu\tau}g_{\rho\lambda } ) \right ] \right\ } , \nonumber\\ & & \textrm{tr}_c \left<0\left| g_{\mu\nu } g_{\rho\sigma } g_{\lambda\tau } \right|0\right > \nonumber\\ & & \quad\quad = \frac{i}{4 d ( d-1 ) ( d-2 ) } \left < g_s^3fg^3\right > \nonumber\\ & & \quad\quad \times \left\ { \left [ g_{\mu\rho } ( g_{\nu\lambda}g_{\sigma\tau } - g_{\nu\tau}g_{\sigma\lambda } ) - g_{\mu\sigma } ( g_{\nu\lambda}g_{\rho\tau } - g_{\nu\tau}g_{\rho\lambda } ) \right ] \right . \nonumber\\ & & \quad\quad \left . - \left [ g_{\nu\rho } ( g_{\mu\lambda}g_{\sigma\tau } - g_{\mu\tau}g_{\sigma\lambda } ) - g_{\nu\sigma } ( g_{\mu\lambda}g_{\rho\tau } - g_{\mu\tau}g_{\rho\lambda } ) \right ] \right\ } , \nonumber\\ & & \left<0\left| g^a_{\mu\nu } g^b_{\rho\sigma;\lambda\tau } \right|0\right > \nonumber\\ & & \quad\quad = \delta^{ab } \left\ { \left ( a \times \sum_{u , d , s } \left < g_s\bar{\psi}\psi\right>^2 - b \times \left < g_s f g^3 \right > \right ) \right . \nonumber\\ & & \quad\quad \times \left [ 2 g_{\lambda\tau } ( g_{\mu\sigma}g_{\nu\rho } - g_{\mu\rho}g_{\nu\sigma } ) + g_{\rho\tau } ( g_{\mu\sigma}g_{\nu\lambda } - g_{\mu\lambda}g_{\nu\sigma } ) \right . \nonumber\\ & & \quad\quad \left . - g_{\sigma\tau } ( g_{\mu\lambda}g_{\nu\rho } - g_{\mu\rho}g_{\nu\lambda } ) \right ] \nonumber\\ & & \quad\quad + \left ( a \times \sum_{u , d , s } \left < g_s\bar{\psi}\psi\right>^2 + b \times \left < g_s f g^3 \right > \right ) \nonumber\\ & & \quad\quad \left . \times \left [ g_{\mu\tau } ( g_{\rho\nu}g_{\sigma\lambda } - g_{\rho\lambda}g_{\nu\sigma } ) + g_{\nu\tau } ( g_{\rho\lambda}g_{\sigma\mu } - g_{\rho\mu}g_{\sigma\lambda } ) \right ] \right\ } , \nonumber\\ & & \textrm{tr}_c \left<0\left| g_{\mu\nu } g_{\rho\sigma;\lambda\tau } \right|0\right > \nonumber\\ & & \quad\quad = 4\left ( a \times g_s^2 \sum_{u , d , s } \left < g_s\bar{\psi}\psi\right>^2 - b \times \left < g_s^3 f g^3 \right > \right ) \nonumber\\ & & \quad\quad \times \left [ 2 g_{\lambda\tau } ( g_{\mu\sigma}g_{\nu\rho } - g_{\mu\rho}g_{\nu\sigma } ) + g_{\rho\tau } ( g_{\mu\sigma}g_{\nu\lambda } - g_{\mu\lambda}g_{\nu\sigma } ) \right . \nonumber\\ & & \quad\quad \left . - g_{\sigma\tau } ( g_{\mu\lambda}g_{\nu\rho } - g_{\mu\rho}g_{\nu\lambda } ) \right ] \nonumber\\ & & \quad\quad + 4\left ( a \times g_s^2 \sum_{u , d , s } \left < g_s\bar{\psi}\psi\right>^2 + b \times \left < g_s^3 f g^3 \right > \right ) \nonumber\\ & & \quad\quad \times \left [ g_{\mu\tau } ( g_{\rho\nu}g_{\sigma\lambda } - g_{\rho\lambda}g_{\nu\sigma } ) + g_{\nu\tau } ( g_{\rho\lambda}g_{\sigma\mu } - g_{\rho\mu}g_{\sigma\lambda } ) \right ] , \nonumber\end{aligned}\ ] ] where @xmath172 and the symbol @xmath173 stands for tracing to the colour matrixes . obviously , @xmath174 our results agree with those of refs.@xcite , except for the matrix element @xmath175 . the result for @xmath175 given ref.@xcite is incorrect . it is noted that the correct equation for @xmath176 is @xmath177 \nonumber\\ & & + g^a_{\mu\alpha;\alpha\nu}- g^a_{\nu\alpha;\alpha\mu},\end{aligned}\ ] ] while in ref.@xcite , the coefficient of the first term has been taken as @xmath178 . t. huang , t. zhong and x. g. wu , phys . d * 88 * , 034013 ( 2013 ) ; t. huang , x. g. wu and t. zhong , chin . . lett . * 30 * , 041201 ( 2013 ) ; x.g . wu , t. huang and t. zhong , chin . c * 37 * , 063105 ( 2013 ) ; x. g. wu and t. huang , phys . d * 82 * , 034024 ( 2010 ) ; x. g. wu and t. huang , phys . d * 84 * , 074011 ( 2011 ) . x. d. xiang , x. n. wang and t. huang , commun . . phys . * 6 * , 117 ( 1986 ) ; s. v. mikhailov and a. v. radyushkin , phys . d * 45 * , 1754 ( 1992 ) ; a. p. bakulev , k. passek - kumericki , w. schroers and n. g. stefanis , phys . d * 70 * , 033014 ( 2004 ) . s. friot , d. greynat and e. de rafael , jhep * 10 * , 043 ( 2004 ) ; y. chung _ et al_. , phys . b * * 102 * * , 175 ( 1981 ) ; nucl . phys . b * 197 * , 55 ( 1982 ) ; h. g. dosch , m. jamin and s. narison , phys . b * 220 * , 251 ( 1989 ) . s. j. brodsky , t. huang , and g. p. lepage , in _ particles and fields-2 _ , proceedings of the banff summer institute , ban8 ; alberta , 1981 , edited by a. z. capri and a. n. kamal ( plenum , new york , 1983 ) , p. 143 ; g. p. lepage , s. j. brodsky , t. huang , and p. b.mackenize , _ ibid_. , p. 83 ; t. huang , in _ proceedings of xxth international conference on high energy physics _ , madison , wisconsin , 1980 , edited by l. durand and l. g pondrom , aip conf 69 ( aip , new york , 1981),p . 1000 .

we report the detection of off - pulse emission from two long period pulsars psr b0525 + 21 and psr b204516 using the interferometric mode of the giant metrewave radio telescope ( gmrt ) in the 325 mhz frequency band . the `` main - pulse '' ( or the equivalent term `` on - pulse '' used in this paper ) is the emission from within the polar - cap region of a pulsar ; and the off - pulse is the emission outside this main - pulse . off - pulse emission from pulsars has been a subject of interest since the discovery of pulsars four decades ago . according to the rotating vector model ( rvm ) proposed by radhakrishnan & cooke ( 1969 ) the observed pulses are due to relativistically beamed radiation along the open dipolar field lines . the plane of linear polarization traces the magnetic field line associated with the emission at every instant . several statistical studies have revealed that pulsar radio emission is in the form of a circular beam arising from the polar cap ( i.e. the main - pulse ) with an opening angle @xmath2 , and from a height above the magnetic pole equal to 12 per cent of the radius of the light cylinder . the width of the observed pulse is a geometrical function of @xmath2 ( which itself is a function of the emission height ) , the pulsar period , the angle between the rotation axis and the magnetic axis ( @xmath3 ) and the angle between the magnetic axis and the line of sight to the observer ( @xmath4 ) . the angles @xmath3 and @xmath4 can be estimated by fitting the rvm to the swing of the polarisation position angle ( ppa ) across the pulse @xcite . the observed main - pulse covers 5 - 10 per cent of the pulse period for 90 per cent of the pulsars . in several pulsars this main - pulse consists of multiple components including low - level bridge emission between such components ( rathnasree & rankin 1995 ) . the pulse width scales as 1/sin(@xmath3 ) , from the geometry of the emission model ; in the rare case of a pulsar with closely aligned magnetic and rotation axes ( i.e. @xmath5 or 180 ) the pulse width can be as high as 100 per cent . several pulsars have low level but detectable bridge emission between two widely spaced pulse components , making them candidates for an aligned rotator geometry . in rare cases when @xmath3 is close to 90 , emission from the opposite pole ( i.e. the inter - pulse ) may be observed at 180phase from the main - pulse . in all these instances the observed emission is believed to lie within the polar cap ( i.e. main - pulse ) as described earlier . in a few pulsars , low level emission components known as pre-/post - cursors ( ppc ) have been observed outside the main - pulse ( e.g. see mitra & rankin 2010 for a discussion ) . in these pulsars , all geometrical evidence indicate that the main pulse is consistent with emission from open field lines . the ppc components appear highly polarized and are far from the main - pulse . the discovery of the ppc component about 60away from the main pulse in psr b0943 + 10 is particularly interesting , as the line of sight almost grazes the emission cone for this pulsar @xcite . hence the ppc emission originates either from a much larger height , where due to spreading of dipolar field lines the ppc component can lie far away from the main - pulse , or from the regions of closed field lines . some pulsars occasionally emit giant pulses which are believed to arise close to the light cylinder rather than the polar cap . interestingly ppc pulsars have low spin - down luminosities ( @xmath6 erg s@xmath1 ) , whereas the giant pulses arise from pulsars with @xmath7 erg s@xmath1 . the ppc components and giant pulses are potential sources of magnetospheric off - pulse emission in pulsars . they challenge the conventional wisdom of pulsar radio emission arising only from open magnetic field lines close to the pulsar polar cap and raise questions about the origin of these emission components . the other possible source of off - pulse emission is the interstellar medium ( ism ) around the pulsar . the pulsar loses most of its rotational energy in the form of a relativistic wind which , when confined by the surroundings , may form a pulsar wind nebula ( pwn ) . several types of pwne are formed depending on the confinement mechanism . young and energetic pulsars are often located in their associated supernova remnant ( snr ) . the pulsar wind streaming into the ambient medium produces standing shocks resulting in a plerionic pwn ( like the one observed in the crab nebula ) . in older pulsars , where the surrounding snr is likely to be dissipated , the relativistic particles may interact with the ism magnetic field and radiate , creating ghost remnants ( proposed by blandford et al . 1973 , but not observed till date ) . finally , a pulsar moving through the ism with supersonic speed can produce a bow - shock nebula , where ram pressure balance is established between the pulsar wind and the ambient medium . several such bow shock nebulae have been detected in h@xmath3 , and some in the radio as plerionic bow shocks , but never in both ( see chatterjee & cordes 2002 for a detailed study ) . several searches for pwne have been conducted in the past with varying sensitivities and resolutions . however only in about 10 young pulsars have radio emission been detected outside the main - pulse and all of these are believed to be associated with pwne . in all such cases the associated pulsars are young ( 10@xmath8 - 10@xmath9yr ) , have high spin - down luminosities ( @xmath10 erg s@xmath1 ) and all except one are associated with snrs @xcite . it is believed that the pulsar wind and the environment change as pulsars slow down and age , making them less likely to harbour pwne . the present study was designed to detect radio emission away from the polar caps and not associated with pwne . therefore we targeted old and less energetic pulsars unlikely to harbour pwne . their profiles did not show any components outside the main - pulse . the targets were neither aligned rotators nor were inter - pulsars . we attempted to detect emission within a pulse phase range of approximately 80 - 250 from the peak in the main - pulse . we describe the gated interferometric observations and the data analysis in section 2 , the results in section 3 , the additional tests we carried out to confirm that the off - pulse emission were not instrumental artifacts in section 4 , and discuss the implications of the detections in section 5 . we chose two pulsars , b0525 + 21 and b204516 , whose pulse profiles did not show any features outside the main - pulse @xcite and were stronger than 50 mjy to be able to put a non - detection upper limit of at most 5 per cent of the pulsed flux . the properties of the targets are listed in table [ pulsprop ] . we targeted long period pulsars because : + 1 . pwne are so far only known from short - period , energetic pulsars in the vicinity of snr . a long period selects against both high energy pulsars and snr association , and hence pwne . we wanted at least 8 bins ( of 131 or 262 ms each ) across the pulse period to cleanly separate the off - pulse and the on - pulse regions . the temporal broadening across the 16 mhz bandwidth due to dispersion by the ism was 183.3 ms for psr b0525 + 21 ( binwidth 262 ms , gate width 1.31 s ) , and 47.8 ms for psr b204516 ( binwidth 131 ms , gate width 0.655 s ) . therefore , we did not have to dedisperse the signal . lccccccc pulsar&period & dm & dist&@xmath11 & @xmath12 & @xmath13&@xmath14 + & s & @xmath15 & kpc & year & erg @xmath16 & km @xmath16 & mjy + b0525 + 21 & 3.7455 & 50.937 & 2.28 & 1.48@xmath17 & 3.0@xmath18 & 229 & 80.5 + b204516 & 1.9616 & 11.456 & 0.95 & 2.84@xmath17 & 5.7@xmath18 & 511 & 169.3 + interferometric observations are better than single dish measurements for this study for several reasons : + 1 . an imaging interferometer is only insensitive to constant flux density background along the spatial axes while a standard pulsar receiver is usually insensitive to the constant background along the time axis ; the detection of off - pulse emission is essentially an attempt to find such a constant background along the time axis . self - calibration of interferometric data can correct instrumental and atmospheric gain fluctuations on very short time scales . the corrections are determined by the flux densities of the constant and bright background sources in the field and hence would not be affected by the pulse variation of the relatively weak pulsar . the higher spatial resolution of an interferometer greatly reduces the coincidence of unrelated sources ; thereby reducing the probability that the off - pulse emission is from an unrelated source within the synthesized beam . we imaged the pulsars with the giant metrewave radio telescope ( gmrt ) , an aperture - synthesis radio interferometer located near pune , india @xcite . the 30 antennas of 45 m diameter provide a maximum baseline of 27 km and can be operated at 6 frequency bands between 50 and 1450 mhz . we observed the pulsars at 325 mhz with a 16 mhz bandwidth split into 128 channels . the frequency was chosen for its optimal combination of resolution ( 10 ) and sensitivity ( few 100 @xmath19jy in 4 hours of observing ) . the high frequency - resolution was useful in flagging narrow - band radio - frequency interference ( rfi ) and to avoid bandwidth smearing . the shortest integration output by the hardware correlator is 0.131 s , though this data is usually averaged and median filtered to provide a standard data output integration of 16.77s . the two pulsars were first observed on 19 may , 2009 with 0.262s integration to keep the data output rate and size manageable . this time resolution ( of 7 bins ) was found to be inadequate in the case of psr b204516 while the data was of poor quality ( @xmath20 missing antennas ) for both targets . we reobserved the sources on 19 january , 2010 , with integrations of 0.131s for psr b204516 ( period 1.9616s ) , and 0.262s for psr b0525 + 21 ( period 3.7455s ) . this allowed us to divide both the pulsar periods into 14 time bins . several earlier studies used the technique of online gating to separate the on- and off - pulse regions using the timing information of pulsars , and accumulated time - averaged on- and off - pulse data @xcite . this resulted in relatively smaller size of data files and hence shorter computation times which was a major consideration 1 - 2 decades ago . the disadvantage was the impossibility of rectifying any errors during the gating process ( see results and discussions in gaensler et al . we recorded the data at sufficiently high resolution and applied offline gating . in the absence of accurate absolute time reference in the interferometric mode we used offline gating to isolate the off - pulse region of the pulsar period . the 30 antennas of gmrt provide 30 self , and 435 cross visibilities at each integration interval ( 0.131s or 0.262s ) . we folded the self - data to obtain the pulsar profile and identify the on - and off - pulse regions . this was used to gate the cross - antenna visibilities into on- and off - pulse data sets . the temporal and frequency gain variations in the self - data were corrected by normalising the instantaneous values by the local median . the timescale for the temporal median was 30 - 50 times the pulsar period to ensure that the pulse variation was retained . the data from all frequency channels and antennas were averaged with robust sigma - clipping and processed using the standard folding algorithm for pulsars ( hankins & rickett 1975 ) to determine the pulse profile . the pulsar profiles and the off- and on - pulse gated are shown in figure [ fig_profilep ] . the off- and on - pulse gates of psr b0525 + 21 were centred on phases 0.24 and 0.64 , respectively . the 5 bins nearest to each were averaged to construct the off- and on - pulse data sets . the corresponding phases for psr b204516 were 0.21 and 0.67 , respectively . the folding and gating procedures were carried out using software developed for this purpose . the on- and off - pulse data were cleaned of radio frequency interference ( rfi ) using the rfix algorithm @xcite . they were separately calibrated , flagged and imaged in a standard manner using the nrao aips package . the flux - density scale of the images was determined from observations of the calibrator 3c48 and the measurements of baars et al.(1977 ) with the latest corrections of perley et al ( 1999 ; in aips ) . the data sets were taken through several rounds of phase self - calibration followed by a final round of amplitude and phase self - calibration . * psr b0525 + 21 : * observations of the calibrator 0521 + 166 were interspersed with the target to correct for the amplitude and phase gain fluctuations . the presence of the extremely strong and extended crab nebula ( flux @xmath21 1000 jy and angular size @xmath22 10 ) at the edge of the primary beam ( 1.5away ) resulted in enhanced noise and strong ripples in the initial image . therefore , subsequent analyses were carried out with a lower uv cut - off at 1.5 k@xmath23 which reduced the artifacts and noise . this eliminated all structures larger than 2.3 from the image ; this did not affect our purpose as our target source was expected to be much smaller than 2 . * psr b204516 : * the calibrator 2137207 was observed for amplitude and phase calibration . a pointing error resulted in the pulsar being located 35 away from the field centre ( 60 per cent gain level of the primary beam ) . the primary beam correction was applied to the image to obtain the correct flux density for the pulsar . ] 10 mjy ) in the on- and off - pulse images . the scatter around unity confirmed that the flux - scale was the same for both . ( b0525 + 21 : top ; b204516 : bottom ) [ fig_onvoff],title="fig : " ] 10 mjy ) in the on- and off - pulse images . the scatter around unity confirmed that the flux - scale was the same for both . ( b0525 + 21 : top ; b204516 : bottom ) [ fig_onvoff],title="fig : " ] lccccccccc pulsar & frequency & time & beam & size & on rms & on flux & off rms & off flux & int . flux + & ( mhz ) & ( min ) & ( ) & ( pc ) & ( mjy@xmath24)&(mjy@xmath24 ) & ( mjy@xmath24)&(mjy@xmath24)&(mjy@xmath24 ) + b0525 + 21 & 332.9 & 160 & 9.5@xmath256.5 & 0.088 & 0.55 & 80.2@xmath265.8 & 0.45 & 3.9@xmath260.5 & 30.0@xmath262.1 + b204516 & 317.1 & 180 & 11.9@xmath257.2 & 0.042 & 2.10 & 305@xmath2622 & 0.65 & 4.3@xmath261.1 & 110.5@xmath267.9 + we detected off - pulse emission in both the pulsars , coincident with the location in the on - pulse image . the results are summarized in table [ results ] . the on- and the off - pulse emissions appeared to be unresolved for both the pulsars . we compared the flux densities of sources in the field of view to check the flux - density calibration . the ratios of on- and off - image flux densities of sources stronger than 10 mjy are plotted in figure [ fig_onvoff ] . there were 43 and 93 comparison sources in the fields of b0525 + 21 and b204516 , respectively . the scatter of values around unity , especially for strong sources ( which have smaller fractional flux errors ) confirmed the similarity of the flux scale of the on- and off - images for both the pulsars . * psr b0525 + 21 : * the pulsar was detected in the off - pulse image at the 8.6@xmath27 level ( flux - density 3.9@xmath260.5 mjy , rms noise 0.45 mjy / beam ) . the flux density of the pulsar in the on - pulse image was 80.2@xmath265.8 mjy ( rms noise 0.55 mjy / beam ) . the resolution element at the distance of the pulsar corresponds to 0.09 pc . the data from 19 may , 2009 , yielded an off - pulse detection of 3.6 mjy . * psr b204516 : * the pulsar was detected in the off - pulse image at the 6.6@xmath27 level ( flux density 4.3@xmath261.1 mjy ; rms noise 0.65 mjy / beam ) . the flux density of the pulsar in the on - pulse image was 305.2@xmath2621.9 mjy ( rms noise 2.1 mjy / beam ) . the resolution element at the distance of the pulsar corresponds to 0.04 pc . the flux - densities averaged over the entire pulsar period is 30.0@xmath262.1 for b0525 + 21 and 110.5@xmath267.9 for b204516 . these values are smaller than expected ( table [ pulsprop ] ) by factors of 1.5 and 2.7 , which can be caused by effects such as refractive interstellar scintillations @xcite . we confirmed the correctness of our flux scale by comparing the flux densities of the phase calibrators to their known values ( less than 1 per cent ) and by comparing the flux densities of 3 - 5 strong sources in each field to the interpolated values from nvss ( 1.4 ghz ) and vlss ( 74 mhz ) . the quoted errors on the flux densities were obtained by adding in quadrature the image rms , calibration errors and the error on the value of the primary flux - density calibrator @xcite . the signal - to - noise ratio in the published profiles of the pulsars @xcite are insufficient to conclude whether the off - pulse emission detected here is a pedestal ( throughout the period ) or a pulse confined to the off - pulse gate . firstly , we detected the off - pulse emission in two data sets of b0525 + 21 observed 8 months apart , and within the noise the two measurements were identical . we can safely rule out the detections being chance occurance of noise peaks . secondly , b204516 was positioned almost at the half - power point of the primary beam while b0525 + 21 was close to the field centre . the detection in both cases suggests that they are not the spurious structures occasionally seen at the centre of the field . we discuss below several effects which could result in spurious sources in the off - pulse data at the location of the pulsar ; and demonstrate that they are unlikely to be responsible for the same . to determine the probability of finding an unrelated source coinciding with the pulsar we used the vla first catalogue @xcite to determine the following relationship between source counts ( n ) and flux - density ( s ) : @xmath28 where @xmath29 is the flux at 1.4 ghz in mjy , @xmath30 is the flux bin width at 1.4 ghz in mjy and @xmath31 is the area of sky under consideration in @xmath32 . the area of the synthesized beam is @xmath22100 @xmath33 at 325 mhz . the flux - density for consideration in the equation is between 3@xmath25image - rms and 5 - 10 per cent of the pulsar flux , translated to 1.4 ghz ( using spectral index @xmath34 ) . the upper value is related to a realistic upper limit to the flux fraction in the off - pulse from the single - dish pulse profile @xcite . this yielded a probability of coincident unrelated sources of @xmath35 . the probability of detecting coincident sources for both pulsars is the square of the above number and therefore highly improbable . this scenario requires that approximately 1 - 4 per cent of the on - pulse data fall within the off - pulse gate . however , any monotonic drift in the clock , without recovery , would have smeared out the pulse obtained by folding the self - data ; on the contrary we see the on - pulse detected with a signal- to - noise ratio of 50 - 100 ( figure [ fig_profilep ] ) . alternatively , the clock could have maintained an accurate long term average but with large excursions in the values of the individual time stamps , i.e. excursions of the order of 1 - 2 s ( half the pulsar period ) for an output data rate of one visibility per 0.131s or 0.262s . this sounds somewhat contrived ; if this were the case we should have expected a histogram of the time difference between adjacent data to show a large scatter . the histograms plotted in figure [ fig_timehist ] show a tight scatter ( rms 0.007s ) around the expected mean values ( 0.262s and 0.131s ) . , title="fig : " ] , title="fig : " ] additionally , any artifact due to time - smearing should have resulted in the shorter - period and faster - sampled b204516 having a higher off - pulse fraction , whereas the ratio is much higher for b0525 + 21 . this scenario requires temporal correlation between data separated by at least half the pulsar period . the basic sampling interval of the gmrt correlator is 31.25 nanoseconds . a cross - spectrum is output every 16@xmath19s by fourier transforming 512 such samples . the data output every 16@xmath19s should be independent from all other such data in the time series . 8188 ( or 16376 ) such independent samples are averaged to obtain the short - term acquisition ( sta ) date output at 0.131 ( or 0.262 s ) . given the independence at the 16@xmath19s level it is difficult to envisage correlation between one sta and the next , even more so across 6 - 10 stas ( half the pulsar period ) . nevertheless this was a possibility . observers using the standard interferometric mode of the gmrt would not be affected as none of their targets vary on sub - second timescales , and their integration time is rarely less than 2s ( more typically 17s ) . the leakage would only redistribute the flux density along the time axis without causing any change in the measured flux density . observers using the pulsar mode would not have detected such an effect because they are only sensitive to the relative flux in excess of the base level . we tested this possibility by estimating the temporal correlation in the 0.131s data series . assuming that the off - pulse was entirely due to leaking of the signal from the on - pulse bin to the off - pulse bins we estimated the required leakage for a spurious detection of the observed level . there was no way of introducing a narrow , clean pulse into the gmrt receiver system to directly measure such a leakage . so , we investigated the consequence of such a correlation on the noise data . first , we ran simulations to estimate the auto - correlation function for noise in the presence of leakage sufficient to generate the observed off - pulse , i.e. we generated a noise series and smeared each data point into subsequent data points according to a particular time profile e.g. constant leakage into the next 13 bins ; or linearly decreasing leakage into the next 13 bins . the value 13 is related to the period of the two pulsars which covered 14 time bins . we then recorded 6@xmath2510 minute scans of noise data ( 0.131s integration ) but with front - end termination which sealed the telescope at the antenna feeds . this ensured that no temporally continuous external source ( cosmic sources and rfi ) , which would be correlated across the entire observing session , was present in the noise data . the auto - correlation for each baseline - channel ( 435@xmath25120 ) data was separately calculated for each scan . the median autocorrelation profile and the scatter are shown in figure [ autocorrfull ] . the worst data are at the 0.1 per cent level but the median is only about 0.04 per cent . the profile does not fall to zero at large time lags , indicating non - stationarity of the noise signal . we obtained a similar behaviour by introducing a time - varying mean level into the noise in our simulations . this time variation in the mean level may reflect the system gain variations as well , which would have been corrected to a greater or lesser extent by self - calibration during imaging . therefore the profile is a firm upper limit to the contribution of temporal leaking of signals to an off - pulse detection . it should be noted that interferometric imaging is an excellent filter of bad data ; a compact source ( like the off - pulse detection ) would require most of the baselines to show high temporal auto - correlation ; a few bad baselines with high auto correlation would only result in noisy ripples across the image and not a localised source . therefore the median line is a more accurate measure of the leakage for our study than the extreme points in the scatter . we note that non - stationarity and gain fluctuation only occur at the 0.05 per cent level in the gmrt , and would not have a discernible effect for most imaging exercises . ] figure [ autocorrmag ] shows a magnified representation of the relevant part of the noise auto - correlation plot . a1 and a2 ( both psr b0525 + 21 ) and b1 ( psr b204516 ) indicate the required noise auto - correlation if the detected off - pulse were due to the leaking of the signal from the on - pulse bin . the analysis recipe for psr b0525 + 21 ( with similar arguments for psr b204516 ) is as follows : + 1 . the pulsar period includes 14 time bins , each of 0.262 s. the on- and off - pulse fluxes measured over 5 time bins are 80.2 and 3.9 mjy , respectively . higher time resolution observations @xcite show that the main - pulse narrower than one of our bins . + 2 . a leakage of 0.9 per cent of the pulse flux from the first bin into each of the next 13 bins will explain the observed off - pulse . this corresponds to an auto correlation level of 0.0087 . the auto - correlation profile calculated for gmrt noise ( the lowest curve n in figure [ autocorrmag ] ) , is smaller than the required amount by a factor of at least 20 . a leakage reducing with temporal distance is more likely . a linearly reducing leakage ( curve a2 ) requires a similar average auto - correlation in the off - pulse region and a higher correlation at short lags . we conclude that temporal leaking of the signal is incapable of explaining the detected level of off - pulse emission . the curve b2 represents the corresponding model for psr b204516 : off - pulse 4.3 mjy ; on - pulse flux 305.2 mjy ; required constant leakage 0.43 per cent ; required correlation in the off - pulse region 0.003 ; discrepancy is a factor of @xmath22 6 . additionally , the argument against time - smearing ( sec . 4.2 , second para ) is also valid against leaking of the signal along the time series . . the plots a1 , a2 ( psr b0525 + 21 ) and b1 ( psr b204516 ) are the required ( simulated ) noise auto - correlation for a spurious off - pulse detection due to temporal leaking of the pulse into subsequent bins . the thick black curve is the observed noise auto - correlation profile ( from figure [ autocorrfull ] ) . [ autocorrmag ] ] a literature survey yielded many reports of searches for `` continuous emission in the direction of pulsars '' ( table [ litrsurvey ] ) . here we shall discuss only those in which gating techniques were applied to target off - pulse emission , which could have been either from the pulsar itself or from pwne . perry & lyne ( 1985 ) reported off - pulse emission from 4 pulsars ( psrs b1541 + 09 , b160400 , b1929 + 10 and b2016 + 28 ) using a gated two - element interferometer at 408 mhz . subsequent studies showed that two of these pulsars , b1541 + 09 and b1229 + 10 , were aligned rotators , undermining the claim of off - pulse origin @xcite . the off - pulse emission in the other two pulsars , b1604 - 00 and b2016 + 28 , were later identified with unrelated background sources ( storm & van someren greve 1990 ; hankins et al . 1993 ) . bartel et al . ( 1985 ) reported a non - detection from psr b0329 + 54 using gated vlbi interferometry at 2.3 ghz . strom & van someren greve ( 1990 ) reported non - detections in 43 pulsars using the wsrt in gated mode at 327 and 609 mhz . stappers et al . ( 1999 ) reported non - detection in 4 pulsars using atca in gated interferometric mode at 1.3 and 2.2 ghz . gaensler et al . ( 2000 ) searched for unpulsed emission from 27 energetic and/or high velocity pulsars using gating interferometry with the vla and atca at 1.4 ghz and found emission in two cases . however they could not conclusively determine whether these detections corresponded to pwne or the pulsar magnetosphere ; they even suggested that the detections could be spurious and a result of improper online gating . these detections were at the level of 2@xmath27 ( psr b1634 - 45 ) and 13@xmath27 ( psr b1706 - 16 ) . there have been detections of several radio pulsars with associated plerionic bow shock nebula : b1951 + 32 @xcite , b175724 @xcite , b1853 + 01 @xcite and b164343 @xcite . all of these are highly energetic pulsars associated with snr @xcite . gaensler et al . ( 1998 ) reported the detection of off - pulse emission from psr b090649 , which appears to be a pwn generated by a slow moving pulsar in the dense ism . llll non detections/ & detections & tentative & pwne detection + subsequent snr & later refuted & detection & + 1 . schnhardt 1973 , 1974 & 1 . gopal - krishna 1978 & 1 . gaensler et al . 2000 ( @xmath36)&1 . frail & kulkarni 1991 + 2 . weiler et al . glushak et al . 1981 & & 2 . frail et al . cohen et al . perry & lyne 1985 ( @xmath36 ) & & 3 . gaensler et al . bartel et al . 1985 ( @xmath36 ) & & & 4 . giacani et al . strom & van someren & & & + greve 1990 ( @xmath36 ) & & & + 6 . hankins et al . 1993 ( @xmath36 ) & & & + 7 . frail & & & & + scharringhausen 1997 & & & + 8 . stappers et al . 1999 ( @xmath36 ) & & & + + we discuss here the unsuccessful search for off - pulse emission in psr b0525 + 21 by weiler et al . ( 1974 ) and perry & lyne ( 1985 ) . weiler et al . ( 1974 ) were looking for extended emission around the pulsar without employing any kind of gating . they had a synthesized beam size of 47@xmath25 133 with a noise rms of 2.5 mjy / b . it is apparent that their setup was inadequate for the present level of detection . perry & lyne ( 1985 ) employed a two - element interferometer with hardware gating at 408 mhz for their studies but could neither image the fields and nor correct for sensitivity variations . they listed considerable emission in the off - pulse bin for b0525 + 21 ( in fact greater than for b1929 + 10 and b2016 + 28 , which were claimed to be positive detections ) but the noise was also very high precluding a positive detection . the major advantages of our effort were : + 1 . low frequency of observation : the steep spectrum pulsar is likely to dominate over emission from any associated nebula with flatter spectrum ; + 2 . a multi - element interferometer : the gmrt is currently the most sensitive low - frequency instrument . the ability to correct gain variations , filter bad data using imaging residuals and rfi excision tools @xcite allowed us to make sufficiently deep images . + 3 . off - line gating of data : this allowed secure separation of the off- and on - pulse sections . the confinement of the relativistic wind from pulsars generates pwne which are luminous across the electromagnetic spectrum in synchrotron , inverse compton , and optical line emission from the shocked regions . all known pwne seen around radio pulsars have spin - down luminosities @xmath7 erg s@xmath1 . they all appear to be young and , with the exception of psr b090649 @xcite , are associated with snr . two classes of pwne can be formed in the absence of associated snr : one , where the pulsar wind is confined by the density of the ism known as static pwn ; and the other , where the wind is confined by the ram - pressure of motion of the pulsar through the ism called bow - shock nebula . our two target pulsars , b0525 + 21 and b204516 , are not associated with snrs . so , we explored the possibility that the detected off - pulse emission was due to a pwn . the efficiency factor @xmath37 is defined as the ratio of the radio bolometric luminosity ( @xmath38 ) of a pwn to the spin - down energy ( @xmath12 ) of the pulsar @xmath39 . if we assume a typical pwn spectral index of @xmath40 0.3 its radio luminosity between @xmath41 hz and @xmath42 hz is given by @xmath43 where @xmath44 is the distance to the pwn in kpc and @xmath45 is the integrated flux of pwn at 325 mhz in mjy . using the measured flux - density ( table [ results ] ) , source distance and the spin - down energies ( table [ pulsprop ] ) we calculate efficiency factors of 2@xmath46 for b0525 + 21 and 2@xmath47 for b204516 . these @xmath37 values are 1 - 2 orders of magnitude higher than those of previously known pwne ( typical @xmath48 ; see frail & scharringhausen 1997 , gaensler et al . 2000 for a discussion ) . [ [ static - pwn ] ] static pwn + + + + + + + + + + the relativistic particles and poynting flux emanating from the pulsar , at rest relative to the ism , will drive through the ambient medium a shock of radius @xmath49 given by @xcite @xmath50 the velocity of the shock front is given by @xmath51 here @xmath52 , where @xmath53 is the proton mass and @xmath54 is the particle density of the ism . using equation ( [ eqn_rs ] ) the required particle density for a pwn is @xmath55 where @xmath56 is the spindown power in units of @xmath57 erg s@xmath1 , @xmath58 is the age in units of @xmath59 yr , @xmath60 is the radius of pwn in units of 0.01 pc and @xmath61 is the ism density in units of 0.01 @xmath62 . in the present exercise we assume that the pwne in our targets are smaller than one synthesized beam width since the off - pulse emission is unresolved in both pulsars . using equation ( [ eqn_nrs ] ) and values in tables [ pulsprop ] and [ results ] we determined the ism density required to drive a pwn to be @xmath63 for b0525 + 21 and @xmath64 for b204516 . the required particle densities are much higher than typical ism densities of @xmath65 , suggesting that these pulsars are too weak and old to power a static pwn through the ism . since our estimate of the sizes of the nebulae are upper limits the particle density estimates are lower limits . [ [ bow - shock - pwn ] ] bow shock pwn + + + + + + + + + + + + + this requires that the shock velocity ( @xmath66 ) be much smaller than the pulsar transverse velocity ( @xmath67 ) . the typical ism density of @xmath68 in equation ( [ eqn_rsv ] ) yields @xmath66 @xmath69 for b0525 + 21 and @xmath70 for b204516 , which are indeed much smaller than the @xmath67 listed in table [ pulsprop ] . the radius of the shock is given by @xcite @xmath71 from equation ( [ eqn_rbs ] ) the particle density required to sustain a bow shock pwn is @xmath72 where @xmath56 is the spin - down power in units of @xmath73 , @xmath60 is the radius of pwn in units of 0.01 pc and @xmath74 is the velocity of the pulsar through the ism in units of 100 @xmath75 . this requires an ism density of @xmath76 for b0525 + 21 and @xmath77 for b204516 . these values are 3 - 4 orders of magnitude lower than the typical ism density . this implies that the pulsars are too weak to drive a bow shock nebula of the size corresponding to the telescope resolution . however , a bow shock interpretation can be salvaged if the nebulae are @xmath222 orders of magnitude smaller . in summary , identifying the off - pulse emission detected here with pwne results in unrealistic values of the ism particle density . this is not surprising because we selected as our targets middle - aged pulsars , which category has hitherto not been known to be associated with pwne . numerous published studies provide estimates of @xmath3 and @xmath4 for the pulsars studied here . some use an empirical / geometrical ( e / g ) approach to establish @xmath3 and @xmath4 ( lyne & manchester 1988 ; rankin 1993 ; see everett & weisberg 2001 for a discussion on the e / g approach ) . one can also obtain @xmath3 and @xmath4 values by fitting the rvm to the ppa traverse @xcite , but these fits yield highly correlated estimates of the two angles . this issue has been discussed in depth by everett & weisberg ( 2001 ) and mitra & li ( 2004 ) . everett & weisberg ( 2001 ) list six different studies of psr b0525 + 21 with @xmath3 ranging from 116 to 159 values should span 0 - 180(and not 0 - 90 ) for a consistent definition of the pulsar rotation axis . everett & weisberg provide a compilation of published @xmath3 and @xmath4 values corrected for this convention . ] . mitra & li ( 2004 ) estimated @xmath78 for psr b204516 . both rvm fit and the e / g approach were used to obtain the above estimates of @xmath3 . these studies provide good evidence that our targets , b0525 + 21 and b204516 , are not aligned rotators and therefore any off - pulse emission must arise far from the magnetic pole . the pulse profiles of both these pulsars show sharp rising and falling edges . psr b0525 + 21 is an example of a classical double ( d ) profile , whereas psr b204516 is classified as a triple @xcite . the main - pulse width for these pulsars span about 5 per cent of the period , and there are no obvious emission components visible outside the main - pulse . based on the pulse widths and the estimates of @xmath3 and @xmath4 , mitra & li ( 2004 ) concluded that the radio emission arises at a distance of 12 per cent of the light cylinder . hence , the radio on - pulse emission of psr b0525 + 21 and b204516 are classical examples of emission from open dipolar field lines . the remaining 95 per cent of the period , i.e. the off - pulse region , were hitherto thought to be radio quiet zones of the pulsar magnetosphere @xcite . our off - pulse phase corresponds to 80 - 208(from the peak of the main - pulse ) for b0525 + 21 and 101 - 229for b204516 . in a few cases pre-/post - cursor emissions have been detected about 60from the main - pulse , posing questions regarding their origin @xcite . our detections are even further away from the main - pulse . it remains to be seen if these off - pulse detections are an extreme example of ppc or if it represents unpulsed emission from throughout the pulse period . it is to be noted that the opening angle of dipolar field lines scales as the square of the emission height . if radio emission from pulsars originate from higher up the light cylinder ( @xmath79 1 - 2 per cent of the light cylinder ) , the polar emission could span a larger longitudinal range . this implies that radio emission at larger distances from the neutron star surface are potential sites of ppc or off - pulse emission in pulsars . alternatively , if the emission arises close to the neutron star surface , the emitting region will need to encroach upon the closed field lines of the neutron star . thus , the existence of magnetospheric off - pulse emission should prove to be an important input for pulsar electrodynamic models which try to establish the relation between magnetospheric currents and coherent pulsar radio emission @xcite . we report gmrt observations at 325 mhz which detected off - pulse emission from two long - period pulsars which have low spin - down rates and are not associated with supernova remnants . we have adduced evidence that the signals are not artifacts of the observing procedure , nor unrelated background sources which happen to be coincident with the pulsar . we have argued that explaining these detections as pulsar wind nebula requires ism particle densities which differ from typical measured values by several orders of magnitude . robust estimates of the geometrical parameters of these two pulsars argue against them being aligned rotators . this leaves the possibility of emission from the magnetosphere . if the off - pulse emission arises from a much higher height than the on - pulse then it could arise from open field lines at the edge of the light cylinder . on the other hand if the off - pulse emission is from the same height as the on - pulse emission then it must be associated with closed field lines . further studies at multiple frequencies , in polarization and higher resolution are needed to establish the nature of this emission which can impose valuable constraints on the pulsar emission mechanism models . we thank the director of ncra for discretionary observing time which made this work possible . we also thank the staff of the gmrt for their help with these non - standard observations possible . the gmrt is run by the national centre for radio astrophysics of the tata institute of fundamental research . we also thank the referee , scott ransom , for useful comments . arons , j. 1983 , , 302 , 301 athreya , r. 2009 , , 696 , 885 baars , j. w. m. , genzel , r. , pauliny - toth , i. i. k. , witzel , a. 1977 , , 61 , 99 backus , i. , mitra , d. , rankin , j. m. 2010 , , 404 , 30 bartel , n. , ratner , m. i. , shapiro , i. i. , cappallo , r. j. , rogers , a. e. e. , whitney , a. r. 1985 , , 90 , 318 becker , r. h. , white , r. l. , helfand , d. j. 1995 , , 450 , 559 blandford , r. d. , ostriker , j. p. , pacini , f. , rees , m. j. 1973 , , 23 , 145 chatterjee , s. , cordes , j. m. 2002 , , 575 , 407 cohen , n. l. , cotton , w. d. , geldzahler , b. j. , marcaide , j. m. 1983 , , 264 , 273 everett , j. e. , weisberg , j. m. 2001 ; , 553 , 341 frail , d. a. , giacani , e. b. , goss , w. m. , dubner , g. 1996 , , 464 , 165 frail , d. a. , kulkarni , s .r . 1991 , , 352 , 785 frail , d. a. , scharringhausen , b. r. 1997 , , 480 , 364 gaensler , b. m. , stappers , b. w. , frail , d. a. , johnston , s. 1998 , , 499 , 69 gaensler , b. m. , stappers , b. w. , frail , d. a. , moffett , d. a. , johnston , s. , chatterjee , s. 2000 , , 318 , 58 giacani , e. b. , frail , d. a. , goss , w. m. , vieytes , m. 2001 , , 121 , 3133 glushak , a. p. , pynzar , a. v. , udaltsov , v. a. 1981 , , 25 , 182 gopal - krishna 1978 , , 185 , 521 goldreich , p. , julian , w. h. 1969 , , 157 , 869 gould , d. m. , lyne , a. g. 1998 , , 301 , 235 hankins , t. h. , moffett , d. a. , novikov , a. , popov , m. 1993 , , 417 , 735 hankins , t. h. , rickett , b. j. 1975 , methods in comp phy . , 14 , 55 hester , j. j. , kulkarni , s. r. 1988 , , 331 , l121 hobbs , g. , lyne , a. g. , kramer , m. , martin , c. e. , jordan , c. 2004 , , 353 , 1311 lorimer , d. r. , yates , j. a. , lyne , a. g. , gould , d. m. 1995 , , 273 , 411 lyne , a. g. , manchester , r. n. 1988 , , 234 , 477 mitra , d. , deshpande , a. a. 1999 , , 346 , 906 mitra , d. , li , x. h. 2004 , , 421 , 215 mitra , d. , rankin , j. m. 2010 , submitted to perry , t. e. , lyne , a. g. 1985 , , 212 , 489 radhakrishnan , v. , cooke , d. j. 1969 ; aph . lett , 3 , 225 rankin , j. m. 1993 , , 405 , 285 rathnasree , n. , rankin , j. m. 1995 , , 452 , 814 ruderman , m. a. , sutherland , p. g. 1975 , , 196 , 51 schnhardt , r. e. 1973 , nature phy . sci , 243 , 62 schnhardt , r. e. 1974 , , 35 , 13 spitkovsky , a. 2006 , , 648 , l51 stappers , b. w. , gaensler , b. m. , johnston , s. 1999 , , 308 , 609 stinebring , d. r. , condon , j. j. 1990 , , 352 , 207 strom , r. g. , van someren greve , h. w. 1990 , , 171 , 351 swarup , g. , ananthakrishnan , s. , kapahi , v. k. , rao , a. p. , subrahmanya , c. r. , kulkarni , v. k. 1991 , current sci . , 60 , 95 taylor , j. h. , manchester , r. n. , lyne , a. g. 1993 , , 88 , 529 weiler , k. w. , goss , w. m. , schwarz , u. j. 1974 , , 35 , 473

in @xcite barot and marsh provided presentations of finite weyl groups in terms of arbitrary quivers or diagrams of finite type . in brief , the construction works as follows : one needs to consider the underlying unoriented labeled graph of a quiver ( or diagram ) as a coxeter diagram of a ( usually infinite ) coxeter group , and then impose some additional relations on this group that can be read off from the quiver / diagram . it is proved in @xcite that the resulting group depends on the mutation class of a quiver / diagram only , see section [ group ] for the details . the goal of this paper is to use these presentations to construct hyperbolic manifolds , in particular ones of small volume . the most common way to construct hyperbolic manifolds is by considering finite index torsion - free subgroups of cofinite hyperbolic reflection groups ( see e.g. @xcite ) . the approach used in the present paper is completely different : we start with a _ symmetry group _ of a manifold , more precisely , we are looking for a manifold whose symmetry group contains a given finite weyl group . in @xcite everitt and maclachlan describe an algebraic framework for this approach , and in @xcite kolpakov and slavich explicitly construct an arithmetic hyperbolic @xmath0-manifold with given finite isometry group . we use quite different techniques based on _ mutations of quivers and diagrams _ defined by fomin and zelevinsky in the context of cluster algebras . this technique allows us to obtain manifolds having large symmetry groups and , at the same time , relatively small volume , i.e. two properties of quite opposite nature . mutation classes of quivers and diagrams of finite type are indexed by finite weyl groups . our interpretation of the result of @xcite is that , starting from a quiver or diagram @xmath1 of finite type , one can present the corresponding finite weyl group @xmath2 as a quotient of some ( usually infinite ) coxeter group @xmath3 by the normal closure @xmath4 of additional relations . if this coxeter group @xmath3 is hyperbolic , then the weyl group @xmath2 acts properly on the quotient of the hyperbolic space by the group @xmath4 . _ a priori _ , this quotient may have singular points . however , this is not the case . the group @xmath4 is torsion - free . as a corollary , if the coxeter group @xmath3 is hyperbolic and cofinite , we obtain a hyperbolic manifold of finite volume , and the symmetry group of this manifold contains @xmath2 . we list the manifolds that we found in this way in tables [ hyp ] and [ hyp - n ] . in dimension @xmath0 we construct a manifold with euler characteristic @xmath5 , i.e. it has the second minimal volume amongst all @xmath0-dimensional manifolds . the symmetry group of this manifold contains the group @xmath6 as a subgroup . we note that , since this manifold has small volume and large symmetry group at the same time , it does not appear in @xcite . in @xcite we generalized the results of @xcite to quivers and diagrams of affine type and certain other mutation - finite quivers and diagrams . in particular , every affine weyl group also admits presentations of a similar type with slightly more complicated groups @xmath4 . using this result , in section [ aff ] we provide the following theorem . the manifold property holds for quivers and diagrams of affine type . for quivers and diagrams originating from unpunctured surfaces or orbifolds @xcite , as well as for exceptional mutation - finite quivers and diagrams @xcite , the construction in @xcite also provides a group assigned to a mutation class . in contrast to the finite and affine cases , this group is not a coxeter group , however , it also has presentations as a quotient of a coxeter group @xmath7 ( with more types of generators for @xmath8 required ) . using these presentations , for every quiver or diagram @xmath1 of all types discussed above , we construct a cw - complex ( actually being a quotient of the davis complex of a coxeter group @xmath3 ) with a proper action of the group @xmath9 , where the group @xmath10 depends on the mutation class of @xmath1 only . we also define mutations of these complexes , the mutations agree with quiver ( or diagram ) mutations . all the manifolds discussed above are partial cases of these complexes . the paper is organized as follows . section [ background ] contains basic definitions and some essential facts on coxeter groups and quiver mutations . in section [ group ] we recall the construction from @xcite assigning to a quiver of finite type a presentation of the corresponding weyl group . for simplicity , in sections [ group][davis ] we restrict ourselves to simply - laced weyl groups ( and thus , we work with quivers rather than diagrams ) , we then consider the general case in section [ sec diagr ] . in section [ various actions ] we describe the construction highlighted above of various actions of a given finite weyl group . in section [ examples ] we first treat in full details our construction applied to the group @xmath11 ; then we list other geometric manifolds with weyl group actions resulted from the algorithm described in section [ various actions ] . in particular , table [ hyp ] contains the list of hyperbolic manifolds we obtain . in section [ davis ] , after defining the davis complex of a coxeter group , we prove the manifold property . we also define mutations of quotients of davis complexes agreeing with quiver mutations . section [ sec diagr ] is devoted to generalizations of the results to all finite weyl groups . finally , in section [ inf ] we describe how the construction generalizes to affine weyl groups and some other infinite groups related to quivers and diagrams of finite mutation type . we briefly remind some basic properties of coxeter groups . for the details see @xcite . a group @xmath2 is called a ( finitely generated ) _ coxeter group _ if it has a presentation of the form @xmath12 where @xmath13 and @xmath14 for all @xmath15 . here @xmath16 means that there is no relation on @xmath17 . a pair @xmath18 of a coxeter group @xmath2 and its set of generators @xmath19 is called a _ coxeter system_. the cardinality @xmath20 of @xmath21 is a _ rank _ of the coxeter system . an element of @xmath2 is said to be a _ reflection _ if it is conjugated in @xmath2 to an element of @xmath21 . a coxeter system @xmath22 may be depicted by its _ coxeter diagram _ : * the vertices of the coxeter diagram correspond to the generators @xmath23 ; * @xmath24-th vertex is connected to @xmath25-th by an edge labeled @xmath26 with the following exceptions : * * if @xmath27 then there is no edge connecting @xmath24 to @xmath25 ; * * if @xmath28 then the label 3 is usually omitted ; * * if @xmath16 the edge is drawn bold . given a coxeter system @xmath22 , one can write a symmetric @xmath29 matrix @xmath30 as follows : @xmath31 an @xmath29 matrix @xmath32 is _ decomposable _ if there is a non - trivial partition of the index set as @xmath33 , so that @xmath34 whenever @xmath35 , @xmath36 . a matrix is _ indecomposable _ if it is not decomposable . for some coxeter systems @xmath22 the group @xmath2 has a natural discrete action by reflections on a space of constant curvature ( here by an _ action by reflections _ we mean an action where the reflections of @xmath2 are represented by orthogonal reflections with respect to hyperplanes ) . more precisely , there are three cases of interest : * @xmath37 is positive definite : then @xmath2 is a finite group and @xmath2 acts on a sphere @xmath38 ; * @xmath37 is indecomposable and has signature @xmath39 : then @xmath2 acts cocompactly on euclidean space @xmath40 ; * @xmath37 is indecomposable and has signature @xmath41 : then @xmath2 acts on an @xmath42-dimensional hyperbolic space @xmath43 . see @xcite for the details . a fundamental domain of such an action is a connected component of the complement to the set of all hyperplanes fixed pointwise by reflections of the group . this fundamental domain is a polytope ( which may be of infinite volume ) with facets indexed by generators @xmath44 and all dihedral angles equal to @xmath45 ( if @xmath16 the corresponding facets do not intersect ) . such polytopes are called _ coxeter polytopes_. [ root ] if a coxeter group @xmath2 is a finite or affine weyl group , then the outer normal vectors to the facets of the fundamental polytope can be identified with the simple roots of the corresponding root system . in the three cases listed above the group @xmath2 is called _ geometric _ ( note that our class of geometric coxeter groups is wider than in ( * ? ? ? * chapter 6 ) ) . in fact , every coxeter group acts properly cocompactly by reflections on some specially constructed space @xcite , we will discuss this action in section [ davis ] . let @xmath46 be a quiver ( i.e. a finite oriented multi - graph ) containing no loops and no 2-cycles . if there are @xmath47 arrows pointing from @xmath24-th vertex to the @xmath25-th , then we draw one arrow with a weight @xmath48 . for every vertex @xmath49 of the quiver @xmath46 one can define an involutive operation @xmath50 called _ mutation of @xmath46 in direction @xmath49_. this operation produces a new quiver denoted by @xmath51 which can be obtained from @xmath46 in the following way ( see @xcite ) : * orientations of all arrows incident to the vertex @xmath49 are reversed ; * for every pair of vertices @xmath52 such that @xmath46 contains arrows directed from @xmath24 to @xmath49 and from @xmath49 to @xmath25 the weight of the arrow joining @xmath24 and @xmath25 changes as described in figure [ quivermut ] . + @xmath53 given a quiver @xmath46 , its _ mutation class _ is a set of all quivers obtained from @xmath46 by all sequences of iterated mutations . all quivers from one mutation class are called _ mutation - equivalent_. quivers without loops and 2-cycles are in one - to - one correspondence with integer skew - symmetric matrices @xmath54 , where @xmath55 is the number of arrows from @xmath24-th vertex to @xmath25-th one . in terms of the matrix @xmath56 the mutation @xmath50 can be written as @xmath57 , where @xmath58 this transformation is called a _ matrix mutation_. the procedure of matrix mutation is well - defined for more general case of _ skew - symmetrizable _ integer matrices . matrix mutations for non - skew - symmetric matrices correspond to _ diagram mutations _ ( see section [ sec diagr ] for the definitions and generalizations of the results to the case of diagram mutations ) . a quiver is of _ finite type _ if it is mutation - equivalent to an orientation of a simply - laced dynkin diagram . all orientations of a given dynkin diagram are mutation - equivalent , so any quiver of finite type is of one of the following mutation types : @xmath59 , @xmath60 , @xmath61 , @xmath62 , @xmath63 . a quiver @xmath46 is of _ finite mutation type _ ( or _ mutation - finite _ ) if there are finitely many quivers mutation - equivalent to @xmath46 . it is shown in @xcite that each quiver of finite mutation type is either of order @xmath5 , or a quiver arising from a triangulated surface ( see @xcite for details ) , or mutation - equivalent to one of the 11 exceptional quivers . in this paper we deal with quivers of finite mutation type only . in @xcite barot and marsh show that for each quiver of finite type there is a way to explicitly construct the corresponding weyl group . let @xmath46 be an orientation of a simply - laced dynkin diagram with @xmath20 nodes , let @xmath2 be the corresponding finite coxeter group , and @xmath64 be any quiver mutation - equivalent to @xmath46 . denote by @xmath65 the group generated by @xmath20 generators @xmath23 with the following relations : * @xmath66 for all @xmath67 ; * @xmath68 for all @xmath69 , where @xmath70 * * ( cycle relations ) * for every chordless oriented cycle @xmath71 given by @xmath72 we take the relation @xmath73 note that for a cycle of length @xmath74 we have a choice of @xmath74 relations of type ( r3 ) , we take any one of them . it is shown in ( * ? ? ? * theorem a ) that the group @xmath65 does not depend on the choice of a quiver in the mutation class of @xmath46 . in particular , it is isomorphic to the initial coxeter group @xmath2 . the results of @xcite are proved in more general settings , we come to the general case in section [ sec diagr ] . generators of @xmath65 satisfying relations ( r1)(r3 ) can be expressed in terms of standard generators of @xmath2 . it is shown in @xcite that these generators of @xmath65 can be found inductively in the following way : * for @xmath75 the generators @xmath76 are standard coxeter generators ( indeed , dynkin diagrams contain no cycles , so , we will get no cycle relations and obtain standard presentation of the coxeter group ) . * let @xmath76 be the generators of @xmath65 , and let @xmath77 . then the generators @xmath78 of @xmath79 satisfying relations ( r1)(r3 ) are @xmath80 as before , let @xmath46 be an orientation of a dynkin diagram , and let @xmath2 be the corresponding weyl group . choose any quiver @xmath64 mutation - equivalent to @xmath46 , and denote by @xmath81 the group defined by relations of type ( r1 ) and ( r2 ) only ( note that @xmath81 is a coxeter group ) . let @xmath82 be a collection of all cycle relations for @xmath64 ( we take one relation for each chordless oriented cycle ) . denote by @xmath83 the normal closure of the union of the elements @xmath82 in @xmath81 . by the definition of @xmath65 we have @xmath84 . theorem a from @xcite says that @xmath65 is isomorphic to @xmath85 which , in particular , means that @xmath83 has finite index in @xmath81 . the group @xmath81 depends on the quiver @xmath64 chosen in the mutation class of @xmath46 . moreover , depending on the choice of @xmath64 , the group @xmath81 may be finite or infinite . as a consequence , it may act naturally by reflections on different spaces ( sphere , euclidean space , hyperbolic space or , in general , the davis complex of the group @xmath81 , see section [ davis ] ) . since @xmath86 , this results in actions of the initial coxeter group @xmath2 on different spaces . in section [ davis ] we show that the group @xmath83 is torsion - free ( we call this _ manifold property _ ) . in particular , this implies that if the coxeter group @xmath81 acts properly on a space @xmath87 or @xmath88 by isometries with a fundamental domain of finite volume , then the quotient @xmath89 is a finite volume manifold with a symmetry group containing @xmath2 . in the next section we show examples of the construction of hyperbolic manifolds in this way , as well as of actions of weyl groups of type @xmath90 on flat tori . we now present the first ( and the easiest ) example showing how different actions of the same coxeter group may by produced via mutations of quivers . we start with the quiver @xmath46 of type @xmath11 ( see fig . [ stereo](a),(b ) ) . the corresponding dynkin diagram of type @xmath11 determines a finite coxeter group @xmath2 acting on the @xmath5-dimensional sphere by reflections : @xmath91 the fundamental domain of this action is a spherical triangle with angles @xmath92 , see fig . [ stereo](c ) for the stereographic projection image of the action and the fundamental domain . applying the mutation @xmath93 to the quiver @xmath46 , we obtain a quiver @xmath64 ( see fig . [ plane](a ) ) . the group @xmath94 is the group generated by reflections in the sides of a regular triangle in the euclidean plane ( the corresponding dynkin diagram is of type @xmath95 ) , see fig . [ plane](b ) . note that the plane is tessellated by infinitely many copies of the fundamental domain , each copy @xmath96 labeled by an element @xmath97 . to see the action of the initial group @xmath98 we need to take a quotient by the cycle relation @xmath99 which means that in the tessellated plane we identify @xmath100 with the fundamental triangle labeled by the @xmath101 . now , let us see what the transformation @xmath102 means geometrically . the element @xmath103 acts on the plane as the reflection in the side @xmath104 of @xmath100 , @xmath105 is the reflection with respect to the line @xmath106 parallel to @xmath104 and passing through the vertex of @xmath100 not contained in @xmath104 . denote by @xmath74 the distance between @xmath104 and @xmath106 . then the transformation @xmath107 is a translation of the plane by the distance @xmath108 in the direction orthogonal to @xmath104 . hence , the element @xmath102 is a translation by @xmath109 in the direction orthogonal to @xmath104 . clearly , @xmath110 in @xmath2 , and @xmath111 in @xmath2 . this implies that we also need to take a quotient of the plane by the actions of the elements @xmath112 and @xmath113 , i.e. by the translations by @xmath109 in directions orthogonal to the lines @xmath114 and @xmath115 . the quotient of the plane by the group @xmath83 generated by three translations @xmath116 and @xmath117 is a flat torus , see fig . [ plane](c ) . thus , we obtain an action of the quotient group @xmath98 on the torus : the fundamental domain of this action is a regular triangle , @xmath118 copies of this triangle tile the torus ( as @xmath118 copies tile the hexagon ) . [ non - reflection ] the action described above is not an action by ( topological ) reflections : the fixed sets of the generators @xmath119 are connected , so they do not divide the torus into two connected components ( cf . * chapter 10 ) ) . now , we will show how the action of @xmath11 on the torus is related to the initial action on the sphere . recall from section [ pres ] that the presentation for @xmath65 is obtained from the presentation for @xmath120 by the following change of generators @xmath121 in the above sections we considered the action of the @xmath122-generators on the sphere and of the @xmath123-generators on the plane and the torus . however , the @xmath123-generators may be also considered directly on the sphere . namely , let @xmath124 be the reflections with respect to the corresponding sides @xmath125 of the fundamental spherical triangle @xmath126 , see fig . [ cover](a ) . then the elements @xmath127 and @xmath128 are reflections with respect to @xmath114 and @xmath115 respectively . the element @xmath103 is also a reflection as it is conjugate to the reflection @xmath129 ; more precisely , @xmath103 is a reflection with respect to the ( spherical ) line @xmath130 ( see fig . [ cover](b ) ) . consider the triangle @xmath131 bounded by @xmath132 , and @xmath133 ( there are 8 possibilities to choose such a triangle , see remark [ rem choice ] for the justification of our choice ) . in contrast to the triangle @xmath126 bounded by @xmath125 , the triangle @xmath131 is not a fundamental triangle for the action : @xmath131 has angles @xmath134 , and the group generated by @xmath127 and @xmath128 has order 6 , but the whole neighborhood of the vertex @xmath135 is tiled by @xmath136 copies of @xmath131 only . to resolve this inconsistency we consider a @xmath5-sheet covering of the sphere branched in @xmath137 and in all @xmath2-images of @xmath137 ( the four bold points in fig . [ cover](b ) ) . clearly , the covering space is a torus ( moreover , one can easily see that the covering is given by weierstrass @xmath138-function ) . it is also not difficult to see that the action of @xmath11 on this covering torus is exactly the same as its action on the torus in section [ torus ] : the covering torus consists of @xmath118 triangles , and every vertex is incident to exactly six triangles . [ rem choice ] consider the outer normals @xmath139 to the sides of @xmath126 ( cf . remark [ root ] ) . to get the outer normals @xmath140 to the sides of @xmath131 we use the following rule : @xmath141 here @xmath142 is the vector obtained from @xmath143 by the reflection @xmath144 . now we will obtain the action of @xmath11 on the sphere from its action on the torus . let @xmath131 be the fundamental triangle for the action of @xmath11 on the flat torus ( see fig . [ back](a ) ) . let @xmath145 be the canonical projection from euclidean plane to the torus ( corresponding to the action of @xmath83 on the plane ) , denote by @xmath146 any preimage of @xmath131 under @xmath145 . the generators @xmath147 can be thought as reflections with respect to the sides of @xmath146 . the generators @xmath23 can be obtained from the generators @xmath119 as follows : @xmath148 these are reflections with respect to the sides of the dashed domain @xmath149 ( see fig . [ back](b ) ) , where @xmath149 is an unbounded strip on the euclidean plane . we now glue a manifold from @xmath118 copies of the strip ( according to the group action , as we glued @xmath118 triangles above ) , denote by @xmath150 the space obtained via this gluing . note that due to the cycle relations four strips attached one to another along the `` long '' ( unbounded ) sides will result in an ( unbounded ) cylinder . the manifold @xmath150 consists of exactly @xmath151 cylinders . compactifying @xmath150 by adding a limit point at the end of each of the @xmath152 cylinders , we get exactly the tiling shown in fig . [ stereo](c ) . the next series of examples is provided by the finite coxeter group @xmath153 , @xmath154 ( see fig . [ dn](a ) for the dynkin diagram ) , which naturally acts on @xmath42-dimensional sphere ( for @xmath155 the dynkin diagram and weyl group @xmath156 coincides with @xmath11 , so the example from the previous section can be considered as a partial case ) . it is easy to check that the quiver @xmath46 of type @xmath60 ( see fig . [ dn](b ) ) is mutation - equivalent to the oriented cycle @xmath64 shown in fig . [ dn](c ) ( in contrast to @xmath11-example , one needs more than one mutation for this ) . note that the group @xmath81 ( i.e. coxeter group without cycle relations ) is the affine coxeter group @xmath157 , thus , a group acting on the euclidean space @xmath40 . a fundamental domain of this action is a coxeter simplex @xmath158 with the coxeter diagram @xmath157 . the generators for this action are the reflections @xmath159 with respect to the facets @xmath160 of this fundamental simplex . our next aim is to take a quotient by the cycle relation @xmath161 the element @xmath162 is a conjugate of @xmath163 , so it is a reflection with respect to the hyperplane @xmath164 parallel to @xmath165 and passing though the vertex of @xmath158 opposite to @xmath165 . if @xmath74 is the distance between the hyperplanes @xmath166 and @xmath165 , then the element @xmath167 is a translation by @xmath108 in the direction orthogonal to @xmath165 , and its square @xmath168 is a translation by @xmath109 . so , we need to take the quotient by the action of this translation . the element @xmath169 has @xmath170 conjugates of type @xmath171 , ( @xmath172 ) . they all are translations by @xmath109 in the direction orthogonal to @xmath173 . the group generated by @xmath174 is a free abelian group of rank @xmath170 . hence , taking a quotient of @xmath175 by the translations @xmath174 we get an @xmath42-dimensional torus @xmath176 . the action of @xmath81 on @xmath40 turns into a faithful action of @xmath177 on the torus @xmath176 . the coxeter diagram @xmath178 is the unique affine dynkin diagram containing a cycle . this implies that our construction gives no further examples of actions on non - trivial quotients of euclidean space . we now consider the case when the group @xmath81 acts on a hyperbolic space , and hence , the group @xmath179 acts on its quotient @xmath180 ( recall that @xmath83 is the normal closure of all cycle relations ) . as it is shown below ( theorem [ manifold ] , corollary [ man ] ) , in this case @xmath150 is always a manifold . in this section we list all known actions on hyperbolic manifolds obtained in this way ( table [ hyp ] ) . in table [ hyp ] we omit orientations of arrows of quivers @xmath46 and @xmath64 . here arrows of @xmath46 may be oriented in arbitrary way , and the only requirement on the orientations of arrows of @xmath64 is that all chordless cycles are oriented . the underlying graphs of @xmath46 and @xmath64 are the coxeter diagrams of fundamental polytopes of the action of @xmath2 and @xmath81 on the sphere and the hyperbolic space respectively . all the volumes except the last two are computed in @xcite , the euler characteristic ( and thus the volume ) of the fundamental polytope of the action of @xmath181 was computed by the program coxiter written by rafael guglielmetti . the volume of the fundamental polytope of the action of @xmath182 is not known to the authors . the euler characteristic is computed for all even - dimensional manifolds . the precise value of the volume of @xmath150 for four - dimensional @xmath150 is equal to @xmath183 , and for six - dimensional @xmath150 equals @xmath184 . to 0.96 the manifold @xmath150 obtained in the case @xmath185 is the manifold @xmath186 studied in @xcite , it has minimal known volume amongst @xmath187-cusped hyperbolic @xmath136-manifolds . it can be obtained via a ramified @xmath5-fold covering over the @xmath136-sphere @xmath188 branching along @xmath189-skeleton of a @xmath0-simplex . this manifold can be also obtained as the complement of the five - chain link in the @xmath136-sphere . the symmetry group of @xmath150 always contains @xmath2 . further , if the underlying graph of the quiver @xmath64 has symmetries , then the fundamental polytope of the action of @xmath65 on @xmath150 also has these symmetries , and thus the symmetry group of @xmath150 also contains a semidirect product of @xmath2 and the group of symmetries of the graph . in particular , one can note from table [ hyp ] that the underlying graph of @xmath64 always has an additional symmetry of order @xmath5 , therefore we actually get the group @xmath190 acting on @xmath150 . moreover , in the cases @xmath191 or @xmath181 the symmetry group of the fundamental polytope is @xmath192 , which results in the action of the group @xmath193 on the manifold @xmath150 . [ var ] * in all cases the fundamental domain of the action @xmath65 on @xmath150 is a non - compact hyperbolic polytope , i.e. the manifold @xmath150 has cusps . * the combinatorial type of the fundamental domain in all but the last two cases is a simplex . in the last two cases it is a pyramid over a product of two simplices . * we do not know how to compute the volume of the @xmath187-dimensional hyperbolic pyramid which is used to construct the manifold whose symmetry group contains @xmath194 ( the second last line of table [ hyp ] ) . at the same time , one can note that this pyramid consists of four copies of a pyramid with coxeter diagram shown on figure [ in ] ( the pyramid in the table [ hyp ] can be obtained from one on fig . [ in ] by two consequent reflections in the facets corresponding to the leaves of the diagram ) . this smaller pyramid belongs to _ napier cycles _ , a class of polytopes classified by i m hof @xcite . in particular , it is a simply truncated orthoscheme , and thus there is a hope to compute its volume using the technique developed by kellerhals @xcite . the list in table [ hyp ] is complete in the following sense . any two nodes of any quiver of finite type are connected by at most one arrow , therefore all dihedral angles of the corresponding fundamental polytope of the group action are either @xmath195 or @xmath196 . all hyperbolic coxeter polytopes satisfying such property are classified by prokhorov in @xcite . one can see that the only coxeter diagrams whose orientations are quivers of finite type are those listed in table [ hyp ] . on the other hand , our approach allows to construct more hyperbolic manifolds . in @xcite , parsons introduced a notion of a _ companion basis _ of a finite root system . in our terms , all companion bases ( where every vector is considered up to plus / minus sign ) can be described as follows . take a quiver @xmath64 of finite type , construct the corresponding group @xmath81 , and construct the isomorphism between @xmath81 and the weyl group @xmath179 by iterated changes of generators ( see section [ pres ] ) . denote by @xmath197 the images of generating reflections of @xmath81 in @xmath65 . now , take the root system of @xmath2 , and an @xmath20-tuple of roots corresponding to reflections @xmath197 . this @xmath20-tuple forms a companion basis . thus , all the manifolds constructed above originate from companion bases . below we present an example of a hyperbolic manifold obtained from a generating set which does not correspond to a companion basis . as is mentioned in @xcite , not every generating set of reflections of @xmath2 corresponds to a companion basis . the easiest example is the quadruple of generators @xmath198 of @xmath185 given by roots @xmath199 respectively , where the root system @xmath200 consists of vectors @xmath201 in @xmath202 with standard basis @xmath203 . the reason is the following : every two of these four roots form an angle @xmath204 , so if there were a corresponding quiver , then its underlying graph should have been the complete graph on four vertices . however , it is easy to see that any such quiver is mutation - infinite , so it can not be mutation - equivalent to any orientation of dynkin diagram @xmath200 . now we may note that the coxeter group @xmath205 with coxeter diagram shawn on fig . [ a4complete ] is generated by reflections in the facets of an ideal hyperbolic @xmath136-dimensional simplex . further , since @xmath206 generate @xmath185 and satisfy the same relations as @xmath207 above , we can write @xmath2 as a quotient of @xmath7 by a normal subgroup @xmath8 which is the normal closure of additional relations . it is easy to see that we can take @xmath8 as the normal closure of elements @xmath208 and @xmath209 . now , one can check that @xmath8 is torsion - free , so the quotient of @xmath210 by the action of @xmath8 will be a finite volume hyperbolic manifold ( it has @xmath211 cusps and symmetry group containing @xmath212 , where @xmath200 is the symmetry group of the coxeter diagram ) . in this section we first recall the construction of davis complex @xmath213 for each coxeter system @xmath22 . then , given a dynkin quiver @xmath46 and a quiver @xmath64 from the same mutation class , we consider the davis complex @xmath214 for the group @xmath81 together with the action of the coxeter group @xmath85 on the corresponding quotient space @xmath215 and prove the manifold property for @xmath150 ( which states that @xmath83 is torsion - free ) . in particular , we prove that all quotients listed in table [ hyp ] are hyperbolic manifolds . for any coxeter system @xmath216 there exists a contractible piecewise euclidean cell complex @xmath217 ( called _ davis complex _ ) on which @xmath2 acts discretely , properly and cocompactly . the construction ( based on results of vinberg @xcite ) was introduced by davis @xcite . in @xcite moussong proved that this complex yields a natural complete piecewise euclidean metric which is @xmath218 . we first give a brief description of this complex following @xcite , and then follow @xcite to define convex polytopes in @xmath217 . for a finite group @xmath2 the complex @xmath219 is just one cell , which is obtained as a convex hull @xmath220 of the @xmath2-orbit of a suitable point @xmath137 in the standard linear representation of @xmath2 as a group generated by reflections . the point @xmath137 is chosen in such a way that its stabilizer in @xmath2 is trivial and all the edges of @xmath220 are of length 1 . the faces of @xmath220 are naturally identified with davis complexes of the subgroups of @xmath2 conjugate to standard parabolic subgroups . if @xmath2 is infinite , the complex @xmath219 is built up of the davis complexes of maximal finite subgroups of @xmath2 glued together along their faces corresponding to common finite subgroups . the 1-skeleton of @xmath219 considered as a combinatorial graph is isomorphic to the cayley graph of @xmath2 with respect to the generating set @xmath21 . the action of @xmath2 on @xmath219 is generated by reflections . the _ walls _ in @xmath219 are the fixed points sets of reflections in @xmath2 . the intersection of a wall @xmath221 with cells of @xmath219 supplies @xmath221 with a structure of a piecewise euclidean cell complex with finitely many isometry types of cells . walls are totally geodesic : any geodesic joining two points of @xmath221 lies entirely in @xmath221 . since @xmath222 is contractible and @xmath218 , any two points of @xmath222 can be joined by a unique geodesic . any wall divides @xmath219 into two connected components . all the walls decompose @xmath219 into connected components which are compact sets called _ any chamber is a fundamental domain of the @xmath2-action on @xmath219 . the set of all chambers with appropriate adjacency relation is isomorphic to the cayley graph of @xmath2 with respect to @xmath21 . in what follows , if @xmath2 and @xmath21 are fixed , we write @xmath223 or @xmath222 instead of @xmath219 . if the group @xmath2 acts cocompactly by reflections on spherical , euclidean or hyperbolic space @xmath224 , then the davis complex @xmath222 is quasi - isometric to @xmath224 . for any wall @xmath221 of @xmath222 we denote by @xmath225 and @xmath226 the closures of the connected components of @xmath227 , we call these components _ halfspaces_. a _ convex polytope _ @xmath228 is an intersection of finitely many halfspaces @xmath229 , such that @xmath230 is not contained in any wall . clearly , any convex polytope @xmath228 is a union of closed chambers . in what follows by writing @xmath229 we assume that the collection of walls @xmath231 is _ minimal _ : for any @xmath232 we have @xmath233 . a _ facet _ of @xmath230 is an intersection @xmath234 for some @xmath235 . for any @xmath236 a set @xmath237 is called a _ face _ of @xmath230 if it is not empty . we can easily define a dihedral angle formed by two walls : if @xmath238 there exists a maximal cell @xmath239 of @xmath222 intersecting @xmath240 . we define the angle @xmath241 to be equal to the corresponding euclidean angle formed by @xmath242 and @xmath243 . clearly , any dihedral angle formed by two intersecting walls in @xmath222 is equal to @xmath244 for some positive integers @xmath49 and @xmath245 . a convex polytope @xmath230 is called _ acute - angled _ if each of the dihedral angles of @xmath230 does not exceed @xmath246 . a convex polytope @xmath230 is called a _ coxeter polytope _ if all its dihedral angles are integer submultiples of @xmath145 . in particular , a fundamental domain of any reflection subgroup of @xmath2 is a coxeter polytope in @xmath222 . conversely , any coxeter polytope in @xmath222 is a fundamental chamber for the subgroup of @xmath2 generated by reflections in its walls ( see e.g. ( * ? ? ? * theorem 4.9.2 ) ) . let us briefly recall the procedure described in section [ various actions ] : * start with a dynkin quiver @xmath46 with corresponding weyl group @xmath2 . let @xmath64 be any quiver mutation - equivalent to @xmath46 . * consider the coxeter group @xmath81 defined by the relations ( r1 ) and ( r2 ) ( its coxeter diagram coincides with the underlying graph of the quiver @xmath64 ) . * let @xmath214 be the davis complex of the group @xmath81 . by construction , @xmath81 acts on @xmath214 by reflections . * let @xmath247 be the collection of all defining cycle relations for @xmath64 , define @xmath83 to be the normal closure of all these relations in @xmath81 . * define @xmath248 . as is shown in @xcite , @xmath65 is isomorphic to @xmath2 . * since @xmath249 , @xmath83 acts on @xmath214 . consider @xmath250 . then @xmath179 acts properly on @xmath251 . so , for each quiver in the mutation class we have assigned an action of @xmath2 on the space @xmath251 ( the latter heavily depends on the quiver @xmath64 ) . [ manifold ] the group @xmath83 is torsion - free . [ man ] if @xmath214 is a manifold then @xmath251 is also a manifold . let @xmath252 be a fundamental domain for the action of @xmath65 on @xmath214 . @xmath253 of a face @xmath254 of @xmath252 is the union of all fundamental domains @xmath255 of @xmath222 containing @xmath254 : @xmath256 we will prove the following statement which is equivalent to the manifold property ( cf . * corollary 2 ) ) . [ manifold - eq ] let @xmath252 be a fundamental domain for the action of @xmath65 on @xmath214 . then for any face @xmath254 of @xmath252 the interior @xmath257 of the star of @xmath254 in @xmath251 is isometric to the interior of the star of @xmath254 in @xmath214 . equivalently , if a face @xmath254 of @xmath252 is contained in @xmath47 fundamental domains of the action of @xmath81 on @xmath214 , then @xmath254 is also contained in @xmath47 fundamental domains of the action of @xmath65 on @xmath251 . to prove the manifold property we will use the following observation . [ eq - dihedral ] let @xmath258 and @xmath259 be two generating reflections of @xmath81 , and let @xmath119 and @xmath260 be their images in @xmath179 under the canonical projection @xmath261 . then the orders of the elements @xmath262 and @xmath263 coincide . the dihedral group generated by @xmath119 and @xmath260 is a quotient of the dihedral group generated by @xmath258 and @xmath259 . the order of the latter group equals @xmath152 or @xmath0 , therefore , if we assume that the two groups are distinct , the order of the former equals two , i.e. @xmath264 , which is impossible since a coxeter group @xmath2 of rank @xmath20 can not be generated by less than @xmath20 reflections ( see e.g. ( * ? ? ? * lemma 2.1 ) ) . let @xmath254 be a face of @xmath252 . denote by @xmath265 the index set such that @xmath254 is the intersection of the walls fixed by @xmath266 . to prove theorem [ manifold - eq ] ( and thus the manifold property ) it is sufficient to prove that the group @xmath267 generated by @xmath266 has the same order as the group @xmath268 generated by @xmath269 . denote @xmath270 . the group @xmath271 can be considered as a group generated by reflections in the facets of a spherical coxeter @xmath272-simplex @xmath273 with dihedral angles @xmath45 , where @xmath26 is the order of @xmath262 . furthermore , since every @xmath274 is a reflection on the @xmath275 , the group @xmath276 can also be considered as a group generated by reflections in the facets of a spherical @xmath272-simplex @xmath230 . the simplex @xmath230 may not be coxeter , however , by lemma [ eq - dihedral ] , its dihedral angles are equal to @xmath277 for some @xmath278 being coprime with @xmath26 . since @xmath27 or @xmath136 , this implies that either @xmath279 , or @xmath280 . according to @xcite , this means that reflections in the facets of @xmath273 and @xmath230 generate isomorphic groups . in this section we describe how to obtain @xmath281 from @xmath251 , where @xmath282 . the construction generalizes the examples shown in sections [ torus ] and [ back - s ] . * let @xmath283 be the generators of @xmath2 corresponding to @xmath64 ( i.e. , if @xmath76 with relations ( r1 ) and ( r2 ) are generating reflections of @xmath81 composing a coxeter system , and the @xmath76 with relations ( r1)(r3 ) are their images under canonical homomorphism @xmath284 ) , let @xmath159 be the generators corresponding to @xmath285 . as it is shown in @xcite , @xmath119 can be obtained from @xmath23 as follows @xmath286 + denote by @xmath287 , @xmath288 , the canonical projection @xmath289 . according to the construction , a fundamental chamber @xmath290 of the action of @xmath2 on @xmath291 is an image of a fundamental chamber @xmath292 under the projection . the elements @xmath23 act on @xmath214 by reflections in the facets of @xmath293 , let walls @xmath294 be the mirrors of the reflections @xmath283 . similarly , the elements @xmath119 are reflections in the facets of @xmath295 , let walls @xmath296 be the mirrors of the reflections @xmath159 . * consider the action of the element @xmath119 on @xmath214 . it is a reflection with respect to a wall @xmath297 satisfying @xmath298 * the fundamental chamber @xmath293 is an intersection of @xmath20 halfspaces bounded by walls @xmath299 . denote these halfspaces by @xmath300 , i.e. @xmath301 . note that the dihedral angles of a polytope @xmath293 equal @xmath45 , where @xmath26 is the order of @xmath302 in @xmath81 . in particular , as all @xmath27 or @xmath136 , every two facets of @xmath293 have a non - empty intersection . * consider the domain @xmath303 defined as follows : + @xmath304 is the intersection of @xmath20 halfspaces @xmath305 , where the walls @xmath306 are defined in step @xmath5 and the half - spaces @xmath307 , @xmath308 , are chosen by @xmath309 ( cf . remark [ rem choice ] ) . + note that @xmath304 is well - defined , i.e. @xmath304 indeed has exactly @xmath20 facets . this follows from the fact that the facet @xmath310 of @xmath293 ( which is also a facet of @xmath304 ) has exactly @xmath170 facets @xmath311 , and by construction of @xmath304 all these faces of @xmath293 are also faces of @xmath304 . * the domain @xmath304 is a convex ( possibly , non - compact ) polytope in @xmath214 with @xmath20 facets . the reflections with respect to the facets of @xmath304 are precisely @xmath119 , they generate the group @xmath81 , however , in general , @xmath304 is not a fundamental domain of the action of @xmath81 on @xmath214 . * now , we will use @xmath304 to construct a new cell complex @xmath312 ( such that the group @xmath2 acts on @xmath312 with fundamental chamber @xmath304 ) . + consider @xmath313 copies of the polytope @xmath304 , labeled by the elements of @xmath2 ( denote by @xmath314 the polytope corresponding to @xmath315 ) . + glue the polytope @xmath314 to the polytope @xmath316 along the facets @xmath317 of @xmath314 and @xmath316 so that the reflection in the common facet swaps the polytopes ( and takes the facet @xmath318 of @xmath314 to the facet of @xmath316 having the same number ) . denote by @xmath312 the space obtained by the gluing . note that by `` reflection '' here we mean a map from one polytope to another , this may not extend to a reflection on the whole @xmath312 ( cf . remark [ non - reflection ] ) . * denote by @xmath26 the order of the subgroup @xmath319 ( note that the coxeter group @xmath320 is the group @xmath321 ) . consider the following completion @xmath322 of @xmath312 . * * first , we complete the fundamental domain @xmath304 ( obtaining a completed fundamental chamber @xmath323 ) in the following way . for every pair @xmath324 we add an @xmath325-dimensional face @xmath326 if it is not a face of @xmath304 yet ; we define the dihedral angle composed by facets @xmath327 and @xmath328 of @xmath323 as @xmath45 . then , for each finite subgroup @xmath329 ( where @xmath330 is an index set ) we add an @xmath331-dimensional face @xmath332 if it is not a face of @xmath304 yet . the procedure is well - defined since any subgroup of a finite group is finite . the resulting polytope @xmath323 can be identified with a fundamental chamber of @xmath333 , i.e. with @xmath295 . * * now , we do this for each @xmath314 , @xmath315 ( a new face @xmath334 of @xmath314 is identified with the similar face of @xmath335 if @xmath336 ) . the procedure results in a new complex @xmath322 . * as @xmath323 can be identified with @xmath295 , and @xmath322 is obtained from copies of @xmath323 by gluings along the action of @xmath2 ( see step 6 ) , the complex @xmath322 can be identified with @xmath337 . in all the considerations above we started with a dynkin quiver @xmath46 . in this section we consider mutations of _ diagrams _ instead of quivers , which allows us to deal with non - simply - laced dynkin diagrams . recall that quivers are in one - to - one correspondence with skew - symmetric integer matrices . an integer @xmath29 matrix @xmath56 is _ skew - symmetrizable _ if there exists a diagonal @xmath29 matrix @xmath338 with positive integer entries such that the product @xmath339 is skew - symmetric , i.e. , @xmath340 . given a skew - symmetrizable matrix @xmath56 one obtains a _ diagram _ @xmath341 in the following way : @xmath1 is an oriented labeled graph with vertex set @xmath342 , where vertex @xmath24 is connected to vertex @xmath25 by an arrow labeled by @xmath343 if @xmath344 . one diagram may correspond to several distinct matrices . any diagram @xmath1 constructed by a skew - symmetrizable matrix satisfies the following property : for any chordless cycle in @xmath1 the product of labels along this cycle is a perfect square ( cf . * exercise 2.1 ) ) . since skew - symmetric matrices are also skew - symmetrizable , we may understand quivers as a partial case of diagrams : to make a diagram out of a quiver one needs to square all the labels . a _ mutation _ @xmath50 of a diagram @xmath1 is defined similarly to the mutation of quivers @xcite : * orientations of all arrows incident to a vertex @xmath49 are reversed ; * for every pair of vertices @xmath52 such that @xmath1 contains arrows directed from @xmath24 to @xmath49 and from @xmath49 to @xmath25 the weight of the arrow joining @xmath24 and @xmath25 changes as described in figure [ diagrammut ] . + @xmath345 as for quivers , a diagram is _ of finite type _ if it is mutation - equivalent to an orientation of a dynkin diagram . the construction is generalized straightforwardly , but the definition of the group defined by a diagram is a bit longer than the one in the simply - laced case . let @xmath1 be an orientation of a dynkin diagram with @xmath20 nodes , let @xmath2 be the corresponding finite weyl group , and let @xmath346 be any diagram mutation - equivalent to @xmath1 . denote by @xmath347 the group generated by @xmath20 generators @xmath23 with the following relations : * @xmath66 for all @xmath67 ; * @xmath68 for all @xmath69 , not joined by an edge labeled by @xmath0 , where @xmath348 * ( cycle relation ) for every chordless oriented cycle @xmath71 given by @xmath349 and for every @xmath350 we compute @xmath351 , where the indices are considered modulo @xmath74 ; now for every @xmath352 such that @xmath353 , we take the relation + @xmath354 where @xmath355 ( this form of cycle relations was introduced by seven in @xcite ) . according to @xcite , all the cycle relations for a given chordless cycle follow from one with @xmath356 ( such a relation always exists ) . thus , as for quivers , we may take exactly one defining relation per cycle . it is shown in @xcite that the group @xmath347 does not depend on the choice of a diagram in the mutation class of @xmath1 . in particular , it is isomorphic to the initial coxeter group @xmath2 . the further part of the construction generalizes to the diagram settings verbatim . we obtain three manifolds , see table [ hyp - n ] @xmath2 & @xmath1 & @xmath346 & @xmath313 & @xmath357 & .actions on hyperbolic manifolds , non - simply - laced case . [ cols="^ " , ] + & & & & & & & + & & & @xmath358 & 2 & @xmath359 & compact & -4 + @xmath360 & & & @xmath361 & 3 & @xmath362 & 16 & + @xmath363 & & & @xmath364 & 3 & @xmath365 & compact & + the manifold @xmath150 constructed for the group @xmath360 coincides with the manifold constructed in section [ act - hyp ] for the group @xmath366 ( see table [ hyp ] ) . in contrast to the simply - laced case , we do not know whether the list of hyperbolic manifolds ( table [ hyp - n ] ) that can be obtain by our construction is complete : there is no known classification of finite volume hyperbolic coxeter polytopes with angles @xmath367 , so potentially the construction may provide other examples of finite volume hyperbolic manifolds . [ lemma - n ] the proof of the manifold property works for diagrams , but we need to update the proof of lemma [ eq - dihedral ] . let us proceed by induction on the number of mutations required to obtain @xmath346 from @xmath1 . assume that the lemma holds for a diagram @xmath368 mutation - equivalent to @xmath1 , and we want to deduce the lemma for @xmath369 . in the notation of lemma [ eq - dihedral ] ( after substituting @xmath46 by @xmath1 ) this means the following : if we denote by @xmath370 the reflections corresponding to nodes @xmath371 of @xmath368 and by @xmath372 their projections to @xmath2 , we want to deduce the equality of orders of elements @xmath373 and @xmath374 from the equality of the orders of @xmath375 and @xmath376 . if one of @xmath69 ( say , @xmath24 ) equals @xmath49 the statement is obvious : in this case the label of the arrow between @xmath24 and @xmath25 is the same in @xmath346 and @xmath368 ( and thus the orders of @xmath373 and @xmath375 are the same ) , and @xmath377 ( since either @xmath378 or @xmath379 ) , so the orders of @xmath263 and @xmath380 are also the same . if the other case we can restrict our consideration to the subdiagrams of @xmath346 and @xmath368 spanned by three vertices @xmath371 . a subdiagram of a diagram of finite type also has finite type @xcite , so if it is not simply - laced then @xmath346 ( and @xmath368 ) is mutation - equivalent to one of the three diagrams @xmath381 , @xmath382 and @xmath383 . now , a short straightforward check verifies the lemma . consider the diagram @xmath384 ( see fig . [ b3](a ) ) . the corresponding dynkin diagram of type @xmath381 determines a finite coxeter group @xmath2 acting on the @xmath5-dimensional sphere by reflections : @xmath385 the fundamental domain of this action is a spherical triangle with angles @xmath386 , see fig . [ b3](b ) for the action and a fundamental domain in stereographic projection . [ b3 ] now , apply the mutation @xmath93 to the diagram @xmath1 obtaining the diagram @xmath346 ( see fig . [ b3](a ) ) . the corresponding group @xmath387 acts on the hyperbolic plane by reflections , and the fundamental domain of this action is a triangle @xmath252 with angles @xmath388 . we need to take a quotient of this action by the normal closure of the element @xmath389 . as we expect from manifold property , we will get some finite volume hyperbolic surface @xmath21 tiled by 48 images of the triangle @xmath252 . it is easy to see that @xmath21 is a genus @xmath136 surface . similarly to section [ sph to tor ] describing @xmath11 case , one can see that @xmath21 can be obtained as a ramified degree @xmath0 covering of the sphere branching at six points composing the orbit of the `` central point '' ( see fig . [ b3](c ) ) . the result of @xcite allows to generalize the construction to infinite groups . more precisely , it is shown in @xcite that to a diagram @xmath1 of finite mutation type that either is exceptional ( see @xcite ) or arises from a triangulation of a unpunctured surface or orbifold ( see @xcite ) one can assign a coxeter group @xmath3 ( using exactly the same rule as above ) . this implies that we can repeat verbatim the construction of actions on factors of the davis complex @xmath394 , so that we get an action of an ( infinite ) group @xmath395 on the quotient space @xmath396 . furthermore , as before , a mutation of @xmath1 induces a mutation of @xmath397 . a diagram @xmath1 is called a diagram of _ affine type _ if @xmath1 is mutation - equivalent to an orientation of an affine dynkin diagram different from an oriented cycle . in particular , if @xmath1 is of affine type , the group @xmath398 is an affine coxeter group . mutation classes of diagrams of affine type may be rather complicated ( see @xcite for the detailed description of mutation classes of non - exceptional diagrams of finite and affine type ) . in particular , a diagram @xmath1 in the mutation class may contain one arrow labeled by @xmath0 ( this corresponds to two reflections in a coxeter system of @xmath3 generating an infinite dihedral group , or , equivalently , to two non - intersecting facets of a fundamental chamber of @xmath399 ) . we are currently unable to drop the assumption that @xmath1 is of affine type . our proof of manifold property uses an induction on the number of mutations one needs to perform to obtain a diagram without oriented cycles ( such diagrams are called _ acyclic _ , and diagrams from their mutation classes are _ mutation - acyclic _ ) . if @xmath1 is not of finite or affine type , then either @xmath1 is not mutation - acyclic , or it is not mutation - finite ( see @xcite ) . we are grateful to v. emery , m. feigin , a. king , r. marsh , r. kellerhals and m. shapiro for helpful discussions . we also thank r. guglielmetti for computing the volume of one of the hyperbolic pyramids . the work was initiated during the program on cluster algebras at msri in the fall of 2012 . we would like to thank the organizers of the program for invitation , and the institute for hospitality and excellent research atmosphere . , constructing hyperbolic manifolds , _ computational and geometric aspects of modern algebra _ ( edinburgh , 1998 ) , 7886 , london math . soc . lecture note ser . , 275 , cambridge univ . press , cambridge , 2000 . , skew - symmetric cluster algebras of finite mutation type , _ 14 ( 2012 ) , 11351180 . , cluster algebras of finite mutation type via unfoldings , _ int . math . res . notices _ ( 2012 ) , 17681804 . , cluster algebras and triangulated orbifolds , _ adv . math . _ 231 ( 2012 ) , 29533002 , on polyhedra of finite volume in the lobachevskii space with dihedral angles @xmath195 and @xmath196 ( russian ) , _ lectures in mathematics and its applications _ , vol . 2 ( russian ) , 151187 , ross . nauk , inst . steklova , moscow , 1988 .

charge transfer is ubiquitous in physical processes affecting biological , chemical , and materials systems . the representation of charge transfer is of intense current interest throughout the physical sciences . a powerful concept in both modeling and understanding how charges redistribute themselves during a physical process is chemical potential equalization.@xcite to apply chemical potential equalization successfully , it is essential to use a charge - dependent energy model which behaves correctly for all configurations encountered in the process of interest . consider a diatomic molecule ab . atom a is assumed to be more electropositive than atom b. we are interested in charge disproportionation reactions typified by @xmath0 and @xmath1 where the charge @xmath2 is not necessarily an integer . charge disproportionation and transfer reactions are important in electron transfer in biophysical systems,@xcite enzyme catalysis reactions,@xcite and such processes as electron - hole production and recombination in organic semiconductors.@xcite for dimers ( b = a ) , eq . ( [ i.0a ] ) corresponds to broken charge - symmetry states . furthermore , eqs . ( [ i.0a ] ) and ( [ i.0b ] ) have been the prototype reactions for a wide variety of theoretical studies of chemical bonding.@xcite the most prevalent model potential for charged species is the quadratic expansion of the classical electrostatic potential,@xcite @xmath3 the @xmath4 are expansion coefficients which are atom - type dependent and @xmath5 is an interionic potential representing the classical electrostatic contribution . in general all of these functions depend on the separation @xmath6 . the charge @xmath2 is usually , but not always , independent of @xmath6 . @xmath5 approaches a @xmath7 dependence at sufficiently large @xmath6 . @xmath8 contains the charge - independent short - range and dispersion interactions between the atoms . the other expansion coefficients are frequently interpreted in terms of physical quantities such as chemical potential and hardness.@xcite the quadratic form has the virtue of simplicity and works adequately when the range of @xmath6 is small enough to prevent @xmath2 from changing appreciably . if eq . ( [ i.1 ] ) is used at all separations , the atoms in the ab molecule will remain ionic even at separations where they are supposed to return to neutral states . for those situations where charge transfer does occur , alternative functional forms of charge dependence need to be invoked . for instance , alavi et al.@xcite and grochowski and coworkers@xcite use phenomenological switching functions to effect charge transfer . at very large separations , though , it is known that the charge dependence becomes piecewise linear ( fig . [ fig : v2u]).@xcite comparison of piecewise linear and quadratic models of the charge dependence of the energy for a diatomic . the solid line corresponds to the energy of the _ isolated _ atoms as a function of acquired charge @xmath2 ( measured with respect to charge transfer from the cation ) . the dashed curve corresponds to a quadratic representation of the charge dependence . the energy minimizes at @xmath9 for the piecewise linear model , but not for the quadratic model . at large separations , the piecewise linear model is correct . traditionally the quadratic model is considered to be more correct at finite separations . ] morales and martinez@xcite ( hereafter referred to as mm ) conclude that the `` grand canonical '' ( gc ) approach@xcite can not describe a realistic charge transfer process . in the gc approach each atom is considered to be an open system with respect to exchange of energy and numbers of electrons . they use a 3-state model with integer charge resonances representing the states of the ensemble for each atom . for atom a for instance , the gc energy is expressed in the form of an ensemble average @xmath10 where @xmath11 are the occupation numbers and @xmath12 are the energies for the integer charge species @xmath13 + , 0 , and @xmath14 . the @xmath15 depend on @xmath2 and are equal to or greater than zero . ( see refs . , , and for the detailed expressions . ) mm show that the gc energy as expressed by eq . ( [ i.1a ] ) minimizes to integer charges , and never to fractional charges . when the two atoms are well separated and only weakly interacting , the covalent and ionic resonance states can be regarded as being close to eigenfunctions and the mixing terms between the states can be ignored . the nonlinearities in the occupation numbers as functions of charge introduced through temperature at physically reasonable temperatures are not sufficient to produce states with fractional charges . thus charge transfer as represented by an ensemble of weakly interacting integer - charge states is too simplistic to describe fractionally charged states . mm also examine a 3-state valence bond ( vb ) approach usqing the same resonance states as in the gc approach . this is equivalent to a 0 k ensemble where the states are allowed to interact . the vb approach is able to represent charge transfer processes with fractional charges . however , the representation of the energy depends on resonance matrix elements . the issue becomes one of defining sets of coefficients for the state wavefunctions that allow one to recover quadratic and gc - like energy expressions . using coulson charges@xcite as an approximation and a maximum entropy valence bond approach ( mevb ) , mm derive energy expressions as functions of @xmath2 with the forms of eqs . ( [ i.1 ] ) and ( [ i.1a ] ) . in the mevb approach , decomposition into atomic contributions is accomplished through an examination of the various matrix elements appearing in the total energy expression . however , mm were unable to find a general expression linking the gc and classical electrostatic forms . the three main difficulties in deriving a charge dependent potential are the same ones facing mm and others . first is to define the charge , second is to evaluate or eliminate the resonance energy matrix elements , and third is to define the energy for a particular value of @xmath2 . first , to define @xmath2 , we invoke a density - functional - theory@xcite ( dft ) motivated atom - in - molecule@xcite pseudo - atom concept within the context of the evb approach . we assume the availability of a practical density decomposition strategy@xcite to define the pseudo - atom densities . the charge is defined as an average over the difference between two pseudo - atom densities . no restriction to coulson@xcite or other definitions of charge@xcite is necessary . the dependence of the evb wavefunction on @xmath2 is then deduced , which in turn yields an energy for arbitrary @xmath2 , not just the optimum @xmath2 . second , to evaluate resonance energy matrix elements , we retain the vb approach of mm , but use an empirical valence bond ( evb)@xcite strategy , rather than explicitly evaluating a model hamiltonian or using the mevb averaging procedure of mm . specifically , a reference energy is separated out for a fixed value of @xmath2 . to define the energy for an atom in the molecule , we require consistency between the evb wavefunction energy and dft energy for each atom . in so doing we are able to cast the energy of each atom into a form suitable for constructing empirical potential energy surfaces that could be used in simulations of larger systems . third , the typical situation is that there are many wavefuctions and many electron densities that are consistent with a particular @xmath2 . to define a unique energy from among the possible choices , we adhere to constrained search density functional theory ( cs - dft)@xcite rather than appealing to a maximum entropy principle . to the best of our knowledge , the evb method has not previously been combined with an atom - in - molecule approach.@xcite von szentply and coworkers define atomic charges , but their definition is implicitly limited to a quadratic dependence.@xcite grochowski and coworkers@xcite use evb parameterized by _ ab initio _ calculations for key molecular fragments and for defining the charges . the charges are defined with a phenomenological spatial dependence that does maintain correct dissociation to neutrals . furthermore , the charge dependence of the final potential is purely electrostatic . the inconsistency between the gc and classical electrostatic forms is explained in ciosowski and stefanov@xcite using a different definition of @xmath2 , which is based on the total system wavefunction . the charge is expressed as a perturbation on the molecular hamiltonian @xmath16 . computed atom - in - molecule equilibrium charges are used in subsequent `` charge - constrained '' calculations to study the energy and electronegativity dependences about the ground - state atom - in - molecule charges . nalewajski@xcite provides a simplified rendition of ciosowski and stefanov , but only for orthogonal resonance states . while physically correct , these efforts have not been cast in the form of general purpose potential energy surfaces . in particular , neither attempts to resolve the resonance energy issue in a way that is tractable for large - scale simulations . here we derive a charge - dependent empirical potential which faithfully represents charge transfer as a function of separation between atom a and an entity b. in the simplest case , entity b is another atom . more generally , b represents a collective embedding environment or reservior.@xcite the charge may be fractional . our primary interest is in deriving a general functional form with correct physical and chemical behavior at all interaction strengths rather than providing an exact treatment of particular terms or systems . the three difficulties just outlined are addressed . we then analyze these potentials in special limiting cases and in the light of the results of ciosowski and stefanov@xcite and nalewajski.@xcite finally , we derive general models for both pair potentials and atom - in - molecule energies . the evb method is a much more general technique than described here . reviews of evb are available from warshel _ et . al_.@xcite here we confine our discussion to a 2-state model of the ab molecule . there are fixed covalent and ionic resonance states represented by wavefunctions @xmath17 and @xmath18 , respectively.@xcite no assumption is made about the quality of these wavefunctions . however , @xmath17 retains @xmath19 electrons on atom a and @xmath20 electrons on atom b. atoms a and b are neutral in the covalent resonance state . consequently @xmath19 and @xmath20 are equal to their respective nuclear charges , @xmath21 and @xmath22 . similarly , @xmath18 retains @xmath23 electrons on atom a and @xmath24 electrons on atom b. in each resonance state , the total number of electrons is @xmath25 . @xmath17 and @xmath18 are assumed to be normalized to unity . the wavefunction of the system @xmath26 is the combination@xcite @xmath27 where @xmath28 determines the relative ionic character . by normalization , @xmath29 where @xmath30 for the ab molecule with hamiltonian @xmath16 , the mixed - state energy takes the form @xmath31 where @xmath32 , @xmath33 and @xmath34 are the associated energy matrix elements , @xmath35 and @xmath36 minimizing @xmath37 with respect to @xmath28 gives the optimized values , @xmath38 where @xmath39 and @xmath40 . the @xmath41 signs in eq . ( [ ii.7 ] ) correspond to ground ( gs ) and excited ( xs ) states , whose energies are designated as @xmath42 and @xmath43 , respectively . from eq . ( [ ii.7 ] ) , one can see that the off - diagonal matrix elements , @xmath34 and @xmath44 , control charge transfer in evb . when @xmath45 , @xmath46 . depending on the root , @xmath47 ( @xmath14 root ) or @xmath48 ( + root ) varies from @xmath141 to + 1 . note that as @xmath49 goes to zero , that is , as the covalent and ionic curves cross , charge transfer or the coulson - fischer transition@xcite becomes more abrupt . if @xmath50 and @xmath51 are nondegenerate eigenfunctions of @xmath16 , then @xmath52 . there is either no charge transfer or there is complete charge transfer . in either case , the states do not mix . the coefficient @xmath28 governs the ionic contribution to @xmath26 . as @xmath28 increases , @xmath26 migrates toward more complete charge transfer . even neglecting overlap,@xcite the ionic strength @xmath28 interpolates between covalent and ionic states in a physically reasonable way.@xcite the state - to - state interpolating behavior of @xmath28 in eq . ( [ ii.7 ] ) is the essential behavior that we wish to emulate in developing a more broadly applicable charge dependent potential . later in the paper , two other relationships will become useful , which we provide here . first , in the spirit of evb,@xcite one always wants to know the resonance energy in the terms of the ground - state @xmath42 , @xmath44 , @xmath32 , and @xmath33 . that is , @xmath53 where @xmath54 is either `` gs '' or `` xs '' . the second relationship expresses @xmath47 as a function of these same parameters . this is done by substituting eq . ( [ ii.8 ] ) into the expressions for @xmath55 and @xmath56 . the general expression is @xmath57 \ , \nonumber \\ \label{ii.9}\end{aligned}\ ] ] where @xmath58 is either `` @xmath59 '' or `` @xmath60 '' and either the ground or excited state is selected . we begin with the question of how to define charge in our model . there is no unique definition of the charge on an atom in a molecule . to assign charges to individual atoms , we assume that it is possible to decompose the total electron density into pseudo - atom densities . however , we do not need to specify a particular decomposition procedure at this time . we only need to know that some procedure is available.@xcite to define the charge , it is convenient to use @xmath25-electron density matrices and 1-electron densities . in terms of density matrix language , the state of ab is represented as @xmath61 where @xmath62 are the 3@xmath25 dimensional spatial electronic coordinates for the full ab system and @xmath26 is given in eq . ( [ ii.2 ] ) . spin is ignored at this point and , for simplicity , the matrix elements are assumed to be real . ( [ iv.1 ] ) corresponds to the pure - state representation of the density matrix for the 2-state model . we can expand eq . ( [ iv.1 ] ) in terms of the covalent and ionic states , resulting in the relationship @xmath63 where @xmath64 and @xmath65 the total electronic energy expression analogous to eq . ( [ iv.2 ] ) is @xmath66 eq . ( [ iv.3 ] ) corresponds to a non - diagonal representation of the energy . its advantage is that the component contributions of the essential states that are thought to represent the physical system are delineated . a diagonal representation of @xmath67 when state mixing is important can be devised by diagonalizing @xmath16 . for the purposes of extending the present treatment to finite temperature ensembles,@xcite this particular diagonalization of @xmath67 would be useful . the @xmath25-electron densities of interest are @xmath68 , @xmath69 , @xmath70 , and @xmath71 . they bear the usual relationships to the respective @xmath25-electron density matrices , @xmath72 : @xmath73 where @xmath74 are the 3(@xmath25 - 1 ) dimensional spatial electronic coordinates and @xmath75 is either `` @xmath59 '' , `` @xmath76 '' , `` @xmath60 '' , or no subscript . @xmath68 , @xmath69 , and @xmath71 are normalized to @xmath25 . the relationship of central interest is @xmath77 recall the assumption that the covalent and ionic state wavefunctions are given and fixed . thus the total density @xmath68 is determined solely by the value of @xmath28 . the energies of @xmath69 and @xmath71 are well - defined in a conventional dft sense.@xcite however , the `` interference density''@xcite @xmath70 does not have a well - defined energy in dft . nevertheless , its energy may be inferred from the energies of @xmath68 , @xmath69 , and @xmath71 , as we will show below . next we assume that all of the @xmath78 can be decomposed into corresponding pseudo - atom densities , @xmath79 and @xmath80.@xcite the pseudo - atom densities @xmath81 and @xmath82 integrate to non - integer values , @xmath83 and @xmath84 , whereas @xmath85 , @xmath86 , @xmath87 , and @xmath88 are constrained to integrate to integer numbers of electrons . we use asterisks throughout to indicate atom - in - molecule quantities . with these definitions in place , we define @xmath2 from either pair of total and covalent pseudo - atom densities , which others have sometimes referred to as pseudo - atom distortion densities.@xcite we choose atom a : @xmath89 the density decompositions must be constrained to yield the correct number of electrons prescribed by eq . ( [ i.14]).@xcite finally , we note that the present definition of @xmath2 falls into truhlar s class ii category.@xcite now we want to eliminate @xmath28 in favor of @xmath2 . assuming that a component definition of @xmath90 based on eq . ( [ iii.1 ] ) is possible , @xmath91 where @xmath92 @xmath93 is the number of electrons on atom a when it is in a neutral state , and the relationship @xmath94 has been used . the quantity @xmath95 is determined by the difference between @xmath96 and the atom a component of the decomposition of @xmath70 with locally unbiased , electron - number decomposition.@xcite clearly , different density decomposition strategies will yield somewhat different values of @xmath97 . one possibility for determining @xmath97 is to require consistency with the ground - state value of @xmath28 . for instance , if @xmath9 is optimum as for a dimer , @xmath28 equals either 0 or @xmath98 . for eq . ( [ iii.3 ] ) to be applied successfully , @xmath99 would have to correspond to a lower energy state than @xmath100 and the value of @xmath28 would have to be determined from a separate calculation , such as represented by eq . ( [ ii.7 ] ) . in such an approach , one would be effectively modeling @xmath95 via a correspondence with the coulson - fischer transition.@xcite alternatively , by analogy with @xmath68 , we assume that single - particle determinants ( e.g. kohn - sham determinants@xcite ) can be calculated for the @xmath101.@xcite in a kohn - sham based approach , one would be effectively estimating @xmath96 from the overlaps of these determinants . additional ambiguity in @xmath102 arises from the choices for @xmath50 and @xmath51 . these ambiguities lie behind the designation of the present approach as an empirical one . however , these ambiguities can be mitigated by using a reference state as discussed in the next subsection . to that end , it is useful to invert eq . ( [ iii.3 ] ) so that @xmath97 becomes a function of @xmath2 and @xmath28 . that relationship is @xmath103 it should be understood that , in the limit that @xmath104 , @xmath105 also . to construct a potential energy surface for ab , one option is to use eq . ( [ iii.3 ] ) to model the dependence of the charge on separation for some reference state . some empirical potentials such as evb and the modified embedded atom method ( meam)@xcite utilize reference states as a model calibration method . the methods of mcdonald and coworkers,@xcite mccammon , grochowski , and coworkers,@xcite and broughton and mehl@xcite effectively make @xmath2 bond - length dependent . ( [ iii.3 ] ) provides a basis in evb theory for their phenomenological charge transfer switching functions . a more attractive option is to solve for @xmath106 , @xmath107 the coefficient @xmath28 determines the strength of the contribution of @xmath51 to @xmath26 . ( [ iii.5 ] ) states how the charge governs that strength . this expression is consistent with the results of ciosowski and stefanov,@xcite which are derived from a perturbative technique . as noted previously , even if @xmath9 , @xmath28 equals either 0 or @xmath108 . this is because the evb model describes state mixing even when there is no charge transfer . for example , in the @xmath109 molecule , the covalent and ionic wavefunctions mix at all finite separations , but the ground state never involves charge transfer . significantly , this formula also describes deviations from the ground - state charge . ( [ iii.5 ] ) can be substituted into eq . ( [ ii.6 ] ) , and the variational procedure repeated . the result is the same as solving for @xmath2 in terms of the resonance and overlap matrix elements obtained by equating eq . ( [ iii.5 ] ) and ( [ ii.7 ] ) . the evb strategy is to use experimental information to eliminate the resonance energy.@xcite here the analogous procedure is to choose a particular value of @xmath110 and solve for @xmath34 in terms of @xmath111 and the diagonal matrix elements for each @xmath6 . ( the @xmath6 dependence is suppressed . ) the result is @xmath112 substituting eq . ( [ iii.5a ] ) into eq . ( [ iv.3 ] ) , the total energy for arbitrary @xmath113 has the form @xmath114 this form might be used in lieu of classical electrostatic potentials like eq . ( [ i.1 ] ) that have been in common use . it has the structure of an ensemble average,@xcite but the coefficients of @xmath32 and @xmath33 are not necessarily positive semidefinite . ( [ iii.5b ] ) is constructed to possess the proper changes in atomic charges in the limit of molecular dissociation . the construction of a proper ensemble representation is discussed in section 5.a . the foundations for the quadratic dependence of the energy on charge , such as eq . ( [ i.1 ] ) , must stem from eqs . ( [ iii.5 ] ) and ( [ ii.6 ] ) . however , even this simplest example of charge transfer has a considerably more complex dependence than quadratic . that dependence is clearly carried through the overlap contributions . once @xmath115 has been chosen , the procedure for determining a point on the potential energy surface for arbitrary values of @xmath6 and @xmath2 is as follows . from some other source(s ) of information , one must have available five reference values : @xmath111 , its associated ionicity @xmath116 , @xmath32 , @xmath33 , and @xmath44 . for a chosen @xmath6 , one first evaluates * @xmath117 , etc . from eq . ( [ iv.2 ] ) ; * @xmath34 from eq . ( [ iii.5a ] ) ; and * @xmath97 from eq . ( [ iii.10 ] ) . + then , for each @xmath2 of interest , one evaluates * @xmath118 from eq . ( [ iii.5 ] ) using values from steps ( a ) , ( b ) , and ( c ) ; and * @xmath119 from eq . ( [ ii.5 ] ) . the procedure is repeated for each value of @xmath6 of interest . an important question is what to choose for @xmath115 . one possible choice for @xmath115 is the optimum value @xmath120 . however , @xmath120 depends on @xmath34 through eqs . ( [ ii.7 ] ) and ( [ iii.3 ] ) . this variant is equivalent to using the ground - state wavefunction as one of the basis functions in the original formulation of the problem . ( [ iii.5b ] ) then characterizes deviations of the energy from the ground - state energy , @xmath121 , as a function of @xmath2 . note that by choosing @xmath111 to be consistent with @xmath121 , this variant and eq . ( [ iii.5b ] ) will be identical . when @xmath122 , the procedure for determining a point on the potential energy surface is substantially the same as the first . again five reference values are needed , except that knowledge of @xmath123 replaces knowledge of @xmath116 . relative to the first procedure , steps ( a ) and ( b ) become to evaluate * @xmath34 from eq . ( [ ii.8 ] ) ; and * @xmath124 from eq . ( [ ii.7 ] ) . all of the other steps remain the same . one can extend this model to @xmath2 between @xmath125 and 0 . the entire procedure with atom a assumed to be anionic in @xmath51 is repeated . in eq . ( [ i.14 ] ) , for @xmath126 , one simply replaces @xmath2 with @xmath127 and adjusts the partitioning of @xmath70 so that atom a is anionic . requiring continuity in the energy at @xmath9 dictates that @xmath128 as @xmath129 . ( [ iii.5b ] ) remains the same structurally . equivalently , one could replace the subscript `` a '' with the subscript `` b '' everywhere in the procedure . the potential over the entire range of @xmath2 is then represented in a piecewise fashion . while not as rigorous as a 3-state model , the present treatment does retain substantially greater simplicity . another possible extension is to apply the above procedure when the two resonance states both correspond to charged species . for cao , for example , the effective @xmath2 on ca near equilibrium would be almost + 2 . as the cao bond is stretched , @xmath2 would decrease until it passed through a region of undetermined length where it would range between + 1 and 0 . then there would be one form of eq . ( [ iii.5b ] ) covering the range in @xmath2 between 0 and + 1 and a second form covering the range between + 1 and + 2 range . a final extension of this model would allow the entity b correspond to a more general environment than just one other atom . most of the model presented here does not explicitly invoke the specific properties of a diatomic model . however , this extension requires separate considerations not pursued here . we now show how to define pseudo - atom energies for the 2-state model of the previous section . we do this by requiring consistency between the energies based on the density decompositions that were assumed in the previous section and the energies that would result from the corresponding wavefunction expressions . as with @xmath68 , we assume that there is a decomposition @xmath67 into pseudo - atom density matrices . in conformance with rychlewski and parr,@xcite the decomposition applies to @xmath67 rather than to @xmath16 . thus , for @xmath130 the total energy decomposes into @xmath131 analogous expressions are assumed to exist for each of the covalent and ionic contributions to the pseudo - atom energies . the pseudo - atom densities @xmath81 and @xmath82 correspond to the pseudo - atom density matrices @xmath132 and @xmath133 . in the following relations , the expressions for atoms a and b are analogous . only the expressions for atom a will be given . we want the energies of each pseudo - atom defined via density matrices to be equal to the energies defined via densities . consequently , we require that @xmath134 \ . \label{iv.5a}\ ] ] this identification places a new constraint on @xmath132 . rigorously speaking , it should be optimal in the sense of levy cs - dft.@xcite the constraint is that @xmath132 should yield the lowest energy for all other ensemble density matrices which integrate to @xmath81 : @xmath135 \equiv \langle h,\gamma_{\rm a}^*\rangle = \min_{\gamma_{\rm a } \rightarrow \rho_{\rm a}^ * } \langle h,\gamma_{\rm a}\rangle \ . \label{iv.5c}\ ] ] in terms of valence bond resonance states @xmath136 similar to @xmath137 , the pure - state terms in eqs . ( [ iv.6 ] ) , @xmath138 and @xmath139 , can be represented in conventional dft language . the resonance energy can not . ( [ iv.6 ] ) represents pseudo - atom energies for any @xmath2 . to evaluate the resonance energy , we again follow the evb strategy of determining them from some particular value of @xmath110 . the result is @xmath140 if @xmath115 is set equal to the optimum @xmath2 for each value of atomic separation , then @xmath141 will correspond to the experimental potential energy for ab . note that all of the energies in traditional evb are 0 k values . some variations of evb incorporate temperature - dependent solvent effects.@xcite we do not include these variations here . finite temperatures are not required to establish the model . on the other hand , there is nothing here that precludes extending the analysis to a finite temperature ensemble.@xcite substituting eq . ( [ iv.7a ] ) into eq . ( [ iv.6 ] ) , an expression for the pseudo - atom energy is achieved . for any @xmath2 , @xmath142 eq . ( [ iv.8a ] ) has the same structure as eqs . ( [ iii.5b ] ) , but refers to an individual atom . all of the quantities on the right - hand sides of these two equations can be deduced from experiment and/or decomposition calculations on resonance states and overlaps . using the present formulations in simulations of larger systems naturally invokes consideration of chemical potential equalization . one can obtain a statement of chemical potential equalization from eq . ( [ iv.8a ] ) . the total energy is given in eq . ( [ iv.4a ] ) . if @xmath37 is minimized with respect to @xmath2 , then small deviations from @xmath120 will not change the total energy to first order : @xmath143 or @xmath144 where @xmath145 is the chemical potential . charge balance requires that the charges on a and b be exactly opposite . the minus sign in front of @xmath145 comes from the fact that @xmath2 is related to the negative of the change in the number of electrons on a. here we introduce approximations consistent with the dissociation limits of ab , present a general definition of the energy for a given charge , and discuss @xmath146 , hf , and lih as examples . it is insightful to introduce a concept of neglect of differential overlap between resonance states , @xmath147 , which we will refer to as ndol . this leads to simple analytical expressions whose behavior can be examined in detail . ndol is to be distinguished from zero differential overlap ( zdo),@xcite which refers to overlap between orbitals on different atomic centers . in the ndol approximation , @xmath147.@xcite under special conditions , zdo implies ndol . in the ndol approximation , @xmath148 and @xmath44 are zero . from eq . ( [ iii.5 ] ) , the dependence of @xmath28 on @xmath2 becomes @xmath149 from eq . ( [ ii.6 ] ) , the energy dependence on @xmath2 becomes @xmath150 the @xmath151 construct ensures that @xmath152 corresponds to the ground state . here we have used the fact that , by definition of the ionization potential ip and electron affinity ea , @xmath153 . these atom - in - molecule quantities include some electrostatic contributions . this quantity is also called a `` bond hardness''.@xcite thus , eq . ( [ iii.6.1 ] ) is consistent with previous energy expressions obtained with coulson charges@xcite and with the 2-state model of nalewajski@xcite . in coulson,@xcite murrell al_,@xcite and mcweeny@xcite a fraction of ionic character is defined instead of a charge . that fraction is identical to @xmath154 which is @xmath2 in the ndol approximation . clearly , at large @xmath6 , eq . ( [ ii.5 ] ) in combination with eq . ( [ iii.5 ] ) approaches eq . ( [ iii.6 ] ) , eq . ( [ iii.6 ] ) becomes linear in @xmath2 , and @xmath33 , @xmath32 , and @xmath34 approach asymptotic values ( in @xmath6 ) analogous to mm s eq . ( 2.40).@xcite the linear behavior in the asymptotic regime is consistent with the conclusions of pplb . even more importantly , eq . ( [ iii.6 ] ) is expressly non - analytical ( _ i.e. _ , it can not be expanded in a taylor series ) about @xmath9 and @xmath155 , whereas eq . ( [ ii.5 ] ) in combination with eq . ( [ iii.5 ] ) is analytical at both points away from the ndol limit ( @xmath156 ) . coulson s fig . 5.7 is a graph of @xmath157 , which clearly shows the non - analytical behavior , although he did not comment on it.@xcite the non - analytical behavior of the energy as a function of @xmath2 is also seen in gc - dft.@xcite eq . ( [ iii.6 ] ) embodies the evb representation of that behavior . the non - analytical behavior results in a derivative discontinuity in the energy as a function of charge at integer values of the charge . perhaps because orthogonality between resonance states is assumed , nalewajski does not comment on the behavior of the derivative at integer charges.@xcite ciosowski and stefanov do see their version of the ndol limit as connected with pplb.@xcite likewise , pplb,@xcite perdew,@xcite and ciosowski and stefanov@xcite note that the derivative discontinuity disappears once the systems in the gc representation begin to interact significantly . ( [ iii.5 ] ) embodies that behavior as well . examples of these charge dependences are illustrated below . starting from eq . ( [ iii.6 ] ) , we evaluate @xmath34 at the optimum ndol @xmath2 , @xmath158 where @xmath159 . the ground state corresponds to the negative root of @xmath118 in eq . ( [ iii.6.1 ] ) , since as @xmath160 , @xmath2 must also go to zero , assuming that @xmath161 . for convenience , we call the ground state value of eq . ( [ iii.6 ] ) @xmath162 . the solution for the resonance energy is equivalent to the well - known evb expression@xcite @xmath163 substituting eq . ( [ iv.13 ] ) into eq . ( [ iii.6 ] ) , we find @xmath164 the dependence of @xmath165 on @xmath32 and @xmath33 appears to be different from that implied by eq . ( [ iii.5b ] ) . in fact , by setting @xmath166 in eq . ( [ iii.5b ] ) , the two expressions become identical . we can gain further insight from eq . ( [ iv.14 ] ) by solving for @xmath32 and @xmath33 in terms of @xmath167 , @xmath43 , and @xmath168 . in a typical diatomic , @xmath43 corresponds to the first electronic state which dissociates to the ions @xmath169 and @xmath170 . because of the ndol approximation , these three pieces of information are sufficient to specify the energy . to achieve the desired result , we first use the fact that @xmath171 can be derived from either eigenvalue to find that @xmath172 . next we solve for @xmath32 from eq . ( [ iv.12 ] ) assuming that we know @xmath173 . the result is that @xmath174 and @xmath175 in the ndol approximation , the pure state energies are simple linear combinations of the eigenenergies . making all of the necessary substitutions and rearrangements in eq . ( [ iv.14 ] ) , we achieve the ensemble representation : @xmath176 where the occupation number is @xmath177 to see that eq . ( [ iv.15c ] ) has the desired properties , note that @xmath178 lies between 0 and 1 over the interval [ 0,1 ] in @xmath2 , it is 0 at @xmath179 , and @xmath180 at @xmath179 is also 0 . these properties are illustrated in fig . [ fig : omega - ndol ] . taking into account our restriction to a 2-state model , if we apply atom decomposition to the eigenenergies , we would obtain the same form as eq . ( [ i.1a ] ) , with @xmath181 , @xmath182 , and @xmath183 . importantly , the representation is in terms of eigenenergies instead of energy matrix elements . furthermore , as @xmath173 approaches 0 , as in the assumed dissociation limit for ab , @xmath165 becomes linear in @xmath2 . 2-state evb - ndol occupation number , eq . ( [ iv.15e ] ) , in the ensemble representation of @xmath184 , eq . ( [ iv.15c ] ) . charges are dimensionless . ] another representation of the occupation number , eq . ( [ iv.15e ] ) , is significant . by using the relationship @xmath185 from eqs . ( [ iii.6.1 ] ) , we obtain @xmath186 as is necessary physically , the occupation number is 0 when @xmath187 , where @xmath188 corresponds to @xmath189 . the complete generalization of eq . ( [ iv.15f ] ) is equivalent to following the process steps outlined above . ( [ iv.15a ] ) and ( [ iv.15b ] ) can be inverted . inversion gives @xmath165 in terms of @xmath32 and @xmath33 : @xmath190 where @xmath191 and @xmath192 consistent with the conclusions of mm and physical necessity , the coefficients are not positive - semidefinite . [ fig : omega - h ] illustrates eq . ( [ iv.15h ] ) , the coefficient for the covalent state . in order to cover the range of energies between @xmath42 and @xmath43 , the coefficients of @xmath32 and @xmath33 can not possibly be positive - semidefinite . the coefficients are not defined for @xmath193 . at that value of the ground - state charge , @xmath32 must equal @xmath33 . 2-state evb - ndol coefficient for the covalent state given by eq . ( [ iv.15h ] ) . the ionic state coefficient is obtained by reflecting this figure through @xmath194 . charges are dimensionless . ] it is interesting to see how the present results connect with classical electrostatic potentials ( eq . ( [ i.1 ] ) ) . we can expand eq . ( [ iv.15e ] ) locally as a function of @xmath2 as long as the expansion point is not zero or one . first , @xmath195 the obvious value about which to expand is @xmath173 . by construction , @xmath196 if we evaluate eq . ( [ iv.17 ] ) at @xmath179 , as @xmath197 0 or 1 , the expansion behaves badly . alternatively , we might try expanding about @xmath198 , for the physically appealing reason that @xmath199 . then @xmath200 to second order , @xmath201 this quadratic expansion has the form of eq . ( [ i.1 ] ) and behaves well physically under dissociation , even to the extent of preserving the dissociation limit . the electronegativity , hardness , and electrostatic contributions are embedded in the eigenenergies . the equivalent expansion of eq . ( [ iv.15 g ] ) might be more revealing in displaying these contributions . a key observation is that the quadratic term vanishes completely at infinite separation ( @xmath197 0 ) and only the linear dependence survives . again , the survival of the linear charge dependence is consistent with pplb . there is no residual atomic hardness contribution as has appeared in many implementations of eq . ( [ i.1]).@xcite we speculate that a hardness contribution might be missing because we have considered only a 2-state model instead of a 3-state model . our reasoning behind this speculation is discussed below . we have ruled out the possibility that introducing the ndol approximation prior to making the expansion is a factor . nevertheless , adding a third state must not change the fact that the coefficient of the quadratic term must go zero in order for the results to be consistent with pplb.@xcite likewise , any hardness contributions to the linear term must be scaled by a coefficient that goes to zero at large @xmath6 . clearly , one can not approximate either eq . ( [ iii.5 ] ) or ( [ iii.6 ] ) quadratically to arbritrary accuracy in a global sense . next we consider the densities and their energies in the ndol approximation . from eq . ( [ iii.1 ] ) we have @xmath202 thus , from eq . ( [ iii.6.1 ] ) , @xmath203 by symmetry in eq . ( [ iv.20 ] ) , @xmath204 , but , from eq . ( [ ii.5 ] ) , @xmath205 . to address this difficulty , mm appeal to a maximum entropy principle . we prefer instead to appeal to cs - dft.@xcite accordingly , cs - dft instructs us to place the @xmath206 into groups defined by the density that they produce . since @xmath207 and @xmath208 both yield the same density , @xmath209 , they are grouped together . the energy assigned to @xmath210 $ ] is the minimum of the energies for these two @xmath67 s : @xmath211 = \min \ { \langle h,\gamma^{\rm ndol}(\gamma(q))\rangle , \nonumber \\ \langle h,\gamma^{\rm ndol}(-\gamma(q))\rangle\ } \ . \label{iv.23}\end{aligned}\ ] ] the ground state energy is then the minimum over all @xmath2 of @xmath212 $ ] . when differential overlap ( @xmath213 ) is included , the densities appear to become unique for @xmath28 s of different signs.@xcite note that a similar procedure could be followed for mm s 3-state case at the zdo level that they assume . see also ref . . even with this comparatively simple extension , the situation is less clear than in the 2-state case . many densities may have the same charge on the atoms . by again appealing to cs - dft , one can assign @xmath119 for a given @xmath2 by minimizing over the energies of all densities with the same @xmath2 . that is , as a straightforward extension of cs - dft , we very generally define @xmath214 \ , \label{iv.24}\ ] ] where @xmath215 represents all of the other undetermined parameters of the density @xmath216 . by minimizing over @xmath215 , one may introduce dependencies on the energy matrix elements that are absent from the present 2-state model.@xcite this is the reason behind our speculation that the isolated - atom hardnesses do not appear in eq . ( [ iv.19 ] ) because of the limitations of the 2-state model . one advantage of the charge - generalized cs - dft approach over a maximum entropy principle is that the true ground state properties can be preserved in the same way that the ground state energy can be recovered by minimizing over all allowed densities in conventional cs - dft.@xcite first we discuss modeling the polar molecules hf and lih in the ndol approximation . then , for the nonpolar molecule @xmath146 , we compare the ndol and general cases . there we utilize the weinbaum wavefunction.@xcite it provides an excellent illustration of the ambiguities encoountered in defining the resonance state wavefunctions . we also examine various approximations for @xmath148 and examine the change in the charge dependence as a function of @xmath6 . for hf and lih , we use rkr curves@xcite to define @xmath42 and @xmath43 . for the x@xmath218 and b@xmath218 states of hf , the rkr data are from di lonardo and douglas,@xcite ; for the x@xmath218 and a@xmath218 states of lih , the rkr data are from chan _ et al._@xcite and pardo _ et al._@xcite we use the calculations of ref . to define @xmath173 . these data are shown in figs . [ fig : hf - fit ] and [ fig : lih - fit ] . the b@xmath218 and a@xmath218 states dissociate to ions , h@xmath219 and f@xmath220 and li@xmath219 and h@xmath220 , respectively . to allow matching of the different spatial ranges of the data , analytical fits for the energy curves were made with the rose and rose+ionic functional forms.@xcite the charge data were fit with the functional form , @xmath221 , where the @xmath222 s are fitting parameters . @xmath32 and @xmath33 were computed from eqs . ( [ iv.15a ] ) and ( [ iv.15b ] ) , respectively . in both cases , @xmath32 and @xmath33 cross at @xmath223 and meet their respective states at the dissociation limits . representative shapes of the charge dependence for given values of @xmath6 are shown in figs . [ fig : hf - cuts ] and [ fig : lih - cuts ] . the correct dissociation - limit behavior is observed in both cases . in simulations where each atom remains within a unit charge interval , the ndol model might therefore prove to be useful , although it is unlikely to be quantitative . the difference in charge transfer characteristics between the present results and pplb can be seen by contrasting the @xmath224 curve of fig . [ fig : lih - cuts ] with the @xmath225 curve of their fig . 1.@xcite the transfer is more gradual here and passes through fractional charge states , compared to pplb which is very sharp and passes directly from a completely covalent state to a completely ionic one . next we consider @xmath146 . the simplest valence bond form for the covalent state @xmath226 , where @xmath227 is the atomic orbital overlap.@xcite our nave inclination for the ionic state is to use the familiar @xmath228 . the total wavefunction is @xmath229 . it turns out that the total densities for both states are identical : @xmath230 . ( designation of the electronic coordinate is suppressed in the densities for readability . ) the interference density is @xmath231 . evb model of hf . all energies are in ev , distances in , and charges dimensionless . rkr data from ref . . charges from ref . . ] evb model of hf . all energies are in ev , distances in , and charges dimensionless . rkr data from ref . and . charges from ref . . ] eq . ( [ iv.15c ] ) at discrete values of @xmath6 for hf . all energies are in ev , distances in , and charges dimensionless . ] ( [ iv.15c ] ) at discrete values of @xmath6 for lih . all energies are in ev , distances in , and charges dimensionless . ] if we assume a simple hirshfeld partitioning as our total density decomposition strategy,@xcite then it is natural to assume that each component of the density is likewise scaled by @xmath232 in order to obtain the atom a contribution . however , we find that because of symmetry , @xmath233 , which can not be correct . of course , the problem is that the assumption about the hirshfeld partitioning form is incorrect . this partitioning can not lead to a value of @xmath234 that integrates to @xmath235 . a different partitioning must be chosen to force all of the density in @xmath71 to belong to atom b. in addition , we know the optimum values of @xmath28 for these two resonance states@xcite and that these values correspond to @xmath9 . for instance , at the equilibrium @xmath146 separation , @xmath236 , with the resonance energy defined as positive.@xcite from eq . ( [ iii.3 ] ) , we also can deduce that @xmath237 . this analysis hints at the subtle properties that the density decomposition must possess if one implements explicit decomposition of @xmath70 . another immediate insight is that any fixed value of @xmath28 applied within @xmath26 leads to an acceptable covalent state . clearly , the energy change caused by a polarization of the @xmath146 charge density will differ depending on one s choice for the resonance state . similarly , the density decomposition will show some sensitivity to this choice . of course , a more sensible choice for the ionic state is @xmath238 because , for a homonuclear diatomic , the charged states correspond to broken charge - symmetry states . even in this simple case , the partitioning of @xmath70 may be nontrivial . as a final consideration , we illustrate the influence of overlap on ionicity . for this purpose , we utilize our original choices for covalent and ionic wavefunctions , @xmath239 and @xmath240 . the overlap integral is estimated in terms of @xmath227 as given above . taking @xmath241 , the atomic overlap is @xmath242 , where @xmath243 . the effect of overlap is shown in fig . [ fig : h2-ion ] as @xmath244 . there it can been seen that the ndol approximation becomes accurate beyond approximately 2 . the relatively large range of @xmath6 for which the ndol approximation is accurate in this case is due in no small part to the fact that @xmath44 depends on the square of @xmath227 . this broad range of accuracy and the fact that eqs . ( [ iv.15c ] ) and ( [ iv.15e ] ) are well - behaved for any @xmath6 makes it tempting to use ndol in general . however , near equilibrium bond lengths , fig . [ fig : h2-ion ] indicates that overlap effects should not be ignored . ionicity scaled by @xmath245 at discrete values of @xmath6 for @xmath146 . distances are in and charges are dimensionless . ] in order to apply chemical potential equalization in a simulation that involves conditions far from reference states , the potential energy must be defined for arbitrary values of the charges . to address this challenge , we have derived a new charge - dependent pair potential from a 2-state empirical valence bond model . the charge is defined from a decomposition of the density into constituent contributions . the explicit charge dependence is deduced by requiring consistency between the density decomposition and the wavefunction descriptions of the ground state . the energy expression can be made valid for any range of charge of interest . the decomposition theme is further extended to define the energy of individual constituents , again by requiring consistency between the density decomposition and the wavefunction descriptions . the energy of the system for a given value of charge is made unique by appealing to constrained search density functional theory . an examination of the model shows linear dependence on charge at the dissociation limit , as well as discontinuous behavior in the derivative of the energy as a function of charge at integer values of the charge . this behavior is consistent with the analysis of perdew _ et al._@xcite and others.@xcite the potential energy for arbitrary charge and separation of the constituents can be represented as an ensemble average of the eigenenergies with a nonlinear , analytical dependence of the occupation number on charge . the representation of the potential energy in terms of the pure - state energy matrix elements is possible , but , by physical necessity , can not be expressed as an ensemble average with positive semidefinite coefficients . to determine a pair potential for all values of @xmath2 and @xmath6 with this method , one needs five reference or calibration curves . these most often will be computed values of the ground - state energy , the charges along the ground - state energy curve , the resonance state overlap integral , the covalent energy and ionic energy . measurements of ground and ionic - excited state energies and charges can also be used . the covalent energy and ionic energy are determined for integer charges only . to determine a point on the potential energy curve , a well - defined five step procedure is followed using the five input curves . an immediate application of the method could be to construct the reference potential curves for the @xmath246 , @xmath247 , and ab systems . simulations on arbitrary mixtures of these three types of systems under nonequilibrium initial conditions would then be possible . chemical potential equalization would be used to dynamically adjust the charges of the constituents . more generally , we envision the present approach as forming the basis for a new class of charge - dependent empirical potentials for use in large - scale simulations of reactive systems . the work of s.m.v . was performed in part at los alamos national laboratory under the auspices of the u. s. department of energy , under contract no . w-7405-eng-36 , and funded through its center for semiconductor modeling and simulation , a crada program performed jointly with the semiconductor research corporation , and through the advanced fuel cycle initiative . s.m.v . thanks the university of new mexico , department of physics and astronomy for its hospitality during the 2003 - 2004 academic year . s.r.a . would like to thank the national science foundation for support during the initial stages of this work ( dmr-9520371 ) . this work was supported by national science foundation grant no . che-0304710 . m. i. page , `` the mechanism of chemical catalysis used by enzymes , '' in _ new comprehensive biochemistry , vol . 6 the chemistry of enzyme action _ edited by m. i. page , ( elsevier , amsterdam , 1984 ) , pp . 229 - 270 . van der horst , p. a. bobbert , p. h. l. de jong , m. a. j. michels , g. brock , and p. j. kelly , phys . b * 61 * , 15817 ( 2000 ) ; m. rohlfing and s. g. louie , phys . rev . b * 62 * , 4927 ( 2000 ) ; m. rohlfing , m. l. tiago , and s. g. louie , synth . metals * 116 * , 101 ( 2001 ) . u. w. schmitt , and g. a. voth , j. phys . b * 102 * , 5547 ( 1998 ) ; u. w. schmitt , and g. a. voth , j. chem . phys . * 111 * , 9361 ( 1999 ) ; m. cuma , u. w. schmitt , and g. a. voth , j. phys . chem . a * 105 * , 2814 ( 2001 ) . l. jaroszewski , b. lesyng , and j. a. mccammon , j. mol . ( theochem ) * 283 * , 57 ( 1993 ) ; p. grochowski , b. lesyng , p. baa , and j. a. mccammon , int . j. quant . chem . * 60 * , 1143 ( 1996 ) ; j. trylska , p. grochowski , and m. geller , int . j. quant . * 82 * , 86 ( 2001 ) . j. h. rose , j. ferrante , and j. r. smith , phys . 47 * , 675 ( 1981 ) ; j. h. rose , j. ferrante , and j. r. smith , phys . b * 28 * , 1835 ( 1983 ) ; j. h. rose , j. r. smith , f. guinea , and j. ferrante , phys . b * 29 * , 2963 ( 1984 ) ; j. ferrante , and j. r. smith , phys . b * 31 * , 3427 ( 1985 ) ; j. r. smith , h. schlosser , w. leaf , j. ferrante , and j. h. rose , phys . a * 39 * , 514 ( 1989 ) .

disordered media may homogenize in various ways . for example , in a microscopically inhomogeneous material , the solution to the heat equation might satisfy a local limit theorem when viewed on larger scales . this might be on the entire space or within a bounded domain with certain boundary conditions . in bounded domains we can furthermore ask whether the solution to a poisson equation homogenizes as the domain s diameter , i.e. , the macroscopic scale , grows to infinity . or , alternatively , we can wonder whether the dirichlet spectrum of the associated laplace operator converges in some sense or whether the occupation time measures of the corresponding diffusion fulfill a large deviation principle . when we say that we let the macroscopic scale grow to infinity , this is always in comparison to a microscopic scale @xmath1 , which might tend to zero instead . although all these aspects of homogenization are _ a priori _ different , intuition suggests that they should somehow be related . especially if the associated laplace operator is self - adjoint , i.e. , it is the generator of a reversible random walk , then the homogenization of the poisson equation is strongly linked to spectral homogenization ( see ( * ? ? ? * chapter 11 ) ) . spectral homogenization in turn is linked to the validity of a large deviation principle for the occupation time measures of a random walk in bounded domains ( * ? ? ? * theorem 5 ) . it is therefore plausible that these aspects of homogenization should hold under similar conditions . for self - adjoint laplace operators , a crucial condition for many kinds of asymptotic homogenization is apart from ergodicity the validity of a poincar inequality with an optimal constant independent of @xmath1 ( uniform poincar inequality ) . for spectral homogenization this is immediately evident since the optimal poincar constant is the inverse of the principal dirichlet eigenvalue of the laplacian ( see remark [ rem : optimality ] ) . in the situation of the present paper , we will see that the uniform poincar inequality is necessary and carries us quite far , although it is not completely sufficient for our results . however , it leads us to conditions that are optimal ( up to a critical case , see remark [ rem : optimality ] ) . in the present paper , we examine a discrete disordered medium that belongs to a class of random conductance models on the lattice @xmath0 with stationary and ergodic conductances on nearest - neighbor and unbounded - range connections . random conductance models are of high mathematical and physical interest ( see @xcite and references therein ) . for these models , ( * ? ? ? * theorem 1.11 , remark 1.12 ) already used the poincar inequality and a related sobolev inequality to prove the validity of a local limit theorem for the heat kernel in the case where only nearest neighbors are connected . as we explained above , a uniform poincar inequality is also necessary for spectral homogenization . in this model , its validity depends on the integrability of the tails of the conductances ( see e.g. proposition [ prop : localpoincare ] and ( * ? ? ? * proposition 2.1 ) ) . to be more precise , let @xmath2 denote the random conductance on the edge @xmath3 and define @xmath4<\infty \}\end{aligned}\ ] ] where @xmath5 denotes the expectation with respect to @xmath2 . for the moment , let us assume that only nearest neighbors are connected . then the crucial assumption is @xmath6 see assumption [ ass : integ ] . additionally we require that @xmath7<\infty$ ] ( cf . assumption [ ass : environ][ass : momcond ] ) . as we explain in remark [ rem : optimality ] , up to the critical case @xmath8 , the condition in is optimal . in fact , @xmath9<\infty$ ] is sufficient for the poincar inequality but not for the moser iteration , which we use in section [ subsec : maximalinequalities ] . if @xmath10 , then it is possible ( and in the i.i.d . case even almost sure @xcite ) that trapping structures as in figure [ fig : trap ] appear , which immediately contradict a uniform poincar inequality . in addition to poincar and sobolev inequalities , our proofs rely on stochastic two - scale convergence , an analytic method that is based on the ergodic theorem and was introduced in @xcite . for the random conductance model with conductances @xmath11 $ ] ( @xmath12 ) restricted to nearest - neighbor connections , faggionato @xcite already used stochastic two - scale convergence in order to prove homogenization of the laplace operator with a spectral shift on the infinite connected component of @xmath0 . the spectral shift compensates for the lack of a poincar inequality . as a consequence of the homogenization of the poisson equation on bounded domains , the homogenization of the top of the dirichlet spectrum follows by the methods of ( * chapter 11 ) , see theorem [ thm : spectrum ] . for this result , remark [ rem : optimality ] explains in which sense condition is sharp . for i.i.d . conductances even decides between a completely homogenizing phase , which we cover in this paper , and a completely localizing phase of the principal dirichlet eigenvector , which was studied in @xcite . we thus extend the results of faggionato @xcite and boivin and depauw @xcite . faggionato showed spectral homogenization in dimension @xmath13 under condition and @xmath7<\infty$ ] , whereas boivin and depauw proved spectral homogenization for conductances that are uniformly bounded from above and away from zero ( uniform ellipticity ) . as an application of the spectral homogenization , we prove a quenched large deviation principle for the occupation time measures , given that the random walk stays in a slowly growing box , see proposition [ prop : ldp ] . we extend the results of ( * ? ? ? * theorem 1.8 ) , where the authors use the deep connection between the dirichlet energy of the laplace operator and the donsker - varadhan rate function of the occupation time measures of the associated random walk . in the recent paper @xcite , the authors prove that under the same integrability conditions as in the present paper , the dirichlet energy of the random conductance laplacian @xmath14-converges to a deterministic , homogeneous integral ( see their corollary 3.4 and proposition 3.18 ) . together with their compactness result ( * ? ? ? * lemma 3.9 ) , theorem 13.5 of @xcite implies the homogenization of the poisson equation for models where the connections are of finite range . with the method employed in the present paper , however , we can allow for unbounded - range connections ( see assumption [ ass : environ][ass : momcond ] ) . further , we identify the corresponding limit operator ( see theorem [ thm : twoscale - strong ] ) . let us consider a graph with vertex set @xmath0 and edge set @xmath15 i.e. , we assume that there exists an undirected edge between any two sites @xmath16 . we further assume that each edge @xmath3 carries a nonnegative random variable @xmath2 , which we call the _ conductance _ of the edge @xmath3 . moreover , we call the family @xmath17 _ environment _ or _ landscape_. if @xmath18 for @xmath19 ( @xmath20 ) , we will also write @xmath21 instead of @xmath2 . moreover , @xmath22 denotes the translations by a vector @xmath23 , i.e. , we write @xmath24 since the edges in are undirected , we have @xmath25 . in the variable - speed random conductance model , we consider the laplacian @xmath26 , which acts on real - valued functions @xmath27 as @xmath28 note that @xmath26 is @xmath29-a.s . well - defined under assumption [ ass : environ][ass : momcond ] . since the conductances are symmetric , i.e. , @xmath25 , the associated markov process is reversible . as we explain in section [ sec : discrder ] , the laplacian @xmath26 is the discrete analogue of a divergence - form operator with random weights . we denote the probability space that governs the environment @xmath30 by @xmath31^e , \mathcal{b } ( [ 0,\infty ] ) ^{\otimes e } , { \mathbb{p}}\right)\ , , \label{eq : def - prob - space}\end{aligned}\ ] ] and the expectation with respect to the law @xmath29 by @xmath5 . for any @xmath32 , we denote the set of _ open edges _ by @xmath33 further , @xmath34 denotes the set of all undirected nearest - neighbor bonds . in this paper we will usually assume that the law @xmath29 fulfills the following conditions . [ ass : environ ] the law @xmath29 is stationary and ergodic with respect to spatial translations @xmath35 . [ ass : momcond ] @xmath36 < \infty$ ] . [ ass : fulllattice ] for @xmath29-a.e . @xmath30 , the set @xmath37 of open edges contains the set @xmath38 of nearest - neighbor edges of @xmath0 . in addition to assumption [ ass : environ ] , our main results rely on an integrability condition for the lower tails of the conductances , for which we need to define the notion of paths in @xmath39 . a path of length @xmath40 between @xmath41 and @xmath42 in @xmath39 is a sequence @xmath43 with the property that @xmath44 , @xmath45 and @xmath46 for any @xmath47 . if @xmath48 is a path and there exists @xmath49 such that @xmath50 , then we use the shorthand notation @xmath51 . for any @xmath52 and @xmath53 , let @xmath54 be a collection of paths in @xmath39 between the vertices of the edge @xmath3 with length at most @xmath40 such that no two paths in @xmath54 share an edge . we define the measures @xmath55 and @xmath56 on @xmath0 by @xmath57 where @xmath58 we let @xmath59 denote the minimizer of the rhs of . [ ass : integ ] if @xmath13 , then @xmath60<\infty$ ] for any @xmath61 . in addition , if @xmath62 , then [ ass : integ : dhalf ] there exists @xmath63 and a path set @xmath54 such that @xmath64<\infty$ ] for any @xmath61 . there exists @xmath63 , a path set @xmath54 and @xmath65 such that @xmath66<\infty$ ] for any @xmath61 . r3.3 cm [ rem : iid ] generally , @xmath67<\infty$ ] is sufficient for assumption [ ass : integ][ass : integ : dhalf ] . this can even be improved if the conductances @xmath2 @xmath68 are independent and identically distributed ( i.i.d . ) and @xmath62 . for example , on the nearest - neighbor lattice @xmath69 with i.i.d . conductances , assumption [ ass : integ][ass : integ : dhalf ] holds if @xmath70<\infty$ ] for any edge @xmath61 . similarly , assumption [ ass : integ][ass : integ : q ] holds if there exists @xmath71 such that @xmath72<\infty$ ] for any edge @xmath61 . this follows because any two sites in @xmath73 are connected through @xmath74 independent nearest - neighbor paths ( see figure [ fig : indeppaths ] , cf . 2 ) , ( * ? ? ? * fig . 2.1 ) ) . if we added further links to the set @xmath38 , the number of independent paths between any two sites would increase whence the critical exponent @xmath75 would decrease . if we assumed that @xmath38 would contain all the links of @xmath76 , then it would even be sufficient to assume that there exists @xmath77 such that @xmath72<\infty$ ] . note that in order not to violate assumption [ ass : environ][ass : momcond ] , we would assume in this case that @xmath78 where the @xmath79 are i.i.d . , @xmath80 and @xmath81 is the euclidean length of the edge @xmath3 . we aim to consider the behavior of the operator @xmath26 in boxes of the form @xmath82 with zero dirichlet boundary conditions . more precisely , we fix an environment @xmath30 on the entire @xmath0 , let the box size @xmath83 grow to infinity and want to characterize the behavior of solutions to the poisson equation and the spectral problem . for this purpose we use analytic techniques as introduced in section [ sec : analytictools ] . regarding these techniques , it is more natural to replace the lattice @xmath0 by the rescaled lattice @xmath84 and the growing box @xmath85 by the box @xmath86 with @xmath87 and @xmath88 . in this context , the laplacian defined in corresponds to the accelerated operator @xmath89 which acts on real - valued functions @xmath90 as @xmath91\ , , \qquad \left ( x\in{\varepsilon}{{\mathbb{z}}^{d}}\right)\ , , \label{equ : defgeneratoreps}\end{aligned}\ ] ] where the conductances @xmath92 remain random variables associated with the links in the edge set @xmath93 , i.e. , the links between sites in @xmath73 . note that if @xmath26 is the generator of a markov process @xmath94 , then @xmath89 is the generator of the diffusively rescaled markov process @xmath95 , which fulfills @xmath96 . for @xmath97 and @xmath98 , we define the function spaces @xmath99 we abbreviate @xmath100 . analogously to @xmath101 , we introduce the hilbert spaces @xmath102 through @xmath103 and let @xmath104 and @xmath105 be equipped with the scalar products @xmath106 for @xmath107 , we let @xmath108 denote the half - open ball @xmath109^d$ ] . we define the local averaging operator @xmath110 acting on functions @xmath111 by @xmath112 a direct calculation shows that its adjoint operator @xmath113 is given by @xmath114 where we write @xmath115 for the characteristic function of @xmath116 . given a function @xmath117 , we are interested in the solution @xmath118 of the poisson problem @xmath119 with zero dirichlet conditions . the above problem has a unique solution because @xmath120 is invertible on @xmath105 . [ thm : twoscale - strong ] let @xmath121 be a sequence of functions such that @xmath122 weakly in @xmath123 for some @xmath124 . then for almost all @xmath32 the sequence of solutions @xmath125 to the problem satisfies @xmath126 strongly in @xmath123 , where @xmath127 solves the limit problem @xmath128 almost everywhere in @xmath129 with @xmath130 defined through . we prove this theorem at the end of section [ ref : proofpoisson ] . in lemma [ lem : non - degeneracy ] we prove that @xmath130 is strictly positive definite and by standard arguments @xmath130 is symmetric . r2.5 cm [ rem : otherlattices ] with our methods , theorem [ thm : twoscale - strong ] can be easily generalized for other lattices than @xmath0 . in order to apply our methods directly , we just have to require that the lattice is translationally invariant ( for the two - scale convergence , see section [ sec : two - scale ] ) and fulfills a sobolev inequality ( as in or ) with isoperimetric dimension @xmath131 ( to obtain the necessary poincar inequalities and make the moser iteration work ) . for example , the triangular lattice in figure [ fig : triangular ] is translationally invariant and has isoperimetric dimension @xmath132 . if we therefore replace @xmath0 by the triangular lattice and the dimension @xmath133 in assumption [ ass : integ ] by the isoperimetric dimension @xmath131 , theorem [ thm : twoscale - strong ] still holds . note that in view of remark [ rem : iid ] , we observe that in the case of i.i.d . conductances on the triangular lattice , assumption [ ass : integ][ass : integ : dhalf ] holds if @xmath134<\infty$ ] . [ rem : perc ] although we focus here on the random conductance model with long - range jumps and positive nearest - neighbor conductances , our arguments do not require the full strength of this assumption . for instance , we can also extend the homogenization result to the _ nearest - neighbor _ percolation case . more precisely , we can relax assumption [ ass : environ][ass : fulllattice ] such that the set of open edges @xmath135 forms a unique infinite cluster that satisfies both a volume regularity condition and a ( weak ) relative isoperimetric inequality on large scales , cf . notice that in the nearest - neighbor percolation setting , similar homogenization results have also been obtained by faggionato in @xcite under the additional assumption that the conductances are bounded from above . let us now consider the spectrum of the operators @xmath136 with an arbitrary bounded , continuous @xmath137 . on the domain @xmath138 with zero dirichlet conditions we can represent @xmath136 as a real symmetric matrix and therefore we can choose the set @xmath139 of dirichlet eigenvectors such that they form an orthonormal system . by virtue of the perron - frobenius theorem ( see e.g. ( * ? ? ? * chapter 1 ) ) the principal dirichlet eigenvalue @xmath140 is unique . thus , we now consider the problem @xmath141[1.5em][c]{0cm}{}}\\ & \lambda_1^{\varepsilon } < \lambda_2^{\varepsilon}\leq \ldots \leq \lambda_k^{\varepsilon}\ldots \ , , { \parbox[0pt][1.5em][c]{0cm}{}}\\ & \langle \psi_k^{\varepsilon } , \psi_l^{\varepsilon}\rangle_{{{\mathcal{h}_{\varepsilon } } } } = \delta_{kl}\ , .{\parbox[0pt][1.5em][c]{0 cm } { } } \label{equ : defeigveceps } \end{aligned}\end{aligned}\ ] ] similarly , we consider the spectrum of the operator @xmath142 , i.e. , @xmath143[1.5em][c]{0cm}{}}\\ & \lambda_1 ^ 0 < \lambda_2 ^ 0 \leq \ldots \leq \lambda_k^0\ldots \ , , { \parbox[0pt][1.5em][c]{0cm}{}}\\ & \langle \psi_k^0 , \psi_l^0 \rangle_{{{\mathcal{h}_{\varepsilon } } } } = \delta_{kl}\ , .{\parbox[0pt][1.5em][c]{0 cm } { } } \end{aligned}\end{aligned}\ ] ] in order to study the homogenization of , we need the following result . [ prop : twoscale - strong ] let @xmath121 be a sequence of functions such that @xmath122 weakly in @xmath123 for some @xmath124 . let @xmath137 be such that @xmath144 . then for almost all @xmath32 the sequence of solutions @xmath125 to the problem @xmath145 satisfies @xmath126 strongly in @xmath123 , where @xmath127 solves the limit problem @xmath146 almost everywhere in @xmath129 with @xmath130 defined through . we prove this proposition in section [ sec : proofspectrum ] . by virtue of ( * lemma 11.3 , theorem 11.5 ) , proposition [ prop : twoscale - strong ] implies the following result , see section [ sec : proofspectrum ] . [ thm : spectrum ] let @xmath147 . if assumptions [ ass : environ ] and [ ass : integ][ass : integ : q ] hold , then @xmath148 further , the following statements are true : 1 . [ cor : spectrum : subseq ] let @xmath147 and let @xmath149 be a null sequence . then there @xmath29-a.s . exists a family @xmath150 of eigenvectors of the operator @xmath151 and a subsequence , still indexed by @xmath152 , along which the vector @xmath153 2 . [ cor : spectrum : lincom ] on the other hand , if the multiplicity of @xmath154 is equal to @xmath155 , i.e. , @xmath156 then there @xmath29-a.s . exists a sequence @xmath157 such that @xmath158 where @xmath159 is a linear combination of the eigenfunctions of the operator @xmath160 corresponding to the eigenvalues @xmath161 . [ rem : optimality ] let us discuss in what sense assumption [ ass : integ ] is optimal for the result of theorem [ thm : spectrum ] with @xmath162 . since the principal dirichlet eigenvalue has the variational representation @xmath163 ( also known as the rayleigh - ritz formula , or the courant - fischer theorem ) , it is necessary for spectral homogenization that @xmath29-a.s . there exists @xmath164 such that @xmath165 and for all @xmath166 ( uniform poincar inequality ) . r3 cm if we assume that @xmath29-a.s . only nearest - neighbor connections carry a positive conductance , i.e. , @xmath167 , then assumption [ ass : integ ] is optimal for the uniform poincar inequality up to the critical case @xmath168<\infty \}=q_{\rm c}$ ] ( cf . ) . this means that if @xmath168<\infty \}<q_{\rm c}$ ] , then it is possible to construct an environment where the uniform poincar inequality does not hold as @xmath1 tends to zero . for @xmath62 , this is due to trapping structures as in figure [ fig : trap ] where @xmath169 can concentrate its entire mass , see e.g. ( * ? ? ? * section 1.4 ) . the construction of stationary , ergodic environments with such trapping structures is analogous to the one of a trap for the constant - speed random walk in ( * ? ? ? * theorem 5.4 ) . in the i.i.d . case and if @xmath168<\infty \}<1/4 $ ] , the traps occur even @xmath29-a.s . for @xmath1 small enough and the principal dirichlet eigenvector localizes @xmath29-a.s . in a single site ( * ? * theorem 1.8 ) . in @xmath13 and if @xmath168<\infty \}<1 $ ] , even an i.i.d . environment contradicts the uniform poincar inequality : by a borel cantelli argument we can show that @xmath29-a.s . for @xmath1 small enough there exist edges @xmath170 and @xmath171 such that @xmath172 and @xmath173 , respectively , and such that both @xmath174 and @xmath175 decay much faster than @xmath1 . when we insert a function @xmath125 into that is 1 on the interval @xmath176 $ ] and zero otherwise , then we see that @xmath177 diverges as @xmath1 tends to zero , which is a contradiction to a uniform poincar inequality . if we assume that @xmath37 is @xmath29-a.s . strictly larger that @xmath38 but contains only connections of bounded length , an analogous construction as in ( * ? ? ? * theorem 5.4 ) shows that @xmath178 is still optimal in the general stationary ergodic case with @xmath62 . for independent conductances however , @xmath75 decreases when the upper bound for the length of the connections increases , see also remark [ rem : iid ] . on the other hand , if we assume that @xmath37 contains connections of unbounded length , all the suggested counterexamples fail and the question about the optimal conditions requires further research . for a fixed realization @xmath30 of the environment , we consider the markov process @xmath179 on @xmath73 , which jumps with rate @xmath21 from a site @xmath41 to the site @xmath180 . since the holding times are site dependent , this markov process is called the variable - speed random walk among random conductances ( see @xcite for a review ) . its generator @xmath26 is given by . our main motivation for this paper is to prove a quenched large deviation principle ( ldp ) for the _ occupation time measures _ or _ local times _ @xmath181 of the random walk among random conductances , given that the random walk stays in a certain growing region of the lattice . more precisely , we define a spatial scaling @xmath182 with @xmath183 and consider the rescaled local times @xmath184 further , let @xmath87 and define @xmath185 . in ( * ? ? ? * theorem 1.8 ) , the authors prove a quenched large deviation principle for the function @xmath186 given that @xmath187 and under the assumption that the conductances are i.i.d . and uniformly elliptic . our aim is to generalize this result to stationary and ergodic conductances and replace the uniform ellipticity condition by a suitable moment condition . let us recall some facts about the local times of the simple random walk . we define the set @xmath188 and equip @xmath189 with the weak topology of integrals against bounded continuous functions @xmath190 . notice that on the event @xmath191 the function @xmath186 is an element of the set @xmath189 and an @xmath192-normalized random step function on @xmath193 . in the case of a simple random walk , i.e. , when @xmath194 , it is known that on the event @xmath191 the function @xmath186 satisfies a large deviation principle on @xmath189 with scale @xmath195 and rate function @xmath196 , where @xmath197 see @xcite for further explanation and @xcite . we prove that under quite general conditions , this is also true for the random conductance model , see proposition [ prop : ldp ] and corollary [ cor : ldp ] . for general stationary and ergodic conductances , however , the resulting rate function reads @xmath198 and the matrix @xmath199 is defined as in . [ ass : heatkernel ] there exists @xmath200 such that @xmath29-a.s . for @xmath201 large enough @xmath202 \;\geq\ ; ct^{-d/2 } \end{aligned}\ ] ] for all @xmath203 with @xmath204 . [ prop : ldp ] let assumptions [ ass : environ ] , [ ass : integ][ass : integ : q ] and [ ass : heatkernel ] be fulfilled . then @xmath29-a.s . the rescaled local times @xmath186 satisfy a large deviation principle with respect to the weak topology of integrals against bounded continuous functions @xmath190 under @xmath205 $ ] on @xmath189 . the scale is @xmath195 and the rate function @xmath206 is defined in . we prove this proposition in section [ sec : proofpropldp ] as a consequence of theorem [ thm : spectrum ] . in the special case where only nearest - neighbor conductances are positive , proposition [ prop : ldp ] together with the heat kernel bounds of ( * ? ? * proposition 4.7 ) respectively , implies the following corollary . [ cor : ldp ] let the conductances be stationary and ergodic with law @xmath29 and let @xmath29-a.s . @xmath207 . for @xmath208 $ ] satisfying @xmath209 assume that @xmath210 < \infty$ ] and @xmath211 < \infty$ ] for any @xmath52 . then the large deviation principle from proposition [ prop : ldp ] holds . in analogy to the definition of @xmath101 in , we define the following space - averaged norms for functions @xmath212 . let @xmath213 be a non - empty set and @xmath214 . then @xmath215 moreover , we let @xmath216 abbreviate the average of @xmath217 over the set @xmath218 . the main objective in this subsection is to prove weighted poincar and sobolev inequalities . the poincar inequalities of proposition [ prop : localpoincare ] and are the main tools in the proof of lemma [ lem : precompact ] , whereas the sobolev inequality of proposition [ prop : sobolev ] with @xmath219 ensures uniform @xmath220-bounds of the solution to the poisson equation ( see section [ subsec : maximalinequalities ] ) . starting point for our further considerations is the fact that the underlying unweighted euclidean lattice @xmath39 satisfies the classical sobolev inequality for any @xmath221 . let @xmath222 be finite and connected and @xmath223 . then , @xmath224 for @xmath13 , whereas for any @xmath225 and @xmath226 we have @xmath227 for @xmath225 this sobolev inequality follows from the isoperimetric inequality of the underlying euclidean lattice , see e.g. ( * ? ? ? * theorem 3.2.7 ) . [ prop : localpoincare ] for any @xmath228 and @xmath229 , let @xmath230 . suppose that @xmath13 . then , there exists @xmath231 such that @xmath232 for any @xmath233 . furthermore , for every @xmath62 and @xmath234 there exist constants @xmath235 and @xmath236 with @xmath237 such that @xmath238 & \mspace{36mu}\leq\ ; c_{\mathrm{pi}}\ , { \ensuremath { \mathchoice{\big\lvert \nu_l^{{\omega } } \big\rvert } { \lvert \nu_l^{{\omega } } \rvert } { \lvert \nu_l^{{\omega } } \rvert } { \lvert \nu_l^{{\omega } } \rvert}_{\raisebox{-.0ex}{$\scriptstyle \frac{d}{2 } , b(n)$ } } } } \ , \frac{n^2}{|b(n)|}\mspace{-3mu } \sum_{x , y \in b(c_{\mathrm{w } } n ) } \mspace{-15mu } { \omega}(\{x , y\})\ , \big| u(x ) - u(y ) \big|^2 , \end{aligned}\ ] ] for any @xmath239 , where the measure @xmath240 is given by with suitable path sets @xmath241 . as in ( * ? ? ? * proposition 2.1 or 6.1 ) , the assertion is an immediate consequence of and hlder s inequality ( see also ( * ? ? ? * lemma 2.3 ) ) . nevertheless , we will repeat the argument here for the reader s convenience . since @xmath242 , the assertion follows from by an application of the cauchy - schwarz inequality . let us now consider , i.e. , the case @xmath225 . for @xmath243 we let @xmath244 denote the difference @xmath245 . for any @xmath61 we observe that by the cauchy - schwarz inequality @xmath246 where we recall the definitions of @xmath247 and @xmath248 in and below . thus , for any @xmath249 , hlder s inequality yields @xmath250 where @xmath251 note that there exists @xmath252 such that @xmath253 for any @xmath254 . in addition , there exists @xmath255 such that @xmath256 if @xmath257 . thus , when we choose @xmath258 , then follows from . we define @xmath259 our next task is to establish the corresponding versions of and on the weighted graph @xmath260 . for this purpose , for @xmath225 and @xmath261 we define @xmath262 notice that @xmath263 is monotonically increasing in @xmath264 and converges to @xmath265 as @xmath264 tends to infinity . moreover , @xmath266 . [ prop : sobolev ] let @xmath228 and @xmath267 . suppose that @xmath13 . then there exists @xmath268 such that @xmath269 for any @xmath233 with @xmath270 . furthermore , for every @xmath62 , @xmath234 and @xmath271 there exists @xmath272 such that @xmath273 for any @xmath239 with @xmath270 , where the measure @xmath240 is given by with suitable path sets @xmath241 . in analogy to , we define @xmath274 we prove proposition [ prop : sobolev ] after the following remark . for @xmath13 , proposition [ prop : sobolev ] implies that @xmath275 for @xmath62 , proposition [ prop : sobolev ] implies that @xmath276 when we insert @xmath277 into , we especially obtain that @xmath278 under assumption [ ass : integ][ass : integ : dhalf ] and by virtue of the ergodic theorem , and imply that for @xmath279 there exists a @xmath29-a.s . finite @xmath280 such that for all @xmath166 and all @xmath125 we have @xmath281 in the sequel we will give a proof only for . the assertion follows by similar arguments . to lighten notation , set @xmath282 and define @xmath283 . the constant @xmath284 appearing in the computations below is independent of @xmath285 but may change from line to line . let @xmath286 and @xmath226 . since @xmath287 for @xmath288 , we have @xmath289 hence , an application of minkowski s inequality yields @xmath290 thus , for any @xmath261 the assertion follows as in the previous proof from combined with by choosing @xmath291 . [ prop : linftypoiss ] let @xmath292 and suppose that @xmath293 is a solution of . for some fixed @xmath294 consider the measure @xmath295 on @xmath0 as defined in . then , for any @xmath296 there exist @xmath297 $ ] , @xmath298 , and @xmath299 such that @xmath300 we prove this proposition after the following remark . [ rem : linftypoiss ] note that if @xmath293 is a solution of , then due to , and the cauchy - schwarz inequality it follows for any dimension @xmath279 that @xmath301 let assumption [ ass : integ][ass : integ : dhalf ] be fulfilled . then @xmath302 implies by the ergodic theorem that both @xmath303 and @xmath304 are bounded as well . thus , implies that in dimension one @xmath305 is bounded . furthermore , if even assumption [ ass : integ][ass : integ : q ] is fulfilled and @xmath306 , then implies that @xmath305 is bounded for @xmath62 as well . we use the moser iteration scheme . let us fix @xmath166 and consider @xmath307 with @xmath308 . we define @xmath309 for any @xmath310 . by virtue of eq . ( a.2 ) in @xcite we obtain the following energy estimate @xmath311 since @xmath169 is a solution to the poisson equation , the energy estimate implies that @xmath312 by the sobolev inequality and jensen s inequality it follows that @xmath313 we define @xmath314 for @xmath315 . further , we set @xmath316 for @xmath317 and @xmath318 for @xmath319 . recall that @xmath320 for any @xmath65 . furthermore , we observe that for any @xmath321 we have @xmath322 . thus , by iterating the inequality and using the fact that @xmath323 , we obtain that there exists @xmath324 such that @xmath325 where @xmath326 and @xmath327 such that @xmath328 . choosing @xmath329 , we complete the proof . the very first step to prove homogenization of the operator @xmath89 is to show that a sequence @xmath330 ( @xmath331 ) has a strongly convergent subsequence if @xmath332 . the dirichlet energy @xmath333 is defined in . [ lem : precompact ] if assumptions [ ass : environ ] and [ ass : integ][ass : integ : dhalf ] are fulfilled , then for @xmath29-a.e . realization @xmath32 the following statement is true : for any sequence @xmath331 ( @xmath334 , @xmath335 ) with @xmath336 , the sequence @xmath337 has a strongly convergent subsequence in @xmath338 . this lemma was also recently shown in ( * ? ? ? * lemma 3.9 ) . first of all we observe that by virtue of , there exists a @xmath29-a.s . finite @xmath280 such that @xmath339 which implies that @xmath340 is finite by assumption . by the banach - alaoglu theorem it follows that there exists a subsequence , which we still index by @xmath1 , and @xmath341 such that @xmath342 for any @xmath343 consider a partition of @xmath129 into @xmath344 congruent open subcubes @xmath345 with side length @xmath346 . we further define @xmath347 . then @xmath348 and @xmath349 as @xmath350 . we estimate @xmath351 where , in analogy to , we abbreviate @xmath352 since @xmath330 converges weakly in @xmath353 to @xmath354 , the sum over the second term on the rhs of vanishes as @xmath1 tends to zero . it remains to show that , as the @xmath355 , the limit superior of the sum of the first and third term is zero as well . we use arguments similar to the ones given in ( * ? ? ? * proposition 2.9 ) , see also ( * ? ? ? * lemma 3.9 ) . let @xmath356 ( @xmath357 ) be the unit base vectors of @xmath193 . by virtue of proposition [ prop : localpoincare ] there exists @xmath358 such that @xmath29-a.s . for @xmath1 small enough the first term in the brackets of the rhs in can be estimated by @xmath359 where for @xmath13 we set @xmath360 . for @xmath62 we set @xmath277 . since any edge @xmath61 is contained in at most @xmath361 cubes @xmath362 , summing over @xmath363 yields @xmath364 note that @xmath365 is independent of @xmath327 and @xmath366 is bounded in @xmath1 by assumption . by assumption [ ass : integ][ass : integ : dhalf ] and the ergodic theorem , there @xmath29-a.s . exists @xmath367 such that for all @xmath368 we have @xmath369^{1/q}\ , . \end{aligned}\ ] ] from and it follows that there exists @xmath164 independent of @xmath327 such that @xmath29-almost surely @xmath370 since @xmath327 might be arbitrarily small and @xmath371 has bounded support , the claim follows . in what follows , we will generalize a result by boivin and depauw . [ thm : boiv - dep ] for every @xmath372 , for @xmath29-almost every @xmath32 it holds @xmath373\int_{q}v(x){\,\mathrm{d}}x\qquad\forall v\in c(\overline{q})\ , , \label{equ : boiv - dep}\end{aligned}\ ] ] and the null - set depends on @xmath217 but not on @xmath374 . [ rem : tempelman]evidently , we can also choose @xmath374 as the characteristic function of any relatively open or compact set @xmath375 and we obtain the tempelman ergodic theorem . we will use both theorem [ thm : boiv - dep ] and remark [ rem : tempelman ] in order to prove the following theorem . [ thm : general - ergodic - thm]for every @xmath372 , for @xmath29-almost every @xmath32 the following holds : let @xmath376 be a sequence of functions from @xmath377 with support in @xmath138 such that @xmath126 pointwise a.e . in @xmath129 . furthermore , let @xmath378 . then @xmath379 and @xmath380\int_{q}u(x){\,\mathrm{d}}x \label{eq : thm : general - ergodic - thm}\ ] ] and the null - set depends on @xmath217 but not on the sequence @xmath169 . first we note that @xmath379 since @xmath378 . now we let @xmath381 and let @xmath382 be a sequence of mollifiers approximating the identity . by egorov s theorem , there exists a compact set @xmath383 with @xmath384 such that both @xmath126 and @xmath385 uniformly on @xmath383 . we now make the following decomposition : @xmath386\int_{q}u(x){\,\mathrm{d}}x\right|\nonumber\\ & \qquad\qquad\leq \left| { \varepsilon}^d \sum_{x\in { \varepsilon}^{-1 } q_{\varepsilon } } \left ( { u^{\varepsilon}}({\varepsilon}x ) - u_\delta ( { \varepsilon}x)\right ) f\left(\tau_x \omega\right ) \right| + \nonumber\\ & \qquad\qquad\phantom{\leq}+\left| { \varepsilon}^d \sum_{x\in { \varepsilon}^{-1 } q_{\varepsilon } } u_\delta ( { \varepsilon}x ) f\left(\tau_x \omega\right ) - \mathbb{e}[f]\int_{q}u_\delta ( x){\,\mathrm{d}}x\right| + \left| \mathbb{e}[f]\int_{q}\left ( u_\delta ( x ) - u(x ) \right ) { \,\mathrm{d}}x\right| \label{equ : pfgenergodichelp2}\end{aligned}\ ] ] since @xmath387 , the second summand on the above rhs converges to zero by virtue of theorem [ thm : boiv - dep ] . for the first summand on the rhs of we estimate that @xmath388 since the function @xmath389 converges uniformly in @xmath1 to @xmath354 on @xmath383 , we can estimate by virtue of remark [ rem : tempelman ] that @xmath390\ , .\end{aligned}\ ] ] we further estimate the second summand on the rhs of by @xmath391\,,\end{aligned}\ ] ] where we have used remark [ rem : tempelman ] . thus , as @xmath355 , we obtain that @xmath392\int_{q}u(x){\,\mathrm{d}}x\right|\\ & \qquad\quad\leq \sup_{x\in k_{\eta}}\left|u_{\delta}(x)-u(x)\right|\,|q|\,{\mathbb{e}}[f ] \;+\ ; 2\eta \sup_{{\varepsilon}>0}\left\vert { u^{\varepsilon}}\right\vert _ { \infty } { \mathbb{e}}[f ] \;+\ ; \left| \mathbb{e}[f]\int_{q}\left ( u_\delta ( x ) - u(x ) \right ) { \,\mathrm{d}}x\right|\end{aligned}\ ] ] as @xmath393 , the uniform convergence @xmath394 on @xmath395 yields @xmath396\int_{q}u(x){\,\mathrm{d}}x\;\right| \;\leq\ ; 2\eta\ , \sup_{{\varepsilon}>0}\left\vert { u^{\varepsilon}}\right\vert _ { \infty}{\mathbb{e}}[f]\,.\end{aligned}\ ] ] since the last inequality holds for every @xmath381 , the claim follows . we first note that the probability space given in is generated from the compact metric space @xmath397^{e}$ ] , and therefore the notion of continuity on @xmath398 makes sense . we say that a function @xmath399 is _ shift covariant _ if it fulfills @xmath400 for all @xmath401 ( cf . @xcite eq . ( 3.14 ) ) . note that shift covariant functions @xmath402 fulfill @xmath403 . then directly implies that @xmath404 we define on @xmath405 the space @xmath406\,.\end{aligned}\ ] ] accordingly , we define the scalar product between @xmath407 by @xmath408\ , .\label{equ : scalarprodcov}\end{aligned}\ ] ] note that @xmath409 is a closed subspace of @xmath410 , where @xmath411 is the measure on @xmath398 defined by @xmath412 . since @xmath398 is a compact metric space , @xmath413 is separable for all @xmath414 and thus also the countable product space @xmath410 and its subspace @xmath409 are separable . further , we note that for all @xmath415 it holds that @xmath416 satisfies @xmath417 . therefore @xmath418 is in @xmath419 . a _ local function _ on @xmath398 is a bounded , continuous function that only depends on finitely many coordinates of @xmath397^{e}$ ] . following the outline of chapter 3 in @xcite , we define the closed subspace @xmath420 let @xmath421 be the orthogonal complement of @xmath422 in @xmath419 and let us define @xmath423 note that since @xmath424 satisfies , the last equation also reads @xmath425 then we have the following lemma . [ lem : divsol ] @xmath426 using the above notation , we define @xmath427 through @xmath428\,:\,\tilde{\chi}\in\left(l_{\rm pot}^2\right)^{d}\right\ } \,,\label{eq : definition - corrector}\ ] ] i.e. , @xmath429 is the orthogonal projection of @xmath430 on the space @xmath431 with respect to the scalar product defined in . we will see below that we can write the homogenized matrix as @xmath432\right)\left({\hat{e}}_{j } \cdot\left[z+\chi({\omega},z)\right]\right ) \right]\,,\label{eq : definition - a - hom}\ ] ] where the @xmath356 , @xmath357 , denote the unit base vectors of @xmath193 . in analogy to ( * ? ? ? * lemma 4.5 ) we know the following result . [ lem : non - degeneracy ] suppose that @xmath433 < \infty$ ] with @xmath56 as defined in . then the matrix @xmath434 is positive definite . in particular , the vectorial space spanned by the following vectors @xmath435 \in { { \mathbb{r}}^{d } } , \qquad b \in l^2_{\mathrm{sol } } \label{eq : non - degeneracy - consequence}\end{aligned}\ ] ] coincides with @xmath193 . first we notice that @xmath436 for any @xmath437 and @xmath357 , provided that @xmath438 < \infty$ ] . indeed , by the cauchy - schwarz inequality and the shift covariance , we observe that @xmath439 & \;\leq\ ; { \mathbb{e}}\big[1/{\omega}_l(0 , { \hat{e}}_i)\big]^{1/2}\ , \biggl({\mathbb{e}}\biggl[\sum_{x , y\in\gamma_l^{\rm opt } } { \omega}(\{x , y\ } ) |\psi ( \tau_x { \omega } , x - y)|^2\biggr]\biggr)^{1/2 } \nonumber\\ & \;\leq\ ; \sqrt{l\;|\gamma_l|}\ , { \mathbb{e}}\big[\nu_l^{{\omega}}(0)\big]^{1/2}\ , { \ensuremath { \mathchoice{\big\lvert \psi \big\rvert } { \lvert \psi \rvert } { \lvert \psi \rvert } { \lvert \psi \rvert}_{\raisebox{-.0ex}{$\scriptstyle l_\mathrm{cov}^2 $ } } } } \ , , \end{aligned}\ ] ] where we abbreviate @xmath440 , recall . moreover , by adapting the argument given in ( * ? ? ? * proof of lemma 4.8 ) , it follows that @xmath441 = 0 $ ] for any @xmath442 and @xmath357 . in particular , @xmath443 = 0 $ ] for any @xmath444 . now let @xmath445 . since @xmath446 = 0 $ ] , it follows that @xmath447\bigr)\big ] \\[.5ex ] & \overset{\eqref{eq : l1:est}}{\;\leq\ ; } |v \cdot { \hat{e}}_i|\ , \sqrt{l\;|\gamma_l|}\ , { \mathbb{e}}\big[\nu_l^{{\omega}}(0)\big]^{1/2}\ , { \mathbb{e}}\,\biggl [ \sum_{z\in{{\mathbb{z}}^{d } } } { \omega}_{0,z } \bigl ( v \cdot [ z + \chi({\omega},z)]^2\bigr ) \biggr]\ , . \end{aligned}\ ] ] thus , by summing both sides over @xmath357 , we obtain @xmath448 ^ 2\bigr ) \biggr ] \;\geq\ ; \frac{|v|_2 ^ 2}{|v|_1}\ , \,\big(l\,|\gamma_l|\ , { \mathbb{e}}\big[\nu_l^{{\omega}}(0)\big]\big)^{-1/2 } \;>\ ; 0\ , . \end{aligned}\ ] ] thus , the matrix @xmath434 is positive definite . by following literally the proof of ( * lemma 4.5 ) we obtain the claim . [ [ bochner - spaces ] ] bochner spaces + + + + + + + + + + + + + + we will use the concept of bochner spaces , which are a special case of the theory outlined in @xcite . let @xmath449 be a normed space with norm @xmath450 with the corresponding topology and borel-@xmath451-algebra and let @xmath452 be a lebesgue - measurable set . then , for @xmath453 , we define the space @xmath454 given a measure space @xmath455 , it turns out that @xmath456 and @xmath457 are isometrically isomorph via the trivial identification @xmath458 . here , @xmath459 denotes the lebesgue measure and @xmath460 denotes the product measure . while not being necessary , this notation has proved useful in homogenization theory since the introduction of two - scale convergence in @xcite . in particular , it gives a clear and intuitive meaning to spaces such as @xmath461 or @xmath462 . if @xmath463 is a family of vectors in @xmath449 , we denote @xmath464 if @xmath465 is a countable dense subset of @xmath449 , i.e. @xmath449 is separable , every element of @xmath466 can be approximated by finite sums of elements of @xmath467 @xcite . with the following definitions of discrete derivatives , we can write the operator @xmath89 in divergence form . for @xmath468 we define the @xmath1-forward derivative in the direction @xmath469 by @xmath470 and the analogous backward derivative , @xmath471 further , we define @xmath472 and write @xmath473 for the function that maps @xmath414 to @xmath474 . accordingly , we define @xmath475 and @xmath476 . moreover , for a function @xmath477 we define @xmath478 we use this notation to clearly distinguish between @xmath479 , an operator on discrete functions , and @xmath480 , an operator on the sobolev space @xmath481 . a direct calculation shows that when @xmath482 maps @xmath483 , then @xmath484 moreover , for @xmath485 we observe that @xmath486 when we compare the divergence form of the operator @xmath89 in with the limit operator in , we better understand the result of theorem [ thm : twoscale - strong ] . furthermore implies that @xmath89 is strictly positive definite on any bounded domain with zero dirichlet conditions at the boundary . we adapt the concept of stochastic two - scale convergence by zhikov and piatnitsky @xcite to our setting . we denote by @xmath487 the function that maps @xmath414 onto its @xmath488th coordinate and observe that , since @xmath489 $ ] is finite , @xmath490 for @xmath491 . since @xmath409 is separable , there exist countable sets @xmath492 and @xmath493 such that @xmath494 is dense in @xmath409 . we can assume that every @xmath495 is the gradient of a local function . furthermore , there exists a countable subspace @xmath496 such that @xmath497 is dense both in @xmath498 and in @xmath499 . we then find that @xmath500 is dense in @xmath501 . we denote by @xmath502 the set of all @xmath32 such that theorem [ thm : boiv - dep ] holds 1 . for all @xmath503 , where @xmath504 , 2 . for all @xmath505 , where @xmath506 , and 3 . and for all @xmath507 , where @xmath508 is a finite subset of @xmath0 , 4 . @xmath509 for all @xmath510 and all @xmath511 . we call @xmath512 the set of _ typical realizations_. note that @xmath513 ( compare to ( * ? ? ? * lemma 4.4 ) ) . let @xmath514 . we say that @xmath515 _ converges weakly in two scales _ to @xmath516 if @xmath517{\,\mathrm{d}}x\end{aligned}\ ] ] for all @xmath518 and all @xmath504 . in this case we write @xmath519 . [ prop : ex - two - scale - limit - farfield]for all typical realizations @xmath520 it holds : if @xmath514 and @xmath164 are such that @xmath521 then there exists a subsequence @xmath522 and @xmath516 such that @xmath523 the proof goes along the lines of classical proofs of two - scale convergence like for example in @xcite , section 5 . we observe that for every @xmath524 and @xmath504 we find @xmath525 where , in the last step , we have also used that @xmath374 has bounded support . it follows that since @xmath497 and @xmath526 are countable , we can choose a subsequence @xmath527 as @xmath528 such that the limit @xmath529 of @xmath530 exists for every @xmath524 and @xmath504 . we notice that the functional @xmath531 is linear in @xmath532 . furthermore , due to ( [ eq : two - scale - exist - estim-1 ] ) , @xmath531 is continuous on @xmath533 . it follows by riesz representation theorem that we can find @xmath534 such that @xmath535\ , { \,\mathrm{d}}x\,.\ ] ] since @xmath500 is dense in @xmath536 , we obtain that @xmath537 is uniquely defined . since , in addition , @xmath497 is dense in @xmath538 , holds for all @xmath539 . [ lem : bdddiscrgrad ] for all typical realizations @xmath540 and all lipschitz functions @xmath541 there exists @xmath542 $ ] , which depends only on @xmath543 and @xmath30 , such that @xmath544 if @xmath543 is bounded , then @xmath280 is @xmath29-a.s . finite . we observe that we can interchange the order of the sums and estimate @xmath545 the first term on the above rhs is finite by virtue of the ergodic theorem . this also holds for the second term after an index shift in @xmath41 and a rearrangement of the two sums . [ lem : testwithgradient ] let @xmath546 . for all typical realizations @xmath520 it holds : @xmath547 let @xmath548 . since @xmath549<\infty$ ] , we can choose a finite point - symmetric subset @xmath550 such that @xmath551 \;<\;\delta\ , .\end{aligned}\ ] ] then we split the sum in into a sum over @xmath552 and a sum over @xmath553 . for @xmath552 we observe that , since @xmath554 , we have @xmath555 uniformly in @xmath511 . further , we observe that there exists @xmath556 such that for @xmath552 and for all @xmath557 , the statement @xmath558 implies that @xmath559 . it follows that for @xmath1 small enough , we have @xmath560 this together with and the ergodic theorem implies that @xmath561 let us now consider the case @xmath553 . as in the proof of lemma [ lem : bdddiscrgrad ] , we interchange the sums and observe that @xmath562 by the ergodic theorem and the choice of @xmath563 it follows that the limit superior of the above rhs is bounded from above by a constant times @xmath564 , which we can choose arbitrarily small . [ cor : testwithgradient ] for all typical realizations @xmath520 it holds : if @xmath565 , then @xmath566\ , { \,\mathrm{d}}x \label{eq : two - sc - conv-2}\end{aligned}\ ] ] for all @xmath567 . first we observe that @xmath568 for @xmath491 and @xmath569 where the @xmath356 , @xmath357 , denote the unit base vectors of @xmath193 . therefore the assumption that @xmath565 implies that @xmath570{\,\mathrm{d}}x\,,\end{aligned}\ ] ] where the @xmath356 , @xmath357 , denote the unit base vectors of @xmath193 . it follows that @xmath571{\,\mathrm{d}}x\,\end{aligned}\ ] ] for all @xmath554 . in order to prove ( [ eq : two - sc - conv-2 ] ) , it thus remains to show that @xmath572 this follows from cauchy - schwarz , i.e. , @xmath573 the first factor on the rhs is bounded by assumption and the second factor converges to zero by virtue of lemma [ lem : testwithgradient ] . let us start with the following auxiliary lemma [ lem : bsolorthgrad ] for all @xmath574 and all @xmath575 the following is true : @xmath576 we write the lhs of as @xmath577 the second term is immediately zero since @xmath574 and @xmath578 by . the first term is absolutely convergent and thus we can interchange the sums . after an additional index shift @xmath579 , we obtain that the above first term is equal to @xmath580 where we have used as well as the symmetry of the conductances , i.e. , @xmath581 . the claim follows from and @xmath575 since @xmath574 . we can now prove the convergence of gradients . [ lem : two - scale - gradient ] for all @xmath520 the following holds . if @xmath582 is a family of functions with @xmath583 for all @xmath1 and @xmath584 then there exists a subsequence @xmath585 , @xmath586 and @xmath587 such that @xmath588 condition together with lemma [ lem : precompact ] implies that there exists a subsequence , which we still index by @xmath1 , and @xmath589 such that @xmath590 in @xmath591 . it remains to show that @xmath592 and to proof the second statement in . by virtue of proposition [ prop : ex - two - scale - limit - farfield ] , condition further implies that there exists a subsequence , which we still index by @xmath355 , and @xmath516 such that @xmath593 in the two - scale sense . we choose @xmath575 and @xmath554 and apply to the discrete product rule @xmath594 to obtain that @xmath595 for the first term on the rhs of , we obtain from the two - scale convergence of @xmath596 that @xmath597{\,\mathrm{d}}x\ , . \label{eq : wo - scale - gradient - help-2}\end{aligned}\ ] ] for the second term on the rhs of , we first notice that the sum is absolutely convergent since @xmath598 where for the last equality we have used the relation @xmath599 , the shift covariance and the substitution @xmath600 . we now use the fact that @xmath169 and @xmath601 are bounded , apply the cauchy - schwarz inequality and the ergodic theorem to obtain that the above sum is indeed finite . it follows that for the second term on the rhs of , we can exchange the order of the sums . by the same arguments as those that led to , we obtain that @xmath602 further we notice that since @xmath169 has support only in @xmath603 , we can estimate @xmath604 the limit superior of the second factor vanishes due to lemma [ lem : testwithgradient ] and the third factor is finite due to the ergodic theorem . thus , for @xmath605 we have @xmath606 to summarize , for the second term on the rhs of , it follows that @xmath607 by the assumptions on @xmath608 and @xmath424 , the last bracket on the above rhs is in @xmath609 . since we already know that the subsequence @xmath590 in @xmath610 , there exists a further subsequence , which we still index by @xmath350 , where @xmath611 converges pointwise a.e . in @xmath129 ( * theorem 4.9 ) . moreover , @xmath169 has support in @xmath138 and @xmath612 by assumption . it follows that we can apply theorem [ thm : general - ergodic - thm ] along the above subsequence and obtain that @xmath613 { \,\mathrm{d}}x\ , .\end{aligned}\ ] ] thus we obtain by that @xmath614{\,\mathrm{d}}x = -\int_{{{\mathbb{r}}^{d } } } u(x ) \nabla v(x)\cdot { \mathbb{e}}\left [ \sum_{z\in{{\mathbb{z}}^{d}}}z{\omega}_{0,z } b(\omega , z ) \right ] { \,\mathrm{d}}x\ , . \label{eq : wo - scale - gradient - help-3}\end{aligned}\ ] ] let us now argue that implies that @xmath615 . by virtue of , for any @xmath616 we can choose @xmath617 such that @xmath618= { \hat{e}}_i$ ] . then implies that @xmath619{\,\mathrm{d}}x\right|\\ & \leq \| b^i \|_{l^2_{\rm cov}}\int_{{{\mathbb{r}}^{d } } } \left|v(x)\right| \left({\mathbb{e}}\left[\sum_{z\in{{\mathbb{z}}^{d}}}{\omega}_{0,z } { { \rm w}}^2(x,{\omega},z)\right]\right)^{1/2 } { \,\mathrm{d}}x\\ & \leq \|v\|_2 \| b^i \|_{l^2_{\rm cov } } \left(\int_{{{\mathbb{r}}^{d } } } { \mathbb{e}}\left[\sum_{z\in{{\mathbb{z}}^{d}}}{\omega}_{0,z } { { \rm w}}^2(x,{\omega},z)\right ] { \,\mathrm{d}}x\right)^{1/2}\ , .\end{aligned}\ ] ] since @xmath620 , there exists @xmath164 such that for any @xmath616 we observe that @xmath621 by virtue of ( * ? ? ? * proposition 9.3 ) it follows that @xmath622 . since @xmath623 , we conclude @xmath586 . we now use integration by parts on the rhs of and obtain that @xmath624{\,\mathrm{d}}x=0\ , . \ ] ] since the last equation holds for all @xmath524 and all @xmath575 , we find that @xmath625 we start with an auxiliary lemma . [ lem : twoscale - weak ] let @xmath626 be a sequence of functions such that @xmath122 weakly in @xmath123 for some @xmath124 and such that @xmath627 . then for almost all @xmath32 it holds : the sequence of solutions @xmath125 to the problem satisfies @xmath126 strongly in @xmath123 , where @xmath127 solves the limit problem . we test equation with an arbitrary test function @xmath628 with @xmath629 and obtain by that @xmath630 we now choose @xmath631 and apply and cauchy - schwarz to obtain that @xmath632 hence , in combination with remark [ rem : linftypoiss ] we conclude that [ equ : pfthm15bddness ] @xmath633 @xmath634 it follows that by virtue of lemma [ lem : precompact ] and lemma [ lem : two - scale - gradient ] , there exists @xmath586 , @xmath635 and a subsequence , which we still index by @xmath1 , such that @xmath636 for all @xmath637 and @xmath638 . let us choose @xmath554 with @xmath639 and @xmath640 with @xmath641 for some bounded local function @xmath642 . when we insert @xmath643 into , then we observe for all @xmath166 that @xmath644\nonumber\\ & = { \varepsilon}^d \sum_{x\in{{\mathbb{z}}^{d } } } \sum_{z\in { \mathbb{z}}^d } \omega_{x , z } \partial_{z}^{{\varepsilon}}{u^{\varepsilon}}({\varepsilon}x)\ , v({\varepsilon}x)\varphi ( \tau_{x}\omega , z ) \nonumber\\ & \qquad\qquad+{\varepsilon}^d \sum_{x\in{{\mathbb{z}}^{d } } } \sum_{z\in { \mathbb{z}}^d } \omega_{x , z } \partial_{z}^{{\varepsilon}}{u^{\varepsilon}}({\varepsilon}x)\ , { \varepsilon}\tilde{\varphi } ( \tau_{x+z}\omega ) { \partial}_z^{\varepsilon}v({\varepsilon}x)\ , . \label{equ : pfthm15help2}\end{aligned}\ ] ] the second summand on the above rhs vanishes as @xmath350 since @xmath645 by assumption @xmath646 is bounded . the second factor is bounded due to and the third factor is bounded by virtue of lemma [ lem : bdddiscrgrad ] . since the lhs of vanishes as well , and imply that in the limit @xmath350 and along the chosen subsequence we obtain @xmath647 = 0\,.\label{eq : thm - two - scale - lim - help - split-2 } \ ] ] since @xmath648 is dense in @xmath431 and @xmath497 is dense in @xmath649 , equation holds for all @xmath650 and all @xmath651 . let @xmath652 be given through . since @xmath653 is given , the function @xmath654 is the unique solution to . we have thus identified @xmath655 . now we observe that if we test by an arbitrary @xmath656 with support in @xmath129 , we obtain that @xmath657 passing to the limit , we obtain by virtue of corollary [ cor : testwithgradient ] and @xmath658 that @xmath659\ , { \,\mathrm{d}}x = \int_{{\mathbb{r}}^d } 2f(x)g ( x)\,{\,\mathrm{d}}x\ , . \label{eq : thm - two - scale - lim - help - split-1}\end{aligned}\ ] ] when we now insert @xmath660 and @xmath661 for @xmath357 into and add the resulting equations to , then we obtain that @xmath662\ , { \,\mathrm{d}}x = \int_{{\mathbb{r}}^d } 2f(x)g ( x)\,{\,\mathrm{d}}x\ , .\end{aligned}\ ] ] a comparison with the definition of @xmath663 in finally yields that @xmath354 solves @xmath664 since @xmath130 is non - degenerate , we find that is the weak formulation of . hence , from elliptic regularity theory ( * ? ? ? * chapter 6 ) , we obtain that @xmath665 . since the solution @xmath354 of is unique , it follows that holds for the entire sequence . as for the last ingredient for the proof of theorem [ thm : twoscale - strong ] , we observe the following : on the cube @xmath129 the operator @xmath120 with zero dirichlet conditions is strictly positive definite ( see e.g. ) and thus it follows that on @xmath129 its inverse @xmath666 is well - defined . similarly , the inverse @xmath667 of @xmath668 on @xmath129 is well - defined . we have the following lemma . [ lem : condjkoii2 ] the operators @xmath669 are @xmath29-a.s . positive , compact and self - adjoint . the norms @xmath670 are @xmath29-a.s . bounded by a constant independent of @xmath1 . since @xmath663 is positive definite and symmetric , the properties of @xmath671 follow from the theory of elliptic partial differential equations , see e.g. ( * ? ? ? * chapter 6 ) . the operator @xmath672 is uniformly bounded in @xmath1 by virtue of . moreover , @xmath672 is real and symmetric by construction and therefore self - adjoint . finally , its range @xmath105 is finite - dimensional and thus @xmath672 is compact . let us first show that @xmath673 where @xmath665 is the solution to . indeed , since @xmath672 is self - adjoint , we observe that @xmath674 since @xmath675 in @xmath676 and @xmath677 , lemma [ lem : twoscale - weak ] implies that @xmath678 converges strongly in @xmath676 to @xmath679 . it follows that @xmath680 where we have used that the operator @xmath671 is self - adjoint , see lemma [ lem : condjkoii2 ] . we further note that @xmath681 by the same arguments as for . since @xmath682 is dense in @xmath683 , it thus follows that @xmath684 . by virtue of lemma [ lem : precompact ] and we conclude that @xmath126 strongly in @xmath676 . the existence of solutions to follows from positivity of the first eigenvalue for small @xmath1 . hence we can calculate the apriori estimates similar to by testing with @xmath169 and using @xmath144 to obtain @xmath685 since @xmath686 is bounded , this implies that @xmath687 is bounded in @xmath1 . from lemma [ lem : precompact ] it follows that @xmath688 strongly in @xmath353 and hence @xmath689 . hence from theorem [ thm : twoscale - strong ] we obtain that @xmath354 solves . first , we notice that without loss of generality , we can assume that the function @xmath686 is nonnegative . otherwise , we simply substitute @xmath686 for @xmath690 and prove the result for the new @xmath686 . then we notice that the substitution has simply shifted the spectrum by the constant @xmath691 and the new eigenvectors are the same as the old ones . thus , it suffices to prove the claim for @xmath692 . note that directly implies that if @xmath692 , then @xmath140 is positive . then lemmas [ lem : condjko ] and [ lem : condjkoii2 ] ensure that conditions i - iv of ( * ? ? ? * section 11.1 ) are satisfied and theorem [ thm : spectrum ] follows by virtue of ( * ? ? ? * theorems 11.4 , 11.5 ) . as in the paragraph before lemma [ lem : condjkoii2 ] , we now define the operators @xmath672 and @xmath671 as the inverses of @xmath136 and @xmath693 , respectively . for @xmath692 , we further consider the spectrum of the operators @xmath672 : @xmath694[1.5em][c]{0cm}{}}\\ & \mu_1^{\varepsilon}\geq \mu_2^{\varepsilon}\geq \ldots \geq \mu_k^{\varepsilon}\ldots \ , , \quad \mu_k^{\varepsilon}>0 \ , , { \parbox[0pt][1.5em][c]{0cm}{}}\\ & \langle \psi_k^{\varepsilon } , \psi_l^{\varepsilon}\rangle_{{{\mathcal{h}_{\varepsilon } } } } = \delta_{kl}\ , , { \parbox[0pt][1.5em][c]{0 cm } { } } \end{aligned}\end{aligned}\ ] ] as well as the spectrum of the operator @xmath671 @xmath695[1.5em][c]{0cm}{}}\\ & \mu_1 ^ 0 \geq \mu_2 ^ 0 \geq \ldots \geq \mu_k^0 \ldots \ , , \quad \mu_k^0 > 0 \ , , { \parbox[0pt][1.5em][c]{0cm}{}}\\ & \langle \psi_k^0 , \psi_l^0 \rangle_{{{\mathcal{h}_{\varepsilon } } } } = \delta_{kl}\ , .{\parbox[0pt][1.5em][c]{0 cm } { } } \end{aligned}\end{aligned}\ ] ] [ rem : relationbl ] the eigenfunctions @xmath696 of the operator @xmath672 and the eigenfunctions @xmath697 of the operator @xmath671 coincide with the eigenfunctions of the operators @xmath136 and @xmath693 , respectively . their eigenvalues relate to those of @xmath160 and @xmath693 by @xmath698 [ lem : condjko ] 1 . [ lem : condjkoi ] for any @xmath699 , the following is true : @xmath700 further , @xmath701 provided that @xmath702 and @xmath703 and @xmath704 2 . [ lem : condjkoiii ] let @xmath111 and let @xmath705 with @xmath706 . then the following is true : if @xmath707 then @xmath708 3 . [ lem : condjkoiv ] for any sequence @xmath705 such that @xmath709 , there exists a subsequence @xmath710 and a vector @xmath711 such that @xmath712 for ( i ) : let @xmath699 . by jensen s inequality it follows that @xmath713 for we first observe that @xmath714 the second term on the above rhs converges to zero by assumption . for the first term we note that the triangle inequality together with yields @xmath715 which is bounded from above . it follows that the first term on the rhs of converges to zero as well . part ( ii ) follows directly from proposition [ prop : twoscale - strong ] and . similarly , part ( iii ) follows from proposition [ prop : twoscale - strong ] and since @xmath716 implies that there exists a subsequence @xmath717 along which @xmath718 in @xmath676 . this proof is an application of the grtner - ellis theorem and goes along the lines of ( * ? ? ? * theorem 1.8 ) . for the convenience of the reader , we outline the main steps here . let @xmath137 be a bounded , continuous function . we define the generating cumulant function @xmath719 } \subset \alpha_t q\right]\ , .\end{aligned}\ ] ] as in @xcite , it suffices to show that @xmath720 where @xmath721 denotes the principal dirichlet eigenvalue of @xmath151 on @xmath129 with zero dirichlet boundary conditions . then the claim follows by the grtner - ellis theorem . in order to show , we define the operator @xmath722 acting on real - valued functions @xmath723 by @xmath724 } \subset \alpha_t q \ } } f\left ( x_t \right)\right ] \quad ( z\in\alpha_t q\cap { { \mathbb{z}}^{d}})\ , .\end{aligned}\ ] ] since @xmath186 is a step function , @xmath722 admits the semigroup representation @xmath725 \right\}\ , , \end{aligned}\ ] ] where the operator in the exponent is considered with zero dirichlet conditions at the boundary of @xmath726 and @xmath727 } v\left ( \frac{z+y}{\alpha_t } \right ) \ , { \,\mathrm{d}}y \qquad ( z\in \alpha_t q \cap zd)\ , .\end{aligned}\ ] ] let @xmath728 denote the principal dirichlet eigenvalue of @xmath729 on @xmath726 with zero dirichlet boundary conditions . let @xmath730 be the corresponding principal dirichlet eigenfunction . then , in order to show , we have to show that @xmath731 for any @xmath732 . the second equality follows by virtue of theorem [ thm : spectrum ] . it remains to prove the first equality . for this purpose we notice that an eigenvalue expansion together with cauchy - schwarz and parseval s identity yields that @xmath733 on the other hand , since @xmath734 , we can estimate from below @xmath735 since @xmath736 is a normalized eigenfunction . thus , if @xmath737 decays at most polynomially , we have proved the claim . similarly to the proof in @xcite , we obtain that @xmath738\right)\ , , \end{aligned}\ ] ] where @xmath739 is an upper bound for @xmath686 . since @xmath736 is normalized and @xmath740 = \min_{x\in\alpha^2_t q\cap { { \mathbb{z}}^{d } } } { \mathrm{p}}^{{\omega}}_0 \left [ x_{\alpha_t } = x \right]\end{aligned}\ ] ] decays at most polynomially by assumption [ ass : heatkernel ] , the claim follows . the authors thank wolfgang knig , stefan neukamm , mathias schffner , and mark peletier for fruitful discussions and valuable hints . f.f . gratefully acknowledges the support by dfg within the rtg 1845 _ stochastic analysis with applications in biology , finance and physics_. m.h . was supported by dfg through sfb 1114 _ scaling cascades in complex systems _ , subproject c05 _ effective models for interfaces with many scales_. t. kumagai . , volume 2101 of _ lecture notes in mathematics_. springer , cham , 2014 . lecture notes from the 40th probability summer school held in saint - flour , 2010 , cole dt de probabilits de saint - flour . [ saint - flour probability summer school ] .

individual localized spins are promising for the implementation of emerging quantum information technologies in the solid state @xcite . in this field , magnetic dopants in conventional semiconductors present many desirable features , such as reproducible quantum properties , stability , and potential scalability for further applications @xcite . thanks to their point - like character , a longer spin coherence time ( compared to carriers spins ) can also be expected at low temperature , making them potentially good systems to store quantum information . semiconductor quantum dots ( qds ) permit efficient electrical or optical manipulation of individual carrier spins @xcite . it has been shown that the optical properties of a qd can also be used to control the spin state of individual @xcite or pairs @xcite of magnetic atoms . the spin of a magnetic atom in a qd can be prepared by the injection of spin polarized carriers and its state can be read through the energy and polarization of the photons emitted by the qd @xcite . the insertion of a magnetic atom in a qd where the strain or the charge states can be controlled also offers degrees of freedom to tune the properties of the localized spin such as its magnetic anisotropy responsible for the spin memory at zero magnetic field @xcite . positively charged magnetic qds present a large magnetic anisotropy induced by the exchange interaction between the confined heavy - hole and the spin localized on a magnetic atom @xcite . it has been shown in charge tunable mn - doped cdte / znte qds that the hybrid mn - hole spin can be prepared by optical orientation under excitation with circularly polarized light @xcite . it was also demonstrated that a positively charged mn - doped qd forms an ensemble of optical @xmath0 systems that can be independently addressed @xcite . this level structure suggests that the hybrid mn - hole spin states could be coherently manipulated using two resonant optical fields @xcite . for an efficient coherent control and a practical use of this hybrid spin in a quantum device one would require however sufficiently long relaxation and coherence times for the two ground states of the @xmath0 systems . in this article , resonant photoluminescence ( pl ) of the positively charged exciton ( x@xmath1 ) coupled to a single mn atom in charge tunable cdte / znte qds @xcite is exploited to analyze the spin relaxation mechanisms of the coupled hole and mn spins . auto - correlation of the resonant pl and resonant optical pumping experiments reveal an efficient spin relaxation channel of the mn - hole system : a mn - hole flip - flop induced by the interplay of the mn - hole exchange interaction and the lattice deformation induced by acoustic phonons . a model predicts mn - hole flip - flops in the nanosecond range , in agreement with the spin dynamics observed experimentally . we also show that the optical @xmath0 systems are connected by inefficient forbidden spin - flips which are enhanced under a weak transverse magnetic field . the article is organized as follows : the sample and experiments are presented in sec . the spin structure of the positively charged mn - doped qds is shortly described in sec . the dynamics of the resonant pl probed with auto - correlation measurements is described in sec . iv . in sec . v we discuss resonant optical pumping experiments allowing probing the initialization and relaxation of coupled hole and mn spins . in sec . vi we present a model describing acoustic phonon induced mn - hole flip - flops and show that this mechanism is responsible for the strong pl observed under resonant excitation of x@xmath1-mn . finally , in sec . vii we model the spin dynamics of positively charged mn - doped qds under resonant excitation and compare the calculated autocorrelation response and optical pumping signal with experimental results . the sample consists of mn - doped self - assembled cdte qds grown by molecular beam epitaxy on a p - doped znte ( 001 ) substrate according to the procedure dscribed in ref . a bias voltage is applied between a 5 nm gold schottcky gate deposited on the surface of the sample and the p - doped substrate to control the charge state of the qds . under a positive bias , as we already reported in ref . 22 , a single hole is trapped in the magnetic qds and only the emission of the positively charged exciton is observed ( figure [ fig1](a ) ) . individual qds containing one mn are isolated using micro - spectroscopy techniques @xcite . a high refractive index hemispherical solid immersion lens is mounted on the surface of the sample to enhance the spatial resolution and the collection efficiency of single dot emission . the qds are excited with a tunable continuous wave ( @xmath2 ) laser tuned to an excited state of the dots or on resonance with the ground state of the positively charged exciton . the resulting collected pl is dispersed and filtered by a one - meter double monochromator . the temporal statistics of the emission of photons is analyzed through photon - correlation measurements using a hanbury brown and twiss ( hbt ) setup with a resolution of about 0.8 ns . under our experimental conditions , with photon counts rates of a few khz , the measured photon pair time distribution yields , after normalization , to the second order correlation function of the pl intensity , @xmath3 . for optical pumping experiments , trains of resonant circularly polarized light are prepared with electro - optic or acousto - optic modulators with a switching time of about 10 ns and the resonant pl is detected with a fast avalanche photodiode . permanent magnet mounted on a translation stage are used to apply a weak magnetic field in faraday or voight configurations . ) and positively charged ( @xmath4 ) exciton as a function of energy and bias voltage . ( b ) pl of the mn - doped qd under a positive bias voltage of v=5.5v . inset : scheme of the energy levels of the ground ( @xmath5 ) and excited states ( @xmath4 ) in a positively charged mn - doped qd as a function of their angular momentum ( @xmath6 ) . ( c ) experimental ( left ) and calculated ( right ) color - scale plot of the linear polarization dependence of the pl of @xmath4 at b=0 t ( top ) and under a transverse magnetic field b@xmath7=0.42 t ( bottom ) , maximum transverse field used in the spin dynamics experiments . the parameters used in the calculation are listed in table [ paraqd].,width=3 ] a mn atom in a strained self - assembled cdte qd exhibits a fine structure dominated by a weak magnetic anisotropy with an easy axis along the qd axis @xcite . neglecting the tetrahedral crystal field of the cdte matrix @xcite , this fine structure is described by the effective spin hamiltonian @xmath8 here , @xmath9 depicts the effect of the biaxial strain and @xmath10 describes the anisotropy of the strain in the plane of the qd . @xmath9 varies from 0 @xmath11ev ( strain free qd @xcite ) to 12 @xmath11ev ( strained cdte layer matched on a znte substrate @xcite ) and typical values around 7 @xmath11ev are usually observed in cdte / znte qds @xcite . when a hole is trapped in a qd containing a single mn , the spin structure is controlled by the mn - hole exchange interaction that reads @xmath12 with i@xmath13 the exchange energy between the hole and the mn ( @xmath14 ) and @xmath15 the hole spin operator . in the presence of heavy - hole / light - hole mixing , @xmath15 , represented in the basis of the two low energy heavy - hole states , is related to the pauli matrices by @xmath16 and @xmath17 with @xmath18 . @xmath19 is the coupling energy between heavy holes and light holes separated by an effective energy splitting @xmath20 . @xmath21 is the angle relative to the [ 110 ] axis of the principal axis of the anisotropy ( shape and/or strain ) responsible for the heavy - hole / light - hole mixing @xcite . for a weak valence band mixing , the mn - hole energy levels are mainly controlled by @xmath22 and form a spin ladder with a quantization axis along the qds growth direction . these states are labelled @xmath23 . the pl of a positively charged mn - doped qd is presented in figure [ fig1](b ) . when an exciton is injected in the qd loaded with a single hole , one has to consider the x@xmath1-mn complex . the energy levels of such a complex , where the two hole spins are paired in an antiparallele configuration , is dominated by the electron - mn exchange interaction @xmath24 with @xmath25 the electron spin and @xmath26 the exchange energy between the electron and the mn spins . the twelve e - mn states are split into a ground state sextuplet ( total spin m=3 ) and a fivefold degenerated manifold ( total spin m=2 ) ( see inset of fig . [ fig1](b ) ) . these energy levels are labelled @xmath27 . .values of the parameters used to model the positively charged mn - doped qd presented in figure [ fig1 ] . i@xmath28 , i@xmath13 , @xmath29 , @xmath21 , @xmath30 and @xmath31 are chosen to reproduce the linear polarization intensity map of fig . [ fig1](c ) . the other parameters can not be extracted from the pl measurements and values for typical mn - doped qds are chosen for the calculation of the spin dynamics presented in section vii . [ cols="<,^ , > " , ] [ paraph ] considering a linear dispersion of acoustic phonons @xmath32 and in spherical coordinates with @xmath33 , the explicit formula of the decay rate ( [ fermi ] ) is @xmath34 where we used the continuum limit ( @xmath35 with @xmath36 the crystal volume ) and integrated over @xmath37 and @xmath38 . the summation is taken over the acoustic phonon branches @xmath39 of corresponding sound velocity c@xmath40 . the geometrical form factors for each phonon branch , @xmath41 , are given by @xmath42 system associated with the electron - mn state @xmath43 extracted from the full level structure of a positively charged mn - doped qd ( fig . [ fig1](b ) ) . ( b ) relaxation time @xmath44 , between the two mn - hole ground states of the @xmath0 system calculated with the material and qd parameters listed in table [ paraph ] and a temperature t=7k . the vertical line shows the energy splitting in the studied qd of the mn - hole states involved in the @xmath0 systems considered here ( resonances ( 2 ) and ( 3 ) identified in fig . [ fig2]).,width=316 ] in the numerical calculation of the spin flip time @xmath44 presented in fig.[fig7 ] we use the material parameters of cdte or znte and the typical parameters for self - assembled cdte / znte qds listed in table [ paraph ] . the calculated relaxation time strongly depend on the energy separation between the mn - hole levels @xmath45 . this energy dependence is controlled by the size of the hole wave - function given by l@xmath46 and l@xmath47 . the estimated flip - flop time is also strongly sensitive on the phonon induced mixing of the ground heavy - hole states with the higher energy light - hole levels @xcite . in our model , this mixing is controlled by @xmath20 , an effective energy splitting between heavy - holes and light - holes . this simple parameter can indeed describe more complex effects such as a coupling of the confined heavy - hole with ground state light - holes in the barriers @xcite or effective reduction of heavy - hole / light - hole splitting due to a presence of a dense manifold of heavy - hole like qd states lying between the confined heavy - hole and light - hole levels @xcite . from this modelling we deduce that for a hole confined in small cd@xmath48zn@xmath49te alloy qds , the mn - hole flip - flop time @xmath44 can be easily bellow 2 ns for an effective heavy - hole / light - hole splitting @xmath20=15 mev and an energy separation in the mev range , typical for the mn - hole spin splitting in magnetic qds . we will use in the following calculations @xmath44=1.5 ns for the ground states of each @xmath0 system . using the level scheme presented in fig.[fig1](b ) for a positively charged mn - doped qd and the estimated mn - hole flip - flop rates , we can calculate the time evolution of the 24x24 density matrix @xmath50 describing the population and the coherence of the 12 electron - mn states on the excited state and the 12 mn - hole states on the ground state of a positively charged qd . in the markovian approximation , the master equation which governs the evolution of @xmath50 can be written in a general form ( lindblad form ) as : @xmath51+l\varrho}\ ] ] where @xmath52 is the hamiltonian of the complete system ( @xmath53-mn @xmath54 and mn - hole @xmath55 ) : @xmath56 and @xmath57 in ( [ lindblad ] ) , @xmath58 describes the coupling or decay channels resulting from an interaction with the environment @xcite . the population transfers from level @xmath59 to level @xmath60 in an irreversible process associated with a coupling to a reservoir is described by a lindblad term of the form @xmath61 where @xmath62 is the incoherent relaxation rate from level @xmath59 to level @xmath60 . such term can describe the radiative decay of the exciton ( irreversible coupling to the photon modes ) or the relaxation of the carriers or mn spin ( irreversible coupling to the phonon modes ) . it can also be used to describe the optical generation of an exciton in the low excitation regime where the energy shift induced by the coupling with the laser field is neglected . a pure dephasing ( i.e. not related to an exchange of energy with a reservoir ) can also be introduced for the different spins and described by @xmath63 : @xmath64 where @xmath65 is the pure dephasing rate of level @xmath59 . to identify the main spin relaxation channels responsible for the observed spin fluctuations , we first modelled the auto - correlation of the resonant pl using the full spin level structure of a p - doped magnetic qd . for a qualitative description of the observed spin dynamics , we use as an example the mn - doped qd parameters extracted from the linear polarization intensity map listed in table [ paraqd ] and reasonable order of magnitude for the spin relaxation times . as already observed in charged mn - doped qds under pulsed resonant excitation ( ref . 23 ) , we consider that the spin dynamics in the excited state is controlled by the time evolution of @xmath54 , the generation rate of excitons @xmath66=1/@xmath67 and their radiative lifetime @xmath68=0.3 ns . the coherence of the coupled electron - mn spins is limited by a pure dephasing term , @xmath69=0.5 ns , typical value measured in charged mn - doped qds@xcite . for the mn - hole system in the ground state , we take into account a spin relaxation time of the mn in the exchange field of the hole , @xmath70 , describing relaxation channels involving a change of the mn spin by one unit . this spin relaxation channel is introduced for a general description , however its characteristic time ( in the @xmath11s range ) is long compared to the time - scale of the dynamics considered in the resonant pl experiments and does not qualitatively affect the calculated time evolution . because of the presence of valence band mixing in the qds , spin flip of the hole independently of the mn are expected to be more efficient . a spin flip time in the 10 ns range has indeed been calculated for a hole in the exchange field of a mn @xcite . relaxation time of the hole spin around 5 ns has also been measured at zero magnetic field in negatively charged cdte / znte qds @xcite . we then include in the model possible spin flips of the hole by one unit with a characteristic time @xmath71=10ns . the phonon induced mn - hole flip - flops , occurring at @xmath44 , are also introduced between the two mn - hole ground states of each @xmath0 system . for a general qualitative description , an additional pure dephasing time @xmath72 is also included in the dynamics of the mn - hole system with a lindblad term of the form ( [ deph ] ) . we can not extract this parameter from the experiments . we take @xmath72= 5 ns , slightly longer than what was measured for electron - mn , as the mn - hole system is highly split and less sensitive to effective fluctuating magnetic field such as the one produced by nuclear spins for instance @xcite . with the qd parameters listed in table [ paraqd ] and ( unless specified ) @xmath68=0.3ns , @xmath70=5 @xmath11s , @xmath73=10ns , @xmath67=0.25 ns , @xmath44=1.5 ns , @xmath72= 5 ns , @xmath69= 0.5 ns , t=7k and b@xmath74=0 . ( b ) ( c ) and ( d ) illustrate the influence on @xmath75 of @xmath44 , @xmath67 and @xmath76 respectively . note the different vertical scale in ( b).,width=316 ] the transition rates @xmath77 between the different mn - hole states depend on their energy separation @xmath78 . here we use @xmath77=1/@xmath79 if @xmath80 and @xmath77=1/@xmath81 if @xmath82 @xcite . this accounts for a thermalization among the 12 mn - hole levels with an effective spin temperature @xmath83 . the optical excitation ( @xmath67 ) , the exciton recombination ( @xmath68 ) , the mn spin relaxation ( @xmath70 ) , the hole spin relaxation ( @xmath71 ) and the phonon induced transfer time ( @xmath44 ) produce a irreversible population transfer between level @xmath84 and @xmath85 and are described by lindblad terms ( [ inc ] ) . detection and an excitation of @xmath86 and @xmath87 with modulated circular polarization . the qd parameters for the calculations are those listed in table [ paraqd ] and @xmath68=0.3ns , @xmath70=5 @xmath11s , @xmath73=10ns , @xmath72= 5 ns , @xmath69= 0.5 ns , @xmath44=1.5 ns , t=10k and @xmath67=0.25ns . ( a ) influence of a variation of @xmath67 and @xmath44 . ( b ) influence of a transverse magnetic field @xmath76 . the inset presents the transverse magnetic field dependence of the difference of population for a @xmath88 or a @xmath89 excitation.,width=316 ] the time evolution of @xmath90 normalized by @xmath91 and with the initial condition @xmath92 ( mn - hole spin in the state @xmath93 just after the recombination of the exciton ) accounts for the auto - correlation of the emission in @xmath89 polarization of the electron - mn state @xmath94 under @xmath2 resonant excitation . @xmath90 obtained with the qd parameters listed in table [ paraqd ] is presented in fig . [ fig8](a ) . after a fast increase , the calculated population presents a maximum at short delay . with reasonable spin relaxation parameters ( see caption of fig . [ fig8 ] ) the width and the amplitude of the maximum are in good agreement with the photon bunching signals observed experimentally . the width of the calculated bunching is controlled by all the spin - flip terms that can induce an escape out of the resonantly excited @xmath0 system . at zero transverse magnetic field , it is dominated by spin flips in the mn - hole system . as illustrated in fig . [ fig8](a ) , suppressing @xmath73 gives a width of bunching only controlled by the hamiltonian evolution and the decoherence which is slightly larger than what is observed experimentally ( fig . [ fig3 ] ) . the dependence on the excitation intensity , @xmath67 , and transverse magnetic field , @xmath76 , are also qualitatively well reproduced by the model ( fig . [ fig8](c ) and ( d ) respectively ) . at zero magnetic field , the leaks outside the excited @xmath0 systems are dominated by @xmath73 . @xmath95 induces fluctuations in a slightly longer time scale . the situation is different under a weak transverse magnetic field where the electron - mn states are mixed introducing new channel of escape and significantly reducing the width of the bunching . this model is based on a large number of parameters , whose values can not all be extracted precisely from the measurements . let us note however that suppressing the fast flip - flop process connecting the two mn - hole ground states ( @xmath96 in fig . [ fig8](b ) ) still produces a bunching as a weak coherent transfer between the two ground states of the @xmath0 system still exist . however , with this process only , the calculated pl intensity is about 50 time smaller and the behaviour in transverse magnetic field is not compatible with the experimental results . with this model , we can also calculate the population of the electron - mn states under resonant excitation with alternated circular polarization and estimate the efficiency and dynamics of the optical pumping . figure [ fig9 ] presents the calculated time evolution of the population of the electron - mn state @xmath97 under alternated resonant excitation of @xmath86 in @xmath88 polarization or @xmath87 in @xmath89 polarization . this corresponds to the experimental configuration where the qd is resonantly excited with modulated circular polarization at the energy of @xmath86 and @xmath87 ( absorption ( 2 ) in fig . [ fig2](b ) ) and the low energy resonant pl is detected in @xmath89 polarisation . the main features of the time - resolved optical pumping experiments ( see fig . [ fig4 ] and fig . [ fig5 ] ) are well reproduced by the model . the timescale of the pumping transient , in the few tens of nanosecond range , and its excitation intensity dependence are also in good agrement with the experiments ( see figure [ fig9](a ) ) . the influence of a transverse magnetic field , b@xmath74 , on the optical pumping transient can also be described by this model . first , a significant reduction of the pumping time is observed for a weak magnetic field ( b@xmath74=0.2 t in fig . [ fig9](b ) ) . as for the autocorrelation , this acceleration comes from the increase of the leakage out of the @xmath0 system induced by the mixing of the electron - mn states . secondly , the transients obtained when switching the polarization from @xmath98 to @xmath99 and from @xmath99 to @xmath98 become identical for b@xmath100 t , as observed in the experiments ( fig . [ fig5](b ) ) . to understand this behaviour under b@xmath74 , let us remember that we resonantly excite @xmath86 from @xmath101 with @xmath88 light and excite @xmath87 from @xmath102 with @xmath89 photons . in both cases we detect the population of @xmath97 in @xmath89 polarization ( see the excitation / detection configuration illustrated in the inset of fig . [ fig4](b ) ) . if the states @xmath86 and @xmath87 are uncoupled , as it is the case at zero field , we do not detect any light during the @xmath89 excitation . with a sufficiently large mixing of @xmath86 and @xmath87 induced by the transverse magnetic field , for a @xmath89 excitation of @xmath87 , the population can be coherently transferred to @xmath86 during the charged exciton lifetime and @xmath89 light is detected after a recombination towards @xmath103 @xcite . in the optical pumping sequence , we can then observe , in @xmath89 polarization , a transient when the @xmath88 excitation empties the state @xmath101 but also a similar transient when the @xmath89 excitation empties the state @xmath102 . the transverse magnetic field dependence of the difference of steady state intensity observed in @xmath98 and @xmath99 polarization ( inset of fig . [ fig5](b ) ) is also well reproduced by the model ( inset of fig . [ fig9](b ) ) . this depolarization curve is controlled by the anisotropy of the electron - mn spin which is induced by @xmath30 and @xmath9 @xcite . let us note finally that , as expected , suppressing @xmath44 from the model , a very weak average resonant pl and a fast optical pumping are obtained ( fig . [ fig9](a ) , top curve ) . initialized by a sequence of @xmath89/@xmath88 resonant excitation of @xmath87 and @xmath86 . the dashed black line ( shifted for clarity ) is an exponential fit with a characteristic time @xmath104=85 ns . ( b ) corresponding calculated time evolution of the population of @xmath97 . the parameters are those of fig . [ fig9].,width=316 ] including a dark time in the pumping sequence , we can also numerically evaluate the time required for the mn - hole spin to return to the ground state of the excited @xmath0 system . the time evolution of the population of the mn - hole state @xmath105 initially prepared by a sequence of @xmath89/@xmath88 excitation resonant with @xmath94 ( and @xmath106 ) is presented in fig.[fig10 ] . when the optical excitation is switched off , after an abrupt jump due to the optical recombination of the charge exciton , the ground mn - hole state @xmath105 is repopulated in a timescale of about 100 ns , much shorter than the mn spin relaxation time used in the model ( @xmath70=5ns ) . this relaxation is induced by the presence of valence band mixing . in the presence of valence - band mixing , @xmath107 couples two by two the different mn - hole levels . this coupling induces a transfer of population between the different mn - hole levels . the transfer of population becomes irreversible in the presence of dephasing and controls the observed mn - hole spin relaxation @xcite . with @xmath108=1 ( mn - hole spin in the state @xmath109 after a @xmath89 recombination ) for a resonant @xmath88 excitation of the coupled electron - mn states @xmath110 and @xmath111 without and with a longitudinal magnetic field . ( b ) time evolution of @xmath112 under excitation with modulated circular polarization . the parameters used in the calculations are those of fig . [ fig9].,width=316 ] the particular behaviour observed for a resonant excitation of the electron - mn states @xmath110 or @xmath111 ( weak photon bunching and no optical pumping at zero field , fig . [ fig2](a ) and fig . [ fig4](a ) respectively ) is also qualitatively explained by the model ( see figure [ fig11 ] ) . as demonstrated in ref . 23 , the presence of a strain anisotropy term @xmath10 in the @xmath11ev range in the mn fine structure directly couples the states @xmath110 and @xmath111 which are initially degenerated . under circularly polarized resonant excitation we either excite @xmath110 with @xmath88 photons or @xmath111 with @xmath89 photons . at zero magnetic field , the population is transferred between the two states in a time scale of a few hundreds picoseconds @xcite . under circularly polarized resonant excitation , the two @xmath0 systems associated with @xmath113 are simultaneously excited . for alternated circular polarization , a steady state is reached and no pumping transient induced by a leak outside the @xmath0 systems is expected . under a weak longitudinal magnetic field the mn zeeman energy dominates the strain anisotropy term and the coherent transfer is blocked . the states @xmath110 and @xmath111 are decoupled and a large amplitude of bunching and an efficient optical pumping are restored . this behaviour observed in the experiments is qualitatively reproduced by the model . let us finally note that in the modelling of optical pumping at zero magnetic field presented in fig . [ fig11](b ) , fast oscillations are obtained in the first nanoseconds after the polarization switching . these are due to the population transfer between @xmath110 and @xmath111 in the excited state ( directly coupled by e ) and between @xmath114 and @xmath115 in the ground state ( coupled by a combination of non diagonal terms of @xmath55 ) during the coherence time . these oscillations are too fast to be observed in the experiments . the calculated resonant pl intensity in @xmath89 polarization ( proportional to @xmath112 ) is also slightly larger for a @xmath88 excitation than for a @xmath89 excitation . the @xmath89 resonant pl probes the population of @xmath110 which is directly excited by a resonant @xmath88 laser ( see the excitation / detection configuration in the inset of fig . [ fig4](a ) ) . on the other hand , under a @xmath89 laser , one excites @xmath111 and the charged exciton has a probability to recombine before being transferred to @xmath110 and detected in @xmath89 pl . this transfer time results in a slight difference in the steady state resonant pl intensity obtained in a @xmath116 or @xmath117 configuration ( see fig . [ fig4](a ) ) . the model we developed here explains the main behaviour of a positively charged mn - doped qd under resonant optical excitation . it shows in particular the importance of the effective heavy - hole / light - hole splitting @xmath20 on the mn - hole spin dynamics . the studied cdte / znte qds have a weak valence band offset . the resulting small value of @xmath20 is first responsible for the large influence of the qds shape or strain anisotropy on the valence band mixing . the valence band mixing reduces the magnetic anisotropy of the mn - hole system and its spin life time . a small @xmath20 also significantly enhances the coupling of the mn - hole spin with acoustic phonons and the phonon induced mn - hole flip - flops @xmath44 . this fast spin dynamics limits the use of such hybrid spin system in practical quantum information devices . the use of different qd systems with a larger valence band offset and a larger heavy - hole / light - hole splitting @xcite should significantly slow down the mn - hole spin dynamics . using resonant pl of the positively charged exciton , we have identified an efficient spin relaxation channel for the hybrid mn - hole spin in a qd . a modelling confirms that mn - hole flip - flops in a nanosecond timescale are induced by an interplay of mn - hole exchange interaction and the lattice deformation of acoustic phonons . these flip - flops are responsible for the large pl intensity observed under resonant excitation of the @xmath0 systems present in positively charged mn - doped qds . we showed that jumps out of an optically excited @xmath0 system are possible . these leaks induce a large bunching of the resonant pl and are at the origin of the optical pumping of the mn - hole spin observed under circularly polarized resonant excitation . escape out of the excited @xmath0 system can be enhanced by a transverse magnetic field which mixes electron - mn states in the excited state of the charged qd . the fast mn - hole spin dynamics revealed by these experiments make difficult the practical use of this hybrid spin for quantum information devices . however , the use of a different qd systems with a better hole confinement and a larger heavy - hole / light - hole splitting would significantly reduce the influence of valence band mixing on the mn - hole spin relaxation and limit the interaction of the hybrid mn - hole spin with acoustic phonons . this work was realized in the framework of the commissariat lenergie atomique et aux energies alternatives ( institut nanosciences et cryognie ) / centre national de la recherche scientifique ( institut nel ) joint research team nanophysique et semi - conducteurs . the mechanism of mn - hole flip - flop that we identified and modelled in the case of positively charged magnetic qds can also be directly observed in the resonant pl of neutral qds . this is illustrated in fig . [ figappendix ] which presents the pl of the exciton in a qd where all the dark exciton states ( @xmath118 ) are well separated from the bright exciton states ( @xmath119 ) and can be observed at zero magnetic field on the low energy side of the spectra . ( plain line ) and dark @xmath120 ( dotted line ) exciton energy levels calculated with @xmath26=-0.03 mev , @xmath121=0.12 mev , @xmath122 , @xmath30=10 @xmath11ev and an electron - hole exchange interaction @xmath123=-0.78 mev . see for instance reference 32 for a description of the model.,width=316 ] in the pl excitation experiments presented in fig . [ figappendix ] , the detection window is set on the dark states while a circularly cross - polarized laser is scanned on the three high energy levels of the bright exciton . three successive resonances are observed . ( 1 ) : an excitation on @xmath124 produces the dominant pl on a dark states associated with a mn spin state @xmath125 . as a mn spin flip by two units is unlikely , we then attribute this pl to @xmath126 . the transfer between these states involves a mn - hole flip - flop . the second weaker contribution comes from the state @xmath127 . it corresponds to a transfer conserving the mn spin and involving a spin flip of the electron . ( 2 ) : an excitation on @xmath128 produces the dominant pl on a dark states associated with a mn spin state @xmath129 . as discussed above , a mn spin flip by four units is unlikely , we then attribute this pl to @xmath130 . the transfer involve a mn - hole flip - flop . the second contribution comes from the state @xmath131 . it corresponds to a conservation of the mn spin and a spin flip of the electron . ( 3 ) : from an excitation on @xmath132 , the dominant pl comes from the state @xmath133 . it corresponds to a transfer with conservation of the mn spin and a spin flip of the electron . a pl is also observed from @xmath130 after a spin flip of the hole and from @xmath131 . the latter transfer involves an electron - mn flip - flop . let s note that in the state @xmath132 a mn - hole flip - flop is forbidden ( parallel hole and mn spin ) . if we resume the observed resonances in this neutral qd , we can distinguish two types of spin transfer which can occur within the lifetime of the exciton ( typically 300ps ) : first an efficient transfer from a bright to a dark state involving a mn - hole flip - flop which produces a change of one unit of the mn spin state . secondly , a weaker transfer from a bright to a dark state involving a carrier spin - flip with conservation of the mn spin . j. r. petta , a. c. johnson , j. m. taylor , e. a. laird , a. yacoby , m. d. lukin , c. m. marcus , m. p. hanson , a. c. gossard , science * 309 * , 2180 ( 2005 ) . m. veldhorst , c. h. yang , j. c. c. hwang , w. huang , j. p. dehollain , j. t.muhonen , s. simmons , a . laucht , f. e. hudson , k. m. itoh , a. morello , a. s. dzurak , nature ( london ) * 526 * , 410 ( 2015 ) . k. saeedi , s. simmons , j. z. salvail , p. dluhy , h. riemann , n. v. abrosimov , p. becker , h .- j . pohl , j. j. l. morton , m. l. w. thewalt , science * 342 * , 830 ( 2013 ) n. bar - gill , l.m . pham , a. jarmola , d. budker and r.l . walsworth , nat . comm . * 4 * , 1743 ( 2013 ) . p. m. koenraad , m. e. flatte , nat . mater . * 10 * , 91 ( 2011 ) . m. atature , j. dreiser , a. badolato , a. hogele , k. karrai , a. imamoglu , science * 312 * , 551 ( 2006 ) . d. press , t. d. ladd , b. zhang , and y. yamamoto , nature ( london ) * 456 * , 218 ( 2008 ) . b. d. gerardot , d. brunner , p. a. dalgarno , p. ohberg , s. seidl , m. kroner , k. karrai , n. g. stoltz , p. m. petroff , r. j. warburton , nature ( london ) * 451 * , 441 ( 2008 ) . k. de greve , p. l. mcmahon , d press , t. d. ladd , d. bisping , c. schneider , m. kamp , l. worschech , s. hofling , a. forchel , y. yamamoto , nat . phys . * 7 * , 872 ( 2011 ) . l. besombes , y. leger , l. maingault , d. ferrand , h. mariette , j. cibert , phys . lett . * 93 * , 207403 ( 2004 ) . c. le gall , a. brunetti , h. boukari , l. besombes , phys . lett . * 107 * , 057401 ( 2011 ) . a. kudelski , a. lemaitre , a. miard , p. voisin , t.c.m . graham , r.j . warburton , o. krebs , phys . lett . * 99 * , 247209 ( 2007 ) . j. kobak , t. smolenski , m. goryca , m. papaj , k. gietka , a. bogucki , m. koperski , j .- rousset , j. suffczynski , e. janik , m. nawrocki , a. golnik , p. kossacki , w. pacuski , nat . comm . * 5 * , 3191 ( 2014 ) . l. besombes , c.l . cao , s. jamet , h. boukari , j. fernandez - rossier , phys . rev . b * 86 * , 165306 ( 2012 ) . o. krebs , a. lemaitre , phys . lett . * 111 * , 187401 ( 2013 ) . l. besombes , h. boukari , c. le gall , a. brunetti , c.l . cao , s. jamet , b. varghese , nanophotonics * 4 * , 75 ( 2015 ) . d. e. reiter , v. m. axt , t. kuhn , phys . b * 87 * , 115430 ( 2013 ) . m. goryca , t. kazimierczuk , m. nawrocki , a. golnik , j. a. gaj , p. kossacki , p. wojnar , g. karczewski , phys . * 103 * , 087401 ( 2009 ) . j. c. oberg , m. reyes calvo , f. delgado , m. moro - lagares , d. serrate , d. jacob , j. fernandez - rossier , c. f. hirjibehedin , nat . * 9 * , 64 ( 2014 ) . y. leger , l. besombes , l. maingault , d. ferrand , h. mariette phys . b * 72 * , 241309(r ) ( 2005 ) . k. vyborny , j. e. han , r. oszwaldowski , i. zutic , and a. g.petukhov , phys . b * 85 * , 155312 ( 2012 ) . b. varghese , h. boukari , l. besombes , phys . b * 90 * , 115307 ( 2014 ) . a. lafuente - sampietro , h. boukari , l. besombes , phys . b * 92 * , 081305(r ) ( 2015 ) . j. houel , j. h. prechtel , a. v. kuhlmann , d. brunner , c. e. kuklewicz , b. d. gerardot , n. g. stoltz , p. m. petroff , r. j. warburton , phys . rev . lett . * 112 * , 107401 ( 2014 ) . p. wojnar , c. bougerol , e. bellet - amalric , l. besombes , h. mariette , h. boukari , journal of crystal growth * 335 * , 28 ( 2011 ) . c. le gall , r. s. kolodka , c. l. cao , h. boukari , h. mariette , j. fernandez - rossier , l. besombes , phys . b * 81 * , 245315 ( 2010 ) . c. le gall , l. besombes , h. boukari , r. kolodka , j. cibert , h. mariette , phys . rev . lett . * 102 * , 127402 ( 2009 ) . m. goryca , m. koperski , p. wojnar , t. smoleski , t. kazimierczuk , a. golnik , p. kossacki , phys . * 113 * , 227202 ( 2014 ) . m. qazzaz , g. yang , s.h . xin , l. montes , h. luo , j.k . furdyna , solid state communications * 96 * , 405 ( 1995 ) . causa , m. tovar , s.b . oseroff , r. calvo , w. giriat , phys . lett . * a77 * , 473 ( 1980 ) . a. h. trojnar , m. korkusinski , u. c. mendes , m. goryca , m. koperski , t. smolenski , p. kossacki , p. wojnar , p. hawrylak , phys . b * 87 * , 205311 ( 2013 ) . l. besombes , h. boukari , phys . b * 89 * , 085315 ( 2014 ) . s. jamet , h. boukari , l. besombes , phys . b * 87 * , 245306 ( 2013 ) . j. fernandez - rossier , phys . b * 73 * , 045301 ( 2006 ) . y. leger , l. besombes , l. maingault , h. mariette , phys . b * 76 * , 045331 ( 2007 ) . l. besombes , y. leger , l. maingault , d. ferrand , h. mariette , j. cibert , phys . b * 71 * , 161307(r ) ( 2005 ) . l. besombes , k. kheng , l. marsal , h. mariette , phys . b * 63 * , 155307 ( 2001 ) . c. cohen - tannoudji , b. diu , f. laloe , _ mecanique quantique _ ( hermann , paris , 1973 ) . e. tsitsishvili , r. v. baltz , h. kalt , phys . b * 67 * , 205330 ( 2003 ) . k. roszak , v.m . axt , t. kuhn , p. machnikowski , phys . b * 76 * , 195324 ( 2007 ) . mahan , _ many - particle physics _ ( plenum press , new york , 1993 ) . woods , t.l . reinecke , r. kotlyar , phys . rev . b * 69 * , 125330 ( 2004 ) . s. adachi , _ properties of group iv , iii - v and ii - vi semiconductors _ ( john wiley and sons ltd , chichester , 2005 ) . michler , _ single quantum dots fundamentals , applications and new concepts _ ( springer , berlin , 2003 ) . luo , g. bester , a. zunger , phys . b * 92 * , 165301 ( 2015 ) van exter , j. gudat , g. nienhuis , d. bouwmeester , phys . a * 80 * , 023812 ( 2009 ) . c. roy and s. hughes , phys . x * 1 * , 021009 ( 2011 ) . c. l. cao , l. besombes , j. fernandez - rossier , phys b * 84 * , 205305 ( 2011 ) . l. cywinski , phys . b * 82 * , 075321 ( 2010 ) . c. le gall , a. brunetti , h. boukari , l. besombes , phys . rev . b * 85 * , 195312 ( 2012 ) . a. o. govorov , a. v. kalameitsev , phys . b * 71 * , 035338 ( 2005 ) . s. moehl , f. tinjod , k. kheng , h. mariette , phys . b * 69 * , 245318 ( 2004 ) .

the rich variety of conformations which leads to many different intrinsic properties of polymer solutions has continuously drawn considerable interest in soft matter research . computer modeling is increasingly being used as an integral part of theoretical study , in order to both test existing theories and to trigger the development of new concepts . furthermore , computer simulations have also become an essential tool in materials research , especially for predicting and understanding the behavior of complex systems , where a complete theory is not available . it has been proven to be an effective and inexpensive way to study these systems . in order to observe large - scale properties , it is crucial to reduce the computational cost by coarse - graining the details of the atomic structure . this is particularly true for polymer systems and studies of their universal static and dynamic properties @xcite . in this context , using a conventional bead - spring chain model to represent a polymer molecule in molecular dynamics ( md ) simulations is usually sufficient . in the case of dilute and semidilute polymer solutions , a correct model also needs to take into account the effect of solvent molecules . this effect is two - fold : on the one hand , the good solvent quality results in swelling of the random coil ; on the other , the solvent - mediated long - range dynamic correlations between different segments of the chain , known as hydrodynamic interactions ( hi ) , significantly influence the dynamical behavior @xcite . in the present methodological study , we focus on the dilute regime which is theoretically most thoroughly understood . in order to capture hydrodynamic interactions in md simulations , the solvent particles need to be incorporated explicitly . typically , the number of solvent particles required for such a model is of the order of thousands even for a short chain . although such studies are feasible @xcite , they are rather inefficient , for this reason . therefore , a more coarse - grained description of the solvent is highly desirable . two complementary approaches have been developed to do this . `` mesoscopic '' methods keep the solvent degrees of freedom , but describe them in a simplified fashion . these include dissipative particle dynamics ( dpd ) , multi particle collision dynamics ( mpcd ) @xcite , and lattice boltzmann ( lb ) . these approaches are typically one to two orders of magnitude faster than md @xcite . conversely , brownian dynamics ( bd ) simulations remove the solvent degrees of freedom completely , and take their effect into account via non - trivial long - range dynamic correlations in the stochastic displacements . this is possible due to the time scale separation between the fast solvent motion and the slow conformational polymer degrees of freedom . since the number of degrees of freedom is reduced drastically , the method has the potential to save cpu time by additional several orders of magnitude , in particular in the dilute limit . however , a simple implementation of the correlations @xcite leads to an algorithm which scales like @xmath3 , where @xmath4 is the number of brownian particles , and therefore becomes infeasible as soon as @xmath4 exceeds a few hundred @xcite . it is therefore very important to treat hi by means of fixman s algorithm @xcite ( scaling roughly as @xmath5 ) , which we do in the present study . the recently introduced method by geyer and winter @xcite would reduce the necessary cpu effort by roughly one additional order of magnitude for typical chain lengths , and exhibit a more favorable scaling ( @xmath6 ) . however , it is based upon an inexact approximation of the hydrodynamic correlations that can not be improved systematically ( in contrast to fixman s method ) . for this reason , this algorithm was not implemented . none of these approaches is sufficient to reach @xmath7 ; this latter goal is only attainable by the implementation of very recent `` superfast '' bd algorithms based upon fast fourier transforms . these latter algorithms scale as @xmath8 , where the exponent @xmath9 depends on the details of the underlying physics , and is usually substantially smaller than unity . these methods require the study of a confined system , and hence are not used in the present study . while the advantages and disadvantages of the methods are well - known in general terms ( and have resulted in differing methodological preferences in different groups of researchers ) , not much is known _ quantitatively _ in terms of a clear comparison of computational efficiency . the present paper aims at partly filling this gap . recently , one of the present authors @xcite has proposed a new mesoscopic method for simulating polymer - solvent systems . the solvent is represented by a fluid on a grid , simulated via the lattice boltzmann approach , while the motion of the polymer chain is governed by a continuous md model . the two parts are coupled by a simple dissipative force . the lattice boltzmann ( lb ) method was originally developed to simulate hydrodynamics on a grid @xcite . it has been shown that the lb method is a fast and effective method for simulating fluid flows , which has the same speed and accuracy as other navier - stokes solvers @xcite . ladd @xcite successfully applied the lb method to colloidal systems ( originally with a _ conservative _ coupling ) and showed that the cpu cost scales linearly with the number of particles . moreover , he showed how fluctuations can be incorporated into the lb model , which is essential in order to investigate brownian motion @xcite . this procedure has recently been refined and improved @xcite . the dissipative coupling method @xcite was thoroughly tested by applying it to a single polymer chain in solution , for which the data of a previous md simulation @xcite were available , and whose parameters were used as an input for the mesoscopic model . in this work , we study the dynamics of a single chain in a solvent to compare the predictions of the explicit solvent model via the lb method with the predictions of the implicit solvent model by bd simulations . many recent simulation studies have investigated very similar systems , using various mesoscopic solvent models like mpcd @xcite , dpd @xcite , direct solution of the navier - stokes equation @xcite , and smoothed dpd @xcite . these studies were mainly done in order to test and verify the validity of the chosen mesoscopic simulation approach . meanwhile it is fair to say that a single chain in solution is a standard benchmark test bed for mesoscopic simulation methods , the necessary condition for passing being the correct reproduction of the zimm scaling laws within the limitations of finite chain length and finite solvent volume . the present study , however , aims directly at a _ quantititave comparison _ between two different methods , and therefore it is crucial that that the underlying polymer model is exactly identical for both methods . this is the reason why the data of refs . @xcite , though all exhibiting essentially the same physics , are of no direct relevance for the present investigation , since all of them employ slightly different polymer models , and simulate solvents with somewhat different viscosities . one very recent study by ladd et al . @xcite has done a rather similar comparison between lb and bd , and arrived at similar conclusions ; the present paper should be viewed as a complementary study that puts some more emphasis on the issue of computational efficiency . after introducing the models , we show how to map the input parameters of the hybrid model onto the input values of the bd model to directly compare the predicted quantities ( sec . [ sec : models ] ) . section [ sec : results ] then confirms the expected physical equivalence of the two approaches in terms of comparing static and dynamic data . furthermore , this section also presents our comparison on the numerical cost or benchmark data for both methods . finally , in sec . [ sec : conclusions ] we summarize the results and give our conclusions . in this work , a polymer molecule is represented by a conventional bead - spring chain model , which consists of @xmath4 beads that are connected via @xmath10 finitely extensible nonlinear elastic ( fene ) massless springs . the lennard - jones potential , which acts between all monomers , is used to model the excluded volume ( ev ) effect . the two potentials @xmath11 and @xmath12 are given by the expressions @xmath13 where @xmath14 is the bead - bead distance , @xmath15 is the spring constant and @xmath16 the maximum extension of the bond . @xmath17 and @xmath18 are the energy and length parameters of the lennard - jones potential , respectively . in this method , the evolution of the lb variables @xmath19 is governed by the following lattice boltzmann equation @xcite : @xmath20 the variable @xmath21 is the ( partial ) fluid mass density at grid site @xmath22 at time @xmath23 , corresponding to the discrete velocity @xmath24 . @xmath25 is the time step , and the lattice spacing is denoted by @xmath26 . the small set of velocities @xmath24 ( @xmath27 , where the value of @xmath28 depends on the details of the model ) is chosen such that @xmath29 is a vector leading to the @xmath30th neighbor on the grid . @xmath31 is a collision operator for dissipation due to fluid particle collisions , such that the populations always relax toward the local pseudo - equilibrium distribution @xmath32 that depends on the local hydrodynamic variables @xmath33 ( the total mass density ) and @xmath34 ( the local flow velocity ) . the collision process is constructed in such a way that it conserves both @xmath35 and @xmath36 . @xmath37 is the stochastic term , which is essential in order to simulate thermal fluctuations that drive brownian motion . the local pseudo - equilibrium distribution can be represented as a second - order expansion of the maxwell - boltzmann distribution , given by @xcite @xmath38 where @xmath39 are a set of weight factors , which depend on the sublattice @xmath30 ( i. e. the magnitude of @xmath24 ) and @xmath40 is the speed of sound . in this work , we have used the algorithm proposed in ref . @xcite , however , with the modification that the original 18-velocity model ( d3q18 ) was replaced by the d3q19 19-velocity model @xcite . the set of @xmath24 consists of the particle being at rest , the 6 nearest and 12 next - nearest neighbors on a simple cubic lattice . the magnitudes of the velocities corresponding to these three sets of particles are @xmath41 , and @xmath42 , respectively . the weight factors for the d3q19 model are @xmath43 , @xmath44 and @xmath45 . the early model of ref . @xcite only considered the thermalization of modes related to the viscous stress tensor . it is important to note that even though this procedure is correct in the hydrodynamic limit , it provides poor thermalization on smaller length scales @xcite . adhikari et al . @xcite have shown that by applying thermalization to all nonconserved modes one gets a significantly improved numerical behavior at short scales ; the theoretical background is now thoroughly understood @xcite . in this work , we have also investigated the effects of thermalization of the kinetic modes on various dynamic properties . the coupling to the beads is done via simple interpolation of the flow velocity from the surrounding sites , and by introducing a phenomenological stokes friction coefficient @xmath46 of the beads . this gives rise to a friction force on the particles , plus a langevin force that balances the frictional losses . the total momentum is conserved by subtracting the corresponding momentum transfer from the surrounding fluid . it can be shown that this procedure satisfies the fluctuation - dissipation theorem @xcite . for further technical details on this method and its theoretical analysis , we refer the reader to ref . @xcite . the configuration of a bead - spring chain is specified by the set of position vectors @xmath47 ( @xmath48 ) . the time evolution of this configuration is governed by the it stochastic differential equation @xmath49 where summation over repeated indices is implied . here the symbols @xmath50 , @xmath51 , @xmath52 , and @xmath53 denote the time step , boltzmann s constant , the temperature , and the diffusion tensor , respectively , where the latter decribes the hydrodynamic interactions between the beads . the forces @xmath54 and @xmath55 are the spring and excluded - volume contributions , repectively . @xmath56 are random variables representing a discretized wiener process , such that @xmath57 and @xmath58 , where @xmath59 is the kronecker delta and @xmath60 is the unit tensor . finally , the tensor @xmath61 is related to the diffusion tensor such that @xmath62 @xcite . the frictional properties of the chain and the hydrodynamic interactions between the beads are modeled via the diffusion tensor @xmath53 . its diagonal elements contain the bead friction coefficient @xmath63 , where @xmath64 is the solvent viscosity and @xmath65 is the stokes radius of the bead . the off - diagonal elements represent the hydrodynamic interactions via the tensor @xmath66 , for which we take the regularized rotne - prager - yamakawa ( rpy ) tensor @xcite with solvent viscosity @xmath64 and stokes radius @xmath65 . taken together , the diffusion tensor is given by @xmath67 further details can be found in ref . @xcite . the computationally most intensive part is to determine the matrix @xmath61 . generally , cholesky decomposition of @xmath53 is used to obtain @xmath61 as an upper ( or lower ) triangular matrix , and the computational cost for this method scales as @xmath68 @xcite . fixman made use of the fact that there are many possibilities to define @xmath61 as some square - root matrix of @xmath53 , and , noting that it is the vector @xmath69 that is required rather than the matrix @xmath61 , applied a truncated chebyshev polynomial expansion to obtain @xmath69 with a lower computational cost , scaling roughly as @xmath70 @xcite . in the present paper , we make use of this accelerated technique as well . for implementation on a computer , physical quantities must be represented in certain units , i. e. in terms of suitable dimensionless ratios . this is typically done by choosing a natural unit system where three independent elementary quantities are set to unity . in the coupled md / lb simulation approach , this is usually done by choosing the lennard - jones parameters @xmath17 , @xmath18 , and @xmath71 ( @xmath72 , where @xmath73 is the mass of the monomer ) as the units of energy , length , and time , respectively ; this choice enables one to make direct contact with md simulations with explicit solvent @xcite . conversely , bd simulations have traditionally @xcite used a unit system where one chooses @xmath74 as the energy unit and @xmath75 as the length unit . this is particularly useful for the simulation of pure gaussian ( harmonic ) chains where the interaction potential has neither an energy scale nor a length scale built in . the time unit in bd simulations is usually chosen as @xmath76 . of course , a meaningful comparison of results requires that all data are represented in one common unit system . at this point , one realizes that this is less straightforward than one might think at first glance . while the conversion of length and energy units is trivial , and directly facilitated by the fact that both methods use the same molecular model for the polymer chain , and perform the simulations at the same temperature , the conversion of time units is not . this is so because of the different time scales underlying the basic updating algorithms : the md / lb method is based upon simulating the system on _ inertial _ time scales , while bd focuses directly on the larger _ diffusive _ time scales . this is directly reflected by the occurence of the intertial parameter @xmath73 ( monomer mass ) in the md / lb time unit , which does occur in the bd model , and the diffusive parameter @xmath77 ( monomer friction constant ) in the bd time unit . it is important to notice that @xmath77 is _ not _ an input parameter to the md / lb model . rather , one needs to carefully distinguish between the short - time friction coefficient @xmath46 that is indeed an input parameter it describes the stokes coupling of the monomer to the lb fluid in its immediate vicinity and the long - time friction coefficient @xmath78 describing the particle s long - time response that is modified by solvent backflow effects . it is this latter parameter that must be identified with the bd @xmath77 , and it is esentially an _ output _ parameter . fortunately , the relation between @xmath46 and @xmath78 is well understood ahlrichs and dnweg @xcite have shown that @xmath79 where @xmath80 is a numerical prefactor and @xmath26 is some measure for the range of interpolation to the surrounding lattice sites . for linear interpolation to the nearest sites , one finds @xmath81 , if @xmath26 is the lb lattice spacing . for highly accurate results , one should also take into account a small correction for the finite size of the simulation box @xcite ; this has however not been done in the present paper . rather , we took the mapping determined in ref . @xcite to calculate @xmath78 from @xmath46 and identified this with the @xmath77 parameter of the bd calculations . although the physical mapping done in this way is essentially correct , it is important to notice that this aspect introduces a certain amount of numerical inaccuracy when it comes to quantitative comparisons . given the fact that it is intrinsically impossible to run the two simulations with the same unit system , we chose to keep the previously established systems of the two respective methods , and to map the results _ a posteriori _ by the procedure outlined above . furthermore , we chose to present all results in md / lb units . technically , this means that for length and time unit conversions we need to set @xmath82 and @xmath83 , where the `` * '' superscript denotes bd non - dimensionalization , while `` - '' denotes a non - dimensionalization for md / lb . the corresponding factors for the conversion from bd to md / lb units are then trivially found to be @xmath84 note that @xmath85 and @xmath86 are @xmath15 and @xmath87 in non - dimensional md / lb units , respectively , while the ratio @xmath88 is just the non - dimensionalized temperature in md / lb units . the dimensionless hydrodynamic interaction parameter @xmath89 used in the bd simulations is essentially a non - dimensionalized stokes radius . we thus find @xmath90 or @xmath91 we therefore parameterized our simulations by first picking simulation parameters for the md / lb model ( which were then directly used for md / lb ) , then converting these to bd units using the procedure outlined above , and then running the thus - obtained equivalent bd model . for these latter simulations , a time step size @xmath92 ( in bd units ) was found to produce sufficiently accurate results . the physical input values for the present model are chosen from the benchmark values developed in ref . @xcite , which have been shown to reproduce the results of a typical pure md simulation @xcite . as in the comparison between lb and md simulations , we study a system of a single polymer chain of length @xmath93 monomers immersed in a fluid with temperature @xmath94 , density @xmath95 , and kinematic viscosity @xmath96 . the lattice spacing @xmath97 is set to unity , which is roughly identical to the bond length ; this is necessary to resolve the hydrodynamic interactions on small length scales with sufficient accuracy . furthermore , following ref . @xcite , the coupling parameter @xmath98 was set to @xmath99 . the values of the fene spring potential parameters are @xmath100 and @xmath101 . the time step size for the polymer ( the md part of the simulations ) is set to @xmath102 . it should be noted that such a large time step is possible since the inclusion of dissipation and noise lead to a substantial stabilization , compared to purely microcanonical md . the value of the time step that updates the fluid should be chosen in a way such that the lb variables @xmath19 do not become negative too often . here , we choose @xmath103 , where such a case rarely occurred during the observation time . it is important to mention another free input parameter which governs the time scale for the evolution of hydrodynamic interactions , known as the schmidt number @xmath104 , where @xmath105 is the diffusion constant of the single monomer . this parameter can be set arbitrarily in the lb method by choosing @xmath106 and @xmath105 ( which can be tuned by choosing @xmath107 ) accordingly . ideally , the value of @xmath108 should be chosen such that hydrodynamic interaction evolves much faster than the diffusion of a monomer . in our case , we have @xmath109 , which has been shown to result in zimm - like behavior @xcite . for the lb simulations , the polymer chain moves within a cubic box of length @xmath0 with periodic boundary conditions , while it is drifting freely in an infinite medium for the bd simulations . in order to accurately compare various properties between the two systems , one must understand the effects of the box length @xmath0 on any observable of interest in the lb simulations . thus it is essential to only compare quantities under identical conditions ( i. e. independent of the box length ) . hence various box lengths @xmath0 ranging from 10 to 35 lennard - jones units were investigated ; this allows us to extrapolate to the @xmath2 limit . the mean square radius of gyration and the mean square end - to - end distance are given by @xmath110 with @xmath111 being the inter - particle distance . these two quantities are both related to the number of monomers by the expression @xmath112 where @xmath106 is the flory exponent . for a self - avoiding walk ( saw ) , the flory exponent @xmath106 is 0.588 @xcite . in principle , @xmath106 can be obtained from simulations , using the scaling law in eq . [ eq : flory ] . however , this method would require simulations for a wide range of @xmath4 values . alternatively , one can use the static structure factor @xmath113 to obtain @xmath106 much more efficiently . in the scaling regime @xmath114 ( @xmath115 being a microscopic length of the order of the bond length ) , a power law relation between the static structure factor and the wave vector @xmath116 holds : @xmath117 figure [ fig : skstat ] shows the static structure factor as a function of wave vector @xmath116 for the lb simulations with the presence of thermalization of all modes , and the bd simulations . it can be clearly seen that the values of the static structure factor obtained from the lb simulations are exactly the same as those obtained from the bd simulations , indicating that they have the same static conformations . from eq . [ eq : svsk ] , the value of @xmath106 can be extracted from the linear region of the log - log plot of @xmath118 vs @xmath116 . as expected , the values for @xmath106 obtained via this method are the same for both the lb and the bd simulations , as reported in table [ tab : props ] . however , they are approximately 5% higher than the asymptotically correct value , which is a consequence of the finite chain length . the results for the mean square radius of gyration and the mean square end - to - end distance in table [ tab : props ] further confirm this agreement with regard to static conformations between the two methods . however , at small box length ( @xmath119 ) , the results for these static properties for the lb method deviate from their asymptotic values . the discrepancy observed here always arises when the box length is too small compared to the chain size , where the chain is more likely to wrap over itself due to spatial restriction and hence alter its static conformations . we also found that the two versions of lb thermalization ( `` stresses only '' vs `` full '' thermalization ) yield identical results for the chain conformational statistics . in general , we only quote values obtained for full thermalization , unless indicated otherwise . the hydrodynamic radius for a single chain in an infinite medium is given by @xmath120 for a chain in a finite box , as is the case here in the lb method , it has been shown that the hydrodynamic interactions of the chain with its periodic images effectively increases @xmath121 @xcite . in order to account for this finite - size effect , a finite - size correction of order @xmath1 for most dynamic properties , resulting from the slow @xmath122 decay of hydrodynamic interactions , is required @xcite . the results for the infinite - box value @xmath123 agree excellently with each other for all simulations ( see table [ tab : props ] ) . since the overwrapping effect is more sensitive to large inter - particle distances , it turns out that the deviation in the inverse hydrodynamic radius is too small , for the range of box lengths used , for it to be distinguishable . as can be seen in table [ tab : props ] , the deviation is more pronounced for the radius of gyration , and even more for the end - to - end distance . according to dynamic scaling , the longest relaxation time @xmath124 of the chain is , by order of magnitude , identical to the time that the chain needs to move its own size , i. e. @xmath125 , where @xmath126 is the diffusion constant of the chain s center of mass . this leads to a dynamic scaling law @xmath127 , where @xmath128 is the dynamic scaling exponent . for a chain with hydrodynamic interactions , this relaxation time is known as the zimm time @xmath124 . for this case , @xmath129 in the limit of long chains . this implies that @xmath130 , which gives a dynamic exponent of @xmath131 for models with hi . for the rouse model ( i. e. chains without hydrodynamic interactions ) , where @xmath132 , one finds a dynamic exponent of @xmath133 . these quantities will be referred to in the discussion below . the mean - square displacement of the chain s center of mass @xmath134 for both methods is depicted in fig . [ fig : g3 ] . from the figure , it can be clearly seen that @xmath135 strongly depends on the box length @xmath0 for the lb simulations . moreover , they seem to converge to the value predicted by the bd simulations ( @xmath136 ) in the limit of large @xmath0 . effects of thermalization of the kinetic modes in lb simulations on this property will be discussed subsequently . the chain s center of mass diffusion constant @xmath126 can be determined by the slope of the @xmath135 vs @xmath23 curve , where the relationship @xmath137 holds . by fitting a power law to the simulation data , we obtain the exponents and the diffusion constants shown in table [ tab : props ] . these exponents support the prediction of simple diffusive behavior ( @xmath138 ) . theoretically , one would expect that two diffusive regimes exist : on the one hand , there should be a short - time diffusive regime , corresponding to time scales well below the zimm time , @xmath139 , but also well above the ballistic regime , @xmath140 ; note that @xmath141 only in the lb case , since the bd equation of motion is overdamped . on the other hand , there should be free diffusion for times @xmath142 . both these regimes exhibit @xmath138 behavior , but with different prefactors , with a smooth crossover around the zimm time @xcite . in principle these two different diffusion constants can be obtained via fits to the corresponding regimes . in practice , however , it turns out that the values are very close to each other , and hence the crossover is very smooth @xcite . therefore its unambiguous identification is very difficult , i. e. impossible within the resolution of our data . the mean - square displacement of a single monomer @xmath30 is given by @xmath143 here , only the two innermost monomers near the center of the chain are evaluated to eliminate end effects ; the results are plotted in fig . [ fig : g1 ] . the values of @xmath144 behave similarly to those of @xmath135 . in the sub - diffusive time regime , corresponding to the short - time diffusive regime for @xmath135 , here evaluated between @xmath145 and @xmath146 , the scaling behavior @xmath147 is predicted @xcite . the corresponding exponents obtained from a power - law fit are listed in table [ tab : props ] and indicate a value of @xmath148 as @xmath149 . regardless of the finite - size corrections due to the box length and the effects of thermalization , these values clearly favor the zimm model compared to the rouse model , which predicts @xmath150 . the deviation from the asymptotic zimm value @xmath151 ( or @xmath152 ) is mainly a result of finite chain length . figure [ fig : g2 ] shows the mean - square displacement of a single monomer in the center of mass system ( i. e. the two innermost monomers to eliminate end effects ) @xmath153 \\ & - & \left [ \mathbf{r}_i(t_0 ) - \mathbf{r}_\text{cm}(t_0 ) \right ] ) ^2 \rangle . \nonumber \label{eq : g2}\end{aligned}\ ] ] interestingly , when viewed within the center of mass system , all the results lie on top of each other , regardless of the box length @xmath0 . this result also holds for lb simulations without full thermalization . this shows that the global center - of - mass motion of the chain is actually the primary contribution to the deviations between lb and bd results . in the data of fig . [ fig : g2 ] this contribution is suppressed : in terms of rouse modes , only the internal modes remain . for these modes , however , it has been shown @xcite that the hi with the periodic images is much weaker , while the leading - order @xmath122 hi cancels out . therefore , the corresponding finite - size effect scales as @xmath154 instead of @xmath1 , and this is so small that it is invisible in fig . [ fig : g2 ] . theoretically , these data can also be used for estimating the zimm time as the time where the crossover to the long - time plateau occurs . however , the crossover is quite extended and smooth , making it difficult to extract . we therefore estimated the zimm time via @xmath155 strictly speaking , this definition is only valid for a single chain in an infinite medium , where there is no finite box size effect . in the presence of finite box size , it becomes the definition for the _ translational _ time ( @xmath156 ) rather than the zimm time : the former is subject to an @xmath1 finite - size effect , due to the strong @xmath0-dependence of @xmath157 , while the latter , being defined via the relaxation of internal modes , is only subject to an @xmath154 size effect , as discussed above . the translational times obtained from eq . [ eq : tauz ] ( as shown in table [ tab : props ] ) are indeed different for different box lengths @xmath0 , as expected . conversely , the results displayed in fig . [ fig : g2 ] indicate that the systems with different box sizes have ( essentially ) all the same ( internal - mode ) zimm time , since their data all lie on top of each other . next , we focus on the leading order @xmath1 finite size correction for the long - time diffusion constant of the chain s center of mass , @xmath157 . in principle , a plot of @xmath157 vs @xmath1 should give a straight line for large @xmath0 , and an extrapolation to the limit @xmath2 should yield the same value as predicted by the bd simulations . figure [ fig : dcmvsl ] shows the values of @xmath157 for the lb simulations with and without thermalization of the kinetic modes at various box lengths @xmath0 plotted together with the value obtained from the bd simulation at @xmath158 . it is worth mentioning that the bd value of @xmath157 can be obtained from the mean square displacement of the chain center of mass or via fixman s expression @xcite . the latter method has been shown to produce a much more reliable result and is easier to carry out @xcite . the value reported here has been cross checked by both methods and the results are almost the same within error bars . for the lb simulations without thermalization of the kinetic modes , the value of @xmath157 at the asymptotic limit @xmath158 is different from that predicted by the bd simulations by about @xmath159 . however , when all the kinetic modes in the lb simulations have been thermalized , the deviation in @xmath126 reduces to @xmath160 . this result clearly indicates that it is very important to thermalize all the kinetic modes in order to obtain correct values for dynamic properties . the reason for the remaining small discrepancy between lb and bd is not completely clear , since there are numerous possible sources . firstly , it should be noted that the underlying equations of motion are quite different : lb works with inertia , while bd employs overdamped dynamics . this results in different schmidt numbers @xmath108 and different mach numbers @xmath161 , the latter being defined as the ratio of the flow velocity to the speed of sound : both are finite in the lb method , while in the bd case they are strictly infinite ( @xmath108 ) and zero ( @xmath161 ) , respectively . furthermore , the shape of the hi function at small interparticle distances is somewhat different for the two methods : in the bd case , we employ the rpy tensor , while the nearest - neighbor interpolation for lb results in a short - range hi that differs somewhat from the rpy tensor ( see also the discussion in ref . finally , it should be noted that the value of the constant @xmath80 in eq . [ eq : zeta_eff_vs_zeta_bare ] , which is crucial for the mapping between the lb friction parameter @xmath46 and the bd friction @xmath78 , is only known with some numerical inaccuracy . for highly accurate mappings , it is also necessary to include a finite - size correction in the definition of @xmath80 @xcite ; this was not done in the present study . in order to examine whether the thermalization of the lb kinetic modes is also important for the internal modes of chain motion , we have performed a rouse mode analysis . the rouse modes for a discrete chain are defined as @xcite @xmath162 \label{eq : rmode}\ ] ] for @xmath163 . within the approximation of the zimm model , the autocorrelation function of the modes should decay exponentially @xcite @xmath164 where @xmath165 is the relaxation of the @xmath166-th mode . to validate our rouse mode analysis routine , we have carried out extensive simulations for a ( gaussian ) rouse chain of @xmath167 in the absence of hi and ev ; the results for @xmath165 are in excellent agreement with the analytical predictions @xcite . figure [ fig : rmodeall ] shows the normalized autocorrelation function for @xmath168 for the lb model with box length @xmath169 , and the bd model . for nonzero times , there is a small deviation between lb and bd , the latter exhibiting again a slightly faster dynamics . this deviation systematically becomes smaller upon increasing the mode index @xmath166 . since high mode index means essentially relaxation on a rather small length scale , it is tempting to attribute the deviation to the finite propagation of hi in the lb model , i. e. to retardation effects , or effects of finite schmidt number , which are more important on large length scales than on small ones . nevertheless , this hypothesis is not proven . in ref . @xcite it was shown that the autocorrelation function is only subject to an @xmath154 finite size effect , in contrast to the usual @xmath1 behavior . figure [ fig : r1modevsl3 ] shows the value of the autocorrelation function of the first rouse mode @xmath170 at a fixed finite time @xmath171 , for lb simulations at various box lengths @xmath0 and bd simulations at @xmath172 . within our numerical resolution , the data indeed confirm this @xmath154 finite size effect , both with and without thermalization of the kinetic modes . furthermore , they demonstrate again that thermalization of all the kinetic modes in lb simulations improves the accuracy of the dynamic properties and brings them closer to the bd prediction : the deviation in the extrapolated limit @xmath2 is reduced from @xmath160 down to @xmath173 . the reasons for the remaining discrepancies are probably of the same nature as in the case of @xmath157 . we have also evaluated the dynamic structure factor , which is defined as @xmath174 \right ) \right > .\ ] ] when both wave number and time are in the scaling regime ( i. e. @xmath175 and @xmath176 ) , @xmath177 is predicted @xcite to exhibit the scaling behavior @xmath178 a plot of @xmath179 against @xmath180 should collapse to a single curve @xcite . the results for both methods are shown in fig . [ fig : skz ] . the data were restricted to the scaling regime @xmath181 and @xmath182 . these ranges were obtained from the single monomer mean - square displacement , fig . [ fig : g2 ] , and from the static structure factor , fig . [ fig : skstat ] , respectively . here , we particularly focus on adjusting the exponent @xmath128 such that it would produce the best total data collapse for a chain in an infinite medium ( i. e. in the bd model ) . obviously , the results from the simulations show zimm - like rather than rouse - like behavior . even though we have suppressed the finite box size effect , a dynamic exponent of @xmath148 yields the best data collapse , which is somewhat smaller than the correct asymptotic one . this result is also consistent with the value of @xmath128 obtained earlier via the exponent of @xmath144 in the sub - diffusive scaling regime ( i. e. @xmath183 ) . the deviation from the asymptotic value is due to the finite chain size used here , and one can expect @xmath151 only in the long chain limit @xmath184 . more detailed comparisons of the structure factor @xmath185 are shown in fig . [ fig : skallbd ] ( @xmath186 dependence at constant time ) and fig . [ fig : stbyst0 ] ( time dependence for the _ normalized _ structure factor @xmath187 at constant @xmath186 ) . figure [ fig : skallbd ] shows the structure factor for bd simulations for a wide range of @xmath186 at three different times and the data clearly indicate that the structure factor decays rapidly with time . the normalized structure factor @xmath187 for three different @xmath186 values are shown in fig . [ fig : stbyst0 ] , and the data seem to indicate that the lb results approach the bd data as @xmath0 is increased , as expected . for the ultra - dilute system considered here , the lattice boltzmann part of the hybrid lb method uses up most of the computational resources as the cpu cost for the md part for the polymer chain is negligible . since the dynamic properties predicted by the lb model are subject to a finite - size correction of order @xmath1 , extrapolation is required to obtain these properties in the asymptotic limit @xmath149 . to perform this extrapolation , together with checking that indeed the asymptotic @xmath1 behavior has been reached , one needs the results of at least three different box lengths . moreover , the box length should be large enough compared to the chain size such that it does not alter the static properties . the data displayed in table [ tab : props ] indicate that it is safe to choose @xmath0 such that @xmath188 . the three different box lengths @xmath0 chosen here are @xmath189 , and @xmath190 . in this work , we set the total cpu time required for the lb simulations to be the sum of all the cpu times required to run 1000 md time steps for each of the chosen box lengths . for bd , we take the cpu time needed to observe the system for the same time span in physical units . each of the simulations performed for the cpu time comparison was run on an itanium 2 processor of a 1.6 ghz sgi altix server 3700 . all the parameters used to carry out this comparison are the optimal values for both methods . several chain sizes ranging from @xmath191 to @xmath192 have been used to obtain the cpu cost for comparison . the results are shown in fig . [ fig : cpu ] . for the lb method , it is clear that the cpu cost scales linearly with the number of particles , i. e. the number of grid points that the solvent lives on , or @xmath193 . since the ratio @xmath194 is kept constant , or @xmath195 , this leads to a cpu cost scaling as @xmath196 . this is indeed found in our benchmarks , see fig . [ fig : cpu ] . similarly , our data also confirm the predicted @xmath70 cpu cost scaling for bd . though the lb exponent is lower than bd , the large prefactor ensures that the total cpu cost for lb is much more expensive compared to bd for the typical chain lengths used in the literature . it is only when the chain length is excessively large ( i. e. @xmath4 of the order of @xmath197 or higher ) that lb will become superior to bd for a single - chain system . the situation completely changes if one studies a semi - dilute solution instead , as has been done in ref . for such a system , we have not done a comparison between lb and bd in terms of actual simulations ; however , by means of scaling considerations one can roughly estimate what the likely result of such a comparison would be . a semidilute solution comprises @xmath198 chains of @xmath4 monomers each , such that the total number of monomers is @xmath199 . therefore the bd cpu cost scales as @xmath200 , while the lb cpu cost depends on the density . within the blob picture of semidilute solutions , one views a chain as a sequence of `` blobs '' , each comprising @xmath201 monomers , and having size @xmath202 , which can be viewed as the typical correlation length of density fluctuations , or the typical distance from which point on chain - chain interactions become important . since the conformation statistics within the blob is that of a self - avoiding walk , one has @xmath203 , where @xmath26 is the monomer size . the sequence of blobs forms a random walk , hence @xmath204 . this gives the minimum size of the simulation box , i. e. @xmath205 , or @xmath206 . we thus see that the cpu effort for the lb method is even slightly _ decreased _ by the factor @xmath207 compared to the single - chain case at the same @xmath4 , due to the shrinkage of the chains resulting from excluded - volume screening . in order to estimate the number of chains @xmath198 , we note that the arrangement of blobs is space - filling , i. e. @xmath208 . comparing this with the previous expression for @xmath193 , one finds @xmath209 . therefore the bd effort , compared to the single - chain case , is increased by a factor of @xmath210 . taken together , this means that the ratio between lb effort and bd effort is changed by a factor of @xmath211 in favor of lb . for @xmath212 , which is needed as a minimum to resolve the gaussian statistics of the chains as a whole , one obtains a factor of @xmath213 , which more or less compensates the two orders of magnitude seen in fig . [ fig : cpu ] . taking into account that for such a system the bd simulation would have to calculate the hi with the periodic images , e. g. , via ewald sums , which is much more complicated than the present single - chain simulation , one sees that for a semidilute solution clearly lb is more efficient , unless a `` superfast '' bd algorithm @xcite is used . for the latter case , the answer is not yet known . the results of ref . @xcite indicate that lb / md may be favorable for a rather small number of monomers ; however , this study was done ( i ) in a non - trivial geometry , which implies a more complicated bd method , and ( ii ) under complete neglect of thermal fluctuations in the lb simulations , resulting in a substantial reduction in cpu effort . such a ( partial or complete ) neglect of thermal noise is sometimes justified in strong nonequilibrium situations such as studied in ref . @xcite ; in that particular case , the justification was checked by additional tests @xcite . another possible situation where lb noise is negligible is the case of strong coarse - graining , where a single lattice site can already be considered as a macroscopic thermodynamic system ( for a detailed discussion , see ref . however , in the general case , and certainly in thermal equilibrium or weak nonequilibrium , the proper inclusion of thermal noise is necessary , as demonstrated theoretically in detail in ref . @xcite , and also corroborated by the present numerical results . for the general case , the estimate of the lb cpu effort given in ref . @xcite is therefore too optimistic . the present study has shown that brownian dynamics simulations are capable of reproducing various properties predicted by a hybrid lb / md model ( or vice versa ) . we have demonstrated how to obtain the input values for the bd simulations from the physical input parameters of the lb model such that both models would produce the same static and dynamic properties . for the lb model , most dynamic properties are subject to a finite - size correction of order @xmath1 . in addition to this , it is very important to thermalize all the kinetic modes in order to obtain the correct dynamic properties . those results that are not affected by @xmath1 finite size effects , such as the mean square displacement in the center of mass system , or the rouse mode autocorrelation function , agree very favorably with each other . for highly dilute systems where the simulation of a single chain is sufficient , bd is usually the method of choice , as it is much more efficient than the coupled lb / md approach , and finite box size effects are absent . the situation changes however in the semidilute case , where it is easy to estimate that bd will not be able to compete , unless `` superfast '' algorithms are used . moreover , one should take into account that the hybrid lb / md algorithm is rather easily adaptable to complicated boundary conditions , and can even be applied to flows at high reynolds numbers , where the fluid degrees of freedom become _ intrinsically _ important , and can not be handled in terms of a green s function . this work was supported by the australian research council under the discovery projects program and the mainz max planck international research school . computational resources were provided by the australian partnership for advanced computation ( apac ) and the victorian partnership for advanced computation ( vpac ) . stimulating discussions with a. j. c. ladd are gratefully acknowledged . .properties for a single chain of length @xmath214 obtained from lattice boltzmann simulations at various finite box lengths and brownian dynamics simulations in infinite medium . @xmath215 exponent obtained by fitting a power law in the sub - diffusive scaling regime of the chain in lattice boltzmann simulations , @xmath216 $ ] . @xmath217 exponent obtained from lattice boltzmann simulations without thermalization of all the kinetic modes . [ cols="<,^,^,^,^",options="header " , ]

glassy systems like spin - glasses and structural glasses below the glass transition temperatures are out of equilibrium even on the macroscopic time scale . thus , the slow dynamics of glassy systems has been a subject of continuous interest in the past years @xcite . experimentally , the anomalous dynamical behaviors are characterized by slow relaxation with long time tail and aging phenomena . among them , aging phenomena are the most striking dynamical behaviors as follows : two - time quantities like the correlation functions @xmath2 explicitly depend on the time elapsed after the quench @xmath3 ( the waiting time ) . if the waiting time @xmath3 is of microscopic time scale , the phenomena are merely transient on the way of relaxation to equilibrium . however , dependence on the waiting time continues even when @xmath3 is so large that one - time quantities like the magnetization are asymptotically close to time - independent values @xcite . since these phenomena mean that the dynamics is not stationary , aging is a sign showing that these systems are out of equilibrium even in macroscopic time scale , i.e. several days or weeks . these aging phenomena appear also in mean - field models of spin - glasses and do not disappear even in the infinite waiting time limit @xcite . in addition , the phase space of these models decomposes into a large number of areas separated with infinitely high free energy barriers . thus , glassy systems never reach true equilibrium and hence they are beyond the scope of thermodynamics and equilibrium statistical mechanics . hence , in order to describe thermodynamic properties of glassy systems , out - of - equilibrium thermodynamics based on a dynamical description without the assumption of ergodicity is necessary . in investigation of such anomalous dynamical behavior of glassy systems , it was found that aging phenomena have some universal properties . theoretical analyses have suggested that there are two time regimes characterized by different dynamical properties of the two - time correlation function @xmath2 and the associated linear response function @xmath4 @xcite . in the first time regime where the time difference @xmath5 is short compared to @xmath3 , the dynamics looks stationary and the usual fluctuation - dissipation theorem ( fdt ) holds . on the other hand , in the second time regime where the time difference is comparable to @xmath3 , aging phenomena occur , i.e. @xmath2 depends on @xmath3 apart from dependence on @xmath5 . in addition , it is known that the usual fdt between the correlation @xmath2 and the response function @xmath4 should be modified in a well - defined way which involves the rescaling of the temperature @xcite . the modification was found to be valid not only for mean - field models but also for other glassy systems : spin - glass models with finite - range interactions @xcite , real spin - glasses @xcite , structural glasses @xcite and a model of phase separation @xcite . in addition , it is known for some glassy systems that the correlation function @xmath2 obeys the scaling law that it depends on @xmath6 and @xmath3 only through the value of @xmath7 , where @xmath8 is a system - dependent increasing function of time . aging of the correlation function and the modification of fdt imply that properties of the work done by modulating an external field in an isothermal process are completely different from properties predicted by traditional thermodynamics . in addition , the existence of the two time scales implies that thermodynamic properties must strongly depend on experimental time scales . hence , the experimental time scale dependence of the thermodynamic properties of glassy systems should be investigated to construct out - of - equilibrium thermodynamics for glassy systems . in order to describe our results precisely , we summarize thermodynamics for an isothermal process . thermodynamics tells that when one quasistatically changes the external field the work needed for the change is independent of a path of changing the external field . in addition , the quasistatic work is equal to the change of the helmholtz free energy . when the process is not quasistatic , the work is larger than the quasistatic work . this fact is called the principle of the minimum work and is derived from the second law of thermodynamics . we show that the properties described above do not hold in systems with aging . more precisely , there are two experimental time regimes characterized by different thermal properties . the first regime is a time - domain where the waiting time is much longer than the time lapse of the process . we call the time lapse the measurement time . in this regime , the principle of the minimum work holds even though a system is out of equilibrium . more precisely , when the process is not infinitely slow , the work needed for the process is larger than the work for an infinitely slow process . in addition , value of the work for the infinitely slow process depends only on the initial state and the final state and hence it can play a role of a free energy . the second regime is the experimental time regime where the length of the measurement time are comparable to that of the waiting time . in this regime , for the single - correlation - scale systems such as @xmath0-spin spherical spin - glasses the work done in an infinitely slow operation depends on the path of changing the field even when the waiting time is infinite . this property form a striking contrast to the consequence of traditional thermodynamics described above . on the other hand , for the multi - correlation - scale systems such as sherrington - kirkpatrick model , the work done in an infinitely slow operation is independent of the path . in sec . [ section2 ] , we describe an isothermal process considered in this paper and introduce the two time regimes which characterize the experimental time scales and play a significant role in this paper . in sec . [ section3 ] , we see that in the first time regime , usual thermodynamic relations hold even though the system is out of equilibrium . the only difference is that the value of the work in an infinitely slow operation @xmath9 is different from that of the change of the free energy calculated from equilibrium statistical mechanics . in sec . [ section4 ] , we present general discussion on properties of the work @xmath9 in an infinitely slow operation in the second time regime and derive conditions when @xmath9 depends on the path of changing the external field . in sec . [ section5 ] , by using the results obtained in the previous section , we show that for the single - correlation - scale systems the work in an infinitely slow operation depends on the path of changing the external field as a consequence of aging . possibility of observation of this path - dependence is also discussed . in addition , we show that for the multi - correlation - scale systems @xmath9 is not path - dependent . our results are summarized in sec . [ section6 ] , where implications of our results on thermodynamics of glassy systems and experimental protocols to observe quasiequilibrium properties are discussed . we describe an isothermal process to consider thermodynamic properties when aging occurs . a simple way to observe aging phenomena is through the following field cooling process ( fig . [ protocol ] ) ; the temperature of the heat bath is decreased in a small field @xmath10 to a sub - critical temperature at time 0 . after a waiting time @xmath11 the field begins to change according to a given time - dependence @xmath12 . this change of the field continues for a period of @xmath13 . the initial and final values of @xmath14 are @xmath15 and @xmath16 . we refer to @xmath13 , the time lapse of the change of the field , as the measurement time . the waiting time @xmath11 and the measurement time @xmath13 characterize the experimental time scales . when the field is so weak that the response is linear , the work @xmath17 done on the sample during the process is given in terms of the response function @xmath4 by @xmath18 where the `` magnetization '' @xmath19 is given by @xmath20 since we are interested in long waiting time behavior , the contribution of the first term of eq . ( [ m ] ) is ignored . for feasibility of showing long time behavior , we rewrite eq . ( [ m ] ) by integration by part to express @xmath19 in terms of a susceptibility @xmath21 instead of the response function ; @xmath22 where the susceptibility is defined as @xmath23 in order to discuss the dependence of the work on the waiting time and the measurement time , we rewrite the expression of the work by the transformations : @xmath24 , @xmath25 . we assume that @xmath26 and @xmath27 are finite , in order to exclude the unrealistic cases where the speed and the acceleration of changing the field is infinity . for example , the path such as @xmath28 is excluded , since @xmath26 and @xmath27 are infinite at @xmath29 . thus , the work is reduced to @xmath30 which implies that the dependence of the work on the experimental time scales is determined by that of the susceptibility . since the first term of eq . ( [ w_s ] ) becomes a constant @xmath31 in the long waiting time limit which we are interested in , we will analyze properties of the second term of eq . ( [ w_s ] ) , i.e. the work done in the zero - field cooling process ( @xmath32 ) , in the rest of this chapter for simplicity ; @xmath33 in order to discuss behavior of the susceptibility when aging occurs , we recapitulate the long - time behavior of the response function referred to in the previous section . it is known that there are two time regimes characterized by different behavior of the correlation function @xmath2 and the fdt @xcite ; * at long times @xmath6 and @xmath3 such that @xmath34 , the correlation function is the function of only the time difference @xmath5 , i.e. the time - translational invariance ( tti ) holds . in addition , although the sample is out of equilibrium , the usual fdt holds @xcite as @xmath35 where @xmath36 is defined as @xmath37 . since the properties of two - time quantities in this time regime are the same as that in equilibrium , this time regime is called the quasiequilibrium regime . * whereas , at long and well - separated times such that @xmath38 , aging occurs , i.e. the two - time correlation function @xmath2 depends on @xmath3 even in the long time limit @xmath39 . this implies no tti and this time regime is called the aging regime . in addition , it is known that in the aging regime the fdt is modified as @xmath40}{k_b t}\frac{\partial c(t , t')}{\partial t ' } , \label{generalized_fdt}\ ] ] where the fdt violation factor @xmath41 is a function which depends on @xmath42 only through the dependence of the correlation function @xmath2 @xcite . thus , a system can not be considered to be in a quasiequilibrium state , since the usual fdt is strongly violated . when only one correlation scale exists apart from the quasiequilibrium regime , it is known that the fdt violation factor is a constant . in order to clarify the meaning of the time region of the aging regime ( @xmath38 ) , we give an explicit expression of aging of the correlation function as @xmath43 \equiv \hat{c}(\mu ) , \label{explicit_expression_aging}\ ] ] where @xmath44 is a system - dependent function which characterizes the aging regime and @xmath45 is the time difference @xmath5 . the correlation function @xmath46 depends on the waiting time @xmath3 through the value of @xmath47 . occurrence of aging means that the limiting function @xmath46 takes a non - trivial value such that @xmath48 . here , @xmath49 is the dynamical e - a order parameter . thus , the time region of the aging regime is the time region where @xmath50 is finite . here , we assume that @xmath47 and @xmath51 are positive . this definition of aging is illustrated in terms of contour plot of the correlation function on @xmath52 plain ( fig . [ contour ] ( a ) ) . the plot when aging occurs is completely different from that when aging does not occur ( fig . [ contour ] ( b ) ) . the contour lines give the system dependent function @xmath44 which characterizes the aging regime , since the contour lines are given by @xmath53 in the contour plot . we show two examples of the function @xmath44 and the aging regime for systems like spherical spin - glasses and real spin - glasses . in these systems , it is known that the correlation function behaves as a function of @xmath7 , where the scaling function @xmath8 is a system - dependent function @xcite . the function @xmath44 is given in terms of @xmath8 as @xmath54-t$ ] , where @xmath55 is a constant larger than unity . the inverse function of @xmath8 exists since @xmath8 is a monotonically increasing function . for example , 1 . when @xmath56 , @xmath57 thus , @xmath58 is a function of @xmath47 and hence the aging regime is the time regime where @xmath59 is finite . when @xmath60 ( 0<\alpha < 1)$ ] , @xmath61 thus , @xmath62 is a function of @xmath47 and hence the aging regime is the time regime when @xmath63 is finite . from these results , it is shown by the definition of the susceptibility eq . ( [ chi ] ) that the behavior of the susceptibility depends on the time regimes according to the dependence of the behavior of the response function . in the quasiequilibrium regime , the susceptibility depends only on the time difference @xmath64 and is given by the correlation function as @xmath65 \label{chi_f}\ ] ] where @xmath66 . whereas , in the aging regime , time translational invariance ( tti ) does not hold and the susceptibility is given by the correlation function as @xmath67 where @xmath41 is the fdt violation factor defined by eq . ( [ generalized_fdt ] ) . it implies that the susceptibility is a function of the correlation function . since the difference of the two arguments of the susceptibility in eq . ( [ w_s_2 ] ) is @xmath68 , there are also two time regimes characterized by different dependence of the work on the experimental time scales @xmath13 and @xmath69 . * when the waiting time is much longer than the measurement time ( @xmath70 ) , the susceptibility which appears in eq . ( [ w_s_2 ] ) obeys eq . ( [ chi_f ] ) since @xmath71 holds . hence , from eqs . ( [ w_s_2 ] ) and ( [ chi_f ] ) the work in this regime is written as @xmath72 \label{w_f}.\end{aligned}\ ] ] in the derivation , @xmath73 for any path @xmath74 is used . * on the other hand , when both the waiting time and the measurement time are long ( @xmath75 ) , the susceptibility obeys eq . ( [ chi_a ] ) since @xmath76 holds @xcite . in this regime , it is seen from eq . ( [ chi_a ] ) that the work is a functional of the correlation function which shows aging . the ambiguous relation @xmath77 is described explicitly in sec . [ section4 ] with the function @xmath44 defined in eq . ( [ explicit_expression_aging ] ) . we refer to the former case as the quasiequilibrium regime and the latter case as the aging regime without any confusion with the time regimes characterized by the behavior of the correlation and the response function . in this section , we investigate the properties of the work done on the sample when the measurement time @xmath13 is short compared to the long waiting time @xmath11 ( the quasiequilibrium regime ) . first , we derive the work done in an slow operation such that the measurement time @xmath13 is so long that the correlation function relax to a time - independent value . this process is formulated by taking the infinite measurement time limit @xmath78 . glassy systems are out of equilibrium even in such an infinitely slow process . hence , we call a process where the external field is changed infinitely slowly a `` slow process '' instead of a `` quasistatic process '' . since for the slow process the time lapse of the process is infinity , the work done in the slow process in the quasiequilibrium regime is given by taking the limit @xmath78 after taking the infinite waiting time limit of eq . ( [ w_f ] ) . since the integrand , the correlation function , is finite , the order of limit and integration can be changed . thus , we can write the work @xmath9 for the slow process as @xmath79.\end{aligned}\ ] ] here , in order to give the expression of the work @xmath9 , we introduce the dynamical edwards - anderson ( e - a ) order parameter defined as @xmath80 using eq . ( [ integration ] ) and the above definition , the work @xmath81 for the slow process is given by @xmath82 \label{w_qs_f } \label{quasistatic_work_f}.\ ] ] therefore , the work for the slow process in the quasiequilibrium regime is the difference of a state function , since the right hand side of eq . ( [ w_qs_f ] ) is independent of the path of changing the field @xmath74 and dependent only on the thermodynamic variables ( @xmath83 and @xmath84 ) and constants intrinsic to the system ( @xmath85 and @xmath49 ) . it is important to note that this property is derived by using only the fdt . on the other hand , for usual systems apart from glassy systems , one may expect that equilibrium is usually achieved when @xmath69 is long ( @xmath86 ) . so , thermodynamics can be applied and the work for the slow process is equal to the change of the helmholtz free energy , @xmath87 , calculated by equilibrium statistical mechanics . however , we show below that this naive expectation fails for glassy systems . more precisely , the work for the slow process , eq . ( [ quasistatic_work_f ] ) , is different from the change of the helmholtz free energy calculated by equilibrium statistical mechanics . it is because the glassy systems are out of equilibrium even when the waiting time is infinite . assuming that @xmath88 is a physical quantity coupled to the external field @xmath14 linearly , we see from statistical mechanics with the assumption of ergodicity that the isothermal susceptibility @xmath89 is given by @xmath90 where @xmath91 denotes average over the gibbs - boltzmann distribution and @xmath92 denotes disorder - average . thus , the free energy difference @xmath87 due to change @xmath93 of the external field is @xmath94 where @xmath95 is the usual e - a order parameter . since ergodicity is broken for glassy systems , the phase space decomposes into many pure states . assuming that @xmath96 denotes the thermal average in local equilibrium in the pure state @xmath97 and @xmath98 denotes the free energy at the pure state @xmath97 , we see that @xmath99 where @xmath100 is the probability that the system is found at the pure state @xmath97 in true equilibrium and is given by @xmath101 . in order to compare @xmath9 , eq . ( [ w_qs_f ] ) , with @xmath87 , eq . ( [ free_energy_difference ] ) , we rewrite eq . ( [ w_qs_f ] ) in terms of pure states . since the correlation function is given by @xmath102 and the local equilibrium in pure states is achieved in the long time limit , @xmath103 where @xmath104 is the probability that the system is found in pure state @xmath97 at time @xmath105 . since the system is out of equilibrium at time @xmath105 , @xmath106 . thus , @xmath107 is not equal to @xmath85 . in addition , from eqs . ( [ ea_parameter_pure_state ] ) and ( [ q_pure_state ] ) , @xmath108 is not equal to @xmath49 . consequently , we conclude that the work @xmath9 for the slow process does not coincide with the free energy difference calculated by statistical mechanics with assumption of ergodicity . we discuss the properties of the work when the measurement time is finite . from eqs . ( [ w_f ] ) and ( [ quasistatic_work_f ] ) , the difference between the work when the process is not slow and the work for the slow process is given by @xmath109-q \ } \label{devi_work}.\ ] ] @xmath110>q$ ] when @xmath13 is finite , since the quasiequilibrium regime is considered . hence , the difference from the work for the slow process is positive for any path of changing the field , when the measurement time is finite . this implies the principle of minimum work ; @xmath111 where the equality holds only when the measurement time is infinite , i.e. the slow - process limit . our result is very similar to the consequence of thermodynamics which tells the work for non - quasistatic process is larger than the change of the helmholtz free energy . however , our result is different from that and beyond the scope of thermodynamics since in our discussion the initial state and the final state of the process are out of equilibrium and the value of the work for the slow process is different from the value of change in the helmholtz free energy derived by equilibrium statistical mechanics . the long measurement time behavior of @xmath112 is analyzed in this subsection . we show that the behavior of @xmath112 when the measurement time @xmath13 is long but finite is determined by the long time behavior of the correlation function . we describe below the results for four types of behavior of the correlation function which include almost all types of relaxation of the correlation , e.g. , the exponential , the power law @xcite , the logarithmic @xcite and the stretched exponential relaxation @xcite . the derivations are given in appendix [ appendix ] . 1 . when the correlation function behaves as @xmath113=0 $ ] , the difference is given by @xmath114 where @xmath115 \label{k}\ ] ] and @xmath81 denotes the work for the slow process . this case includes the power law relaxation such that @xmath116 when @xmath117 and the stretched exponential relaxation ( @xmath118 ) as well as the exponential relaxation @xcite . when the correlation function behaves as @xmath113=c$ ] , the difference is given by @xmath119 these two results show that @xmath120 is proportional to the inverse of the measurement time only when @xmath113=0 $ ] . in these two cases , the difference from the work for the slow process takes the minimum value when the field increases linearly as @xmath121 , since we assume that @xmath26 and @xmath27 is finite . when the correlation function obeys the power law relaxation , such that @xmath116 ( @xmath122 ) , the difference is given by @xmath123 in this case , the difference from @xmath9 obeys the power law whose exponent is equal to the exponent of the correlation function . when the correlation function obeys the logarithmic relaxation @xmath124 , the difference also obeys the logarithmic relaxation as @xmath125 \label{log}.\ ] ] in this case , the difference does not depend on the path @xmath74 . experimentally , these results tell how long the measurement should take and the suitable path of changing the external field in order to determine the value of the work for the slow process , i.e. the difference of a state function in quasiequilibrium regime . in this section , we discuss the properties of the work in the isothermal slow process when both the waiting time and the measurement time are long , i.e. in the aging regime , by using the modified fdt and aging of the correlation function introduced in eqs . ( [ generalized_fdt ] ) and ( [ explicit_expression_aging ] ) . condition when the work for the slow process depends on the path changing the external field is obtained . by using the results obtained in this section , the path - dependence of the work for particular systems is discussed in the next section , sec . [ section5 ] . as shown in sec . [ section2 ] , in the aging regime the work is a functional of the correlation function which shows aging . the slow process of the aging regime is given by taking the long measurement time limit @xmath78 with holding the relation @xmath77 , which guarantees that the slow - process limit is taken within the aging regime . the slow - process limit has always this meaning in this section . the order of the slow - process limit and the integration can be changed since the susceptibility @xmath126 is finite . from eq . ( [ w_s_2 ] ) the work @xmath81 for the slow process in the aging regime is given by @xmath127 \label{qs_w_a}. \end{aligned}\ ] ] thus , the dependence of @xmath128 on @xmath129 and @xmath130 determines the dependence of the work for the slow process on the path of changing the field @xmath74 ; @xmath9 is independent of @xmath74 if @xmath131 does not depend on @xmath129 and @xmath130 . on the other hand , @xmath9 is a functional of @xmath74 if @xmath131 depends on @xmath129 and @xmath130 . we derive conditions when @xmath9 depends on @xmath74 in the rest of this section . in order to analyze @xmath128 , we clarify the meaning of @xmath77 or the aging regime by using @xmath51 in eq . ( [ explicit_expression_aging ] ) . by substitution @xmath132 in eq . ( [ explicit_expression_aging ] ) , we see that the following equation holds in the region except for the point @xmath133 which does not contribute to the value of the work by itself ; @xmath134 } } } c(s_1 \triangle t+t_w , s_2\triangle t+t_w ) = \hat{c}(\mu ) . \label{2}\ ] ] this equation implies that the aging regime is the experimental time regime where @xmath135 $ ] is finite and the slow - process limit of the correlation function in the integrand in eq . ( [ qs_w_a ] ) is a function of @xmath136 $ ] . we discuss below the aging regimes and dependence of @xmath131 on @xmath129 and @xmath130 in four cases of the different long - time behaviors of @xmath44 which exhaust all possibilities ( see fig . [ g ] ) . 1 . when @xmath137 , @xmath50 is finite only in the region @xmath138 . in the region of integration of eq . ( [ qs_w_a ] ) except for the point @xmath139 , @xmath140\rightarrow\infty \label{aging_infty}\ ] ] as @xmath141 since @xmath142 . an example of such @xmath51 is @xmath143 . when @xmath143 and @xmath144 , @xmath145=\infty$ ] as shown in eq . ( [ aging_infty ] ) . since the point @xmath139 does not contribute to the value of the work by itself , the experimental time regime where @xmath146 $ ] is finite does not exist in this case . if there is no other aging regime except for that characterized by such @xmath51 exists , only the quasiequilibrium regime contributes to the value of the work and hence the work for the slow process is independent of @xmath74 and is given by eq . ( [ quasistatic_work_f ] ) . 2 . when @xmath147 , @xmath148 } = \frac{s_2 \triangle t+t_w}{(s_1-s_2)\triangle t } = \frac{s_2+t_w/\triangle t}{s_1-s_2}.\ ] ] thus , from eq . ( [ 2 ] ) , @xmath149 is a function of @xmath150 ; @xmath151 } } } c(s_1\triangle t+t_w , s_2\triangle t+t_w ) \nonumber \\ & = & \hat{c}(\frac{s_2+\mu'}{s_1-s_2 } ) . \label{t}\end{aligned}\ ] ] it implies that the aging regime is the region where @xmath152 is finite . in this aging regime the slow - process limit of the correlation function in the integrand of eq . ( [ qs_w_a ] ) depends on @xmath129 and @xmath130 if @xmath153 in eq . ( [ 2 ] ) is not a constant . it means that the work for the slow process , @xmath9 , depends on the path of change of the external field , @xmath74 . 3 . when @xmath154 and @xmath155 exists , @xmath156}= \frac{s_2 \triangle t+t_w}{g(\triangle t)}\frac{g(\triangle t ) } { g[(s_1-s_2)\triangle t ] } \nonumber \\ \rightarrow \frac{t_w}{g(\triangle t)}g_r(s_1-s_2 ) , \label{aging_regime}\end{aligned}\ ] ] as @xmath78 . an example of such @xmath51 is @xmath157 . when @xmath157 , eq . ( [ aging_regime ] ) holds ; @xmath158}\rightarrow \frac{t_w}{g(\triangle t)}\frac{1}{(s_1-s_2)^{\alpha}}\ ] ] as @xmath141 . from eqs . ( [ 2 ] ) and ( [ aging_regime ] ) , @xmath159 } } } c(s_1\triangle t+t_w , s_2\triangle t+t_w ) \nonumber \\ & = & \hat{c}[\mu'g_r(s_1-s_2 ) ] . \label{st}\end{aligned}\ ] ] it implies that the aging regime is the region where @xmath160 is finite . in this aging regime the slow - process limit of the correlation function in the integrand of eq . ( [ qs_w_a ] ) is a function of @xmath129 and @xmath130 if @xmath153 in eq . ( [ 2 ] ) is not a constant . it means that the work @xmath9 depends on @xmath74 . 4 . when @xmath154 and @xmath161 , we prove below that the aging regime for experimental time scales does _ not _ exist . we assume that the aging regime exists . it means that a function @xmath162 exists such that @xmath163 $ ] is finite . then , from the condition @xmath164 , @xmath165 } \nonumber \\ & = & \lim_{{\triangle t \rightarrow \infty } \atop{\ ; \mu=\frac{t_w}{p(\triangle t ) } } } \frac{t_w}{g[(s_1-s_2)\triangle t ] } \nonumber \\ & = & \lim_{\triangle t \rightarrow \infty } \frac{\mu p(\triangle t)}{g[(s_1-s_2)\triangle t]}.\end{aligned}\ ] ] assuming that @xmath166 , we see from the condition @xmath167 that @xmath165}/ \lim_{{\triangle t \rightarrow \infty } \atop{\ ; \mu=\frac{t_w}{p(\triangle t ) } } } \frac{s'_2\triangle t+t_w}{g[(s'_1-s'_2)\triangle t ] } \nonumber \\ & = & \lim_{\triangle t \rightarrow \infty } \frac{\mu p(\triangle t)}{g[(s_1-s_2)\triangle t]}\frac{g[(s'_1-s'_2 ) \triangle t]}{\mu p(\triangle t ) } \nonumber \\ & = & \lim_{\triangle t \rightarrow \infty}\frac{g[(s'_1-s'_2)\triangle t ] } { g[(s_1-s_2)\triangle t ] } \nonumber \\ & = & \infty . \label{mujun}\end{aligned}\ ] ] since eq . ( [ mujun ] ) contradicts the assumption that @xmath163 $ ] is finite , the aging regime for experimental time scales does _ not _ exist in this case . hence , if only this aging regime characterized by such @xmath44 exists , @xmath168 is equal to @xmath105 or @xmath49 . it implies that the work is independent of @xmath74 . an example of such @xmath44 is @xmath169 . consequently , we conclude that _ if there are aging regimes characterized by @xmath51 of case 2 . or case 3 . and the function @xmath46 is not a constant and the fdt violation factor @xmath41 is not equal to @xmath105 , the work for the slow process depends on the path of changing the external field @xmath74 . in this case , contrary to the fundamental assumption of thermodynamics , the work for the slow process is not the difference of a state function . _ it has been explicitly checked on several disordered models that in the long - time limit two situations with different dynamical behavior seem to exist @xcite : * there are systems with only one correlation scale apart from the quasiequilibrium regime , which we call `` single - correlation - scale systems '' . for these systems , the correlation function scales as @xmath170 $ ] and the fdt violation factor @xmath41 is a constant . equilibrium states of these systems are solved by a one step replica symmetry breaking ansatz . examples are the @xmath0-spin spherical spin - glasses @xcite and a lennard - jones binary mixture which is a model of structural glasses in a glassy state @xcite . the real spin - glasses such as agmn seem to belong to this class since it is known that the correlation function is scaled with single scaling function @xmath8 . * there are systems such as sherrington - kirkpatrick ( s - k ) model which have an infinite number of correlation - scales apart from the quasiequilibrium regime @xcite , which we call `` multi - correlation - scale systems '' . for these systems , ultrametricity in time holds for any correlation such that @xmath171 ; @xmath172 \ \mbox{when } \ t_1>t_2>t_3 \ \mbox{and } \ t_3\rightarrow \infty$ ] . the fdt violation factor @xmath41 is a nontrivial function of @xmath173 . equilibrium properties of these systems are solved by a full replica - symmetry breaking ansatz . in this section , we analyze the path - dependence of the work for the slow process in the aging regime for these two classes of systems by using the general results obtained in the previous section , sec . [ section4 ] . it is important to note that only these two classes of systems seem to exist @xcite . for the single - correlation - scale systems , scaling of the correlation function @xmath170 $ ] holds . hence , the work for the slow process in the aging regime is given as @xmath174\}. \label{w_qs_a_2}\end{aligned}\ ] ] three explicit choices of the scaling function @xmath8 have been proposed so far ; @xmath175 $ ] and @xmath176 $ ] @xcite . we analyze below the properties of the work for the slow process for each three cases . 1 . when @xmath56 , the correlation function decays in power - law as found in the trap model @xcite and the @xmath0-spin spherical spin - glasses @xcite . the slow - process limit of the correlation function in eq . ( [ w_qs_a_2 ] ) depends on @xmath129 and @xmath130 as @xmath177 this is an example of eq . ( [ t ] ) in previous section . 2 . when @xmath178 ( \alpha < 1)$ ] @xcite , the slow - process limit of the correlation function in eq . ( [ w_qs_a_2 ] ) depends on @xmath129 and @xmath130 through the scaling function which behaves as @xmath179.\ ] ] this is an example of eq . ( [ st ] ) . 3 . when @xmath180 ( a > 1)$ ] @xcite , the slow - process limit of the correlation function which appears in eq . ( [ w_qs_a_2 ] ) depends on @xmath129 and @xmath130 through the scaling function which behaves as @xmath181.\ ] ] this is another example of eq . ( [ st ] ) to my knowledge , only three types of the scaling function @xmath8 discussed above are known so far for single - correlation - scale systems . for each of three cases the slow - process limit of the correlation function in the integrand of eq . ( [ qs_w_a ] ) is a function of @xmath129 and @xmath130 . it means for the single - correlation - scale systems that the work for the slow process in the aging regime depends on the path of changing the external field @xmath74 and contrary to the assumption of thermodynamics the work for the slow process is not a difference of a state function . in order to see more explicit form of the work for the slow process in the aging regime , we introduce the modified fdt eq . ( [ generalized_fdt ] ) with constant violation factor @xmath41 which has been verified to be valid for the single - correlation - scale systems @xcite . from eq . ( [ generalized_fdt ] ) , the susceptibility is given by @xmath182 where @xmath183 . hence , from eq . ( [ qs_w_a ] ) , the work for the slow process is given by @xmath184\ } \label{qs_w_a_2}.\end{aligned}\ ] ] since @xmath185 when @xmath186 , the second term on the right hand side is non - zero . hence , if @xmath187 is finite the work for the slow process is not the difference of a state function since the work depends on the path of changing the field @xmath74 . in addition , the value of the work is different from the value of the work for the slow process in the quasiequilibrium regime : the first term on the right hand side of eq . ( [ qs_w_a_2 ] ) . in addition , it is also shown that since the correlation function is positive and smaller than the dynamical e - a order parameter in this regime , there are bounds for the value of the quasistatic work as @xmath188 eq . ( [ bound ] ) shows that the work @xmath9 in the aging regime is independent of the path of changing the field and coincides with the work @xmath9 in the quasiequilibrium regime when the effective temperature @xmath187 is infinite , which holds for the @xmath189-spin spherical spin - glass model @xcite and a model of phase separation @xcite . in the previous subsection , we saw that the work for the slow process depends on the path of changing the external field when @xmath75 . we show here that the path - dependence of the work for the slow process is strong enough to be observed in experiments by evaluating the path - dependence of the @xmath190-spin spherical spin - glass model . first , we describe the several pieces of information needed to evaluate eq . ( [ qs_w_a_2 ] ) . it is known that the scaling of the correlation function , such that @xmath191 and the modified fdt with the effective temperature , i.e. eq . ( [ generalized_fdt ] ) , hold for the model @xcite . when @xmath192 @xcite , @xmath193 has been obtained . the dynamical e - a parameter @xmath49 and the effective temperature are given by @xmath194 and @xmath195 when @xmath192 . in computation of these values , we uses the formulae : @xmath196 in order to evaluate the magnitude of the path - dependence , we compute the relative difference between the works for two paths ( @xmath121 and @xmath197 ) which is defined as the difference between the works for the two paths divided by the average of the two works . the results obtained from eq . ( [ qs_w_a_2 ] ) are as follows ; when the ratio of the measurement time to the waiting time is @xmath198 , the relative difference is equal to @xmath199 , i.e. about @xmath200 ; when the ratio is @xmath201 , the relative difference is equal to @xmath202 , i.e. about @xmath203 . hence , we can say that the path - dependence of the work in the slow process can be observed for the @xmath190-spin spherical spin - glass model . the relative difference is plotted against the ratio of the waiting time to the measurement time in fig . [ figure1 ] . it tells that contrary to intuition the path - dependence becomes larger , as one increases the measurement time with keeping the waiting time fixed . for the multi - correlation - scale systems such as s - k model , it is known that the ultrametricity in time holds for all correlations such that @xmath171 in the long - time limit @xcite ; @xmath204 \ \ \mbox{when } \ t_1 > t_2 > t_3 . \label{ultrametric_relation_2}\ ] ] by using this ultrametric relation eq . ( [ ultrametric_relation_2 ] ) , we prove below that the correlation function in the aging regime is a constant in the two cases ( case 2 and case 3 in sec . [ section4 ] ) where the work for the slow process can depend on the path of changing the external field @xmath74 . thus , for these two cases , the work for the slow process is independent of the path @xmath74 . it has been proven in sec . [ section4 ] for the other two cases that the work for the slow process can not depend on @xmath74 . thus , the work for the slow process for the multi - correlation - scale systems is independent of the path of changing the field in all the experimental time regimes . 1 . when @xmath147 , by substitution @xmath205 in eq . ( [ ultrametric_relation_2 ] ) and assuming that @xmath206 , we see that @xmath207 . \label{ult2}\ ] ] hence , @xmath208 where @xmath209 . thus , from eq . ( [ ult2 ] ) , @xmath210.\ ] ] since it implies that @xmath211 @xmath212 is a constant and hence @xmath213 $ ] is a constant . 2 . when @xmath154 and @xmath155 exists , the ultrametric relation eq . ( [ ultrametric_relation_2 ] ) is written as @xmath214 \label{ult3}\end{aligned}\ ] ] where @xmath97 is a constant such that @xmath215 . the second argument of @xmath216 in eq . ( [ ult3 ] ) is rewritten as @xmath217}}c(\tau+t',t ' ) \nonumber \\ & = & \lim_{{\tau\rightarrow\infty}\atop{\ ; \mu g_r(a)=t'/g(\tau)}}c(\tau+t',t ' ) \nonumber \\ & \equiv & \bar{c}[\mu g_r(a ) ] , \label{cat}\end{aligned}\ ] ] where @xmath218 . similarly , the first argument of @xmath216 in eq . ( [ ult3 ] ) is rewritten as @xmath219 \nonumber \\ & = & \bar{c}[\mu g_r(1-a)]\end{aligned}\ ] ] where the last equality is shown from eq . ( [ cat ] ) . consequently , @xmath220,\bar{c}[\mu g_r(a)]\}.\ ] ] since @xmath221 , @xmath222 is a constant . we proved above only that the correlation function is a constant within a correlation - scale since ultrametricity in time is valid within a correlation - scale . on the other hand , even for the multi - correlation - scale systems , the value of the correlation function becomes @xmath105 when the time difference is much larger than the waiting time . so , value of the correlation function must depend on an infinite number of correlation - scales . hence , there is possibility that there is an infinite number of experimental time regimes characterized by different values of the work for the slow process . summarizing our results , for systems with aging we have considered the experimental time scale dependence of the work exerted by modulating the external field in an isothermal process . we have shown that there are two experimental time regimes characterized by different behavior of the work . in the quasiequilibrium regime @xmath223 , the work for the slow process is independent of the path of changing the external field and the value of the work for the slow process is different from the change of the helmholtz free energy obtained by statistical mechanics with the assumption of ergodicity . when the process is not infinitely slow , the difference between the work for non - slow process and the work for the slow process is positive for any path of changing the filed , i.e. the principle of minimum work holds . in addition , we derived the dependence of the difference on the measurement time from the long - time behaviors of the correlation function . in the aging regime @xmath224 , whose precise meaning is given in sec . [ section4 ] , contrary to a fundamental assumption of thermodynamics , the work exerted in the slow process on the single - correlation - scale systems for which the fdt violation factor @xmath41 is not equal to @xmath105 depends on the path of changing the external field . it was examined that for the @xmath190-spin spherical spin - glass model the magnitude of the path - dependence is strong enough to be observed in experiments . the conditions when the work for the slow process depends on the path is also derived generally . on the other hand , for the multi - correlation - scale systems , the work for the slow process is independent of the path of changing the field . however , the work for the slow process still may depend on history of a system since there is possibility that there is an infinite number of experimental time regimes characterized by different values of the work for the slow process . in order to attack on this problem , information on time scale of the aging regime of multi - correlation - scale systems beyond time reparametrization invariance is necessary @xcite . it is important to note that these results have a much broader range of validity , because they were derived without any assumption on microscopic properties of a system . we consider the implications of our results on thermodynamics for systems with aging . in the quasiequilibrium regime , our results are the same as the consequences of thermodynamics , where the work for the slow process is independent of the path of changing the external field and the work for a non - slow process is larger than that for the slow process . only discrepancy is difference between the value of the work for the slow process and that of the free energy obtained by statistical mechanics with the assumption of ergodicity . however , this discrepancy does not mean that the framework of usual thermodynamics is invalid since the free energy can be defined by the work for the slow process . this suggests the usual framework of thermodynamics is valid in the quasiequilibrium regime even though the system is out of equilibrium . we considered only the properties of the work in the linear response regime . thus , in order to prove validity of usual thermodynamics in the quasiequilibrium regime , the relation between the heat and the entropy has to be considered and extension of our analysis outside the linear response regime has to be performed . for these purposes , it would be interesting to analyze langevin dynamics of spin - glass models with stochastic energetics @xcite . since the work for the slow process in the quasiequilibrium regime is independent of history of a system in linear response regime , it gives a fundamental state function for thermodynamics in the quasiequilibrium regime and is worth being measured in experiments . we discussed an experiment to measure it . our results in sec . [ section3 ] for the measurement time dependence of the work shows how long measurement time is needed to reach the expected accuracy . however , since the waiting time can be long but should be finite in reality , the measurement time has to be finite in order to keep the experimental time scale in the quasiequilibrium regime . otherwise the value of the work does not come close to that of the work for the slow process in the quasiequilibrium regime since the experimental time scale is in the aging regime . hence , one can not expect that the value of the measured work coincides with that of the work for the slow process with arbitrary accuracy . in other words , the accuracy of measurement of the state function is bounded by the length of the waiting time . on the other hand , in the aging regime , our results imply that in contrast to the usual framework of thermodynamics it is essential to consider the experimental time scales such as the waiting time and the measurement time . furthermore , history of a system as a state variable is necessary to describe thermodynamic properties of glassy materials . although constructing thermodynamics for the aging regime is a tough work , it is a fruitful and challenging problem to be investigated since some universal properties in the aging regime are known @xcite . recently , a framework of thermodynamics for the aging regime when the temperature changes in time is proposed @xcite . the framework avoids treating the problem of path - dependence by fixing a path to change the temperature . it is not clear whether thermodynamics for a single path to change the temperature is useful . its validity and usefulness should be checked . i wish to acknowledge valuable discussions with takashi odagaki , akira yoshimori , hiizu nakanishi , kazuo kitahara , osamu narikiyo , miki matsuo , ken sekimoto , shin - ichi sasa , yoshihisa miyamoto , hajime takayama , koji hukushima and hitoshi inoue . in this appendix , we derive the long measurement time behavior of the deviation from the work for the slow process for four types of relaxation of the correlation function described in sec . [ difference ] . at first , we consider the deviation when @xmath225 $ ] is equal to @xmath105 . here , in order to prove that the deviation is proportional to the inverse of the measurement time @xmath13 , we show that the proportional constant is finite . the proportional constant is given as @xmath230.\end{aligned}\ ] ] since @xmath231 exists in the case , @xmath227 is finite . hence , the order of the limit @xmath78 and the integration can be changed . thus , the proportional constant is given by @xmath232 where @xmath233 is defined by eq . ( [ k ] ) . consequently , it is proved that the proportional constant is finite and the deviation is proportional to @xmath234 . next , we consider the deviation when @xmath225 $ ] is equal to a finite value @xmath226 . in order to prove that the deviation is proportional to @xmath235 , we show that the proportional constant is finite . dividing the region of integration of eq . ( [ app : dev1 ] ) into five regions , the proportional constant is given by @xmath236 \label{app : dev2}.\end{aligned}\ ] ] we discuss the behavior of the function @xmath237 in each region . when @xmath238 , from the definition of @xmath227 , eq . ( [ k ] ) , it is shown that @xmath239.\ ] ] thus , @xmath237 is finite even when @xmath78 . on the other hand , when @xmath240 , we express the function @xmath237 by using a function defined as @xmath241 : @xmath242 \nonumber \\ & = & \int_0 ^ 1 d\tau [ c(\tau)-q]+\int_1^{s \triangle t } d\tau [ \frac{c}{\tau}+f(\tau ) ] \nonumber \\ & = & \int_0 ^ 1 d\tau [ c(\tau)-q]+c \log s+c \log \triangle t+ \int_1^{s \triangle t } d\tau f(\tau ) . \end{aligned}\ ] ] since @xmath243 , the fourth term on the right hand side is finite even when @xmath13 is infinite . thus , the behavior of the function is given as @xmath244 when @xmath240 . by evaluating the each term of eq . ( [ app : dev2 ] ) with using the above behavior of the function @xmath237 , it is straightforward to show that @xmath245 the case where @xmath225 $ ] does not exist ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ in order to find out the leading term in the behavior for the long measurement time , we divide the region of integration of eq . ( [ devi_work ] ) into three regions as @xmath248 where we assume that the parameter @xmath97 satisfies @xmath215 . since all the integrands ( the time derivative of the field and the correlation function ) are finite , it can be shown that the first two terms on the right hand side are bounded as @xmath249 and @xmath250 .\ ] ] in addition , we notice that since @xmath251 in the region of integration of the third term , @xmath252 is large when the measurement time @xmath13 is large . thus , the third term is determined by the long - time behavior of the correlation function . at first , we treat the case where the correlation function behaves at a long time as @xmath253 . we divide the region of integration of the third term of eq . ( [ division ] ) into three regions as @xmath254 since all the integrands are finite , it is shown that the last two terms are bounded by @xmath255 and @xmath256 on the other hand , by using the long - time behavior of the correlation function , the first term is given as @xmath257 thus , we find that this term is @xmath258 $ ] . assuming the free parameter @xmath97 is larger than the exponent @xmath259 , one sees that the leading term is eq . ( [ power1 ] ) by comparing the orders of all the terms in eqs . ( [ division ] ) and ( [ division2 ] ) . hence , the long measurement time behavior of the deviation is given by eq . ( [ power ] ) . next , we consider the case where the long - time behavior of the correlation function is given by @xmath260 . again , we start by evaluating the third term on the right hand side of eq . ( [ division ] ) . by using the long - time behavior of the correlation function , the term is given by @xmath261 since in the region of integration the following inequality holds : @xmath262 by neglecting numbers of @xmath263 the third term of the right hand side of eq . ( [ division ] ) is reduced to @xmath264 furthermore , by neglecting numbers of @xmath265 $ ] , the term is reduced to @xmath266 consequently , one sees that the leading term is the above one by comparing the orders of all the terms in eqs . ( [ division ] ) and ( [ division2 ] ) . thus , the long measurement time behavior of the deviation is given by eq . ( [ log ] ) . finally , in order to determine the order of the neglected term we determine the value of the free parameter @xmath97 , since the neglected term contain the free parameter @xmath97 . since in terms of @xmath97 the order of the neglected terms are given as @xmath263 and @xmath267 $ ] , we can determine the value of @xmath97 so that the value of the neglected terms are minimized . thus , assuming that the neglected term is given by @xmath268 where @xmath259 and @xmath269 are constants , the value of @xmath97 is determined by @xmath270 = 0.\ ] ] since the solution is @xmath271 , one sees that the order of neglected term is @xmath272 $ ] as shown in eq . ( [ log ] ) . the present address is research institute for applied mechanics , kyushu university , kasuga 816 - 8580 , japan . electronic address : mituhiro@riam.kyushu-u.ac.jp for recent reviews , see e.g. e. vincent , j. hammann , m. ocio , j .- bouchaud , and l.f . cugliandolo , in _ proceedings of sitges conference on glassy systems _ , edited by e. rubi ( springer , berlin , 1996 ) ; j .- bouchaud , l. cugliandolo , j. kurchan , and m. mzard , in _ spin glasses and random fields _ , edited by a.p . young ( world scientific , singapore , 1997 ) . these articles are found also at e - print cond - mat/9607224 and cond - mat/9702070 . l. lundgren , p. svedlindh , p. nordblad , and o. beckman , phys . * 51 * , 911 ( 1983 ) ; m. alba , j. hammann , m. ocio , and ph . refregier , j. appl . phys . * 61 * , 3983 ( 1987 ) . l.f . cugliandolo and j. kurchan , phys . * 71 * , 173 ( 1993 ) . cugliandolo and j. kurchan , j. phys . a * 27 * , 5749 ( 1994 ) . cugliandolo , j. kurchan , and l. peliti , phys . e * 55 * , 3898 ( 1997 ) . s. franz and h. rieger , j. stat . phys . * 79 * , 749 ( 1995 ) ; e. marinari , g. parisi , f. ricci - tersenghi , and j.j . ruiz - lorenzo , j. phys . a * 31 * , 2611 ( 1998 ) . cugliandolo , d.r . grempel , j. kurchan , and e. vincent , europhys . lett . * 48 * , 699 ( 1999 ) . g. parisi , phys . * 79 * , 3660 ( 1997 ) . barrat and w. kob , europhys . lett . * 46 * , 637 ( 1999 ) . grigera and n.e . israeloff , phys . lett . * 83 * , 5038 ( 1999 ) . l. berthier , j .- l . barrat , and j. kurchan , eur . j. b * 11 * , 635 ( 1999 ) . the reason why the usual fdt holds in the quasiequilibrium regime though a system is out - of - equilibrium is that the change of the free energy stops at the long time limit , since the fdt violation is bounded by the time - derivative of the free energy @xcite . cugliandolo , d.s . dean , and j. kurchan , phys . lett . * 79 * , 2168 ( 1997 ) . when @xmath273 , @xmath274 does not hold even if @xmath275 since the time difference @xmath252 is zero . however , the case can be ignored , since the susceptibility is finite there and the point @xmath273 does not contribute to the value of the integral by itself . ogielski , phys . b * 32 * , 7384 ( 1985 ) ; a. ito , h. aruga , e. torikai , m. kikuchi , y. syono , and h. takei , phys . rev . lett . * 57 * , 483 ( 1986 ) . m. ocio , h. bouchiat , and p. monod , j. mag . mag . * 54 * , 11 ( 1986 ) . ngai , comments solid state phys . * 9 * , 127 ( 1979 ) ; * 9 * , 141 ( 1980 ) ; a.t . ogielski , phys . b * 32 * , 7384 ( 1985 ) ; m.h . cohen and g.s . grest , phys . b * 20 * , 1077 ( 1979 ) . it has been shown from the massless langevin equation without the assumption of a linear response that @xmath276 @xcite . since the correlation function derived from the massless langevin equation obeys the exponential relaxation , our result agrees with that of ref . k. sekimoto and s. sasa , j. phys . jpn . * 66 * , 3326 ( 1997 ) . bouchaud , j. phys . i ( france ) * 2 * , 1705 ( 1992 ) . this form of the scaling function was proposed to account for the results of experiments in polymer glasses @xcite , then used in analyses of aging effects in decay of the thermoremanent magnetization @xcite , and found in the exact solution of the asymmetric spherical sk model @xcite . struik , _ physical aging in amorphous polymers and other materials _ ( houston , 1978 ) . m. ocio , m. alba , and j. hammann , j. physique lett . ( france ) * 46 * , l-1101 ( 1985 ) ; m. alba , m. ocio , and j. hammann , europhys . lett . * 2 * , 45 ( 1986 ) . s. franz and m. mzard , europhys . lett . * 26 * , 209 ( 1994 ) : physica a * 209 * , 1 ( 1994 ) ; l.f . cugliandolo , j. kurchan , p.le doussal , and l. peliti , phys . lett . * 78 * , 350 ( 1997 ) . this form of the scaling function is suggested by the numerical data from the analysis of a point particle in a random potential with infinite dimension @xcite and is used to scale data for the thermoremanent magnetization and the out - of - phase susceptibility . cugliandolo and p.le doussal , phys . e * 53 * , 1525 ( 1996 ) ; l.f . cugliandolo , j. kurchan , and p.le doussal , phys . lett . * 76 * , 2390 ( 1996 ) . cugliandolo and d.s . dean , j. phys . a * 28 * , 4213 ( 1995 ) . the energy is scaled with a constant @xmath277 which determines the variance of the coupling constant between spins as @xmath278 the equations of motion of @xmath2 and @xmath4 of the multi - correlation - scale systems in the long waiting time limit have an infinite set of invariance called `` time reparametrization invariance '' @xcite ; if we perform an arbitrary reparametrization of time @xmath279 and we redefine the two - time quantities as @xmath280 \ \ \hat{g}(\hat{t},\hat{t}')=g[\xi(t),\xi(t')].\ ] ] then , the redefined functions @xmath281 and @xmath282 satisfy the same equations of motion . it is important to note that this invariance is a consequence of having neglected the time derivatives in taking the long waiting time limit . the full equations of motion have no such invariance and the solution is unique . k. sekimoto , j. phys . 66 * , 1234 ( 1997 ) . m. nieuwenhuizen , phys . e * 61 * , 267 ( 2000 ) .

due to the observations of cosmic microwave background ( cmb ) radiation @xcite , large scale structure ( lss ) @xcite , and ia supernovae @xcite we known that the universe is actually in accelerated expansion . a direct way to incorporate this accelerated expansion of universe into einstein general relativity theory @xcite is to include the cosmological constant @xmath6 . the inclusion of this constant has the same effect of a non - zero vacuum energy density , @xmath7 , in such way that the pressure @xmath8 and the density @xmath9 are related by a state equation with the form @xmath10 . on the other hand , the accelerated expansion of universe can also be described by adding to the universe content a homogeneous fluid with energy density @xmath11 , the so called _ dark energy _ this fluid has positive energy density @xmath12 , and negative pressure @xmath13 , in such way that @xmath14 . this pressure is then responsible for the accelerated universe expansion . it is a remarkable fact that @xmath15 of universe content must be in the dark energy form . also , dark energy would be uniformly distributed in all space , and so , its density is constant in all points and times . this is in contrast with the time evolution of baryonic matter density @xmath16 , that diminishes due to the expansion of the universe . it is called _ the cosmic coincidence _ to the fact that today , the baryonic matter density , @xmath16 has approximately the same value of dark energy density @xmath12 @xcite . to compare different cosmological models , it is common to define the density parameter @xmath17 that it is the ratio between the density @xmath18 with the critical density @xmath19 of the friedmann universe , @xmath20 . in this sense the cosmic coincidence implies in the equality @xmath21 . this coincidence can be viewed as an indicative of the existence of some dynamical effect that relates both scales of baryonic matter and of the dark energy density . nevertheless , among all the models in the literature that are devoted to explain the accelerated expansion of universe , there is a class of dynamical models that obtain the desired negative pressure due to the inclusion of a scalar field that is the responsible for the variations in the expected value of vacuum energy . as the neutrino squared mass difference , @xmath22 ev@xmath23 , where @xmath24 and @xmath25 are the masses of third and first neutrino mass state , is of same order of dark energy scale , it is straightforward to think a model in which the scalar field couple to neutrinos and hence , the total energy of the fluid can vary slowly as the neutrino density decreases @xcite . in the _ mass varying neutrino _ models the inclusion of a scalar field allow the coupling of neutrino and dark energy densities due to the non - standard neutrino - scalar coupling . the main consequence of this coupling to the neutrino physics is the fact that now the neutrino masses depends on the medium density . this field could couples neutrinos to the baryonic matter and also to the neutrino background @xcite . the consequences in cosmology of mavan s had been studied in last years @xcite and references therein . for the neutrino phenomenology the consequences of mavan s were studied for the solar and atmospheric neutrinos @xcite . we will use the data from super - kamiokande ( sk for now ) experiment @xcite . in the experiment sk , down - going neutrinos , the ones that are produce in the atmosphere immediately over super - kamiokande detector , reaches sk with the the cosine of zenith angle @xmath26 and travels approximately 20 km in the atmosphere . on the other hand , up - going neutrinos are produced in the opposite side of the planet and crosses all the earth before reaches sk in the the cosine of zenith angle @xmath27 direction . the so - called _ up - down _ asymmetry of neutrino events in sk is recognized as the first experimental corroboration of neutrino flavor oscillations . we explicit here that atmospheric neutrinos that arrives at sk from different directions travels different distances and crosses regions of very distinct densities ( see next section for details ) . this fact makes the angular distribution of events in sk a good place to looking for dependence of medium density effects in the propagation of neutrinos . in fact , the sk experiment reported no improvement of data fit due to the inclusion of diagonal mavan s mechanism in the propagation of atmospheric neutrinos @xcite . for our knowledge there is no analysis made for non - diagonal mavan s mechanism for atmospheric neutrinos . we also apply the mavan s model to describe the beam muon neutrinos and anti - neutrinos at minos experiment @xcite , where a muon neutrino or muon anti - neutrino beam with energy between few hundred of mev and few gev . the neutrino or anti - neutrino beam travels a few hundred of kilometers inside earth s crust before reach the far detector . by the comparison between the number of neutrino events in near and far detector minos collaboration has measure precisely the standard neutrino parameters @xmath28 and @xmath29 . see for example @xcite and references therein . in mavan s context , the main difference between minos beam and sk is that in the former neutrinos cross only the upper crust of earth , that can be described by a constant matter density and the latter cross a different non - constant densities . in this way the mavan s effective oscillation for minos , with a constant density and with sk with a variable density allow us to test the essential characteristic of mavan mechanism , the density dependence of the neutrino mass differences . the paper is organized as follows . in the section ( [ sec : model ] ) we describe the framework to mavans that we adopt and in sec . ( [ oscilograms ] ) we show the changes in neutrino oscillations due to mavans . then in sec . ( [ sec : muons ] ) we compare our results for the angular distribution of events in sk for neutrino oscillations without and with mavans . also in section ( [ sec : minos ] ) we show how mavan s mechanism affects the allowed range of oscillations parameters for minos experiment and in section ( [ sec : chi2 ] ) we perform the @xmath30 analysis of sk atmospheric neutrino combined with minos beam data , and the constrains to mavan s model that we obtain . conclusions are in section ( [ sec : conclusion ] ) . in the neutrino mass - mixing formalism , the time evolution of neutrino flavor eigenstates is given in terms of neutrino mixing , in which one has to describe the flavor eigenstate and hence , the time evolution of atmospheric neutrinos , in the two neutrino flavor approximation , is given by the evolution equation @xmath31\left(\begin{array}{c } \nu_{\mu}\\ \nu_{\tau } \end{array}\right)~ , \label{sch}\end{aligned}\ ] ] and the hamiltonian assumes the form , in the mass basis , @xmath32~ , \label{mmatrix}\end{aligned}\ ] ] here @xmath33 is the neutrino energy , @xmath34 is the square difference of mass eigenstates , and @xmath35 is the mixing matrix @xmath36 where @xmath37 . the muon neutrino survival probability is @xmath38 where @xmath39 is the squared mass difference , l is the distance and @xmath33 is the neutrino energy . the formalism that we adopt to include mavans in the neutrino propagation has the standard model of particles plus a light scalar field @xmath40 with mass @xmath41 that couples with neutrinos @xmath42 , i=1,2,3 and with fermion fields f = e , n , p . in such model , the modification due to mavan s is the introduction of a fermion density dependent term in each one of matrix elements in the neutrino mass matrix , eq . ( [ mmatrix ] ) , and so , including mavan s , the flavor neutrino evolution is described by the generalization of eq . ( [ sch ] ) . explicitly we have @xmath43= h^{{\rm flavor}}_{{\rm mavan } } \left[\begin{array}{c } \nu_{\mu}\\ \nu_{\tau } \end{array}\right ] , \label{eqevo - drg}\end{aligned}\ ] ] where the modified hamiltonian in the mavan s framework is @xmath44 where the mixing matrix u is defined in eq . ( [ u ] ) and the mavan hamiltonian has the form @xmath45~. \label{mevo - drg}\end{aligned}\ ] ] here @xmath46 and @xmath47 are the mavan s parameters and @xmath48 is the function of the earth matter density @xmath18 that neutrinos feels while cross the earth . when @xmath49 we recover the standard neutrino evolution given in eq . ( [ sch ] ) . we can classify the behavior of mavan mechanism given in eq . ( [ mevo - drg ] ) in two types : ( a ) when @xmath50 and @xmath51 and ( b ) when @xmath52 and @xmath53 . in the former case the neutrino survival probability , for constant density , is given by @xmath54 where the effective mass scale is given by @xmath55 $ ] . in this case , when @xmath51 the amplitude of oscillations , @xmath56 is the same as in the standard neutrino oscillations in eq . ( [ standardoscillation ] ) , and the phase of the oscillations , proportional to @xmath57 , have now a matter density dependence . this was the case mostly worked in the literature @xcite . the latter case , the hamiltonian in the mass basis is non - diagonal and for our knowledge it was _ not explored _ in the literature for atmospheric neutrino phenomenology . the probability , for constant density @xmath18 , can be written as @xmath58 where the amplitude , @xmath59 and the phase of oscillations , @xmath60 are different from the usual standard oscillation scenario . the mavan mass difference is given by @xmath61 where the mavan mass difference depends on the medium density and the explicit expression for the amplitude is @xmath62 ^ 2 \label{eq : sineff}\ ] ] that also have a dependence on the medium density . the angle @xmath63 is defined as @xmath64 ^ 2 } { 1 + \left [ 2\alpha^{2}_{32}g(\rho)\right]^{2 } } \label{eq : neta}\ ] ] in mavan s case for @xmath65 induce that the mass basis it is not diagonal and the parameter @xmath63 is the mixing angle that diagonalizes the mass basis as shown in reference @xcite . this made that the non - diagonal mavan s would induces neutrino oscillations even if the vacuum mixing angle was zero , @xmath66 . in the case of standard oscillations , see eq . ( [ standardoscillation ] ) we have the symmetry @xmath67 , but in mavan case with @xmath65 we broken this symmetry and the eq . ( [ eq : sineff ] ) have different results for @xmath68 compared when @xmath69 . for vanishing mavan parameters , @xmath70 , we have @xmath71 , and we recover the standard neutrino oscillation . for very large values of mavan parameters we have @xmath72 , that implies that @xmath73 . if we have large mixing angles @xmath74 we will have suppression of the amplitude , but for smaller mixing angles will have a enhancement of the amplitude . most of the previous analysis of mavan works for the first two generations @xcite and then theirs constrains are related to parameters of the first generation @xmath75 and @xmath76 similar to the parameters defined in eq . ( [ mevo - drg ] ) . from the reference @xcite that provide a upper bound for the elements of matrix @xmath77 ev at 90 % c.l . the only case that works in the mavan scenario for the second and third families are the references @xcite . in this works they assume the non - zero parameters @xmath78 and the medium matter dependence is given by @xmath79 , where @xmath18 is the matter density , @xmath80 and _ n _ is a free parameter . it was ruled out diagonal mavan to be the dominant oscillation scenario for atmospheric neutrinos and they constraint the _ n _ parameter to be in the range @xmath81 at 90 % c.l . @xcite . now we are going to compare the oscillation probabilities without mavan s and with mavan s to understand the changes in the oscillation probabilities . most of previous analysis use a model for the mavan s mass @xmath82 given as a function of parameter with dimensions of energy such as @xmath83 , with the index i denoting the diagonal mass eigenstates i=1,2 and the non - diagonal mass eigenstates i=3 ; with different functions @xmath48 as function of the matter density @xmath18 , and a parameter @xmath84 with dimension of energy . we decide to adopt a dimensionless parameters in this work , the @xmath85 as described in eq . ( [ mevo - drg ] ) , but we can related our parameters @xmath85 with the previous analysis by the replacing of @xmath86 , where @xmath87 is the relevant mass difference of the analysis . in this work we decide to study the case when for non - diagonal mavan parameter , @xmath52 and @xmath53 , vide eq . ( [ mevo - drg ] ) . as an example e will use in this work the mavan density dependence as used in ref . @xcite , given by @xmath88 in this work we will use the matter profile of earth that we take from @xcite . where @xmath18 is the matter density that neutrino crosses and @xmath89 g/@xmath90 is the matter density at earth core . this choice is motivate to generates a soft and well behaved function even with the abrupt variations of earth density profile , other choices made the computation numerical unstable as reported in ref . also it have a finite limit for @xmath91 and @xmath92 always . in this work we will use the matter profile of earth that we take from @xcite . we solve numerically compute the muon neutrino survival probability , @xmath93 for the general case of a neutrino crossing the earth from different chords . we can related the traveled distance by neutrino _ l _ to the zenith angle @xmath94 by @xmath95 , where @xmath96 km is earth radius and _ h _ is the point of atmosphere were neutrinos are produced , approximately @xmath97 km . for @xmath98 we have the maximum ( mininum ) distance . we shown in fig . ( [ fig : pmumu - cz ] ) the muon neutrino probability as a function of cosine of zenith angle , @xmath99 for fixed values of the amplitude of the mixing angle @xmath100 , for the mass difference @xmath101 ev@xmath23 and for a fixed energy @xmath102 gev and several values of @xmath47 : @xmath103 we have respectively the mavan probabilities as dashed curves and without mavan as solid curves in fig . ( [ prob - cz-04 ] , ( [ prob - cz-01]),([prob - cz-02 ] ) and ( [ prob - cz-03 ] ) . for values of @xmath104 , the neutrino traveled in vacuum only and the oscillation probability with and without mavan are the same . for @xmath105 , the neutrino cross inside the earth and the mavan effect begin to pop up . to see more clearly the effect of mavan parameters we can compare the fig . ( [ prob - cz-04 ] ) that have @xmath106 , with fig . ( [ prob - cz-03 ] ) that have @xmath107 . when can see when we increase the mavan parameter @xmath47 , for maximal mixing , the mavan amplitude get smaller and the oscillation phase increases giving more fast wiggles that appear in the probability . the increase of wiggles made the maximums and minimums move to smaller values of @xmath108 . the concept of oscillograms is interesting tool to understand the complete behavior of neutrino probability in some model for neutrino oscillation . we plot in fig . ( [ fig : pmumu_alpha0.1 ] ) the oscillograms of muon neutrino survival probability , denoted by @xmath109 , as function of neutrino energy @xmath110 and the cosine of zenith angle , @xmath111 . in the fig ( [ oscillogram-1-mavan ] ) we show for the standard neutrino case and in the others plots for increasing bigger values of mavan parameter . for the cosine of zenith angle , @xmath112 , when the neutrino did not cross the earth , we have zero mavan effect and for @xmath113 , the muon survival probability is modified due mavan mechanism . for comparison we show in the upper - left panel , in the upper - right panel , lower - left panel lower - right panel of n fig . ( [ fig : pmumu_alpha0.1 ] ) for mavan parameters @xmath47 respectively equal to 0.0 , 0.1 , 0.5 and 1.0 . an enhancement of @xmath114 also implies that the positions of maximums and minimums is dislocated to higher values of neutrino energy . as an example of this let we look to the first minimum in @xmath109 for @xmath115 . in the former three panels of fig ( [ oscillogram-1-mavan ] ) we see that the first minimum ocurs for @xmath116 gev . in the lower - left panel , where @xmath107 this minimum in @xmath109 had its intensity reduced ( due to the increase of @xmath117 but also we see that the minimum was dislocated to @xmath118 gev . at @xmath116 gev now we see the first maximum of oscillation that for s.o . occur at @xmath119 gev . the same kind of displacement is found for all the maximums and minimums in the lower - left panel of fig ( [ oscillogram-1-mavan ] ) when compared with the cases in which @xmath120 . atmospheric neutrinos , composed muon neutrinos and electron neutrinos , are produced all around the earth atmosphere and traveled to super - kamiokande from all directions . once in the detector they interact producing muons and electrons and super - kamiokande measure the zenith angle dependence of these muons and electrons . the rate for these events can be computed as @xmath121\theta[e^{{\rm max}}_{\mu}-e_{\mu}(\cos\theta_{z},e_{\nu},\cos\theta_{\mu } ) ] ~ , \label{nmu}\end{aligned}\ ] ] where @xmath122 is the number of targets in sk , @xmath123 is the livetime , @xmath33 is the neutrino energy , @xmath124 is the cosine of zenith angle ( @xmath94 ) of the neutrino , @xmath125 is the azimuth angle of the neutrino , @xmath126 is the cosine of zenith angle of the muon , @xmath127 is the muon neutrino flux , @xmath128 is the muon neutrino survival probability , @xmath129 is the differential charged current muon - neutrino cross - section . for the integration boundaries , @xmath130 and @xmath131 the initial and final neutrino energies ; @xmath132 and @xmath133 the bin of zenith angle distribution of super - kamiokande experiment , in this we use 10 equal bins of muon zenith angle between @xmath134 . we compute this expression for the 2 different types of events in sk experiment : the so - called _ sub - gev _ data set and _ mult - gev _ data set . they correspond respectively to the intervals of @xmath135 gev , @xmath136 gev ) and @xmath137 gev , @xmath138 gev ) . we use the kinematics of reaction , defined by @xmath139 and @xmath33 variables to set - up the allowed range of muon energy , given by the function @xmath140 and we constrain to be in the _ sub - gev _ and _ mult - gev _ energy range . notice that the zenith angle of leptons it is a function of zenith angle of neutrino , the scattering angle and the energy of neutrino and this produce a stronger averaging effect on the original neutrino direction . for the electron - like zenith distribution of events in sk we can write @xmath141\theta[e^{\rm max}_{e}-e_{e}(\cos\theta_{z},e_{\nu},\cos\theta_{e } ) ] ~ , \label{ne}\end{aligned}\ ] ] which it is very similar to the muon events , with the exception that the electron neutrino did not oscillate , due our initial assumption only have oscillations between muon and tau neutrinos . the muon and electron neutrino fluxes are taken from @xcite . the differential cross section follows @xcite , whe ere we divide the cross section in three parts , first the quasi - elastic neutrino scattering with finite mass corrections @xcite ; second the one pion contribution and third the dis contribution @xcite . due the super - kamiokande did not discriminate between particles and anti - particles , we sum over neutrino and anti - neutrino types . the muon and electron rate for atmospheric neutrinos have the uncertainties from the prediction of the atmospheric neutrino fluxes , @xmath127 and @xmath142 , computation that can be @xmath143 and the relative error @xmath144 of 5% @xcite . because of this error in the absolute normalization we will use as the physical observable is the zenith distribution of number of events and the absolute value of our prediction will be scaled with the experimental data . also the smallness of the relative error of muon and electron neutrino fluxes compared the error in the absolute number implies a stronger correlation between the fluxes of muon and electron neutrinos , that we should take into account . we use this formalism to describe two energy ranges of sk data , the so - called _ sub - gev _ and _ mult - gev _ data set for electrons and muons , eq . ( [ nmu ] ) and eq . ( [ ne ] ) . we compute the muon neutrino survival probability , given in eq . ( [ standardoscillation ] ) , with the oscillation parameters given by @xmath101 ev@xmath23 and the amplitude @xmath145 shown as the black curve in fig . ( [ fig : muonsub ] ) and fig . ( [ fig : muonmulti ] ) respectively for sub - gev and multi - gev sample . for these parameters we have no oscillation for @xmath146 and average out oscillation for @xmath147 . our results match the theoretical curves for the number of events for no - oscillation and standard oscillation of sk experiment . for our computation with mavan s probability , we use the numerical solution of eq . ( [ eqevo - drg ] ) using the matter profile given in eq . ( [ dmevo1 ] ) and we compute the rate for electron and muon ( given by eq . ( [ nmu ] ) and eq . ( [ ne ] ) ) , the result is the dashed curve in fig . ( [ fig : muonsub ] ) and fig . ( [ fig : muonmulti ] ) . the effect of non - zero mavan , using the same oscillation parameters and the mavan parameter @xmath148 it to distort the muon distribution making a small oscillation for cosine of lepton zenith @xmath149 , coming from neutrino zenith angle @xmath150 , due the averaging effect mentioned after the eq . ( [ nmu ] ) ; and suppressing the averaging of neutrino oscillations for @xmath147 , both behavior are disfavored by the sk data and from this we expect to have a constrain in the mavan parameter from this data . in atmospheric neutrinos , neutrino can came from different directions and they probe different medium density making the mass difference @xmath151 to be different in each point of the travel . on the other hand , for minos , the neutrino travel only cross the upper crust , and we can assume that density along this short chord is constant , we will assume @xmath152 g/@xmath90 from the prem model . we can use the results of section ( [ sec : frame ] ) of mavan mechanism for constant medium density where the muon survival tprobability is given by eq . ( [ case2mavan ] ) with the mavan mass difference given by eq . ( [ eq : effdm ] ) and the amplitude by eq . ( [ eq : sineff ] ) . the interesting is that the functional form of mavan oscillation probability is exactly like the standard oscillation probability , with the replacement @xmath153 and @xmath154 . are indicated in the figure . also we assume @xmath155 and @xmath156 ev@xmath23 . points refers to minos data from @xcite . ] until now we present a phenomenological framework for off - diagonal mavan s mechanism and show it changes the @xmath157 atmospheric neutrino oscillation pattern in the sk experiment . however , the sk detector is not the only one sensitive to the neutrino oscillations in the @xmath157 sector and we can use the minos data @xcite to constrain the non - diagonal mavan parameter . we show in fig . ( [ minoscte ] ) the comparison between the standard neutrino oscillation and the mavan s probability for the setup of minos experiment ( using the l = l@xmath158 and the density @xmath152 g/@xmath90 ) . the oscillations parameters are fixed as @xmath159 ev@xmath1 and maximal mixing @xmath160 . when we increase the mavan parameter we have for maximal mixing that the mavan amplitude get suppressed @xmath161 , as you can see that the minimum , in fig . ( [ minoscte ] ) , is less deeper for higher values of non - diagonal mavan parameter @xmath47 . at same that the oscillation phase , @xmath162 is bigger for higher values of @xmath47 , that it implies that the minimum should be for higher values of energy . we also show for comparison in fig . ( [ minoscte ] ) the ratio of experimental number of events over the theoretical prediction without oscillation as data points to emphasize that when we increase the value of non - diagonal mavan parameter we became further apart from the experimental data . so far we have shown examples of mavan effects for maximal vacuum mixing , @xmath160 , e.g. in fig . ( [ fig : pmumu - cz],[fig : pmumu_alpha0.1],[fig : neve],[minoscte ] ) . a subtle effect can happen for the constant medium density case for larger values of mavan parameter , @xmath47 , in this case we have that the mavan amplitude @xmath163 as commented in section ( [ sec : model ] ) . for small vacuum mixing angles @xmath164 this implies that the amplitude is enlarged compared without mavan and for larger vacuum mixing angles @xmath164 the situation is the opposite , and you have suppression of oscillation . to give a example we display in fig . ( [ sineff - ii ] ) , the vacuum mixing angle @xmath165 as a function of mavan parameter @xmath47 that gives a fixed value of mavan amplitude @xmath166 ( in another words the inverse function of eq . ( [ eq : sineff ] ) ) . we can see that it have two solutions : one , the black curve , with small vacuum mixing angle @xmath165 and another solution given by the dashed red curve with large vacuum mixing angle . this implies that we can have a small mixing angle and large @xmath167 parameter or with can begin a large mixing angle and small parameter @xmath167 and both give a effective large mavan amplitude . for @xmath168 , both curves coincide and we have full suppression of oscillation amplitude . we will see the consequences of subtle effect later in our analysis in section ( [ sec : chi2 ] ) . parameter that can made a fixed valur for the mavan amplitude @xmath166 ] here we give the details of data analysis . first we begin with the analysis of minos experiment and later to the super - kamiokande analysis for the standard vacuum oscillation scenario and for the mavan scenario . c.l . for oscillations parameters @xmath170 without ( s.o ) , dotted blue line , and with mavan s , dashed red line for the minos analysis . ] the minos experiment made a likelihood analysis of theirs data for _ neutrinos and anti - neutrinos _ for standard neutrino oscillation and produces a table with the values of @xmath171 , where @xmath172 , where @xmath173 is the likelihood value and @xmath174 is the likelihood for the best fit . this table is publicly available in ref . we can translated the likelihood language into @xmath175 language using @xmath176 . for the mavan analysis , we can use the property that the mavan probability have the same functional form of standard oscillation , as discussed in section ( [ sec : minos ] ) , any function of probability also should have similar behavior . therefore the @xmath177 function given in ref . @xcite as a function of @xmath178 and @xmath179 , should be equal to mavan @xmath180 as a function of the @xmath181 and @xmath162 parameters . numerically we have @xmath182 where for minos experiment we can use the expression for mavan amplitude @xmath181 and mass difference @xmath162 respectively eq.([eq : sineff ] ) and eq.([eq : effdm ] ) that give the mavan parameters as functions of the vacuum oscillation parameters @xmath178 and @xmath179 . we use this procedure to get the allowed region for minos only , for the standard oscillation scenario and for the mavan case . the results is shown in fig . ( [ minosonly - regi ] ) where the dotted blue curve correspond to standard neutrino oscillations region of @xmath183 of c.l .. as we plot the region of allowed region in the standard scenario as an function of @xmath184 ( and not in function of @xmath185 ) then two degenerated minimums ( denoted by up triangles ) do appear . when we include mavans and minimize with respect to @xmath47 we get the allowed region is given by the red dashed curve which has the best fit for non - zero @xmath47 , and for vacuum mixing angles smaller then maximal @xmath186 and mass differences @xmath187 ev@xmath1 . in the mavan s allowed region the mixing parameters can have larger values of @xmath188 and smaller values of @xmath28 that are not allowed in standard oscillation scenario . smaller values of @xmath189 ev@xmath1 are not allowed in the standard scenario due it implies smaller oscillation effect that is in contradiction with the minos data , but in mavan mechanism , the effective mass difference @xmath162 can be larger then vacuum oscillation mass difference compensating the smaller value of @xmath179 . also the value of @xmath190 in the mavan solution , if we are working in standard scenario will implies a smaller oscillation amplitude that also it is not compatible with minos data , but also we can allow values @xmath184 far for maximal for non - zero @xmath47 . now we will work with the analysis of sk data , where we will use the muon and electron data for sub - gev and multi - gev samples . the sample of atmospheric neutrino data is specially interesting for the mavan oscillation effect because it is composed by events produced by neutrinos traveling in vacuum and in matter from the use the muon and electron data for sub - gev and multi - gev sample as discussed in section iii . to settle the basics of our analysis we should be aware that the predicted atmospheric neutrino flux have a uncertainty of 30 % in the normalization and also a stronger correlation between the fluxes , the relative error between the muon and electron flux is around 5% @xcite . as said before , due this normalization error we will test the shape of muon and electron distribution and not the absolute number of events . we made the following way , we are going to analysis the shape of atmospheric neutrino data and not include the comparison of the absolute value of data . to do this we will change our theoretical prediction of oscillation @xmath191 , where @xmath192 is the normalization parameter with a error of @xmath193 , with @xmath194 for each flavor . also we should use a correlation between the electron and muon number of events due the correlation of the neutrino fluxes . we made two analysis : 1 . for the standard oscillation scenario to test our ability to reproduce the results of sk analysis for the oscillation parameters @xmath179 and @xmath195 . we reproduce the main characteristics of with @xmath196 ev@xmath1 and near maximal mixing ; 2 . the mavan scenario for non - diagonal parameter @xmath167 , with @xmath197 , where our goal is reproduce the angular distribution predicted by sk experiment using our computation made independently of the sk experiment and from this to do a reliable analysis of mavan phenomena . our choice of @xmath175 function is @xmath198 where @xmath199 is the number of events in the bin @xmath200 measured by sk , @xmath201 is our prediction for the number of events for the bin @xmath200 that depend on the oscillation model used ; the non - diagonal matrix @xmath202 such the diagonal entries have error of @xmath203 as said before , and the non - diagonal entries fixed by the correlation between the muon and electron flux @xcite . the sum is over the 40 bins : @xmath204 , counting 2 flavors and 2 samples : sub - gev or multi - gev , bins . we add the second term in eq . ( [ xq2-new ] ) to introduce a penalty factor when @xmath205 assume values to far from @xmath206 with respect to the error in normalization , @xmath207 . . summary of our @xmath30 analysis for combined data from super - kamiokande and minos experiment . the first line corresponds to pure standard neutrino oscillations(s.o . ) and the second line corresponds to mavan scenario . in each case are show the best fit ( b.f . ) values of the parameters . [ cols="^,^,^,^,^,^",options="header " , ] the minos experiment test the mavan scenario for a constant density that implies that the mavan parameters are fixed , and combining with sk experiment that test the mavan for variable density made specially adequate the main hypothesis of mavan idea , to have a mass difference and the mixing amplitude that depends on the local density . to achieve this we combine the two analysis , from minos experiment and from sk experiment , we will use the @xmath30 test for both experiments super - kamiokande and minos , @xmath208 with this @xmath175 with three oscillation parameters , first we analyze the standard oscillation scenario and second the mavan scenario for this combined analysis of minos and sk . w show the best fit values for both scenarios in tab . ( [ tab : tab01 ] ) . from this information we can see that there is a milder improvement of the mavan solution over the standard neutrino oscillation for the combining fit . for the mavan analysis of the combination of minos and sk , we have found that the best fit is for a non - zero value of mavan parameter ( @xmath209 and the mixing parameters @xmath210 ev@xmath23 and mixing angle @xmath211 and the best fit parameters for standard oscillation are very similar @xmath212 ev@xmath23 and @xmath213 ( see tab . ( [ tab : tab01 ] ) . for the cases with ( dashed red curve ) and without mavan s effect ( solid black line ) is compared with the result from minos collaboration ( dotted blue curve ) . we also show the values of best fit points respectively in each analysis . ] we shown in the _ central panel _ of fig . ( [ fig : deltachi - dmsin-2 ] ) the allowed region of @xmath214 and @xmath188 parameters for the standard oscillation scenario ( s.o . for now ) and for the mavan scenario for the following cases 1 . first for analysis of the _ standard oscillations ( s.o . ) _ in minos experiment alone shown the dotted blue curve , 2 . second , for the _ standard oscillations ( s.o . ) analysis _ for the _ combination _ of minos and sk experiments shown by the black curve . we can compare with the minos only result the improvement on the determination of range of @xmath188 for the combination . the recent values from global fits for these parameters @xcite are included in our @xmath215 allowed region for @xmath28 and @xmath216 , 3 . third , for the _ mavan scenario _ analysis for the _ combination _ of minos and sk experiments shown by dashed red curve , all plots shown the 90 % c.l . allowed region . comparing the mavan allowed region for minos experiment only shown in fig . ( [ minosonly - regi ] ) and for the combination of minos and sk data ( shown in central panel of fig . ( [ fig : deltachi - dmsin-2 ] ) ) we see that combining the two experiments we constrain more the allowed region of parameters . another good tool to understand our solution is the plot of the projection of @xmath217 function with respect the one of three oscillation parameters , @xmath218 and @xmath47 . when we show the projection , e. g. for example @xmath219 we have minimized over the other parameters , @xmath188 and @xmath47 and so on . we begin with the the _ top panel _ of fig . ( [ fig : deltachi - dmsin-2 ] ) we shown the plot of the projection of @xmath220 for the standard oscillation ( black curve ) and mavan ( dashed red curve ) together with the 66 % , 90 % and 95% c.l . respectively @xmath221 . we can see that the combination of minos and sk suppress the high values of @xmath222 that are present in the minos only analysis , they appear only at high c.l . ( not shown ) when we have @xmath223 . these high values are only possible in minos analysis for @xmath224 , however this large value will induce stronger changes in oscillation probabilities for the sk experiment due larger density that the neutrino feels when cross the earth , typically we have @xmath225 . we can notice in fig . ( [ fig : pmumu - cz ] ) that for @xmath224 we expect to see a stronger suppression of the muon neutrino oscillation that will conflict the sk data . in the _ right panel _ of fig . ( [ fig : deltachi - dmsin-2 ] ) we shown the plot of the projection of @xmath226 for also both standard oscillation and mavan scenario respectively by black and dashed red curve . we can see a slightly increase of allowed region for @xmath179 when we have a non - zero mavan parameter . for @xmath227(dotted - dashed blue line ) , @xmath183(dashed red line ) , and @xmath228(solid green line ) c.l .. black square is the best fit point . the auxiliary upper plot is the same of fig . ( [ fig : deltachi - dmsin-2 ] ) . the left auxiliary plot refers to @xmath229 as function of @xmath47 and minimized with respect to the other mixing parameters , @xmath230 and @xmath179 . ] we show in central panel of fig . ( [ fig : deltachi_aisin ] ) the allowed region of parameters @xmath231 and minimized with respect to @xmath28 at @xmath227 , 90 % and 95 % c.l . you can see from this plot that there is a correlation between the @xmath184 and @xmath167 for range of values allowed , for the highest @xmath47 value we have the widest range for @xmath184 . this is left - over of behavior , discussed in section ( [ sec : minos ] ) , that we can have a smaller vacuum mixing amplitude for higher @xmath47 or a large vacuum mixing amplitude for smaller @xmath47 . in the right panel we show the plot of the projection of @xmath232 where we can see the standard oscillation scenario ( when @xmath51 ) is compatible at @xmath233 and the more interesting information from this plot is that there is no allowed region for @xmath234 at 90 % c.l . . the solutions for higher @xmath235 present in the minos analysis and only appear now for the combined analysis at @xmath236% c.l . for the same reasons discussed in the previous paragraph . at end , to completeness , in central panel of fig . ( [ fig : deltachi_dmai ] ) we show the allowed region of parameters as function of @xmath237 . in this central panel , we can see that the correlation between the values of @xmath238 and @xmath47 are much milder then between the @xmath239 as shown in fig . ( [ fig : deltachi_aisin ] ) . the reason for this is the change in @xmath162 due @xmath47 it is not so stronger compared with the change in @xmath240 . for @xmath227(dotted - dashed blue line ) , @xmath183(dashed red line ) , and @xmath228(solid green line ) c.l .. black square is the best fit point . the auxiliary upper(right ) plot refers to @xmath229 as function of @xmath28(@xmath47 ) and minimized with respect to other parameters . ] in this work we test the possibility of neutrinos have their mass dependent on the medium density . the so - called mavan s model includes this new feature through a new neutrino interaction mediated by a scalar field . we investigate the consequences of such model in the phenomenology of atmospheric neutrino data from super - kamiokande experiment and for the data from minos experiment using for the first time to test the existence of non - diagonal mavan parameter . using the fact the mavan for the constant matter density , as it is the case for the minos experiment , have the same functional form of standard oscillation scenario we use the minos analysis for the standard neutrino oscillation and we extend for the mavan scenario . in this analysis larger values of non - diagonal mavan parameter , @xmath241 , are allowed and we have significant changes in the allowed region for oscillation parameters , @xmath214 and @xmath242 . we compute by ourselves the event rate for super - kamiokande experiment , involves the correct description of the _ sub - gev _ and _ multi - gev _ neutrino energy samples and we also use the analysis of the minos experiment to made a combined fit of these two experiments . in minos experiment the neutrino crosses only a small chord of the earth and then the neutrino feels a constant matter density and for other side the super - kamiokande have neutrinos coming from different directions and therefore feels different medium densities . then by combining the two data , we can test the essence of mavan hypothesis , the mass dependence on the medium density . we have found that the best fit values for @xmath0 ev@xmath1 , @xmath2 and mavan parameter @xmath3 and the best fit values for the standard oscillation give similar values(see table ( [ tab : tab01 ] ) ) . the allowed region for the oscillation parameters , @xmath214 and @xmath242 is shown in fig ( [ fig : deltachi - dmsin-2 ] ) that shows that the allowed region is very stable with the addition of mavan scenario . although the best fit is for non - zero mavan parameter we have all values of mavan parameter @xmath243 are ruled out at 90% c.l that allow us to conclude that the non - diagonal mavan coupling should be give smaller contribuition to neutrino oscillation . a. g. riess _ et al . _ [ supernova search team collaboration ] , astron . j. * 116 * , 1009 ( 1998 ) [ astro - ph/9805201 ] . s. perlmutter _ et al . _ [ supernova cosmology project collaboration ] , astrophys . j. * 517 * , 565 ( 1999 ) [ astro - ph/9812133 ] . p. astier _ et al . _ [ snls collaboration ] , astron . astrophys . * 447 * , 31 ( 2006 ) [ astro - ph/0510447 ] . p. gu , x. wang and x. zhang , phys . d * 68 * , 087301 ( 2003 ) [ hep - ph/0307148 ] . x. -j . bi , p. -h . gu , x. -l . wang and x. -m . zhang , phys . d * 69 * , 113007 ( 2004 ) [ hep - ph/0311022 ] . n. afshordi , m. zaldarriaga and k. kohri , phys . d * 72 * , 065024 ( 2005 ) [ astro - ph/0506663 ] . m. honda , r. takahashi and m. tanimoto , jhep * 0601 * , 042 ( 2006 ) [ hep - ph/0510018 ] . v. barger , p. huber and d. marfatia , phys . * 95 * , 211802 ( 2005 ) [ hep - ph/0502196 ] . m. cirelli , m. c. gonzalez - garcia and c. pena - garay , nucl . b * 719 * , 219 ( 2005 ) [ hep - ph/0503028 ] . m. c. gonzalez - garcia , p. c. de holanda and r. zukanovich funchal , phys . d * 73 * , 033008 ( 2006 ) [ hep - ph/0511093 ] . p. c. de holanda , jcap * 0907 * , 024 ( 2009 ) [ arxiv:0811.0567 [ hep - ph ] ] . f. rossi - torres , m. m. guzzo , p. c. de holanda and o. l. g. peres , phys . d * 84 * , 053010 ( 2011 ) [ arxiv:1012.5304 [ hep - ph ] ] . m. f. carneiro and p. c. de holanda , advances in high energy physics , * article i d 293425 * , ( 2013 ) ; available at http://dx.doi.org/10.1155/2013/293425 doi:10.1155/2013/293425 . k. k. shiraishi , _ super - kamiokande atmospheric neutrino analysis of matter - dependent neutrino oscillation models _ , phd . thesis , at http://www - sk.icrr.u - tokyo.ac.jp / doc / sk / pub/. k. abe _ et al . _ [ super - kamiokande collaboration ] , phys . rev . d * 77 * , 052001 ( 2008 ) [ arxiv:0801.0776 [ hep - ex ] ] . p. adamson _ et al . _ [ minos collaboration ] , phys . rev . lett . * 110 * , 251801 ( 2013 ) [ arxiv:1304.6335 [ hep - ex ] ] . m. c. gonzalez - garcia , h. nunokawa , o. l. g. peres , t. stanev and j. w. f. valle , phys . d * 58 * , 033004 ( 1998 ) [ hep - ph/9801368 ] . d. r. gratieri , phd thesis , appendix e , http://webbif.ifi.unicamp.br/tesesonline/teses/if1564.pdf a. d. dziewonski and d. l. anderson , physics of the earth and planetary interiors * 25 * , 297 ( 1981 ) . m. honda , t. kajita , k. kasahara , s. midorikawa and t. sanuki , phys . d * 75 * , 043006 ( 2007 ) [ astro - ph/0611418 ] .

a bose - einstein condensate ( bec ) is a state of matter in which a large number of bosons occupy the same quantum mechanical ground state . as such , becs present the opportunity to study quantum systems which display large scale ( macroscopic ) collective behaviour . recently there has been much interest in multi - component becs because of their importance to quantum optics @xcite . multi - component becs are most often formed in a multi - well potential in which the components are spatially separated @xcite . alternatively , the multi - component formalism can be used to model a single component bec that possesses several internal degrees of freedom , such as varying amounts of spin @xcite . in general , many - body systems are of significant importance in physics . for example , quantum information processes require the manipulation and control of systems with large numbers of particles . however , many - body systems are often difficult to treat exactly and are most often studied using numerical or approximate methods . the realm of applicability of these methods is limited by the number of degrees of freedom of the system . two common approximate models are the bose - hubbard model of quantum optics @xcite and closely related to it , the lipkin - meshkov - glick model of nuclear physics @xcite , both used to describe the two - body interactions of spin-@xmath8 systems . a family of exactly solvable many - body systems was introduced in @xcite and studied in greater depth in @xcite . these models can be used to describe the physics of a two component bec where both elastic and mode - exchange inelastic collisions are present . mode - exchange collisions are known as general nearest neighbor interactions in the context of multi - well becs @xcite and as inelastic collisions in the case of a single condensate consisting of particles in two hyperfine levels @xcite . the family of models is parameterized by a positive integer @xmath9 , and hence the specific models are called @xmath9-models . this terminology is used because the @xmath9-model contains @xmath10body interactions . analytic solutions can be found for the @xmath9-model when the strengths of the various particle interactions obey specific constraints . while the @xmath9-models are of interest to many - body physics in general , in this paper we will largely be concerned with the @xmath11-model where only single body interactions and two - body collisions are considered . microscopic calculations show that mode - exchange collisions occur in becs as a result of the interaction of a laser field with the system @xcite . the 2-model includes the usual josephson - type interactions , but also allows the effects of mode - exchange collisions to be studied analytically , and as such provides a more realistic framework for studying two - mode becs than the canonical josephson hamiltonian @xcite . the analytic solution found in @xcite requires that the strengths of the various interactions meet certain constraints ; if these constraints are not met the solution is invalid . while in some experimental situations the rates of inelastic and elastic mode - exchange collisions can be controlled externally @xcite , this is not always the case . moreover , even in if the rate of collisions can be manipulated , the constraints will likely still only be approximately satisfied . it is thus of interest to extend the solution space of the system to the case in which these constraints are only approximately satisfied . this naturally leads to the consideration of small parameter perturbations in the model to study its robustness . along with perturbations within the model it is of interest to include additional interaction terms as perturbations , such as inelastic collisions resulting in particle loss . particle loss , often suppressed in experimental settings , is studied theoretically by considering classical rate equations @xcite . more recently , a quantum treatment of 3-body loss has been given in @xcite . including external perturbations extends the predictive power of the model . additional interactions of primary interest include three - body recombination , background collisions , spin exchange and dipole relaxation @xcite . to our knowledge , this is the first study which analyzes particle loss as a perturbative effect in becs . motivated by the above considerations , in this paper we carry out a full perturbative analysis of the solvable bec 2-model proposed in @xcite . we begin by reviewing this model and the solutions derived when the constraint equations on the parameters are satisfied @xcite . we then introduce perturbations to the parameters in the model and analyze the effects on the particle distribution , entanglement and the evolution of the relative number operator . in later sections we discuss the effects of state degeneracy and the inclusion of a general loss term as a perturbation , illustrating the formalism by studying background collisions and three - body recombination . the @xmath11-model studied in @xcite is governed by the hamiltonian @xmath12 where @xmath13 denotes the hermitian conjugate of the preceding term . the two modes @xmath14 and @xmath15 are independent bose operators satisfying @xmath16=1=[b , b^{\dagger}]$ ] with the commutators of all remaining pairs vanishing . the term @xmath17 is the free energy of @xmath18 particles in the @xmath14 mode and @xmath19 particles in the @xmath15 mode , with frequency difference @xmath20 between modes . this frequency difference arises because the model describes atoms in different hyperfine levels , or alternatively unsymmetric spatially separated condensates @xcite . a josephson - type or spin flip interaction is included with strength @xmath21 and phase @xmath22 . in practice such an interaction is induced by an external field , such as a laser @xcite . this interaction may also be interpreted as modeling the tunneling of particles between modes with probability proportional to @xmath21 . also included are terms corresponding to number - preserving elastic and mode - exchange collisions , namely those terms containing four bose operators . the interaction with strength @xmath23 is a single dispersive process . the interaction with strength @xmath24 describes a collision where two particles exchange their mode . the elastic interaction has strength @xmath25 and models the collision of a particle from each mode in which the number of particles in each mode is conserved . we see that the mode - exchange collisions preserve the total particle number but not the relative particle number . the hamiltonian given in eq . ( [ becham ] ) can be efficaciously studied by first introducing the simpler hamiltonian @xcite @xmath26 with real constants @xmath27 and @xmath28 . such a hamiltonian models a two - mode condensate with energy difference @xmath27 between modes and elastic scattering probability proportional to @xmath28 . as is clear from the form of @xmath29 the relative number operator @xmath30 is a commuting observable so that the number of particles in each mode is conserved . in particular , there is no probability of spin - flip or tunneling between modes . we also note that the total number operator @xmath31 commutes with @xmath29 . we may thus take the eigenvalues @xmath32 and @xmath33 of @xmath34 and @xmath35 as labels of the eigenstates , for which we write @xmath36 . for fixed @xmath37 the values of @xmath33 are limited to @xmath38 and @xmath32 . the energy of the state @xmath36 is @xmath39 . for a certain choice of parameters in @xmath40 the solutions @xmath41 of @xmath29 can be used to obtain analytic solutions of @xmath40 . to see this , define a two - mode displacement operator by @xmath42 with displacement parameter @xmath43 . it is clear that @xmath44 is unitary . it can then be shown that if the parameters in @xmath40 satisfy [ coeffid ] @xmath45 @xmath46 @xmath47 @xmath48 @xmath49 @xmath50 then @xmath51 , a result shown by computing @xmath52 observe that @xmath53 may be computed by setting @xmath54 in eq . ( [ trans ] ) . since the eigenvectors of @xmath29 are of the form @xmath55 the eigenvectors of @xmath40 when satisfying eqs . ( [ coeffid ] ) are simply @xmath56 with energy @xmath57 . an extensive analysis and discussion of these solutions is included in @xcite . for fixed @xmath37 the ground state of @xmath40 , labeled as @xmath58 , is found by minimizing the energy @xmath59 with respect to @xmath33 . first assume @xmath60 . in this case , if @xmath61 ( @xmath62 ) the minimum occurs at @xmath63 ( @xmath64 ) . secondly , say @xmath65 . if @xmath66 , then @xmath67 is the closest allowable can only take the values @xmath68 and @xmath32 ; see the appendix for details . ] integer to @xmath69 ; otherwise @xmath67 is the closest allowable integer to @xmath70 . in particular , a choice of either @xmath67 or @xmath71 determines the other . , so that a choice of @xmath67 determines @xmath27 . ] it is important to recall that , in the case of a double - well bec , the two - mode approximation must be satisfied @xcite . this means that collisions taking place in the region where the wavefunctions overlap must be less probable than collisions between particles belonging to the same well . it is possible to find an exact analytic solution to the model in this case by considering @xmath72 . it is interesting to observe that it is not possible to find analytical solutions to this model for an exactly symmetric double - well and satisfy , simultaneously , the two mode approximation . fortunately , in the case of slightly asymmetric double wells ( which corresponds to a more realistic situation ) it is possible to find exact analytical solutions in the two - mode approximation @xcite . the most general @xmath9-model hamiltonian discussed in @xcite has as its hamiltonian @xmath73 where @xmath74 . the eigenvectors of @xmath75 are again @xmath76 with the same energy @xmath77 . the same process of matching coefficients in the expansion of @xmath78 that was used to obtain eqs . ( [ coeffid ] ) can be carried out so that @xmath76 is the exact solution of a hamiltonian containing up to @xmath9-body interactions . in the general case , there are @xmath79 free variables ( the @xmath9 @xmath80 as well as @xmath81 and @xmath22 ) in @xmath82 , whereas the number of terms in a general hamiltonian describing such interactions is @xmath83 as shown in appendix b. so while the above method continues to give analytic solutions for all @xmath9 , its range of applicability decreases as @xmath9 grows . it should also be noted that because @xmath84 contains only terms with the same number of creation and annihilation bose operators , this formalism can not be used to study interactions involving an odd number of bose operators , such as the interaction @xmath85 where one particle is lost . we remark that all of the perturbative analysis carried out below is valid not only for the @xmath11-model , but also for the more general @xmath9-models . this is so because the solutions for all @xmath9-models are @xmath86 . we need only change the interpretation of the perturbative analysis in each case . while satisfying eqs . ( [ coeffid ] ) is sufficient to obtain an analytic solution of @xmath40 , the method presented above fails to produce solutions if the parameters in @xmath40 deviate even slightly from these constraints . thus , in order to compare this model to experimental results we would like to study the solutions to @xmath40 when the constraints are only approximately satisfied . we proceed with these perturbation calculations in the section below . we begin by perturbing each of the coupling constants in @xmath40 away from the conditions given by eqs . ( [ coeffid ] ) , i.e. perturbations of the form @xmath87 where @xmath20 satisfies eqs . ( [ coeffid ] ) and @xmath88 is small . we omit the study of perturbations @xmath89 since this simply leads to a shift in the energy of the state by @xmath90 . in what follows , we assume that all eigenstates are non - degenerate ; the degenerate case is discussed in section 7 . throughout the paper , given an operator @xmath91 we define @xmath92 . to perform many of the calculations that follow we have made use of a coordinate transformation between the @xmath93 basis and the @xmath94 basis , @xmath95 where @xmath96 is a eigenvalue of @xmath97 and similarly for @xmath98 . a change @xmath87 results in the perturbation @xmath99 . a calculation then shows that the non - vanishing matrix elements of @xmath100 are @xmath101 and @xmath102 a change @xmath103 results in the perturbation @xmath104 and we find that @xmath105 and @xmath106 a change @xmath107 results in the perturbation @xmath108 and we find that @xmath109 @xmath110 and @xmath111 \left [ n(n+2 ) -(m \pm 2)(m\pm 4 ) \right ] } .\ ] ] a change @xmath112 results in the perturbation @xmath113 and we find that @xmath114 @xmath115 and @xmath116 a change @xmath117 results in the perturbation @xmath118 and we find that @xmath119 @xmath120 and @xmath121 in each of the cases above the perturbation matrix elements simplify significantly in the limiting cases @xmath122 , in which the josephson coupling vanishes ( @xmath123 ) , and @xmath124 in which the potential well is symmetric ( @xmath125 ) . we elaborate below on the effect of @xmath81 on the particle distributions . in the unperturbed case we can use the analytic solution @xmath58 for parameters satisfying eqs . ( [ coeffid ] ) to find an explicit expression for the particle distribution : @xmath126 using the homomorphism described in the appendix to relate the schwinger @xmath127 representation to the angular momentum representation we write @xmath128 where are the wigner rotation matrix elements under the effect of the lie algebra homomorphism discussed in the appendix . to see why these appear , observe that @xmath129 where @xmath130 and @xmath131 is the total angular momentum vector operator . that is , @xmath132 , and hence @xmath133 , are rotations of the algebra generated by @xmath14 and @xmath15 . ] @xcite @xmath134 and @xmath135 the integers @xmath136 are chosen so that the arguments of the combinatorial symbols are non - negative ; explicitly , @xmath137 and @xmath138 . we plot in figure [ partdistplot ] the unperturbed particle distribution for @xmath139 , @xmath140 and @xmath141 and @xmath142 . observe that the particle distribution is independent of phase @xmath22 . we will see that this does not hold in the perturbed case . as @xmath81 varies the distributions shift along the @xmath33-axes ; for @xmath81 small the maxima shift toward @xmath143 , and move toward @xmath144 as @xmath81 grows . the canonical @xmath11-mode bec predicts ( under certain circumstances ) that the ground state solution is a superposition of two peaked distributions . from figure [ partdistplot ] we see that for @xmath145 the ground state is a superposition of more than two distributions , an effect that is due to the mode - exchange collisions not included in the canonical model @xcite . given by eq . ( [ probdens ] ) for @xmath139 , @xmath140 and _ a ) _ @xmath146 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eq . ( [ probdens ] ) for @xmath139 , @xmath140 and _ a ) _ @xmath146 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eq . ( [ probdens ] ) for @xmath139 , @xmath140 and _ a ) _ @xmath146 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] in order to include perturbative effects in the particle distribution we replace the zeroth order ground state wave function @xmath149 in eq . ( [ probdens ] ) with that including first order corrections , @xmath150 recall that in the non - degenerate case a perturbation @xmath151 induces a first order wave function correction given by @xmath152 where the superscripts indicate to what order in the perturbation the term corresponds ; when no superscript is present the result is to zeroth order . with this , we find that to first order the particle distribution is given by with the understanding that we actually mean @xmath153 . summations of this form arise because of the restriction on the eigenvalues of the relative number operator . ] @xmath154 we show in figures [ partdistomega ] - [ partdistmu ] the particle distributions under each of the five parameter perturbations considered above . note that in certain cases @xmath155 may be negative because the first order term in eq . ( [ pertpd ] ) depends linearly on the perturbation coefficients @xmath156 ; this of course is just a reflection that only first order corrections have been taken into account . we see from eq . ( [ pertpd ] ) and the fact that @xmath156 is proportional to the sign of the perturbation that each of the perturbations can be used to either enhance or diminish the particle distributions in the vertical direction for a given @xmath33 . that is , given a fixed relative number @xmath33 , by choosing the sign of the perturbation appropriately we can increase or decrease the probability of the system being in the state @xmath41 . we also see that as @xmath157 decreases the number of local maxima of the particle distribution increases , and the asymmetry of the perturbations becomes more significant , vertically diminishing probability amplitudes on one side of the central maxima while vertically stretching the amplitudes on the opposite side . the figures also show that the system is much less sensitive to perturbations in @xmath20 and @xmath21 than it is to perturbations in the collision coupling constants ; in the figures the perturbations of @xmath20 and @xmath21 are between @xmath158 and @xmath159 times greater than those in the collision perturbations ( @xmath160 and @xmath25 ) , while the corrections in all of the cases are of the same scale . moreover , out of the collision parameters the particle distribution is most sensitive to perturbations in coupling constants of the _ mode - exchange _ collisions ( @xmath23 and @xmath24 ) . it is thus beneficial to maximize the matching of the mode - exchange collision constraints ( [ coeffid ] ) through redefinition of parameters , allowing the perturbation to have a larger effect on the @xmath20 , @xmath21 and @xmath25 terms . the effect on the particle distributions of varying @xmath81 is to shift the position of the central maxima on the @xmath33-axis : for @xmath124 , corresponding to the case in which the josephson - type interaction is maximal ( @xmath161 ) and the energy of each mode is equal ( @xmath162 ) , the distribution is centered around @xmath163 , while for @xmath164 ( @xmath165 ) , in which the josephson - type interaction vanishes ( @xmath166 ) and the energy difference of the modes is maximal ( @xmath167 ) it is centered @xmath143 ( @xmath144 ) . in each of the cases , despite being centered around different values of @xmath33 the perturbations display the same qualitative behaviour for any choice of @xmath81 . throughout the paper we have chosen @xmath140 in the figures as this lies between the extremes of @xmath168 and @xmath169 . as @xmath81 varies the effects of the perturbations vary in size , but show the same characteristics . for example , as @xmath81 decreases from @xmath170 to @xmath171 ( but with @xmath172 fixed ) , the perturbations have a larger effect on the particle distributions , which indicates that for small values of @xmath81 , we must be more careful about the size of the chosen @xmath172 . this can be explained as follows . in the perturbed case the area under the particle distributions sums to unity only to first order in the perturbation . this sum is also a function of @xmath81 ( since the perturbed terms are ) , unlike in the unperturbed case . we have found that as @xmath81 decreases the sum of probabilities decreases too , which explains why there is a difference in the particle distributions when we keep the size of the perturbations @xmath172 fixed . if we decrease both @xmath81 and @xmath172 , then we find that the particle distributions essentially only shift along the @xmath33-axis , but not in their general shape . given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath173 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath173 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath173 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath175 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath175 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath175 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath176 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath176 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath176 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath177 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath177 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath177 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath178 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath178 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ @xmath148.,title="fig:",width=153,height=153 ] given by eqs . ( [ probdens ] ) and ( [ pertpd ] ) for @xmath139 , @xmath140 , @xmath178 and _ a ) _ @xmath174 , _ b ) _ @xmath147 and _ c ) _ entanglement arises in many - body quantum systems because of the superposition principle and the tensor product structure of the hilbert space . ] , a property of utmost importance in quantum control and quantum information . in particular , states with high entanglement are desirable because of their utility in carrying out quantum information tasks @xcite . we would like to determine whether or not by careful choice of the perturbations we can increase the entanglement from that of the unperturbed case . for a bipartite quantum system the von - neumann entropy @xmath179 is a standard measure of the entanglement of the system , @xmath180 being the reduced density matrix . in the case at hand this reduces to @xmath181 where @xmath182 is either the perturbed or unperturbed particle distribution , depending on the situation . as noted in the section above , the first order particle distribution given by eq . ( [ pertpd ] ) can be negative for certain choices of perturbations . to resolve this problem we could either replace @xmath155 with @xmath183 or we could include the second order term @xmath184 in @xmath155 ; we use the former approach to keep the calculations strictly to first order . when the particle distribution contains first order corrections , @xmath185 , the entanglement is given by @xmath186 for the case at hand we read off that @xmath187 . hence we have a criterion for increasing the entanglement of the system . namely , we increase the entanglement ( to first order ) precisely when @xmath188 note that @xmath189 is proportional to the perturbation strengths , so that we can increase or decrease the entanglement by choosing the sign of the perturbation appropriately . continuing , we note that for each @xmath33 we have @xmath190 , so that to maximize the entanglement we should maximize each @xmath191 . of course , @xmath192 can be made arbitrarily large simply by choosing @xmath172 arbitrary large . however , we are limited in such a choice since we must keep @xmath172 small so that perturbation theory can be trusted . ) for @xmath193.,width=192,height=192 ] we plot in figure [ ent_unpert ] the unperturbed entanglement as a function of @xmath81 and @xmath67 . see @xcite for an extensive analysis . in figures [ ent_omega]-[ent_mu ] we plot the perturbed entanglements @xmath194 for @xmath195 as well as the differences @xmath196 between the perturbed and unperturbed entanglements for @xmath197 ; we choose a different @xmath172 in the latter case so that the differences are more evident . in each of the plots of @xmath198 we see that the largest changes occur approximately along the diagonal lines , connecting the points @xmath199 and @xmath200 and the points @xmath201 and @xmath200 ; these lines in @xmath202 space correspond to certain strengths of the coupling constants in the hamiltonian @xmath40 viewed as functions of @xmath67 . note this type of behaviour also occurs along these lines in the unperturbed entanglement plots , as shown in figure [ ent_unpert ] . comparing the magnitude of the perturbative effects we observe that perturbations to the mode - exchange collision terms ( @xmath203 ) have a much greater effect , their maximum difference being about an order of magnitude larger than those for @xmath21 and @xmath20 . it is also interesting to observe that the coherent states , which correspond to @xmath204 and @xmath205 are the states which are maximally affected by perturbations . in such states @xmath206 is much larger than @xmath207 so that the rate of collisions is relatively small . to further study this we have plotted in figure [ entan_cut_plots ] a two - dimensional cut of figures [ ent_omega]_b)_-[ent_mu]_b ) _ , where we have fixed @xmath208 and allowed @xmath67 to vary . we observe that in each of the plots @xmath198 has both a local maximum and local minimum as @xmath67 approaches @xmath32 . we also observe that the most significant changes occur in the region @xmath209 , where more particles lie in the @xmath14 mode . positive @xmath67 implies that , for fixed @xmath210 , the scattering length for same - mode collisions is positive and as @xmath67 approaches @xmath32 , the collision rate becomes smaller . we then conclude that condensates with negative scattering lengths are more resilient to parameter perturbations and high same - mode collision rates help stabilize the condensate . as a function of @xmath67 and @xmath81 for @xmath211 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath213.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath211 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath213.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath214.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath214.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath215.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath215.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath216.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath216.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath217.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath217.,title="fig:",width=192,height=192 ] for @xmath218 and @xmath23 , respectively , for @xmath208 as a function of @xmath67 . the perturbation strengths used are all @xmath197.,title="fig:",width=115,height=105 ] for @xmath218 and @xmath23 , respectively , for @xmath208 as a function of @xmath67 . the perturbation strengths used are all @xmath197.,title="fig:",width=115,height=105 ] for @xmath218 and @xmath23 , respectively , for @xmath208 as a function of @xmath67 . the perturbation strengths used are all @xmath197.,title="fig:",width=115,height=105 ] for @xmath218 and @xmath23 , respectively , for @xmath208 as a function of @xmath67 . the perturbation strengths used are all @xmath197.,title="fig:",width=115,height=105 ] for @xmath218 and @xmath23 , respectively , for @xmath208 as a function of @xmath67 . the perturbation strengths used are all @xmath197.,title="fig:",width=115,height=105 ] with an analytic solution to the system we may study the evolution of the relative population @xmath219 as a function of time , where the initial state is given by @xmath220 . for simplicity we restrict our study to the case in which @xmath221 . to first order in the perturbation parameter we have @xmath222 we compute @xmath223 where @xmath224 with @xmath225 and @xmath226 + \right . \\ & & \displaystyle \left . \sum_{l =- n+2}^{n } a_{m , l } c_m c_{l-2 } \sqrt{n(n+2 ) - l ( l-2 ) } \cos \left [ \phi - ( e_{l-2 } - e_m ) t \right ] \right ) \biggr\}. \label{pertev } \end{array}\ ] ] for @xmath211 . figures _ b ) _ and _ c ) _ plot the perturbed evolution @xmath227 given by eqs . ( [ unpertev ] ) and ( [ pertev ] ) for @xmath211 with @xmath228 and @xmath229 , respectively . , title="fig:",width=192,height=192 ] for @xmath211 . figures _ b ) _ and _ c ) _ plot the perturbed evolution @xmath227 given by eqs . ( [ unpertev ] ) and ( [ pertev ] ) for @xmath211 with @xmath228 and @xmath229 , respectively . , title="fig:",width=192,height=192 ] for @xmath211 . figures _ b ) _ and _ c ) _ plot the perturbed evolution @xmath227 given by eqs . ( [ unpertev ] ) and ( [ pertev ] ) for @xmath211 with @xmath228 and @xmath229 , respectively . , title="fig:",width=192,height=192 ] we plot in figure [ evoplot ] _ a ) _ the unperturbed evolution of relative population given by eq . ( [ unpertev ] ) . we see the rabi - like oscillations with relative population collapse and revival . in figures [ evoplot ] _ b ) _ and _ c ) _ we plot the evolution of relative population under the parameter perturbations @xmath230 and @xmath231 , respectively . we observe that as @xmath88 increases the time - averaged value of @xmath232 decreases ; from eq . ( [ pertev ] ) changing the sign of @xmath88 would have increased this average value . we see in figure [ evoplot ] _ c ) _ that the perturbation has broken down ; the maximum value of @xmath233 is greater than the total particle number . this breakdown reminds us that we must not let the perturbation grow too large for our analysis to be reliable . we also see ( in figure [ evoplot ] _ c _ ) that as @xmath88 grows the time of population collapse significantly decreases . because of the complexity of the correction term eq . ( [ pertev ] ) an analytic study of the effects of the perturbations on population collapse and revival times is not possible . > from eq . ( [ unpertev ] ) we see that in the unperturbed case an initially pure state remains pure . eq . ( [ pertev ] ) shows that the same holds true even in the perturbed cases . hence , @xmath234 has non - trivial time - dependence if and only if the initial state is entangled . throughout the above analysis we have assumed that the unperturbed states in question are non - degenerate . we proceed now to study the degenerate case , which results for specific values of the constants @xmath27 and @xmath28 . for fixed total particle number @xmath32 two distinct states , labeled by relative population numbers @xmath235 and @xmath236 , have the same energy precisely when @xmath237 . from the analysis of the perturbations completed above we know that the perturbation matrix elements are non - vanishing only if @xmath238 , where we have assumed without loss of generality that @xmath239 . combining these two observations we see that there are at most two pairs of degenerate states . let us consider as an example the perturbation @xmath87 with @xmath240 a_2 $ ] , so that the only pair of degenerate states is @xmath241 and @xmath242 ; each have energy @xmath243 . the matrix of interest is @xmath244 its eigenvalues and eigenvectors yield the first - order energy and wave function corrections . using the calculations performed above it remains to find the solutions @xmath245 of the quadratic @xmath246 . the eigenvalues ( energy corrections ) and corresponding eigenvectors ( wave function corrections ) of @xmath247 are @xmath248 @xmath249 with @xmath250 a suitable normalization constant . figure [ epsilonplot_omega ] _ a ) _ plots @xmath245 as a function of @xmath81 for @xmath139 and @xmath251 . we see that the energy is lowered for all values of @xmath81 ; had @xmath88 been negative the opposite would have been true . we see from figures [ epsilonplot_omega ] _ b ) _ and _ c ) _ , which plot the perturbed particle distributions for @xmath252 , that even for very small perturbations in @xmath20 there is a noticeable change in the particle distribution . this is in contrast to the non - degenerate perturbation of @xmath20 , where even for @xmath253 there was not a large change in the particle distribution . the induced change in the particle distribution is also qualitatively different from that in the non - degenerate case . this arises because the correction in the degenerate case is sinusoidal with frequency much greater than that of the unperturbed particle distribution . regardless of the sign of @xmath88 we see that the central maximum of the particle distributions is shifted to smaller values of @xmath33 . we plot in figure [ ent_degen_omega ] _ a ) _ the perturbed entanglement as a function of @xmath81 and @xmath67 for @xmath230 and @xmath193 , while _ b ) _ of the same figure plots the difference @xmath198 for the same configuration . even for a small perturbation strength @xmath230 the perturbation to the entanglement is still significant , again showing that a degenerate system is more sensitive to perturbations than is the non - degenerate case . we also observe that the largest perturbations are present for @xmath67 close to @xmath254 and @xmath255 . for @xmath67 close to @xmath32 , the perturbations become negligible , regardless of @xmath81 . therefore , we find that high collision rates also help stabilize the condensate against perturbations in the degenerate case . however , in this case , condensates with positive scattering lengths are more stable . similarly , we find that for a perturbation @xmath103 , assuming again that @xmath241 and @xmath256 are the degenerate states , the energy corrections are @xmath257 which is just the energy correction equation for perturbations in @xmath20 with the substitution @xmath258 ; we obtain the wavefunction corrections @xmath259 for @xmath260 in the same manner . figure [ epsilonplot_lambda ] plots the perturbed energy and perturbed particle distributions for @xmath261 and @xmath262 . the same comments made above about figure [ epsilonplot_omega ] for perturbations @xmath88 hold in this case as well . in figure [ ent_degen_lambda ] we plot the perturbed entanglement and entanglement difference @xmath198 for @xmath263 . the figure shows similar behaviour to figure [ ent_degen_omega ] . however , we note that in the case of @xmath21 perturbations @xmath198 is strictly positive ( for positive @xmath260 ) . there are again extrema ( this time both maxima ) for @xmath64 and @xmath264 , with @xmath198 vanishing as @xmath67 approaches @xmath32 . also as in the case above , the system is very sensitive to perturbations , with @xmath265 at some points . ( solid ) and @xmath266 ( dot ) from eq . ( [ degen_en_omega ] ) for degenerate perturbations with @xmath139 and @xmath267 as a function of @xmath81 . figures _ b ) _ and _ c ) _ plot the corresponding unperturbed ( dot ) and perturbed ( solid ) particle distributions @xmath182 with @xmath139 , @xmath140 for @xmath251 and @xmath268 , respectively.,title="fig:",width=192,height=192 ] ( solid ) and @xmath266 ( dot ) from eq . ( [ degen_en_omega ] ) for degenerate perturbations with @xmath139 and @xmath267 as a function of @xmath81 . figures _ b ) _ and _ c ) _ plot the corresponding unperturbed ( dot ) and perturbed ( solid ) particle distributions @xmath182 with @xmath139 , @xmath140 for @xmath251 and @xmath268 , respectively.,title="fig:",width=192,height=192 ] ( solid ) and @xmath266 ( dot ) from eq . ( [ degen_en_omega ] ) for degenerate perturbations with @xmath139 and @xmath267 as a function of @xmath81 . figures _ b ) _ and _ c ) _ plot the corresponding unperturbed ( dot ) and perturbed ( solid ) particle distributions @xmath182 with @xmath139 , @xmath140 for @xmath251 and @xmath268 , respectively.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 with @xmath230 for @xmath193.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 with @xmath230 for @xmath193.,title="fig:",width=192,height=192 ] ( solid ) and @xmath266 ( dot ) from eq . ( [ degen_en_lambda ] ) for degenerate perturbations with @xmath139 and @xmath261 as a function of @xmath81 . figures _ b ) _ and _ c ) _ plot the corresponding unperturbed ( dot ) and perturbed ( solid ) particle distributions @xmath182 with @xmath139 , @xmath140 for @xmath269 and @xmath270 , respectively.,title="fig:",width=192,height=192 ] ( solid ) and @xmath266 ( dot ) from eq . ( [ degen_en_lambda ] ) for degenerate perturbations with @xmath139 and @xmath261 as a function of @xmath81 . figures _ b ) _ and _ c ) _ plot the corresponding unperturbed ( dot ) and perturbed ( solid ) particle distributions @xmath182 with @xmath139 , @xmath140 for @xmath269 and @xmath270 , respectively.,title="fig:",width=192,height=192 ] ( solid ) and @xmath266 ( dot ) from eq . ( [ degen_en_lambda ] ) for degenerate perturbations with @xmath139 and @xmath261 as a function of @xmath81 . figures _ b ) _ and _ c ) _ plot the corresponding unperturbed ( dot ) and perturbed ( solid ) particle distributions @xmath182 with @xmath139 , @xmath140 for @xmath269 and @xmath270 , respectively.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 with @xmath263 for @xmath193.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 with @xmath263 for @xmath193.,title="fig:",width=192,height=192 ] the study of degenerate perturbations in the remaining parameters is completed in the same way as is done above , so we omit these . note that high collision rates help stabilize the condensate against perturbations in @xmath21 for both positive and negative scattering lengths . as mentioned in the introduction it is of interest to study perturbations that model additional interaction terms not included in the original hamiltonian eq . ( [ becham ] ) . we will largely be interested in perturbations that do not preserve the total number of particles in the system . we begin with a general discussion of loss terms and proceed to use this formalism to discuss the effects of background collisions and three - body recombination . interactions in becs that do not preserve the total particle number are often minimized in experimental settings as there is currently no known analytical model that involves such terms . while the hamiltonian @xmath40 we are studying also does not include such terms , in the case that the effects of particle loss terms are expected to be minimal , we may treat loss terms as perturbations to the system . a primary source of particle loss is inelastic collisions @xcite . in magnetic traps particle - type exchange terms dominate loss mechanisms , while in optical traps , these terms may be neglected @xcite . it is thus of interest to obtain predictions of the effects of loss terms by treating them as perturbations to the exactly solvable @xmath11-model . the most general loss term can be written as @xmath271 where @xmath272 are some functions and @xmath273 is the total particle number of the condensate . by probabilistic arguments we expect @xmath274 as @xmath275 grows , so that the cases of primary interest are those with @xmath276 and @xmath277 small . however , it should be noted that higher order collisions ( i.e. not just @xmath278 ) are of physical significance , particularly when the condensate is in its coldest phase and of high particle density @xcite . by suitably choosing @xmath279 we can model specific loss terms . in order to study loss terms we must first make some adjustments to the analysis performed above . the full hilbert space @xmath280 of the hamiltonian eq . ( [ simpham ] ) can be orthogonally decomposed as the fock space @xmath281 where , in the @xmath282 basis , @xmath283 . since the hamiltonian and perturbations we have considered so far have all commuted with the total number operator we have been able to first choose a total particle number @xmath32 for the system , or equivalently the subspace @xmath284 of the total hilbert space , and then proceed with calculations . in order to study loss terms we must enlarge the state space to be @xmath285 where @xmath286 is the set of all accessible total particle numbers . for example , if @xmath287 then @xmath288 . the energy of the state @xmath36 is @xmath289 . although @xmath77 is only functionally dependent on @xmath33 , it has an implicit dependence on @xmath32 since @xmath32 restricts the values of @xmath33 . hence , while @xmath290 is non - degenerate as an element of @xmath291 , it may be degenerate as an element of @xmath292 depending on @xmath33 and @xmath293 . again , considering the example in which @xmath287 we see that @xmath294 is non - degenerate if @xmath295 and degenerate otherwise . in general , let @xmath296 and say that @xmath297 is degenerate if it contains more than one element and say it is non - degenerate otherwise . we examine the effects of the degeneracy @xmath297 below . note that the degeneracy studied in this section is caused by the interactions ( which determine the hilbert space ) , whereas the degeneracy studied in the previous section was caused by a specific choice of the coupling constants @xmath27 and @xmath28 . in order to study the particle distribution of the perturbed state @xmath298 we must modify eq . ( [ probdens ] ) ; if we were to use this formula there would be no perturbative effects on the particle distribution since @xmath299 and @xmath300 are orthogonal if @xmath301 . a suitable generalization is given by @xmath302 observe that in the case of no loss terms @xmath303 so that @xmath304 reduces to eq . ( [ probdens ] ) . if @xmath297 is degenerate it is easy to see that the matrix of interest to degenerate perturbation theory @xmath305 is triangular with zeros along the diagonal . indeed , @xmath306 contains only annihilation terms , and conjugation by @xmath133 , denoted here by @xmath307 , maps annihilation operators to annihilation operators . so @xmath308 is a sum of states with total particle number @xmath9 less than @xmath309 , showing the matrix at hand is triangular , and thus has a trivial spectrum consisting of only zeros . hence we can learn nothing from first order perturbation theory . alternatively , if @xmath297 is non - degenerate we may apply the tools of non - degenerate perturbation theory . although it is straightforward to compute the matrix elements of @xmath306 given by eq . ( [ lossham ] ) in general , the requirement that @xmath297 be non - degenerate severely limits the usefulness of such a calculation . we instead focus on some specific choices of @xmath279 that model interactions of physical interest . background collisions most often occur in becs when particles from the condensate collide with a residual background gas in the condensate chamber , or alternatively , with metastable atoms within the condensate @xcite . background collisions become more important as the density of the condensate increases . as a simple illustration of how we may treat background collisions as perturbations , consider the case in which the initial state is @xmath310 , so that only @xmath14 mode particles can be ejected . the general loss term if @xmath311 and @xmath312 otherwise . ] ( [ lossham ] ) , after conjugation by @xmath133 , is written as @xmath313 where we use @xmath314 to denote terms with more @xmath15 powers than @xmath315 powers . note that any term of @xmath314 annihilates @xmath241 so that the perturbative correction can be found by neglecting all such terms . we compute the desired matrix elements to find that @xmath316 which yields the generalized particle distribution @xmath317 where we set @xmath318 if @xmath319 . from this expression we see that we could have omitted terms that eject an odd number of @xmath14-mode particles . indeed , if @xmath320 is odd , then @xmath321 is identically zero as a function of @xmath33 since if @xmath322 then @xmath323 and @xmath9 all have the same parity . we plot @xmath304 in figure [ bgcplots ] , considering terms that eject @xmath324 and @xmath325 particles for @xmath139 and @xmath140 . in figure [ bgcplots]_a ) _ we set @xmath326 , @xmath327 and @xmath328 . in figures [ bgcplots]_b ) _ - [ bgcplots]_d ) _ we increase each of the @xmath329 by a factor of @xmath11 . the figures show that increasing each @xmath330 decreases the height of the particle distribution . also evident from the figures and the scale of the @xmath329 is that as @xmath331 increases the perturbations have a larger effect . that the perturbations do not blow up reflects the requirement that @xmath274 for large @xmath275 indeed , we have found using numerical simulations that the perturbations that eject @xmath320 particles diverge with increasing @xmath320 . ) , for @xmath139 with _ a ) _ @xmath332 , @xmath333 and @xmath334 , _ b ) _ @xmath335 , @xmath333 , @xmath334 , _ c ) _ @xmath336 , @xmath337 and @xmath334 and _ d ) _ @xmath332 , @xmath333 and @xmath338 . , title="fig:",width=115,height=115 ] ) , for @xmath139 with _ a ) _ @xmath332 , @xmath333 and @xmath334 , _ b ) _ @xmath335 , @xmath333 , @xmath334 , _ c ) _ @xmath336 , @xmath337 and @xmath334 and _ d ) _ @xmath332 , @xmath333 and @xmath338 . , title="fig:",width=115,height=115 ] ) , for @xmath139 with _ a ) _ @xmath332 , @xmath333 and @xmath334 , _ b ) _ @xmath335 , @xmath333 , @xmath334 , _ c ) _ @xmath336 , @xmath337 and @xmath334 and _ d ) _ @xmath332 , @xmath333 and @xmath338 . , title="fig:",width=115,height=115 ] ) , for @xmath139 with _ a ) _ @xmath332 , @xmath333 and @xmath334 , _ b ) _ @xmath335 , @xmath333 , @xmath334 , _ c ) _ @xmath336 , @xmath337 and @xmath334 and _ d ) _ @xmath332 , @xmath333 and @xmath338 . , title="fig:",width=115,height=115 ] we remark here that as a consequence of @xmath321 vanishing for @xmath320 odd we can not learn anything about spin - flip terms from first order perturbation theory . indeed , such terms would be modeled by perturbations of the form @xmath339 and @xmath340 , assuming particles in the @xmath14 mode have greatest energy ; these terms clearly reduce the total number of the system by an odd number . setting @xmath191 equal to the second term on the right hand side of eq . ( [ losspd ] ) we can use eqs . ( [ pertent]-[incent ] ) to compute and increase the entanglement , respectively . since @xmath191 depends linearly on the interaction strengths @xmath341 we see the most obvious manner in which to increase the entanglement is to make @xmath342 large ; the sign of @xmath341 depends on @xmath343 as well as @xmath27 and @xmath28 . , @xmath333 and @xmath334 ( as in figure [ bgcplots ] a ) as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for the same configuration.,title="fig:",width=192,height=192 ] , @xmath333 and @xmath334 ( as in figure [ bgcplots ] a ) as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for the same configuration.,title="fig:",width=192,height=192 ] figure [ ent_bgc ] plots the perturbed entanglement caused by the inclusion of background collisions as a function of @xmath67 and @xmath81 . we see that the largest effect of the perturbation occurs in the region @xmath344 and small ( large negative ) values of @xmath67 . there is also a minimum of smaller magnitude around @xmath345 and @xmath64 . the region with large negative @xmath67 corresponds the case where the scattering length between particles is negative , in which most particles lie in the @xmath15 mode of the condensate . since the background collisions considered here eject particles from the @xmath14 mode , they serve to further decrease the value of @xmath33 . we understand the large effect of the perturbation on the aforementioned region as follows : since most particles lie in the @xmath15 mode , ejecting any particles from the @xmath14 mode has a large effect on the system , since it already has only a small number of @xmath14 mode particles , relative to the number of @xmath15 mode particles . figure [ ent_bgc ] _ b ) _ shows that for values of @xmath67 above the region in question , where the @xmath15 mode particles become more scarce , the perturbation has little effect on the system . again , we understand this as being because ejecting an @xmath14 mode particle from a state with large @xmath33 value is of little significance to the system as a whole . we also see from this figure that the entanglement is decreased , regardless of @xmath67 and @xmath81 , for this specific choice of @xmath329 . perhaps the most physically important type of inelastic collision leading to particle loss is three - body recombination ( tbr ) @xcite . tbr occurs when three particles in a single mode collide to form a diatomic molecule and a particle of the same mode that carries off any excess energy . depending on this energy and the energy of the potential trap , the resultant particle may or may not escape the trap @xcite . a perturbation modeling tbr would be most naturally treated in the @xmath346 formalism where the extra bose operator would correspond to the diatomic molecule . however , in the @xmath127 formalism we have no such third mode . we thus model tbr by the hamiltonian @xmath347 the parameter @xmath348 describes the probability of the emitted particle remaining trapped in the condensate . the term proportional to @xmath349 lowers the total particle number by an odd number and hence , as explained above , will have no effect on the perturbed particle distribution . we thus neglect this term from our analysis and absorb the constant @xmath350 into @xmath348 . with @xmath351 as a perturbation we have @xmath352 . we take as our unperturbed state @xmath241 , which is non - degenerate . proceeding , we find @xmath353 which gives the generalized particle distribution @xmath354 @xmath304 given by eq . ( [ tbrpd ] ) is plotted in figure [ tbrplots ] . as expected the the sign of @xmath348 determines whether the perturbation vertically shrinks or stretches the particle distribution . we also see that the system is very sensitive to three - body recombination terms , as a coupling constant of order @xmath355 causes significant changes to the particle distribution . once again , setting @xmath191 equal to the last term in eq . ( [ tbrpd ] ) , we can use eqs . ( [ pertent ] ) and ( [ incent ] ) to study the perturbed entanglement . the choice of the sign of @xmath348 to increase the entanglement again depends on the values of @xmath81 , @xmath27 and @xmath28 . ) , for @xmath139 and @xmath140 with _ a ) _ @xmath356 , _ b ) _ @xmath357 and _ c ) _ @xmath358.,title="fig:",width=192,height=192 ] ) , for @xmath139 and @xmath140 with _ a ) _ @xmath356 , _ b ) _ @xmath357 and _ c ) _ @xmath358.,title="fig:",width=192,height=192 ] ) , for @xmath139 and @xmath140 with _ a ) _ @xmath356 , _ b ) _ @xmath357 and _ c ) _ @xmath358.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath359.,title="fig:",width=192,height=192 ] as a function of @xmath67 and @xmath81 for @xmath193 and _ b ) _ the difference @xmath212 for @xmath193 and @xmath359.,title="fig:",width=192,height=192 ] we have successfully studied the effects of a number of perturbations to the two - mode bec model considered in @xcite . we have found the corrections to the condensate wave functions , which in turn allowed the determination of the corrections to the particle distribution , time - evolution of the relative number operator and the entanglement . in the non - degenerate case , we have shown that the model of @xcite , in which the coupling constants of the hamiltonian are constrained , is robust to perturbations in these constants . the system is most sensitive to perturbations in the elastic scattering length @xmath25 and in the mode - exchange parameters @xmath24 and @xmath23 . in each of the parameter perturbations the entanglement of the coherent states @xmath310 and @xmath360 is most affected . each parameter perturbation was observed to increase the asymmetry in the particle distributions . we also observed that for specific values of @xmath67 and @xmath81 , the latter corresponding to certain parameter strengths , the entanglement perturbations are especially large . it was found that when the condensate is degenerate ( because of specific choices of @xmath27 and @xmath28 ) it is much more sensitive to perturbations , both in terms of particle distributions and entanglement . the effects on the entanglement are qualitatively different than in the non - degenerate case . in particular , the perturbations to the entanglement are mainly present only in the regions in which @xmath67 is close to @xmath254 which corresponds to a condensate with small negative scattering length . we have also extended the formalism to include the analysis of interactions involving particle loss . from these we can predict corrections to the particle distribution , entanglement and evolution of the relative population . this provides a new class of possible experiments that will allow the model here to be tested . indeed , interactions involving particle loss have been limited experimentally , partially because they create instabilities in the system . with our results these interactions could be allowed to occur , and the results compared with the predictions contained above . the changes induced by both three - body recombination and background collisions are qualitatively different than those induced by parameter perturbations . we find that the system is very sensitive to these external perturbations , which induce large changes in the particle distribution and entanglement from relatively small external coupling strengths , as compared to the induced changes from parameter perturbations . as with the degenerate parameter perturbation , we found that the perturbative effects on the entanglement become negligible as @xmath67 approaches @xmath32 , i.e. when the scattering length is small and positive . in general , we can conclude that higher collision rates make the condensate more stable to perturbations . our results promise to be useful in the experimental realization of two - mode bose - einstein condensates which are stable to parameter perturbations and particle loss . we are planning to extend our analysis to include many - body interactions which are present at cooler stages of the condensate and study the role of such interactions in the stability of the condensate . let @xmath361 and @xmath362 be the usual generators of the lie algebra @xmath127 , satisfying @xmath363=\pm j_{\pm } , \;\;\;\ ; [ j_+ , j_- ] = 2j_z.\ ] ] the eigenstates are labeled as @xmath364 where @xmath365 note that for fixed @xmath366 , @xmath367 may take any of the @xmath368 values @xmath369 . we also have @xmath370 . the schwinger representation of @xmath127 defines a lie algebra homomorphism between the angular momentum representation ( generated by @xmath371 and @xmath361 ) and a bosonic representation . consider two bose operators @xmath14 and @xmath15 with @xmath372 = 1 = \left[b , b^{\dagger}\right]$ ] and all other pairs having vanishing commutator . defining the mapping @xmath373 and extending linearly to the rest of @xmath127 then gives the desired homomorphism @xcite . a short calculation then shows that @xmath374 is mapped to @xmath375 where we have defined @xmath376 . we label the basis states in the bosonic representation as @xmath377 with @xmath32 the eigenvalue of @xmath34 and @xmath33 the eigenvalue of @xmath378 ; this is in complete analogy with the label @xmath379 . note that @xmath380 may take the @xmath381 values @xmath382 . from the definition of the homomorphism it follows that @xmath383 . if we then rescale the state @xmath377 to @xmath384 the above identity is rewritten as @xmath385 that @xmath384 is mapped to a multiple of @xmath386 can be seen directly from the form of the operator @xmath387 , which annihilates a particle in mode @xmath15 while creating one in mode @xmath14 . similarly we have @xmath388 now rescaled , @xmath33 may take on the @xmath389 values @xmath390 . it is of interest to quantify the generality of the @xmath9-model hamiltonian under study . we begin this below by first counting the number of terms in the most general hamiltonian that would be of interest to us . to do this we must define precisely the hamiltonians that are of interest to our study of two - mode becs . first , we limit ourselves to hamiltonians that are polynomials in the bose operators . the interactions under consideration consist of any total number preserving operations . operators corresponding to such interactions must then commute with the total number operator @xmath34 . to restrict the class of interactions , we consider only those that have no intermediate interactions , such as the spontaneous creation and annihilation of a particle . this imposes the restriction that the hamiltonian be normal ordered . as usual , we also require self - adjointness of the hamiltonian . finally , we may decompose the hamiltonian into its homogeneous parts , i.e. terms of degree @xmath391 . so , it is sufficient to first consider only homogeneous hamiltonians , and then construct more general hamiltonians from these . with these assumptions we prove the following proposition . let @xmath392 be a homogeneous , self - adjoint , normal - ordered polynomial in the bose operators @xmath14 and @xmath15 and their adjoints . furthermore , assume that each term in @xmath392 commutes with the total number operator @xmath31 . . then the number of terms in @xmath392 is at most @xmath394 . let @xmath395 denote the number of terms in the hamiltonian . observe that the requirement that each term in @xmath392 commute with @xmath34 ensures that @xmath396 for some @xmath397 . consider now a monomial of @xmath392 with @xmath398 of the operators being either @xmath14 or @xmath399 while the remaining @xmath400 being either @xmath15 or @xmath315 . it is easy to check that , assuming self - adjointness , normal ordering and vanishing commutator with @xmath34 , there are @xmath401 such terms , where for @xmath402 , @xmath403 is the greatest integer less than or equal to @xmath404 . we may group the monomials of @xmath392 into three groups according to whether there are more than , less than , or the same number of @xmath14 mode terms as @xmath15 mode terms . doing so , we find that the maximal number of terms in @xmath392 is @xmath405 since @xmath406 for all @xmath407 we obtain the recursion relation @xmath408 . repeated application of this relation yields @xmath409 , which we may rewrite as @xmath410 since @xmath411 , from which the proposition follows . since we can decompose a general polynomial in terms of its monomials , the above proposition is sufficient to count the maximal number of terms in the most general hamiltonian described above . let @xmath392 be as above without the assumption of homogeneity with @xmath412 . then the number of terms in @xmath392 is at most @xmath413 . again , we can write @xmath396 for some integer @xmath414 . the number of terms in the hamiltonian is the sum of the number of terms in each of its homogeneous parts , i.e. @xmath415 . after some algebra the corollary follows . as noted in the proofs of the above proposition and corollary , the assumptions on the hamiltonian imply @xmath416 is even . we see that the total number of terms in the most general @xmath9-model hamiltonian grows like @xmath417 . we would like to compare this result to the number of terms in the @xmath9-model of @xcite . unfortunately , we have not been able to successfully count the number of terms in the @xmath9-model . in order to do so , one would first need to conjugate @xmath418 by the displacement operator @xmath419 . the number of terms in this calculation grows quickly with @xmath9 which makes the conjecture of a formula for the number of terms difficult . moreover , general arguments to count the number of terms , as in the proposition above , seem difficult to make in this case . regardless , the results for @xmath420 and @xmath421 , summarized in table [ count_tab ] , suggest that there are a significant number of terms missed by the @xmath9-model . whether or not these terms are important in an experimental setting is another question . c c c c @xmath9 & terms in @xmath9-model & terms in general model & missed terms + [ 0.5ex ] 0 & 1 & 1 & 0 + 1 & 3 & 4 & 1 + 2 & 6 & 10 & 4 + 3 & 13 & 20 & 7 + [ 1ex ] [ count_tab ] this work was supported by the natural sciences & engineering research council of canada . m.b.y . was also partially supported by a renaissance technologies fellowship , and would like to thank andrew louca and paul mcgrath for discussions . i. f - s was supported by the alexander von humboldt foundation and would like to thank tobias brandes , carsten henkel and martin wilkens for their hospitality and p. barberis - blostein for useful comments and discussions . g. j. milburn , j. corney , e. m. wright , and d. f. walls , phys . rev . a * 55 * , 4318 ( 1997 ) ; j. i. cirac , m. lewenstein , k. molmer , and p. zoller , phys . rev . a * 57 * , 1208 ( 1998 ) ; a. j. leggett , rev . phys . * 73 * , 307356 ( 2001 ) . m. albiez et al . lett . * 95 * , 200401 ( 2005 ) . l. santos and t. pfau . . lett . * 96 * , 190404 ( 2006 ) ; t. ohmi and k. machida , j. phys.soc . jpn . * 67 * , 1822 ( 1998 ) ; c. k. law , h. pu , and n. p. bigelow , phys . rev . lett . * 81 * , 5257 ( 1998);m . koashi and m. ueda , phys . lett . * 84 * , 1066 ( 2000 ) ; o. e. mstercaploglu et al , phys . a * 68 * , 063616 ( 2003 ) . c. sackett , j. gerton , m. welling and r. hulet . _ atomic physics 16 : proceedings of the sixteenth international conference on atomic physics_. american institute of physics conference proceedings # 477 , new york , 1999 . eds . baylis , w. and drake , g.

the physical phenomenon of entanglement naturally appears in composite quantum systems via interactions among constituents . simple collision models already teach us that different interaction types lead to various types of multipartite entanglement @xcite . systems with local hamiltonians exhibit correlations between the degree of entanglement and eigenenergies @xcite , phase transitions @xcite , and the number of interacting bodies @xcite . multipartite entanglement finds uses in quantum networking applications such as secret sharing @xcite , secret voting @xcite , open - destination teleportation @xcite , etc . for the latter purposes , entanglement can be created within the system not only by interaction among constituent bodies but also by a properly engineered interaction with the environment @xcite . suppose the prepared multipartite entangled state is intended for use in an entanglement - enabled quantum protocol involving remote clients . while transferring the quantum information to recipients , the state will be modified by inevitable noise processes . it can happen that the type of multipartite entanglement received by the clients differs significantly from the original one , and the realization of the desired protocol becomes impossible . similarly , uncontrollable noise processes in quantum memory devices can result in destroying particular correlations within the stored multipartite system and make the released state ineffective @xcite . degradation of entanglement also imposes limitations on the benefit of advanced quantum metrology relying on genuinely multipartite entangled states @xcite . these examples demonstrate the necessity of tracking the multipartite entanglement dynamics and finding noise levels corresponding to the change of entanglement type . previous efforts in this direction relied on specific entanglement measures . negativity @xcite a measure detecting negativity of the density matrix under partial transpose ( npt ) @xcite was originally used by simon _ et al_. @xcite and dr _ et al_. @xcite to analyze ghz , w , and cluster states under local depolarizing noise . then bandyopadhyay _ et al_. @xcite and hein _ et al_. @xcite utilized it to study the behavior of ghz states and graph states , respectively , under general local homogeneous noise . generalized ghz - type states under a local amplitude - damping channel were considered with the help of negativity by man _ et al_. @xcite . et al_. exploited negativity to study effects of local depolarizing , dephasing , and generalized amplitude damping channels on ghz states @xcite and graph states @xcite . those results were obtained for an arbitrary number of qubits ( except for some graph states @xcite and randomly sampled states @xcite ) due to the ultimate simplicity of negativity computation . depolarization and dephasing of qudit ghz states were considered via negativity in @xcite . similarly , concatenated ghz states ( where blocks of a small number of qubits are ghz states themselves ) were considered in @xcite . however , the negativity does not provide comprehensive information about the entanglement structure because it can be sensitive to the entanglement with respect to a particular bipartition only ( remember , e.g. , bound - entangled ppt states @xcite and biseparable but non - triseparable states @xcite ) . the absence of full separability can also be detected by some other measures . for instance , carvalho _ et al_. used the lower bound for a specific generalization of the concurrence and applied it to the dynamics of several - qubit ghz and w states under amplitude - damping and dephasing local channels @xcite . et al_. used the geometric measure of entanglement @xcite to study global dephasing process of four - qubit ghz , cluster , w , and dicke states @xcite . et al_. used the entropic measure for average @xmath0-partite entanglement over quantum trajectories @xcite . et al_. developed the evolution of an averaged sl - invariant entanglement measure for local decoherence @xcite . a similar approach with a lower bound of the concurrence was exploited in @xcite . a non - zero value of these quantities indicates the presence of some entanglement within the quantum system , but gives little information about its particular form and , therefore , the benefit of this entanglement for some applications remains questionable . moreover , vanishing values of the above measures can not guarantee the full separability of the state , and thus , the problem of fundamental noise limits eliminating any form of entanglement ( resulting in fully separable states ) is still open . genuine multipartite entanglement is the exact opposite of full separability : this form of entanglement is intrinsically multiparticle and can not be attributed to the entanglement distributed among smaller subsystems . the detection of genuine entanglement for specific quantum states has been a subject of intensive recent research ( see , e.g. , @xcite and references therein ) . dissipative evolution of genuine multipartite entanglement has been analyzed with the help of some measures . the mean value of a projector - like witness @xcite was used by bodoky _ et al_. to study several qubits within a heuristic model of decoherence based on local relaxation and dephasing times @xcite . et al_. used fidelity- and collective spin - based entanglement witnesses to analyze the dynamics of genuine multipartite entanglement of dicke states under local amplitude - damping , phase - damping , and depolarizing channels @xcite . tripartite negativity and generalized concurrence were also applied to the dissipative dynamics of ghz and w three - qubit entangled states @xcite . let us recall that the above measures are not precise , i.e. , their zero values do not imply in general that the genuine entanglement is lost . on the other hand , precise measures ( based on the convex roof definitions ) are quite hard to compute . this is the main reason why the research in entanglement dynamics is usually restricted to particular initial states ( ghz , w , x , dicke , etc . ) and the use of relatively simple measures . despite existing results for noises preserving genuine entanglement and entanglement on the whole ( absence of full separability ) , the evolution of entanglement structure still remains unexplored . the aim of this paper is to track the transformations of entanglement structure during dissipative processes . by `` structure '' we understand the number of separate components and the number of particles within each of them ( with allowance for convex mixtures ) @xcite . this structure resembles a russian nested doll , and dissipative evolution maps states from the outer to the inner dolls . the evolution of entanglement structure can be seen as a dissociation of the entanglement compound due to interaction with the `` solvent '' ( particles of the environment ) . note that the `` entanglement compound '' refers to a genuinely entangled multipartite component and differs from the concept of an `` entanglement molecule '' whose bonds depict entanglement of reduced two - particle states @xcite . the idea of tracking the entanglement structure was realized for three - qubit ghz states under global depolarization in @xcite and for the restricted hilbert space of single - excitation states in @xcite . we do not restrict ourselves to particular input states and develop a theory of transformations that map _ any _ initial state into a chosen doll . note that mainstream research is focused on showing that a particular state is outside a given doll ( mostly that of biseparable states ) @xcite , whereas ours ensures the opposite and matches the recent approach of ref . our methodology relies on a neat decomposition of the physical map into simpler ( but not necessary physical ) processes involving entanglement - breaking operations @xcite . the criteria obtained are formulated for general quantum channels . to illustrate our approach , we discuss examples of local and global depolarizing noises modelling individual and common baths , respectively . local depolarizing noises are relevant in quantum communication tasks ( exploiting , e.g. , optical fibers ) as well as in purely physical systems such as nuclear spins in molecules @xcite . global depolarizing noise is an appropriate model in experiments where full - rank quantum states are detected @xcite and is argued to be the worst - case scenario of system - environment interactions @xcite . we find the noise levels of corresponding entanglement structure dissociations and reveal differences in the typical dissociation behavior between local and global noises . the paper is organized as follows . in sec . [ section - formalism ] , we precisely describe the multipartite entanglement formalism used , with attention being paid to higher order partitions ( tripartitions , tetrapartitions , etc . ) which are often omitted from consideration . in sec . [ section - dynamics ] , we recall the necessary information about general and local quantum channels . in sec . [ section - ea ] , the problem under investigation ( dynamics of entanglement structure ) is precisely formulated . in sec . [ section - methodology ] , we accomplish the development of methodology and derive the criteria of entanglement dissociation and annihilation . in sec . [ section - application ] , we provide a recipe for applying the obtained criteria to the above - mentioned noises . in sec . [ section - discussion ] , the physical meaning of the results is discussed . in sec . [ section - summary ] , we concisely summarize the ideas , methods , and achieved results . to express the idea of entanglement structure quantitatively , one can make use of the following formalism . whenever we speak about entanglement , we imply a particular partition of the composite system . in general , an @xmath1-body system @xmath2 can be partitioned into @xmath3 subsystems , where @xmath3 ranges from @xmath4 to @xmath1 . if the system is not partitioned at all , we will reckon @xmath5 . one can divide the @xmath1-body system @xmath2 into @xmath3 subsystems ( also referred to as parties ) in @xmath6 different ways , where @xmath7 is the stirling number of the second kind . denote by @xmath8 a set of possible partitions into @xmath3 parties . partitions are ordered in such a way that the parts with fewer bodies go first . then , for a three - body system @xmath9 we have @xmath10 , @xmath11 , and @xmath12 . in the case of a four - body system @xmath13 , the sets of possible partitions are @xmath14 , @xmath15 , @xmath16 , @xmath17 , @xmath18 , @xmath19 , @xmath20 , @xmath21 , @xmath22 , @xmath23 , @xmath24 , @xmath25 , @xmath26 , @xmath27 , and @xmath28 . denote by @xmath29 the @xmath30-th partition of the set @xmath8 , e.g. , @xmath31 . in order to address the @xmath32th subsystem of the partition @xmath29 , we will use the notation @xmath33_m$ ] , e.g. , @xmath34_2 = d$ ] . quantum states of the system @xmath2 are described by density operators @xmath35 ( positive and with unit trace ) acting on the hilbert space @xmath36 and altogether forming the convex set @xmath37 . a state @xmath38 is called separable with respect to a particular partition @xmath29 if the resolution @xmath39_1 } \otimes \cdots \otimes \varrho_i^{[{{\cal p}}_j^k]_k}$ ] holds true for some probability distribution @xmath40 and density operators @xmath41_m}$ ] , @xmath42 . we will denote such a separable state as @xmath43 for brevity . if @xmath44 for any @xmath43 , then @xmath38 is said to be entangled with respect to the partition @xmath29 . the above consideration of partitions is important because the physics of multipartite entanglement can be quite counterintuitive . for instance , the three - qubit state of refs . @xcite is separable with respect to any bipartition @xmath45 but is entangled with respect to tripartition @xmath46 . another example is a four - qubit smolin state @xcite which is separable with respect to bipartitions @xmath47 , @xmath48 , @xmath49 and is entangled with respect to bipartitions @xmath50 , @xmath51 , @xmath52 , @xmath53 , any tripartition @xmath54 , and quartering @xmath55 . now we can define the concept of @xmath3-separability of a quantum state , which indicates that the state can accommodate components each of which has @xmath3 separate parties . namely , the state @xmath38 is called @xmath3-separable and denoted @xmath56 if it adopts the resolution @xmath57 for some probability distribution @xmath58 and separable density operators @xmath43 . note that @xmath56 can still be entangled with respect to partitions @xmath29 if @xmath59 . clearly , if the state is @xmath3-separable , then it is also @xmath60-separable , which implies the inclusion relation @xmath61 for convex sets of @xmath3-separable states . a natural measure of separability appears : @xmath62 : = \max\limits_{\varrho = \varrho_{k\text{-sep } } } k.\ ] ] if @xmath63 = 1 $ ] , then the state @xmath38 is called genuinely entangled ( ge ) . if @xmath63 = n$ ] , then the state @xmath38 is fully separable ( fs ) . one can quantify multipartite entanglement in an alternative way by counting the number of bodies that are actually entangled @xcite . this number would indicate the resources needed to create the state . for instance , the state @xmath64 of a 5-body system @xmath65 is 2-separable but comprises a party @xmath66 which can be genuinely entangled ( @xmath67=1 $ ] ) , i.e. , requires 3 bodies to be entangled . to embody this idea in a precise manner , we introduce the following definition of resource - intensiveness ( compatible with the concepts of entanglement depth @xcite and producibility @xcite ) : @xmath68 : = \!\!\ ! \min\limits_{\varrho = \sum\limits_{k=1}^n \sum\limits_{j=1}^{{\genfrac{\lbrace}{\rbrace}{0pt}{}{n}{k } } } p_j^k \sigma_j^k } \max\limits_{m=1,\ldots , k } \ ! \left\ { \#~\text{bodies within}~ [ { { \cal p}}_j^k]_m \right\ } \ ! .\ ] ] denote by @xmath69 \le r\}$ ] the convex set of @xmath70-entangled states . obviously , @xmath71 . importantly , @xmath72 , @xmath73 , and @xmath74 . depending on the quantum state , the range of @xmath75 can be @xmath76 $ ] for a fixed @xmath77 , and the range of @xmath77 can be @xmath78 $ ] for a fixed @xmath75 . denotes the smallest integer greater than or equal to @xmath79 , and @xmath80 denotes the greatest integer less than or equal to @xmath79 . ] the relations between two families of sets @xmath81 and @xmath82 for a four - body system are shown in fig . [ figure1 ] . schematic of sets @xmath83 ( dashed ) and @xmath84 ( solid ) for a four - body system.,width=321 ] we describe the physical evolution of open quantum systems by the input output formalism of quantum channels : @xmath85 $ ] , where @xmath86 is a completely positive trace - preserving ( cpt ) linear map on trace - class operators @xmath87 . the physical meaning of the evolution via a cpt map @xmath88 can be readily seen from the stinespring dilation @xcite : @xmath89 \equiv { \rm tr}_{\rm env } [ u ( \varrho_{\rm in } \otimes \xi_{\rm env } ) u^{\dag}]$ ] for some state of the environment @xmath90 and some unitary operator @xmath91 . complete positivity ( cp ) of the map @xmath88 acting on a system @xmath92 guarantees that @xmath93 \ge 0 $ ] for all composite states @xmath94 of the system @xmath92 and an arbitrary ancilla , with @xmath95 being the identity transformation . equivalently , the map @xmath88 is cp if it adopts the diagonal sum representation @xmath96 = \sum_k a_k x a_k^{\dag}$ ] . if the kraus operators @xmath97 satisfy @xmath98 ( identity operator ) , then @xmath88 is cpt . in order to define a linear map @xmath88 acting on a system @xmath92 , we will use the choi jamiokowski isomorphism @xcite : @xmath99 , \\ & & \label{map - through - choi } \phi [ x ] = d^s \ , { \rm tr}_{s ' } \ , [ \,\omega_{\phi}^{ss ' } ( i_{\rm out}^s \otimes x^{\rm t } ) \ , ] , \end{aligned}\ ] ] where @xmath100 , @xmath101 is a maximally entangled state shared by system @xmath92 and its clone @xmath102 , @xmath103 is the transposition in some orthonormal basis , and @xmath104 denotes the partial trace over @xmath102 . the linear map @xmath105 is cp if and only if @xmath106 . since our main interest is focused on many - body systems , let us consider a composite system @xmath107 acted upon by some channel @xmath108 . to begin with , @xmath109 , which explicitly shows the separability of the maximally entangled state with respect to the partition @xmath110 . while constructing the choi operator ( [ choi - matrix ] ) , the map @xmath111 can in general entangle these subsystems . suppose a local channel @xmath112 which serves as an adequate model in situations when each particle is sent to a corresponding receiver through an individual quantum cable ( fig . [ figure2]a ) . in this case , clearly , @xmath114 is cp if and only if each of the maps @xmath115 , @xmath116 , @xmath117 , @xmath118 is cp . local channels : ( a ) general ; ( b ) homogeneous.,width=321 ] in quantum communication , the typical scenario is to use a single quantum cable to transmit time - separated parties of a multipartite state from the encoder to the decoder ( fig . [ figure2]b ) . neglecting the memory effects , the evolution of a multipartite system is governed by the homogeneous local channel @xmath119 , which also appears in the definition of channel capacities ( see , e.g. , the review @xcite ) . consider a composite @xmath1-body system @xmath107 that undergoes the physical evolution @xmath120 $ ] determined by some cpt map @xmath88 ( we also assume @xmath121 ) . if @xmath122 is separable with respect to the partition @xmath29 ( i.e. @xmath123 ) , then we say that the channel @xmath88 dissociates the entanglement compound of a given @xmath124 into smaller compounds of @xmath33_1,\ldots,[{{\cal p}}_j^k]_k$ ] and denote by @xmath125 the set of such channels . if the channel @xmath88 dissociates the entanglement of all input states @xmath126 in this way , then we will refer to @xmath88 as dissociating entanglement with respect to the partition @xmath29 and denote @xmath127 . using entanglement measures ( [ k - separability ] ) and ( [ r - entanglement ] ) , we can quantitatively describe the processes of entanglement structure dynamics . namely , denote by @xmath128 a set of channels @xmath88 such that @xmath129 \right ] \ge k$ ] . by construction , @xmath130 is a convex hull of the sets @xmath125 . similarly , @xmath131 is a set of channels @xmath88 such that @xmath132 \right ] \le r$ ] . regarding state - independent properties , we straightforwardly introduce the sets of channels @xmath133 and @xmath134 . the developed formalism of sec . [ section - formalism ] immediately results in the following inclusion diagram for the above sets : @xmath135 we have used a special notation for two distinctive classes of channels : * entanglement annihilating channels ( ea ) transforming any input state into a fully separable one @xcite ; * channels that dissociate genuine entanglement ( dge ) , thus , transforming genuinely entangled states into non - genuinely entangled ones . the problem under investigation is twofold : ( i ) to characterize the sets of channels @xmath128 and @xmath136 as well as state - independent sets from the above diagram , ( ii ) to track how exactly the multiparticle entanglement structure dissociates under particular noises . our special attention is paid to ea and dge channels . before proceeding to the derivation of criteria , we need to clarify the relation between the problem involved and the well known approaches developed so far . consider a ( not necessarily composite ) system @xmath92 acted upon by a channel @xmath137 . if the choi operator @xmath138 is separable with respect to the partition @xmath139 , then @xmath88 is a so - called entanglement - breaking ( eb ) map @xcite , whose peculiarity is that @xmath140 $ ] is separable with respect to the partition @xmath141 for all density operators @xmath142 . in fact , separability of @xmath138 implies that @xmath88 has the holevo form @xmath96 = \sum_k { { \rm tr}[f_k x ] } \omega_k$ ] , where @xmath143 is a positive operator - valued measure and @xmath144 , i.e. @xmath88 is a measure - and - prepare procedure . the latter representation , in its turn , implies @xcite that there exists a diagonal sum representation with rank-1 kraus operators @xmath145 with @xmath146 and @xmath147 . as concerns a composite system @xmath107 , the eb channel @xmath108 disentangles @xmath92 from any other system but can in principle result in any entanglement dynamics within @xmath92 ( among @xmath148 , @xmath149 , @xmath150 , @xmath118 ) . for instance , the output state can be genuinely entangled or fully separable depending on the entanglement of vectors @xmath151 constituting kraus operators . however , the local channel @xmath152 is entanglement breaking if and only if each of the channels @xmath115 , @xmath116 , @xmath117 , @xmath118 is entanglement breaking . this can be readily seen from the requirement of separability of the choi operator @xmath153 with respect to the partition @xmath154 . thus , the local entanglement breaking channel is automatically entanglement annihilating but the converse is not true . these and other differences between entanglement breaking and entanglement annihilating channels are discussed in @xcite . in this section , we provide criteria to detect the different kinds of entanglement dissociation discussed above . we start with a description of our methodology which is based on an extensive use of various convex sets of operators and maps . in addition to quantum states described by positive semidefinite unit trace operators @xmath155 , an important role will be played by block - positive operators @xcite . the operator @xmath156 is called block - positive with respect to the partition @xmath29 if it satisfies @xmath157_1 } \otimes \cdots \otimes \psi_k^{[{{\cal p}}_j^k]_k}| } \xi_j^k { | \psi_1^{[{{\cal p}}_j^k]_1 } \otimes \cdots \otimes \psi_k^{[{{\cal p}}_j^k]_k } \rangle } \ge 0\ ] ] for all vectors @xmath158 . block - positive operators are closely related to entanglement witnesses @xcite and can be used to determine separability : a state @xmath155 is separable with respect to the partition @xmath29 if and only if @xmath159 } \ge 0 $ ] for all block - positive operators @xmath156 . we must emphasize that the concepts of entanglement dissociation and annihilation from sec . [ section - ea ] do not imply any ancillary system besides the multipartite system @xmath107 itself . this allows to relax cpt condition of the physical transformation @xmath88 and construct an extended set @xmath160 $ ] of ( mathematical ) linear maps @xmath161 having the same entanglement behavior as @xmath88 on the corresponding domain of input states . for example , the extended set @xmath162 $ ] consists of linear maps @xmath161 satisfying the only restriction that @xmath163 $ ] is equal to some @xmath43 . similarly , @xmath164 $ ] and @xmath165 $ ] denote the extensions of sets @xmath128 and @xmath166 , respectively . as we show later , the extensions turn out to be useful because they adopt a good characterization . the original set of maps can be found by intersecting with cpt maps , e.g. , @xmath167 $ ] . [ proposition-1 ] suppose a linear map @xmath161 acting on a system @xmath2 . then @xmath168 $ ] if and only if @xmath169 \ge 0 $ ] for all @xmath156 . separability of @xmath163 $ ] with respect to the partition @xmath29 is equivalent to the inequality @xmath170 \xi_j^k \right ] \ge 0 $ ] for all @xmath156 . substituting ( [ map - through - choi ] ) for @xmath163 $ ] concludes the proof . as a result , the cone @xmath171 $ ] is dual to the cone of maps @xmath172 = \xi_j^k { { \rm tr}[\varrho_{\rm in } x]}$ ] . as concerns the state - independent property @xmath173 , the map @xmath161 belongs to the set @xmath173 if its choi matrix satisfies @xmath174 \ge 0 $ ] for all @xmath156 and @xmath175 . the criterion provided by proposition [ proposition-1 ] is not quite operational . to overcome this obstacle we derive sufficient criteria of entanglement dissociation . consider a particular partition @xmath29 . suppose a linear map @xmath176 which transforms the density operator @xmath124 into some hermitian ( but not necessarily positive ) operator @xmath177 $ ] such that @xmath178_1 } \otimes \cdots \otimes \psi_{m-1}^{[{{\cal p}}_j^k]_{m-1 } } \otimes i \otimes \psi_{m+1}^{[{{\cal p}}_j^k]_{m+1 } } \otimes \cdots \otimes \psi_k^{[{{\cal p}}_j^k]_k } \big| \nonumber\\ & & \times \,\ , \xi[\varrho_{\rm in } ] \,\ , \big| \psi_1^{[{{\cal p}}_j^k]_1 } \otimes \cdots \otimes \psi_{m-1}^{[{{\cal p}}_j^k]_{m-1 } } \otimes i \otimes \psi_{m+1}^{[{{\cal p}}_j^k]_{m+1 } } \nonumber\\ & & \quad\quad\quad\quad \otimes \cdots \otimes \psi_k^{[{{\cal p}}_j^k]_k } \big\rangle \ge 0\end{aligned}\ ] ] is fulfilled for some vectors @xmath179 , i.e. @xmath177 $ ] after projection onto these vectors becomes a positive operator from the cone @xmath180_m})$ ] . if this is the case , then for rank-1 kraus operators @xmath181 with arbitrary @xmath182 , the operator @xmath183 ( a_1^{\dag } \otimes \cdots \otimes a_{m-1}^{\dag } \otimes i \otimes a_{m+1}^{\dag } \otimes \cdots \otimes a_k^{\dag})$ ] belongs to a cone of separable states @xmath43 . thus , we obtain the following sufficient criterion of entanglement dissociation . [ proposition-2 ] concatenation of a linear hermitian map @xmath184 and a @xmath60-partite eb operation @xmath185_1 } \otimes \cdots \otimes { \rm id}^{[{{\cal p}}_j^k]_m } \otimes \cdots \otimes { { \cal o}}_{\rm eb}^{[{{\cal p}}_j^k]_k } \big)$ ] belongs to @xmath162 $ ] if @xmath177 $ ] becomes positive after projection on right - singular vectors of the rank-1 kraus operators of the eb operation . the idea of proposition [ proposition-2 ] is shown for a 6-body system in fig . [ figure3 ] . the benefit of the constructed concatenation is that the map @xmath184 does not have to be positive ( in contrast to ref . @xcite ) , which makes the set @xmath162 $ ] even larger . when all possible states @xmath124 are considered , the satisfaction of requirement ( [ requirement ] ) becomes equivalent to the positivity of the map @xmath185_1 } \otimes \cdots \otimes { \rm id}^{[{{\cal p}}_j^k]_m } \otimes \cdots \otimes { { \cal o}}_{\rm eb}^{[{{\cal p}}_j^k]_k } \big ) \circ \xi$ ] . this map is automatically positive if @xmath184 transforms density operators into block - positive operators @xmath156 , which in turn is equivalent to the fact that its choi operator is block - positive of the form @xmath186 . [ corollary-1 ] if @xmath187 is block - positive with respect to the partition @xmath188 , then @xmath189_1 } \otimes \cdots \otimes { \rm id}^{[{{\cal p}}_j^k]_m } \otimes \cdots \otimes { { \cal o}}_{\rm eb}^{[{{\cal p}}_j^k]_k } \big ) \circ \xi \in { { \cal d}}_j^k$ ] for arbitrary eb operations . [ cols="^,^,^,^,^,^,^,^,^,^ " , ] the elementary block of ea channel is obtained by applying entanglement breaking operations on @xmath190 particles ( see fig . [ figure3]a ) . the exact form of chosen eb operations reads @xmath191 = \frac{1}{2 } { | \psi_i \rangle } { \langle \psi_i| } x { | \psi_i \rangle } { \langle \psi_i|}$ ] , where @xmath192 form a symmetric informationally complete positive operator - valued measure ( sic - povm ) for qubits ( see the explicit analytical form of the vectors @xmath193 in @xcite ) . this choice is justified by the fact that @xmath194 . ( the same result would be obtained by using projectors on mutually unbiased bases @xcite instead of sic - povm elements , however , this approach leads to worse results for some input states @xmath124 . ) the suggested decomposition reads @xmath195_1 } \otimes \cdots \otimes { \rm id}^{[{{\cal p}}^n]_m } \otimes \cdots \otimes \phi_{q=1/3}^{[{{\cal p}}^n]_n } \big ) \circ \xi_{a}(m),\ ] ] where @xmath32 is the index of a particle not subjected to eb operations . we have taken into account that each @xmath196 is composed of eb operations @xmath197 and therefore it is convenient to parameterize the map @xmath198 in such a way that the vectors @xmath199 are not included in the parametrization directly . however , the linear map @xmath200 should satisfy the requirement ( [ requirement ] ) for all choices of vectors @xmath201 from the set @xmath193 . to parameterize the map @xmath202 we resort to a so - called diagonal map of the form @xmath203 & = & \frac{1}{2^n } \sum_{i_1,\ldots , i_n=0,\ldots,3 } x_{i_1 \cdots i_n } { { \rm tr}[(\varsigma_{i_1}\otimes\cdots\otimes\varsigma_{i_n})x ] } \nonumber\\ & & \qquad \qquad \qquad \qquad \qquad \times \varsigma_{i_1}\otimes\cdots\otimes\varsigma_{i_n},\end{aligned}\ ] ] where @xmath204 and @xmath205 , @xmath206 , @xmath207 are conventional pauli matrices . let @xmath208 $ ] denote the number of zeros in the sequence @xmath209 . consider diagonal maps @xmath200 such that the coefficients @xmath210 depend on @xmath211 $ ] and @xmath212 $ ] only , i.e. @xmath213,\#_0[i_1 \cdots i_{m-1 } i_{m+1 } \cdots i_n])$ ] , with restrictions on the parameters @xmath214 being imposed by ( [ requirement ] ) . then , the relation ( [ ea - resolution ] ) becomes valid if @xmath215 where @xmath216 is the conventional kronecker delta . for a fixed input state @xmath124 , we find the restrictions on the parameters @xmath214 given by ( [ requirement ] ) and then solve the system of equations ( [ ea - system ] ) numerically . if the system has a solution for some @xmath217 , then it also has a solution for @xmath218 . solutions ( @xmath219 ) are presented for some interesting states @xmath124 ; @xmath220 ; @xmath221 , where @xmath222 is a projector on unextendible product bases for 3 qubits @xcite ; @xmath223 . ] of @xmath224 qubits in tables [ table - local ] and [ table - global ] for local and global noises , respectively . we also consider the case of all possible input states as follows : since @xmath225 linearly depends on the parameters @xmath214 , we check the corresponding block - positivity of @xmath226 ( see corollary [ corollary-1 ] ) for some number of parameters @xmath214 and construct a convex hull of satisfactory parameters ; then we solve the system of equations ( [ ea - system ] ) for @xmath214 from the convex hull ; the maximum value @xmath227 for which the system has a solution is presented in tables [ table - local ] and [ table - global ] in the rows `` all '' . the output state will be @xmath229-separable and @xmath4-entangled if the channel can be decomposed into elementary transformations @xmath230 $ ] from proposition [ proposition-2 ] , each containing @xmath231 eb operations @xmath232 on two qubits ( see fig . [ figure3]b ) . as in the previous subsection , we choose eb operations of the form @xmath233=\frac{1}{4}{| \psi_i \rangle } { \langle \psi_i| } x { | \psi_i \rangle } { \langle \psi_i|}$ ] , where @xmath234 form a sic - povm in @xmath235 ( see the explicit analytical form of the vectors @xmath236 in @xcite ) . then @xmath237 is a depolarizing map acting on two qubits ( @xmath148 and @xmath149 ) simultaneously . the decomposition of channel @xmath88 reads @xmath238_1 } \otimes \cdots \nonumber\\ & & \otimes { \rm id}^{[{{\cal p}}_j^{n/2}]_m } \otimes \cdots \otimes \phi_{q=1/5}^{[{{\cal p}}_j^{n/2}]_{n/2 } } \big ) \circ \xi_{b}(j , m),\end{aligned}\ ] ] where @xmath239 is a set of @xmath240 partitions @xmath241 such that @xmath242_1 = \ldots = \#[{{\cal p}}_j^{n/2}]_{n/2 } = 2 $ ] ( two qubits in each party ) , and the map @xmath243 must meet the condition ( [ requirement ] ) for all choices of vectors @xmath244 from the set @xmath236 . using diagonal maps @xmath243 of the form ( [ diagonal ] ) with the parametrization @xmath245,\#_0[i_1 \cdots i_{l-1 } i_{l+1 } \cdots i_{l'-1 } i_{l'+1 } \cdots i_n])$ ] , @xmath246_m$ ] , we obtain the following system of equations : @xmath247 the maximal values of @xmath227 , for which the system has a solution compatible with ( [ requirement ] ) , are presented for various input states in tables [ table - local ] and [ table - global ] . the output state will be @xmath249-separable and @xmath229-entangled if the channel can be decomposed into elementary transformations @xmath250 $ ] , @xmath251 ( for such @xmath30s , the @xmath1-body system is divided into @xmath229 single - body parts plus one part comprising @xmath229 bodies , see fig . [ figure3]c ) . to find the decomposition for proposition [ proposition-3 ] , we use the single - qubit eb operations @xmath252 , @xmath253 , @xmath254 as in sec . [ section - applicability - ea ] . this yields the following decomposition : @xmath255_1 } \otimes \cdots \otimes \phi_{q=1/3}^{[{{\cal p}}_j^{n/2 + 1}]_{n/2 } } \nonumber\\ & & \qquad\qquad\qquad\quad \otimes { \rm id}^{[{{\cal p}}_j^{n/2 + 1}]_{n/2 + 1 } } \big ) \circ \xi_{c}(j),\end{aligned}\ ] ] where the map @xmath256 must satisfy the requirement ( [ requirement ] ) for all choices of vectors @xmath257 from the set @xmath193 ( vectors corresponding to sic - povm for a qubit ) . using diagonal maps @xmath256 of the form ( [ diagonal ] ) with the parametrization @xmath258,\#_0[i_{l_1 } \cdots i_{l_{n/2}}])$ ] , @xmath259_{n/2 + 1}$ ] , we obtain the following system of equations : @xmath260 the maximal values of @xmath227 , for which the system has a solution compatible with ( [ requirement ] ) , are presented for various input states in tables [ table - local ] and [ table - global ] . the output state will be @xmath4-separable and @xmath229-entangled if the channel can be decomposed into elementary transformations @xmath262 $ ] , @xmath263 ( such choice of @xmath30s corresponds to bipartitions of an @xmath1-body system into equal @xmath229-body parts , see fig . [ figure3]d ) . we use the following eb operations @xmath232 on @xmath229 qubits : @xmath233=\frac{1}{2^{n/2}}{| \psi_i \rangle } { \langle \psi_i| } x { | \psi_i \rangle } { \langle \psi_i|}$ ] , where @xmath264 is a set of normalized vectors such that @xmath265 is a set of sic projectors ( see the explicit form of vectors @xmath264 up to @xmath266 in @xcite ) . [ let us recall that a particular form of vectors @xmath199 is important only for a particular input state @xmath124 . if the input state is arbitrary , i.e. the domain is @xmath267 , then one should not care about the specific form of eb operations . ] the important fact is that @xmath268 is a depolarizing map acting on @xmath229 qubits . the possible decomposition reads @xmath269_m } \otimes { \rm id}^{[{{\cal p}}_j^{2}]_{\{1,2\ } \setminus m } } \big ) \circ \xi_{d}(j , m),\end{aligned}\ ] ] where @xmath270 must satisfy condition ( [ requirement ] ) for all vectors @xmath264 and for the corresponding domain of density operators @xmath124 . for computational reasons let us note that checking the validity of ( [ requirement ] ) is less time - consuming when we justify the positivity of the hermitian operator without revealing its eigenvalues . namely , the eigenvalues of a @xmath271 hermitian matrix @xmath272 are non - negative if and only if @xmath273 for @xmath274 , where @xmath275 is given by the recurrence relation @xmath276}$ ] with initial condition @xmath277 @xcite . we use this technique for @xmath278 . using diagonal maps @xmath270 of the form ( [ diagonal ] ) with the parametrization @xmath279,\#_0[\{i_1 \cdots i_n\ } \setminus \{i_{l_1 } \cdots i_{l_{n/2}}\}])$ ] , @xmath280_m$ ] , we obtain the following system of equations : @xmath281 the maximal values of @xmath227 , for which the system has a solution , are presented for various @xmath124 in tables [ table - local ] and [ table - global ] . scaling of the entanglement degradation properties of an @xmath1-qubit local depolarizing channel @xmath282 with increasing @xmath1 : ( a ) entanglement annihilation , ( b ) dissociation of genuine entanglement.,width=321 ] the elementary blocks of these channels can be obtained by applying an eb operation on a single qubit ( fig . [ figure3]e ) . we use the same eb operations as in subsection [ section - applicability - ea ] . this yields the decomposition @xmath283_1 } \otimes { \rm id}^{[{{\cal p}}_{j = m}^2]_2 } \big ) \circ \xi_{e}(m),\ ] ] where the map @xmath284 must satisfy ( [ requirement ] ) for all vectors @xmath193 corresponding to sic - povm for a qubit . using diagonal maps @xmath284 of the form ( [ diagonal ] ) with the parametrization @xmath285,\#_0[i_1 i_{m-1 } i_{m+1 } \cdots i_n])$ ] , we find that ( [ age - resolution ] ) becomes a valid equality if @xmath286 the maximal values @xmath227 , for which the system has a solution and ( [ requirement ] ) is fulfilled , are presented for various input states in tables [ table - local ] and [ table - global ] . to begin with , the npt criterion gives a little information about the multipartite entanglement structure . indeed , one can observe in tables [ table - local ] and [ table - global ] many situations when @xmath287 $ ] is either negative under partial transpose but not genuinely entangled , or positive under partial transpose but not fully separable . the gap between states @xmath287 $ ] that are surely not genuinely entangled and those that are definitely genuinely entangled is quite narrow for particular input states . this can be treated as an indication of the efficiency of the rather simple decomposition ( [ age - resolution ] ) involving single - qubit entanglement breaking operations . the remarkable fact is that our method enables us to consider _ all _ input states and find channels that transform any of them to a particular entanglement structure . this is what we mean by a `` typical '' behavior . for example , we can detect channels that annihilate entanglement , for which the output state is always fully separable whatever the input state is . note , that the bounds obtained on @xmath227 for entanglement annihilation are higher than those that can be found via the condition of sufficiently small purity @xmath288)^2]}$ ] @xcite . scaling of ea for the local depolarizing channel is shown in fig . [ figure4]a . when @xmath289 , the channel @xmath282 can not be ea if @xmath290 @xcite , however , the question if @xmath291 still remains an open problem . on the contrary , from formula ( [ age - system ] ) one can see that the genuine entanglement of any input state can be dissociated by a negligible noise in the limit @xmath289 ( fig . [ figure4]b ) . the dissipative dynamics under consideration can be described by the gradually decreasing parameter @xmath292 , where the dissipation rate @xmath293 takes , in principle , different values for local and global noises . for our purposes it is enough to know that @xmath227 continuously diminishes from @xmath294 to @xmath295 . for such types of dissipative dynamics , the state evolution through the nested sets of sec . [ section - formalism ] is irreducible : once the state comes into a particular `` doll '' of the structure , it can not escape it in the future . using the data from table [ table - local ] , we may conclude that the dissociation of genuine multiparticle entanglement under local depolarizing noise starts by detaching a single random particle ( i.e. the state becomes @xmath296-entangled ) . then the noise detaches particles one by one resulting in @xmath3-separable @xmath297-entangled states ( @xmath3 increases with decreasing @xmath227 ) . indeed , since the noise is local , once a particle is detached from the entanglement compound , there is no way for it to rejoin ( fig . [ figure5 ] ) . finally , the noisy evolution makes the state fully separable . the analysis of table [ table - global ] shows that the entanglement dissociation progresses in a different way under global depolarizing noise : while the beginning stage also implies detaching of a single random particle from the entanglement compound , in further dynamics this particle can fuse with another one and form a two - particle entanglement cluster that is detached from the main compound ( a convex combination of such states ) . the process continues until the point when the original compound is divided into two clusters ( @xmath4-separable @xmath229-entangled state ) , then the detachment of particles and their successive fusion result in the formation of more entanglement clusters of smaller size ( @xmath3-separable @xmath298-entangled state , @xmath3 increases with decreasing @xmath227 ) , and so on until the full separability ( see fig . [ figure5 ] for the case of 6 qubits ) . our study was motivated by the necessity to know the multiparticle entanglement structure and its vulnerability to noises in physical and quantum - informational applications . we did not restrict ourselves to specific input states and considered the set of all possible states as well . we found criteria for maps dissociating entanglement with respect to a particular partition and developed sufficient conditions for their reliable detection . namely , the channel of interest should adopt a decomposition into ( not necessarily completely positive ) linear maps which give rise to the desired form of the output . one can draw a rough analogy between this decomposition and the path integral formulation of quantum mechanics , where the trajectories can be quite non - physical but this does not affect the resulting physical evolution . for local and global depolarizing @xmath1-qubit channels we provided a simple strategy of constructing decompositions that allowed us to find noise levels guaranteeing the particular form of entanglement structure . our decompositions are not optimal and can in principle be improved by applying modifications of semidefinite programming @xcite and other algorithms @xcite for choi operators . nevertheless , our toolbox allowed us to reveal differences in entanglement structure dynamics under local and global noises : the particles split one by one from the entanglement compound in the case of local noise , and tend to form clusters in the case of global noise . we believe that the obtained results may be extended to other noise models and provide additional information about the general rules of the dynamics of multiparticle entanglement structure . _ acknowledgments_. this work was supported by eu integrated project siqs , cost action mp1006 , apvv-0646 - 10 ( coqi ) and vega 2/0127/11 ( tequde ) . s.n.f . acknowledges support from the national scholarship programme of the slovak republic and the 7th fp project iqit . s.n.f . and a.a.m . acknowledge support from the dynasty foundation and the russian foundation for basic research under project no . 12 - 02 - 31524-mol - a . a.a.m . acknowledges partial support from the austrian science fund ( fwf ) through the sfb foqus : f 4012 . m.z . acknowledges support from 7th fp strep project raquel and gacr project p202/12/1142 .

it is now widely believed that solar flares are powered by magnetic energy which had been stored in the corona through slow stressing applied from the photospheric boundary . in an idealized model the energy builds up as the coronal magnetic field responds without resistance ( every field line line tied and unbroken ) . the flare then occurs as coronal reconnection exchanges those field line footpoints to achieve a lower energy state . in this process , the footpoints are changed by the reconnection , but the vertical photospheric field in which the field lines are anchored is not . the potential field from this fixed photospheric field has the minimum magnetic energy possible . the maximum energy available for release is the amount by which the initial field exceeds this potential field energy , called the _ free energy_. quantitative simulation of the above scenario has proven to be extremely challenging owing to the vast range of scales involved . magnetic fields of even modest complexity develop , when stressed , current structures many orders of magnitude thinner than the global length scale @xcite . a numerical solution therefore requires additional magnetic diffusion to prevent the development of unresolvable current structures . the corresponding diffusive time is necessarily much shorter than actual diffusive times and generally shorter than the multi - day times governing the stressing phase . as a consequence , direct simulation , for example by time dependent mhd solution , includes artificial ( diffusive ) energy losses competing with the energy build - up . few such computations have been capable of demonstrating pre flare energies comparble to those released by the ensuing flare@xcite . an alternative means of estimating pre flare energy storage is offered by the minimum current corona model ( mcc : * ? ? ? * ; * ? ? ? this is a quasi - static technique using equilibira , called flux constrained equilibria ( fce ) , minimizing magnetic energy subject to a set topological constraints composing a subset of all line - tying constraints . rather than constraining every pair of footpoints , the mcc groups footpoints into unipolar photopsheric regions and constrains the net coronal flux connecting each pair of regions , called _ domain fluxes_. since it uses only a subset of the actual constraints , its fields provide a lower bound on the actual free energy . as the photospheric regions move relative to one another the potential field above changes as do the fluxes by which it would link region pairs ( potential domain fluxes ) . since the actual field is constrained from changing these fluxes , it becomes increasingly different from the potential field and thus gains free energy . significantly , this energy is available for release by the violation of topological constraints , a process which can occur on very small spatial scales . the mcc has been used to estimate pre flare energy storage in a number of flares @xcite . a partitioning algorithm was developed to automatically group photospheric flux into distinct regions @xcite . provided it is permissible to neglect submergence or emergence of flux , the regions can be taken to move relative to one another but with constant flux . this simplified scenario was deemed adequate to model pre flare evolution in several cases @xcite , including the landmark halloween event @xcite . when footpoints within a given region move internally , such as during sunspot rotation , additional constraints must be introduced involving the arrangement of footpoints within regions @xcite . flux emergence is a well known precursor for flares and cmes @xcite and is thus likely to play a role in energy storage for many flares . it has not yet been accounted for in a complete mcc energy estimate owing cheifly to technical hurdles . doing so would require the constraints to be somehow modified to account for flux emergence . two prototypical cases were treated by ignoring all but one constraint to which emergence makes an obvious modification @xcite . a step toward a more general application was made by @xcite , who accounted for emergence and submergence in a computation of coronal reconnection times in the quiet sun . changes in the fluxes of photopsheric regions were used to generate a list of flux changes due to emerging and submerging domains . the algorithm used had several drawbacks , including a tendancy to assign both emerging and submerging domains to the same photopsheric region . the present work introduces a new algorithm by which emerging flux regions may be automatically accommodated in the mcc constraints leading to a pre flare energy estimate . flux changes in photospheric regions are used to generate a list of domain flux changes due to emergence and submergence , as in @xcite . in this case , however , a single region is linked to only emerging or submerging domains according to the sense of its own change . the possibility of artificial photospheric flux changes due to variations in sequential partitioning can be accomodated at the same time using a closely related algorithm . ( this step is independent of the partitioning algorithm . ) we also present a new algorithm for automatically associating domains with separators , a crucial step in the generation of an energy estimate . the new methodology is illustrated by applying it to a flare which occured on 16 oct . 2010 and was observed by instruments on the solar dynamics observatory ( sdo ) spacecraft . new flux emerges into an existing active ( noaa ar 11112 ) for two days prior to the flare ( goes class m3 ) . we derive photopsheric magnetic fluxes from a series of 123 line of sight magnetograms from the hmi instrument @xcite at cadences of @xmath2 minutes , as discussed in the next section . in section [ sec : photosphere ] , we detail our algorithms for partitioning the set of magnetograms into unipolar regions . the flux variations in these regions are used to define a set of emerging and submerging domains which are automatically generated by an algorithm explained in section [ sec : change ] . the emerging domains largely resemble those we would have expected based on inspection of the time series , but still contain some physically dubious assignments that must be fixed by hand . in the future , we would like to completely automate each of these algorithms . in section [ sec : erg ] , we use the magnetograms to derive a coronal topology for the post flare potential magnetic field and use this to place a bound on the free energy of the pre flare field . the post flare potential field includes a coronal null point whose fan surface encloses one of the newly emerged polarities . this is a novel feature in the mcc and requires the development of one additional method . we apply our energy estimate to noaa ar 11112 in section [ sec : ergapp ] , which reveals that most of the pre flare energy is due to currents passing through the coronal null point . though the methods presented here are of general utility , we apply them , for concreteness , to noaa active region 11112 . we base our calculations on magnetograms taken by the helioseismic and magnetic imager ( hmi ) on board the sdo spacecraft . figure [ fig : oldflux ] shows an example line of sight ( los ) magnetogram from oct . 13th , prior to flux emergence . lighter pixels show positive flux , dark pixels negative flux , and gray pixels zero flux ; the grayscale saturates at @xmath3 g . the active region originally contains only previously emerged flux as it crosses the eastern limb in the southern hemisphere on october 9th , 2010 . new flux begins emerging around 08:00 ut on the 14th , and is associated with a goes m2.9 flare several days later , at 19:07 ut on the oct . 16 . the region shows little activity in the 6 days before the flare . our dataset consists of 123 line of sight magnetograms , each consisting of @xmath4 pixels with 1.0 arcsecond resolution and approximately 0.5 arcsecond / pixel@xcite . they have a cadence of @xmath5 2 hours for the first 21 timesteps , from 2010 - 10 - 13 00:04 ut to 2010 - 10 - 14 18:20 ut , and @xmath5 half hour timesteps thereafter , from 2010 - 10 - 14 18:20 ut to 2010 - 10 - 16 23:37 ut . flux emergence begins around timestep 21 . as shown by the supplementary media , a movie of the los magnetograms , the photospheric changes are well characterized by these two cadences . our data window has flux imbalance @xmath6 over the three days leading up to the october @xmath7 flare . hmi los magnetogram of noaa ar 11112 prior to new flux emergence . the grayscale saturates at @xmath8 g , and the axes are in arcseconds from disk center . the boxed area encloses the region of flux emergence , which begins @xmath9 hours after this magnetogram . ] the old flux region has a polarity inversion line , running from southeast to northwest , which coincides with a filament as seen in data from the global high resolution h@xmath10 network , as well as sdo euv images at 94 , 131 , 171 , 193 , 211 , 304 , and 335 . this filament appears unaffected by the 19:07 ut flare . to the north of the filament , the diffuse , old flux , negative polarity field includes a curious ring of flux surrounding a strong core , boxed in figure [ fig : oldflux ] . the new flux of both polarities emerges completely within this ring , sweeping the old negative flux out of its way as it carves out a space for itself . this results in the ( zoomed in ) field shown in figure [ fig : newflux ] , roughly 10 minutes before the flare . a zoomed view of los magnetogram of the emerging flux region , boxed in figure [ fig : oldflux ] , on oct 16th , 2010 , 18:53ut , @xmath11 10 minutes prior to the m2.9 flare . the grayscale saturates at @xmath12 g , and axes are in arcseconds from disk center.,scaledwidth=70.0% ] as the new flux emerges it snow plows the old flux in front of it , the old flux concentrates and a strong horizontal gradient in the los field develops between old and new flux , as seen near s 400 , w 410 in figure [ fig : newflux ] . the flare is centered on this strong polarity inversion line . flare loops connect the newly emerged negative flux ( n55 , n58 , n87 ) within the ring to the diffuse , positive flux to the west ( p4 , p15 , p19 ) . we therefore believe that any analysis of this flare s energetics must explicitly account for flux emergence . our analysis naturally splits into two parts . the first parts works directly with the photospheric magnetic field measurements . we first partition the magnetic field at all times into unique , unipolar regions . second , we track these regions and enforce a consistent labeling scheme as they change geometry over time . geometric changes include changes in size , shape , orientation , and total flux ; merging and splitting of regions ; and the complete emergence or submergence of regions . the final step quantifies flux emergence and submergence between pairs of photospheric sources using a novel algorithm . in the second part of our analysis , we use our characterization of the photospheric field as a set of unique unipolar sources to describe the topology of the coronal field . each distinct photospheric region is replaced by a point source of the same flux located at the region s flux weighted centroid . we then find the flux within each potential field domain , and the null points and separators of the potential field . changes in the potential field s domain flux relative to the actual domain fluxes , assumed fixed , indicates the storage of magnetic energy in excess of the potential field . in an energy minimized field , the free energy stored by the field is manifest by current ribbons that develop along separator field lines . for brevity , we will use the phrase `` energy stored in the separator '' as a shorthand for `` free energy stored in the magnetic field which manifests itself as a a current ribbon at the location of a separator . '' the amount of current flowing along these ribbons places a bound on the free magnetic energy of the system . we begin work on the photospheric data by partitioning the magnetogram using the gradient based tessellation algorithm described by @xcite . the initial los magnetograms are converted to a vertical field by assuming a radial field at each pixel , which amounts to dividing by @xmath13 , where @xmath14 is the polar angle from disc center . when determining the flux in each region , we must also account for foreshortening within each pixel , dividing by a second factor of @xmath13 . we then convolve all vertical field , @xmath15 , with the green s function for a potential extrapolation up to a height @xmath16 from an unbounded plane : @xmath17 to reduce the effects of noise , we neglect all the convolved field below a threshold of @xmath18 g . using the smoothed field , we assign a unique label to all local maxima and every pixel strictly downhill with respect to @xmath19 from each maxima . internal boundaries in unipolar regions are eliminated multiple regions are merged into a single region when the saddle point @xmath20 . @xmath21 is the greatest vertical field strength of the saddle s surrounding peaks , and @xmath22 is a threshold value . at the end of this process , each data pixel has an associated label . we call the set of all labels at a given time a mask , and the set of masks at all times a mask array . our goal in this section is to make the mask array consistent from timestep to timestep . in particular , we wish to determine how the newly emerged flux interacts with the old flux , and how that process affects the energetics of the active region . our primary concern is therefore to distinguish between new and old flux , which is a marked departure from previous , similar investigations@xcite . figure [ fig : newflux ] shows an example where the boundary of each unique mask region and its label ( p1 , n1 , p2 , ) have been plotted over the line of sight magnetogram data . all pixels and regions containing flux below our thresholds are given a label of 0 . this figure covers the ring feature of old flux concentrated in regions n1 , n6 , n38 , n49 , n55 , and n67and the newly emerged flux in regions p22 , p29 , p48 , n58 and n87 . the supplimentary material shows that these definitions are not necessarily strict . for instance , n58 , n67 , and n87 initially all break off of n49 . both n58 and n87 are quickly dominated by newly emerged flux , while n67 remains primarily distinct . we refer to `` new flux '' or `` old flux '' regions based on the dominant type of flux determined by inspection of the timeseries ( see supporting media ) . for the present analysis , we have set @xmath23 mm , @xmath24 , and @xmath25 . we also established a minimum flux of @xmath26mx @xmath27 gauss px@xmath28 , after accounting for pixel foreshortening ) for a solitary region . our field of view contains unsigned flux of order @xmath29mx above the 50 gauss threshold , so , a region must contain at least 0.2% of the total unsigned flux before inclusion in our algorithms . for a discussion of how the parameters @xmath16 , @xmath30 , and @xmath22 effect the final partitioning , see @xcite . after partitioning each magnetogram , we attempt to associate the partitions at one timestep to those in the next . to begin with , each region is characterized by its net signed flux and centroid location : @xmath31 for short , we call a region s total flux at the flux weighted centroid the region s associated _ pole_. as a first pass at creating consistently identified regions , we calculate the distance between each pair of centroids at two consecutive timesteps . if centroid a at time @xmath32 , @xmath33 , is closest to centroid b at time @xmath34 , @xmath35 , and b has a as its closest neighbor , and that distance is less than a threshold ( 10 mm ) , then we conclude that region b is region a. both this simple method of association , and exploration of the partitioning parameter space , result in mask arrays that are not to the quality required for mct analysis of an emerging flux region . we therefore developed two more sophisticated automatic procedures . and , described below , provide refinements to the minimum distance identifications . both were heuristically developed to address a prevalent type of inconsistent region labeling . the first algorithm deals with a region that exists for just a single timestep , but is clearly part of another region . the second deals with regions that change labels from one timestep to the next . these problems arise , for instance , when the choice of parameters poorly represents a subset of the timeseries , or when two regions merge or split , and their centroid locations between two timesteps are very different . for clarity , our examples in the following subsections do not directly use masks from our time series , and are only meant to illustrate the action of each algorithm . both algorithms operate directly on pixels in the mask structure , so that there are no physical scales involved . the algorithms only relabel nonzero elements of the mask , so that no external mask boundaries are modified . the first column demonstrates the problem of `` flickering . '' the middle timestep contains a new created region , @xmath36 , inside of the area controlled by region @xmath37 in the surrounding timesteps . @xmath36 only exists for that single timestep , its area reverting back to @xmath37 in the third timestep . the right column shows the effect of running on this region : in the middle timestep , @xmath37 has been completely restored in place of @xmath36.,scaledwidth=50.0% ] attempts to smooth the temporal structure of our mask arrays by better associating regions that last for just a single timestep . this happens when the saddle points in the tessellation algorithm slosh back and forth over the threshold value . figure [ fig : rmvf ] shows an example of this problem . region n2 pops into existence in the middle timestep , occupying some of the territory ascribed to region n1 in the first and third timestep . we see that this is not a `` real '' new region , but should instead be considered part of n1 for the entire time . figure [ fig : rmvf ] column two shows the result of running on this data . the algorithm works on a sliding , three timestep window ( initial , middle , and final timestep ) , and in two steps . in the first step , we find all regions @xmath38 that have nonzero flux only in the middle timestep , and find all mask pixels in the initial and final timestep which overlap @xmath38 . we take each overlapping region @xmath39 and relabel those pixels of @xmath38 which are overlapped by @xmath39 in the initial and final timesteps as @xmath39 . because our source regions change shape and size ( pixel count ) over time , we will usually have leftover pixels after performing the above bulk relabeling . while leftover pixels remain , we relabel the pixel of @xmath38 that currently borders the greatest number of pixels of a single neighbor to that neighbor s label . we have found that , when multiple regions overlap a flickering region , this method divides those pixels between the overlapping regions in decent proportion to their past and future `` control '' of flickering territory . the left column demonstrates the problem of `` vanishing , '' in which a single region n5 in the top left panel cycles through series of different labels . the right column shows the effect of running on this data . the @xmath40 and @xmath41 axes are in arcseconds from disk center , though the physical scale is irrelevant to the algorithm s operation.,scaledwidth=50.0% ] with we dealt with regions that existed for just a single timestep . we now deal with the conceptual inverse problem , where a region exists for some time and then suddenly changes names in the next timestep . figure [ fig : rmvv ] demonstrates the problem , as the region labeled n5 in the top left panel ( and previous , undepicted times ) cycles through a series of names and partitions within a few timesteps . we again begin by considering the mask array in a sliding , three timestep window . we find all non zero flux regions within the window and determine which regions exist in which timesteps . for every region that disappears in the middle timestep , we project its area from the first timestep into the middle and last timesteps . we then find all new regions , in both the middle and final times , that overlap with that projected area . finally , we relabel any new region whose centroid lays within this projected area . figure [ fig : rmvv ] , right column shows the effect of running on data . and each use a sliding , three timestep window , so that it takes many repetitions of each for updated information propagate from the beginning to the end of the mask array . we must therefore repeat each of the algorithms many times . at the same time , acts upon labels that last for just a single timestep ; eventually , all such problems will be fixed . taking these processes into account , we find that the mask arrays best approximate the photospheric field evolution when we switch back and forth between the two algorithms at the beginning of the process , and run just many times at the end of the process . for our noaa ar 11112 data , we ran a total of 35 repetitions , and a total of 1058 repetitions . together , these two algorithms accomplish about nine tenths of the work in creating a consistent set of tessellated masks . part of the remaining tenth may be accounted for by a boundary shift algorithm , described below , between abutting regions of the same polarity . even this does not accurately represent to data , and the final changes to the mask array required for a consistent time series must be done by hand . these changes again mostly involve the placement of the boundary between abutting regions of the same polarity , caused by a failure to distinguish between old flux and actively emerging flux . these boundaries are adjusted manually until each region s flux evolves smoothly . figure [ fig : fluxplot ] top shows the vertical flux in each ( high flux ) region . this includes editing of the masks by both the automatic algorithms and by hand . note that , even after a set of consistent masks have been created , the flux within each region is still rather noisy . to counteract this , we smooth our data with a 7 hour ( 13 timestep ) boxcar function . the result is shown in figure [ fig : fluxplot ] bottom . ( a ) flux in regions with more than @xmath42mx . large , opposing spikes between pairs of curves , for instance between p15 and p19 , show boundary shifts between adjacent regions . ( b ) result of smoothing the data by convolution with a 7 hour boxcar function.,scaledwidth=80.0% ] at this point , we transition from a mostly qualitative treatment of the photospheric field s geometry to a quantitative , topological analysis of a model coronal field : the minimum current corona ( mcc ) model developed in @xcite . we begin by replacing every region in the photospheric field with a magnetic pole defined by the region s total flux and centroid location , as in . with one exception , the rest of this work deals only with the poles and connections between them . the total flux in each pole must be connected to some number of other poles of opposite polarity . this defines the system s connectivity : the undirected , weighted graph , where each vertex of the graph is a pole , and the weight of each edge defines the connectivity between the two vertices . it happens that the graph of noaa ar 11112 is _ simple _ on the eve of the m3 flare , so that each pair of vertices has no more than one associated edge . we refer to edges of the graph as domains . we may additionally define a connectivity graph for the flux _ change _ between two timesteps : if a pole s flux increases , that increase must be distributed among its domains with opposite polarity poles . this is the topological entity we now quantify . the flux in a single photospheric region , and hence that region s associated pole , may vary in two ways . the change is either a true evolution of the field by submergence or emergence through the photosphere , or it is a transfer of flux to / from another like signed region via a shift in the boundary between the regions . that this latter type still exists in our `` consistent '' data can be seen in regions p4 , p15 , and p19 in the smoothed data of figure [ fig : fluxplot ] . these three regions compose the large , diffuse positive region to solar west , as seen in figure [ fig : oldflux ] . if we focus on each of these regions individually , the red , green , and purple lines in figure [ fig : bndry ] , we see several rapid changes in the flux of each . however , if we look at the total flux in these three regions the black line in figure [ fig : bndry]it steadily decreases : the rapid increases and decreases in figure [ fig : fluxplot]b are almost solely due to shifting boundaries between the three . further , the steady decrease shown in the black line is matched by decreases in the old flux part of the flux ring , regions n1 , n6 , n38 , n49 , n55 , and n67 in figure [ fig : newflux ] : the old positive flux is `` submerging '' with the old negative flux . the blue line is a demonstration of the effectiveness of the forthcoming algorithms , and we will revisit it shortly . the submergence of regions p4 , p15 , and p19 ( red , green , purple ) , shown in the total flux of the regions ( black ) . blue shows the calculated submergence by removing flux change due to boundary shifts from the total flux change . ] in order to isolate the different varieties of flux change , we have developed an algorithm which estimates the change in each region s flux due only to boundary displacements between adjacent , like signed regions . these changes necessarily come in pairs : what one region loses in a boundary shift , another gains . our algorithm works as follows . consider a set of like signed poles @xmath43 , and split @xmath43 into submerging and emerging sets , @xmath44 and @xmath45 respectively . each pole has an associated flux change , @xmath46 , with @xmath47 the flux at time @xmath32 given by . we iteratively find the pole with smallest unsigned flux change , @xmath48 , and the pole with closest centroid @xmath49 and opposite sense flux change , @xmath50 . the flux change between these two poles is canceled , @xmath51 and @xmath52 , and the cancellation is recorded in a change in connectivity matrix : @xmath53 . this process is repeated until no more connections can be made . @xmath54 is antisymmetric , and the @xmath55@xmath56 element is the flux change of the edge joining @xmath57 and @xmath58 . each row @xmath57 describes the flux change for a given pole , and each column @xmath58 in a row records how much of pole @xmath57 s total flux change is in partnership with @xmath58 . if all of pole @xmath57 s flux change is due to boundary shifts , then @xmath57 s total flux change is given by summation along the @xmath57@xmath56 row of @xmath54 : @xmath59 in this way , we have paired as much shrinking flux with like signed increasing flux as possible , but have only used the total flux and centroid location of each region . we now refer back to the photospheric mask for the last time . for each region in the mask array we find every like signed region with which it shares a border : call this subset of poles @xmath60 . to allow for regions separated by a few pixels to share a boundary and supporting media : at various timesteps , n38 and n49 are separated by several pixels of low field , yet there is a clear transfer of flux across their shared boundary . ] , we pad each region by 5 pixels and look for overlap with other regions . the value of @xmath54 for each pair of such regions is then the amount of flux they exchanged through a shift in the mask boundary . we have chosen to distribute the flux change and then restict to the subset @xmath60 , rather than the reverse , in order to avoid the ill defined problem of assigning shifts when one region shares boundaries with multiple other regions , which in turn share boundaries with multiple other regions , and so on . we see this , for instance , in the triple boundary between regions p4 , p15 , and p19 . qualitatively , any discrepancy is small given that we connect regions in a nearest to farthest order , and nearby regions tend to share boundaries . the sum of all such changes for a given pole gives that pole s total flux change due to boundary shifts . all other change must be ascribed to submergence or emergence through the photosphere . we may describe this mathematically as @xmath61 the middle term on the rhs is the pole s boundary change , determined in this section ; the final term quantifies the pole s flux change due to submergence and emergence through the photosphere , which we now quantify . our algorithm for determining the flux change due to submergence and emergence rests on two assumptions . first , pairs of poles submerge and emerge together , and second , a single region can either consist of submerging or emerging flux , but not both . these assumptions naturally break our connectivity change graph into two disconnected subgraphs : one consisting of submerging vertices , the other emerging . each subgraph is composed of both positive and negative polarity poles , as flux connects only regions of opposite polarity . in determining the flux change in each domain , we use the same algorithm as for boundary shifts , only with the roles of polarity and sense of change reversed : these domains connect opposite polarity poles with same sense flux change , and represent pairs of poles submerging or emerging together . the graph generated in this manner reflects our physical intuition about how these systems should interact . namely , we expect poles to be more connected to those poles closest to them , rather than those further away . we also expect that poles with little flux change should not have that small change spread between a large number of partners . example map of sub / emerging flux , as generated by the algorithm of [ sec : qsef ] . black lines show submerging regions , white lines emerging . labels mark the centroid locations of the high flux regions of figure [ fig : fluxplot ] , which also contribute to the topological analysis of [ sec : ergapp ] . ] we begin by finding the flux difference for each pole between times @xmath32 and @xmath34 , from which we subtract the change due to boundary shifts , found previously . the remainder is each pole s flux change budget , which must be paired with other poles . to accomplish this , we again iteratively find the least changing pole and cancel its flux change with its closest appropriately signed and sensed neighbor . this cancellation is the weight of each edge and is recorded in the source change matrix , @xmath62 . just like the boundary shift matrix , @xmath62 is defined such that , if all of a pole @xmath63 s flux change between times @xmath32 and @xmath34 is due to real photospheric changes , then @xmath64 ; that is , summation over all columns of a given row returns the total flux change for that row s pole . while the combination of our algorithms dealing with boundary shifts and emergence do quite well in most circumstances , occasionally they produce unphysical connections . consider the timestep shown in figure [ fig : smat ] . here , the emerging positive regions ( p22 , p29 , p48 ) each have some flux change , and our knowledge of the location of emergence says that this change is most likely a combination of boundary shifts and emergence for p48 and p29 , as both n87 and n58 are also actively emerging , and primarily boundary shifts for p22 . however , if the boundary shift between p22 and p29 is larger than the emergence of p48 , then our boundary algorithm ascribes the boundary shift to p29p48 , leaving the emergence mostly between p22 and n87 , which is clearly unphysical . for this reason , we allow for `` preferred connections '' , which accounts for emergence between a subset of poles first , and then proceeds to calculate boundary shifts and submergence and emergence for all poles . for this region , we first pair emergence between regions p22 , n55 , n58 , and n87 for the first 3 days ( 77 timesteps ) , and then between regions p48 , n55 , n58 , and n87 during the remaining time . as with the generation of consistent mask structures in the previous section , we ultimately wish to automate this entire process . however , until we develop algorithms that accurately capture the physical evolution of emerging regions , we will rely on a combination of automatic proceedures and ad hoc prescriptions to most accurately model these systems . figure [ fig : smat ] shows a graphical representation of the changing domain fluxes due to submergence and emergence between 12:04 and 12:35 ut on oct 16th , 7 hours before the flare . black lines show submerging domains , white lines emerging . table [ tab : smat ] shows the flux change in each domain . regions p4 , p15 , p22 , p29 , and n55 have flux change ( for these timesteps ) due solely to boundary displacements with neighboring regions . as discussed above , we begin pairing p48 , which increased by @xmath65mx . the closest ( and only ) negative pole with the correct sense of flux change is n87 , which increased by @xmath66mx . these two poles are paired , the domain flux change is set to @xmath66mx , and p48 s flux budget is reduced to @xmath67mx . there are no more available connections between our preferred poles , so we next account for boundary shifts . after doing so , the least changing region is n38 , which loses @xmath68mx . the closest positive pole with the correct sense of flux change is p19 , which lost @xmath69mx . these two poles are paired , the domain flux change is set to @xmath68mx , and p19 s flux budget is reduced to @xmath70mx . we again select the least changing region ( n49 ) , and the process continues until no more pairings can be made . the final results for this timestep are shown in table [ tab : smat ] . lcr domain & sense & @xmath71mx ) + p19n1 & submerge & 245.764 + p19n38 & submerge & 30.6279 + p19n49 & submerge & 145.569 + p19n58 & submerge & 257.030 + p48n87 & emerge & 849.874 + the pairings represent a compromise between reasonable inference and mathematical necessity . due to their proximity , and common increase , it is reasonable to assume p48n87 are feet of the same emerging flux tubes . p19 , on the other hand , is relatively isolated but has steadily decreasing flux . it is necessary to pair it with decreasing negative poles , which turn out to be arranged along the periphery of the old polarity . being separated by @xmath72 , these regions can not be `` submerging '' in any real sense . however , we believe this decrease represents steady cancellation with the small scale field surrounding these old flux regions . this process should be studied in detail in another investigation , but for the moment we note that it is well represented by a reduction in , for instance , the p19n49 domain flux . we repeat this process for each pair of consecutive timesteps . figure [ fig : p29dom ] provides a representative example , showing the cumulative domain flux changes ( designated by the symbol @xmath73 ) due to the emergence , and slight amount of submergence near the end of our analysis , of p29 , one element of the newly emerging flux . we see here that it emerges primarily with region n58 , part of the newly emerging flux . cumulative changes in each of p29 s domain fluxes , as calculated with the algorithm of [ sec : qsef ] . we find that p29 primarily emerges with n58 and n38 . ] the effectiveness of the combined algorithms can be appreciated in figure [ fig : bndry ] , showing the three componenets of the diffuse western region . the blue line illustrates the effectiveness of our algorithms at capturing the system s behavior due to submergence and emergence . as the field shears , the boundaries between the three subregions have a tendency to discontinuously jump , resulting in large variations in the flux in each region , accounted for by the boundary algorithm . however , the combined flux of the three subregions , shown in black , instead displays a fairly steady decrease : this whole region is submerging . the blue line shows our reconstruction of this submergence using the source change matrix at each time . each pole has an initial flux value , to which we add the elements of @xmath62 , so that each pole s flux at time @xmath32 is given by @xmath74 . this measure completely discounts the ( artificial ) boundary shifts between photospherically adjacent regions . we then sum the flux in each pole thus reconstructed , and this is our estimate for the submergence of diffuse western region . we note that the blue and black lines follow each other reasonably well , indicating that our algorithms accurately capture the submerging trend of this region . in the second section of our analysis , we employ the minimum current corona ( mcc ) model @xcite to use the amount of flux change in each domain to calculate the amount of free magnetic energy stored in our system as it evolves away from a potential configuration . to that end , we must define a set of topological elements . these definitions , briefly summarized below , are described more fully in @xcite and references therein . the general idea is to model distributed photospheric sources as point sources . the potential field generated by these sources will have null points , @xmath75 , where the magnetic vector field vanishes : @xmath76 . we may perform a linear expansion of the field about these points @xcite @xmath77 where the jacobian matrix @xmath78 is both symmetric and traceless because @xmath79 . as such , @xmath80 has three orthogonal eigenvectors with three real eigenvalues : the eigenvectors of the two like signed eigenvalues define a plane ( the fan ) ; the eigenvector of the opposite signed eigenvalue defines a line orthogonal to this plane ( the spine ) . from this foundation the rest of the topological description follows . we define a _ pole _ , the photospheric point source , located at the flux weighted centroid of a mask . _ nulls _ are the zeros of the potential magnetic field , and _ spines _ , the fieldlines connecting a pole to a null along the single eigenvector . a _ fan _ ( synonym _ separatrix _ ) is the set of fieldlines ending in a null s like signed eigenvector plane . domain _ is a simply connected space filled with field lines connecting a given pair of poles . finally , a _ separator _ is the field line connecting two nulls , formed at the intersection of their respective fan surfaces . in a nonpotential field , the separator may broaden into a two dimensional ribbon . coronal current sheets form along the separators of the field . we determine the locations of all nulls in two steps . first , we find those laying in the photospheric plane via the newton raphson method of @xcite . second , we check the euler characteristics given in @xcite , which provide both 2d and 3d relations between the number of nulls of each type and the number of sources of each polarity . when the 3d characteristic is not satisfied , at least one coronal null is missing . when then supply a `` by eye '' list of initial @xmath81 guesses to the same newton raphson root finding method until coronal nulls have been found and the euler characteristics are satisfied . other more automated coronal null detection methods have been developed @xcite , but the ad hoc method described here has worked well in this case . after determining the location of all nulls , we located separators via the method of @xcite , with one modification . those authors only treat photospheric nulls , whose separators lay within separatrices above the photosphere . for coronal nulls , we search for separators in all @xmath82 directions within the null s fan surface . our model photosphere consists of a plane @xmath83 of isolated sources surround by regions where the normal component of the magnetic field is zero . this boundary condition may be satisfied using the method of images , as in @xcite6 . with the normal field reflectionally symmetric in @xmath84 , we have that @xmath85 . whenever we introduce current in the corona along a separator , we must introduce its reflection in the mirror corona to maintain this symmetry . in this way , our separators form closed current loops . topology of noaa ar 11112 on the eve of the flare , depicted in a local tangent plane with coordinates . the point of tangency is taken as the center of charge in the initial magnetogram , translated through solar rotation to the present time , and the axes are in mm . pluses and crosses are positive and negative poles , respectively ; triangles are positive ( @xmath86 ) and negative ( @xmath87 ) nulls ; solid black lines depict spines , dashed lines the trace of fans within the photosphere , and the dotted line is a coronal spine , attached to the coronal null b15 . blue and red lines are projections of separators into the photosphere , with red lines those separators connected to the coronal null.,scaledwidth=75.0% ] once we have created the set of matrices describing domain flux changes due to submergence or emergence , we can create similar matrices describing the flux change due to a changing potential field . we calculate the potential field connectivity at each timestep via the monte carlo method of @xcite . this produces a set of connectivity matrices @xmath88 , where @xmath89 is the flux connecting sources @xmath57 and @xmath58 at time @xmath32 . from these matrices we calculate the change in connectivity due to the changing potential field : @xmath90 summation along rows of the connectivity matrix returns the total flux of the corresponding pole ; eg . we define @xmath92 to be antisymmetric , as with the matrices @xmath62 and @xmath54 before . the mcc model derives an energy from the discrepency between the actual domain fluxes @xmath93 at time @xmath94 and the potential fluxes then , @xmath95 . in the absence of flux emergence or submergence , the actual fluxes are fixed using the potential field domain fluxes at some initial time : @xmath96 , as in @xcite . any flux emergence or submergence through the photosphere , quantified as @xmath62 in the previous section , modifies these actual domain fluxes , so that @xmath97 because we assume a field , initially potential , whose fluxes are fixed under future evolution , we determine how far removed the field is from a potential field configuration by answering the question , `` what flux must be added to a potential field domain @xmath98 at time @xmath99 to get a potential field domain at time @xmath32 ? '' this question is answered by @xmath100 on the right hand side , the first term is the potential field at the initial time . the second is the total flux change through the photosphere . the final term is flux change due to coronal redistribution , which must be achieved by modifying the connectivity matrix . together , the first two terms give the actual domain flux @xmath101 at time @xmath32 . all flux changes from coronal reshuffling are then given by @xmath102 equation holds for all domains , of course , and we represent this as a matrix equation by dropping the domain subscript @xmath98 . @xmath92 is antisymmetric , and we have defined @xmath62 to be antisymmetric , so @xmath103 is also antisymmetric . while antisymmetry of these matrices carries no physical information , we will later show that it does endow @xmath103 with a nice mathematical property . note that the difference in between the actual domain fluxes and the potential field fluxes depends on the choice of the inital time @xmath104 . when dealing with actual data , we must pick an initial time to apply the mcc flux constraint when we believe the field is in a potential configuration in order to find a meaningful difference relative to a later potential configuration . ideally , we would track a single region from its inception up to an event which redistributes the coronal flux , as in a flare . while noaa ar 11112 does not fit this criteria , it likely matters little in this case . as is apparent from our supporting media , and also as we will show below , noaa ar 11112 has a very stable flux configuration for @xmath105hours prior to flux emergence . this newly emerged flux , fixed in set domains as the photospheric field continues to evolve , rapidly diverges from a potential field configuration at later times . so far this discussion only involves the sources themselves and their interconnections . let us now introduce a set of separators @xmath106 . separators are field lines that run from a null point of one type to a null of the opposite type . in the mcc model , they are the sites of current sheets within the corona@xcite . because current may only flow in loops , the separators themselves must close along some path in the mirror corona . we can then speak of separator fluxes : the flux of domains linked by some separator @xmath107 with its closure . we will discuss closures and explain the concept of linking in detail below , but the general idea is that field lines in a linked domain can not reconnect to another domain except by passing through the separator . let @xmath108 be the flux linked by @xmath107 at time @xmath32 in the actual field , and @xmath109 be flux in the potential field domains @xmath110 linked by @xmath107 . for a given closure , a separator always links the same set of domains throughout time . the fluxes in those domains will generally change over time , however . when a domain s flux goes to zero , the domain no longer exists . the flux constrained equilibrium ( fce ) assumption of the mcc model states that domain fluxes are fixed as the field evolves . minimization of the field s energy subject to these constraints shows that the coronal field will be current free except along the separator , where a current ribbon forms @xcite . this occurs whenever @xmath111 . we therefore define @xmath112 the double sum over elements of the redistribution matrix @xmath113 is the difference between the flux linked by the separator in the actual and potential fields at time @xmath32 . it is also the self flux generated by current flowing along the separator , and therefore , the flux that must be added to the real field to relax the fce constraint of mcc : @xmath114 . our method for calculating energies therefore rests on two assumptions . the first is that at some initial time @xmath115 we assume that the domain fluxes are given by the potential field s domain fluxes , @xmath116 , so that all the separator fluxes @xmath117 . the second assumption is that the evolution of the current ribbon s self flux depends on the evolution of the both the actual and potential field configurations : @xmath118 summed over time , this is just the restatement of itself , namely that the current ribbon s flux at some time @xmath32 is the sum of all changes in each linked domain s flux over the life of the separator . when @xmath119 is small we can estimate the properties of the current ribbon , as in @xcite . for a separator of length @xmath120 carrying current @xmath121 , they determine the self flux to be @xmath122 where @xmath123 is a measure of magnetic shear in the separator s vicinity . from this the excess energy of the mcc field relative to the potential field , i.e. the free energy , is@xcite @xmath124 note that and only take into account energy due to the self inductance of each current loop . in general , we must include the effect of mutual inductance , though for physically separated loops we expect self inductance to dominate the total energy . the fce assumption requires that @xmath125 , which allows us to solve for the current in the current ribbon as a function of @xmath126 : @xmath127 where @xmath128 is the inverse of the function @xmath129 . equation provides a method for determining the current residing in , and hence the energy stored by , each separator : it tells us the current required to change the domain flux enclosed by a separator from that of a potential field to some other value . in the present case that value is given by the non zero elements of @xmath103 . a reverse , cumulative sum over the time index @xmath32 for a given flux domain @xmath63@xmath130 gives the time history of departure from a potential field configuration . as is implicit in , each separator may link more than one domain : the appropriate @xmath119 for a separator is the sum of all changes in domain fluxes for each domain linked by that separator . since all field lines in a given domain are topologically equivalent , we may establish separator linkage using a single representative fieldline and the separator itself . after each of these open curves is closed , their linkage is found using the gauss linking number@xcite , @xmath131 , for each field line separator pair : @xmath132 the linking number not only determines whether a separator links a field line ( @xmath133 ) , but also the sense in which it does so ( @xmath134 an integer ) : to calculate the correct @xmath135 , you must sum over all linked domains , multiplied by their respective linking numbers . in order to use gauss linking formula , we must have two closed curves , whereas our curves ( usually ) begin and end at two separate sources ( nulls for separators ) in the photosphere , never penetrating beneath . so , in order to apply the linking formula in this case we must add to each coronal curve some curve in the mirror corona to form a complete loop . there are , in general , many ways to form the closure for each curve , and not all of them will lead to the same linking number . one requirement is that the separator be closed _ above _ the field line s closure : any closure below the field line will always give a linking number of zero . therefore , we have chosen to close the separator with a straight line between the footpoints in the photosphere , and the field lines with a rectangular path formed by following the footpoints down and connecting them with a straight line in the @xmath136 mm plane . another method would be to close the separator , say , with its projection into the @xmath137 ( photospheric ) plane . either method is valid , so long as the same method is used for all field lines and separators . another requirement for using the linking number to find linked domains is that we follow field lines and separators in a consistent direction . in general , changing the direction of _ one _ of the integrals takes @xmath138 . for definiteness , we always trace separators from negative ( a type ) nulls to positive ( b type ) nulls above the photosphere , and thus close separators from b@xmath139a in the photosphere . because the reconnection matrix @xmath103 is antisymmetric , this same care is unnecessary when tracing the field lines , at least for the purpose of the @xmath135 calculation . let us designate by @xmath140 the difference in flux change for the real versus potential field between times @xmath32 and @xmath34 ( as in ) , when we trace a field line from p1 to n2 in the corona , and close the field line from n2@xmath139p1 below the photosphere . then , @xmath141 we have therefore been justified in simply writing @xmath142 , with no designation of tracing a field line from positive to negative or negative to positive in the corona , because @xmath142 is just a summation of multiple such domains . this holds provided we always trace the separator in the same direction . once determined , the linking number for each separator / field line pair provides the correct method for calculating each separator s total @xmath135 , including all linked domains , and hence the energy stored by each separator . we must account for instances where two photospheric regions merge together or split apart , as illustrated in figure [ fig : constrict ] . therein , we focus on 4 poles@xmath143with four flux domains@xmath144and two separators , @xmath145 and @xmath146 , which link the domains in some fashion . suppose two of the poles are each connected to the same null via its spines . if the flux of one pole is zero before(after ) a split(merger ) , then the null between them does not exist at that time , and the flux in the domains linking the sources goes to zero . now , if either @xmath145 or @xmath146 is attached to that null , then the separator will arise or cease to exist with that null , and the flux linked by it is appropriate . then again , if these separators are attached to some other set of nulls and happen to link these flux domains , our calculated linked flux is still appropriate . say flux domain @xmath147 , for instance , either folds into or breaks out from domain @xmath148 . the actual break or merger constitutes a boundary shift ( with @xmath149 having zero flux either before or after the event ) and so does not contribute to the flux linked by either separator . any additional flux change on top of this is either appropriately picked up by @xmath145 or @xmath146 , but not both . topological depiction of two poles either splitting or merging : @xmath150 . dots depict the sources , lines the domains , and red arcs the separators . ] a second complication arises when we encounter a coronal null , as we see in the topology of noaa ar 11112 in figure [ fig : topo ] near local tangent plane coordinates ( 20,30)mm . if a coronal null has any separators connected to it then it must have at least two : one carrying current up , the other down . if many separators connect to the coronal null , then some must carry current up , and others down , but all balanced according to kirchhoff s rules . in order to calculate the flux within a separator induced domain , we must close the separator somehow . above , we addressed this by prescribing a photospheric closure for separators . now , however , we must first stitch pairs of separators together in order to get back to the photosphere . if we have @xmath151 separators connected to the coronal null , we require @xmath152 separator pairs , or `` isolating loops '' in the terminology of @xcite . we might assume that the currents flowing along the @xmath152 loops would distribute to minimize the energy due to self and mutual inductances . this problem remains to be solved , and for the moment we simply designate one separator linked to a coronal null as the `` shared '' separator . we have found that the energies we calculate do not depend heavily on the choice of common leg , laying within @xmath153 of each other . as shown in figure [ fig : topo ] , we have a single coronal null , b15 , with spine lines leading to @xmath154 and @xmath155 . four separators ( in red ) connect to the coronal null : from null a09 between @xmath156 and @xmath157 , a05 between @xmath157 and @xmath158 , a08 between @xmath158 and @xmath159 , and a07 between @xmath159 and @xmath37 . again , because all current flowing from the photospheric nulls up to the coronal null must be balanced by current flowing back down to the photosphere , we mush combine these four separators into three , all of which share one leg ; it does not matter which one . we have chosen a05b15 leg to be common to all three separators . as we will see below , the a05b15a08 separator generates the greatest amount of energy in this system . this is expected , given the high horizontal shear in the photospheric field between @xmath154 and @xmath158 , and the greatly flux deficient domain between these two poles , relative to the potential field configuration at the time of the flare . we may summarize the previous several sections in the following way . in order to perform an energy calculation , we need a set of poles @xmath160 at a given time final time @xmath161 ; a set of nulls at the same time @xmath162 ; a set of separators at the same time @xmath163 ; a time history of the reconnective changes in domain fluxes from the final time , backwards to a previous time @xmath94 where we may assume the field was potential , @xmath164 ; and a set of domains @xmath165 linked by each separator . having all these pieces , we then calculate the total domain flux difference in each domain at time @xmath161 relative to the initial presumed potential field . for each separator , we find all domains linked by that separator . next , each linked separator s time history of reconnective flux changes is summed . this total flux change for each linked domain is then multiplied by the gauss linking number for that domain / separator pair . finally , the total domain changes are summed for each linked domain to fix the total flux difference of the separator at the final time @xmath161 , relative to the potential field at the initial time @xmath94 . mathematically , this is given by , reproduced here : @xmath166 this flux is generated by the current along the separator , , which may be found by the inversion , and plugged into the energy equation to find the energy in excess of the potential field energy at time @xmath161 in the minimum current corona model with the fce constraint applied at time @xmath94 . this energy is a lower bound for any coronal energy model . by time of the flare ( figure [ fig : newflux ] ) the new flux in noaa ar 11112 has bubbled up and expanded within the interior of the old flux ring . the effect is especially pronounced between regions p22 and n49 , the boundary of which has a very strong horizontal gradient in the vertical field . because this gradient is between old and new flux , prior to reconnection during the flare there is little flux in the domain connecting these two regions , and certainly far less than there would be in the potential configuration at that time . we expect one action of the flare will be to increase the flux within the p22n49 domain . because its flux is conserved , any flux p22 adds to its n49 domain must come at the expense of its other domains . the poles with which p22 shares these other domains primarily n58 , n55 , and n87 with which p22 emerged must in turn shuffle their flux into different domains . in this way , a flare originating in one place propagates its effects throughout the entire magnetic structure until it establishes a new , lower energy equilibrium . cumulative domain flux changes for various p22 domains , relative to the potential field at t=0 . arrows indicate the onset of flux emergence and the ut 19:07 m2.9 flare . there is some `` pre emergence '' flux due to our 13 timestep smoothing function . ] we can get a picture of this process by looking at the cumulative domain flux changes for various p22 domains , shown in figure [ fig : p22rmat ] . here , the flux at each timestep is given by the cumulative sum of all flux changes up to that time , @xmath167 . this figure shows by how much flux a domain must be changed to match the potential field configuration of the initial field , plus the contribution due to emergence . p22 emerges in conjuction with n55 , n58 , and n87 , thus all flux is assigned to domains p22n55 , p22-n58 , and p22-n87 . the potential field , however , assigns less flux to these domains and more ( i.e. some ) to domains such as p22n1 and p22n49 . these latter domains are flux deficient relative to the potential field , and appear above the zero line in figure [ fig : p22rmat ] . the former are flux excessive , as a result of emergence , and appear below . note that the potential field distributes the flux among domains differently at different times , and our analysis captures these changes . in particular , p22 emerges to the east of n1 , then migrates westward . as p22 moves past n1 , the p22n1 domain first becomes deficient , then more sufficient as the p22n38 and p22n67 domains become increasingly deficient . the total flux in p22 only slowly increases throughout much this time ( it is mostly emerged by 50hrs , as shown in figure [ fig : fluxplot ] ) , so these later changes in the potential configuration are largely due to shear motion in the photosphere , rather than the initial emergence . aia 171 images during the flare reveal how reconnection works to restore a flux balance closer to potential . at successive times , we see brightenings of small loops progressing west to east within the newly emerged regions . after this we see a series of loop brightenings connecting the new negative regions within the ring to the old diffuse positive region to the west . we believe that this series of loop brightenings is a direct result of domain restructuring due to the emergence of the new flux and its highly nonpotential original field configuration . two cumulative time sums over @xmath168 . the solid line sums from the initial time , 0 , to time @xmath32 and is the flux required to change the domain at time @xmath32 to an assumed potential field at the initial time . the dotted line sums from time @xmath32 to the final time , @xmath169 , and is the flux required to change the final potential field configuration to a flux configuration fixed at time @xmath32 . the difference between the initial and final values is the same for both sums , and they are mirrors of each other about the average of this difference . ] the fluxes shown in figure [ fig : p22rmat ] provide a good description for understanding the evolution of each domain , and directly follow the calculation outlined in [ sec : seps ] . in this paper , however , we are interested in calculating the free energy due to a difference between domain fluxes fixed at some initial time and potential fluxes at the time of the flare . for a completely new active region , we would assume all domain fluxes are zero except those which emerge together ; in this case , our emergence occurs in a region of substantial old flux . as such , our final time is fixed by the flare , and we must choose an appropriate time of initial constraint . combining equations and , we find that @xmath170 , so that@xmath171 is the flux that must be added to domain @xmath98 to account for change in addition to submergence and emergence , which is accounted for within @xmath172 . figure [ fig : p22rmat ] shows the cumulative flux @xmath173 . wishing to vary the time of constraint , we calculate the reverse of this : @xmath174 . this is the amount of flux that must be added to the domain constrained at time @xmath32 , @xmath175 , to match the potential field domain at time @xmath176 . as we will show below , this summation is basically constant prior to emergence : the old flux region is initially stable , and flux emergence drives a departure from this stable configuration . figure [ fig : twosums ] gives an example of both the forward and reverse summations for domain p22n49 . we set the fluxes for the energy calculation in this way because the assumption of the mcc model is that , barring reconnection , the domain fluxes are fixed , and prior to the flare there is no reconnection . in order to compare with a potential field at the time of a flare , we determine our topology immediately before the flare . this fixes all of the poles , nulls , and separators , and the domains linked by each separator . the energy calculation then answers the question , `` _ _ supposing the domain fluxes are fixed at time @xmath32 , by how much must they be changed to match the potential field at the final time ? _ _ '' this difference for each domain linked by a separator then determines the amount of current along that separator , and hence the amount of energy stored by that separator . a reverse cumulative sum of the total domain fluxes linked by each separator relative to the potential field configuration just prior to the flare , identified by nulls to which they attach . separators with three listed nulls are those which pass through the coronal null , b15 . ] as mentioned previously , a single separator will generally link multiple domains . for instance , one of the separators passing through the coronal null b15 in figure [ fig : topo ] , connecting nulls a05b15a07 , links three domains , n58p22 , n67p22 , and n38p29 , with gauss linking numbers -1 , -1 , and + 1 , respectively . the residual fluxes for two of these domains , n58p22 and n67p22 , are shown in green and purple in figure [ fig : p22rmat ] ; the n38p29 domain residual fluxes is similar , peaking at @xmath177mx around 90 hours . this separator @xmath107 s flux , including all three domains @xmath178 with linking numbers @xmath179 , is then given at each time @xmath32 by @xmath180 the sum for this separator is plotted in red in figure [ fig : sepflux ] . the curve shows that , as we look further back in time , the flux linked by this domain increasingly departs from the final potential field configuration , until around @xmath181 hours when all changes level off . this is when region p22 first emerges . note that every separator shows relatively little change in the linked flux before this time . based on figure [ fig : sepflux ] , we might expect the a11/b03 separator to store the most energy , as it has the most linked flux . however , the amount of current flowing in this separator is 12 orders of magnitude less than in the three separators connected to the coronal null : the longer a separator , the less current necessary to generate the required self flux . figure [ fig : energy ] plots the magnetic free energy when applying the fce assumption at varying timesteps . we see that the energy of the three separators connected to the coronal null , shown in magenta , red , and dark green , completely dominate the magnetic free energy buildup . the blue line shows the total for all separators . we also see that , constraining the domain fluxes at any time before the onset of flux emergence , @xmath182 hours , effects little change in the total energy buildup . essentially all of the flux difference relative to the final potential field configuration is due to the emergence itself . the mcc yields a lower bound on the magnetic free energy of a system . after accounting for all flux linked by all separators , and hence the energy stored in each separator , we conclude that noaa ar 11112 stores a minimum of @xmath183 ergs of free magnetic energy over the 23 days leading up to the october 16@xmath56 , 2010 goes m2.9 flare . free magnetic energy relative to a potential field at the final time , and applying the fce assumption at successively earlier timesteps for each separator , and for the total ( blue ) . our energy estimate is then the value of the initial timestep , @xmath184ergs . arrows indicates the onset of emergence and the time of the m2.9 flare . note that there is some `` pre emergence '' due to our boxcar smoothing function . ] in this work we have extended the application of the minimum current corona model to the estimation of energy stored along separators in the coronal field in active regions where flux submergence and emergence plays a significant role in the photospheric field s evolution . this estimation operates in two distinct parts . in the first , we track the evolution of the photospheric line of sight magnetic field using sdo / hmi magnetograms . we found that the higher cadence and greater resolution of hmi , compared to previous analysis using full disk mdi data at a 96minute cadence , greatly improved this portion of the analysis . we found that a half hour cadence was sufficient to capture the evolution of this active region . analysis of this region began before the hmi 720 second averaged data were available . future investigations will use the high quality averaged dataset . the tracking information is stored in a mask array , which assigns unique labels to contiguous groups of data pixels , and matches these labels across all timesteps . this framework accounts for all processes we see in the data : submergence and emergence , merging , splitting , braiding , and spinning . in the present work we focus on submergence and emergence of magnetic flux . all other processes are implicitly addressed in our work except for spinning . @xcite have accounted for spinning explicitly in their work and found that its importance is case dependent . we do not believe spinning plays a significant role in noaa ar 11112 . in the second part of our analysis we use the tracking information in the mask array to characterize the system wholly in terms of its topology . each region s total flux , together with its flux weighted centroid , defines a magnetic pole . a potential field extrapolation using the poles as sources , together with a mirror corona where @xmath185 yields all null points of the system . the potential field extrapolation determines the locations of separators , which are where current sheets form due to free magnetic energy in the coronal field in the mcc model . the amount of current flowing in each separator , and the amount of free energy in the field , is due directly to the nonpotentiality of each flux domain . a lower bound on the free magnetic energy is calculated from the separators and domain fluxes using the method of @xcite , who found that the mcc estimate may be less than other mhd free energy estimates by a factor of ten or more . for noaa ar 11112 we found a lower bound of @xmath186 ergs . this method requires that all domains linked by each separator are known . this has been done by inspection in the past @xcite , but in the present work we have proposed an efficient automatic method using the gauss linking number . provided the specification of some closure for both field lines and separators , and because domains are simply connected spaces , calculating the linking number for each separator and a single field line from each domain will find all linked domains for each separator , and the sense in which they are linked @xmath187 . it must be noted , however , that domains are not equivalent to connections : a single pair of positive and negatives poles may be multiply connected , with different domains in a single connection distinguished by one or more separators . we found no such instances of this in the case of noaa ar 11112 , but this scenario may be simply dealt with by tracing a large number of random field lines between each pair of multiply connected poles . one may then calculate the gauss linking number of each separator with the suite of field lines , and determine the proportion of field lines linked by each separator . the linked flux may then be divided among the separators , accordingly . our free energy estimation does have some limitations . most important , we can only calculate energy storage , not energy release , for instance , in a flare . this is because we fix the domain fluxes at some point in the past and compare them to the potential field domain fluxes at some given time ; however the flare is a coronal phenomenon , while the potential field configuration is derived purely from photospheric data . using coronal data to quantify how domain fluxes change during a flare would allow for an energy estimation before and after a flare , though we are aware of no method for doing so . second , our energy calculation only takes into account the self inductance of each current ribbon . for a set of currents loops , the total energy is given by both self and mutual inductance terms . when the loops are physically separated , the self inductance terms should dominate . however , if some subset of loops share a common leg ( as is the case when multiple separators are connected to a coronal null point ) , then mutual inductance will significantly contribute to the total energy and must be accounted for . this work remains to be done . graham barnes graciously provided code for producing the potential field connectivity matrices using a monte carlo algorithm with bayesian estimates , as described in @xcite . development of the code was supported by the air force office of scientific research under contract fa9550 - 06-c-0019 . we thank the referee for his / her thorough reading and insightful comments which helped improve the manuscript . we also thank our summer 2010 reu student johanna bridge for her work in assessing the performance of the automatic tracking algorithms . this work was supported by nasa lws .

in recent years multiwavelength observations of several pulsars are now allowing the testing and refinement of theoretical models of pulsar emission mechanisms and particle acceleration responsible for high energy electromagnetic radiation . there are now seven pulsars ( eight if psr0656 + 14 is counted ) which are to emit _ pulsed _ gamma - rays observed by the egret instrument aboard the compton gamma - ray observatory ( cgro ) ( see thompson et al 1997 ) , . in addition there are more than two dozen isolated ( rotation powered ) neutron stars which are detected by a number of x - ray satellites such as einstein , rosat and asca ( see e.g. becker and trmper 1997 ) . spectra , pulse shapes , relative arrival times of pulses in different electromagnetic bands and , in some cases , phase resolved spectra offer information which can be useful for identifying the zone of acceleration of the charged particles in the magnetosphere surrounding the pulsar , and the way these particles produce and interact with electromagnetic radiation . however , of the seven gamma - ray pulsars identified , only six have detected pulsations in the x - ray band , and only five ( crab , vela , geminga , psr1055 - 52 and psr1951 + 32 ) have measured x - ray spectra . no x - ray pulsations have been detected from the gamma - ray pulsar psr b1706 - 44 , although steady x - ray emission has been seen by rosat from this object . additionally , asca has taken spectral data of the nebular emission surrounding the pulsar and a combined analysis of this and the higher spatial resolution ( single - band ) rosat hri data has been performed by finley et al ( 1998 ) . psr b1706 - 44 is a young neutron star ( spin down age 1.75@xmath2 yr ) with a spin period of 0.102 s and has a large spin down luminosity of 3.4@xmath3 at a distance of 1.82 kpc . the pulsar which was originally discovered in the radio band by johnston et al ( 1992 ) , ranks fourth in a rank ordered list of spin - down flux ( @xmath4 ) . the soft x - ray emission was detected by the rosat pspc ( becker et al 1992 ) with an upper limit to modulation of the x - ray flux at the 18% level ( becker , brazier and trumper 1995 ) . as most gamma - ray pulsars emit pulsed x - rays , it is of interest to detect the x - ray pulsations from this high spin - down flux pulsar and to study its pulsed spectrum . in this paper we report on our attempt to detect x - ray pulsations from this pulsar using the rossi x - ray timing explorer ( rxte ) . preliminary results from this observation were reported by ray , harding and strickman ( 1997 ) . although we have not detected any x - ray pulsations , the stringent upper limit to the pulsed flux we obtained has implications for magnetospheric particle acceleration and radiation models . in section 2 , we describe the characteristics of psr b1706 - 44 and the theoretical motivations for this study . in section 3 , we describe the observation carried out with rxte and its analysis . in section 4 , we compare the broadband observations of this pulsar with the theoretical models of the pulsed radiation mechanism in the light of our new observation . in section 5 , we discuss our conclusions . x - ray emission from rotation powered pulsars can be a combination of varying amounts of three spectral components : 1 ) a soft blackbody 2 ) a power law and 3 ) a hard component connected with heated polar caps . blackbody - like emission seen at energies between 0.1 and 0.5 kev can result from surface cooling of the neutron star . polar cap heating by energetic particles or internal sources may produce a second thermal component at higher energies .. a power - law component could result from non - thermal radiation of particles accelerated in the pulsar magnetosphere . measurement of the pulsed x - ray flux from polar cap heating can act as a calorimeter of the returned particle flux from a polar cap ( pc ) or outer gap ( og ) vacuum discharge accelerator . the returned particle flux is an indicator of the properties and mechanisms of the non - thermal ( gamma - ray and power law x - ray ) emission from the pulsar . in particular pc models predict a returned flux that is a small percentage of the total accelerated particle flux ( harding and muslimov 1998 ( hm98 ) , arons 1981 ) whereas the og models predict roughly equal numbers of particles accelerated towards the pc as flowing away ( halpern and ruderman 1993 ; wang et al 1998 ( hereafter wrhz ) ) . the latter , therefore , would give rise to comparable x - ray and gamma - ray photon fluxes . these measurements for pulsars could serve as an important separator of the two classes of models . at present , it is not clear how and where in the pulsar magnetosphere the pulsed non - thermal high energy emission originates . polar cap models assume that particles are accelerated above the neutron star surface and that radiation results from a curvature radiation ( daugherty and harding 1996 ) or inverse - compton ( sturner and dermer 1994 ) induced pair cascade in a strong magnetic field . outer - gap models ( cheng , ho and ruderman , 1986 ; chra , b ) assume that acceleration occurs along null charge surfaces in the outer magnetosphere and that the radiation results from curvature radiation / photon - photon pair production cascades ( ray and benford 1981 ) . the computed spectra of the pulsars in both polar cap ( dh96 ) and outer - gap ( ho 1989 ) models predict that the spectrum will extend below 50 kev . in polar cap models this emission is primarily power - law synchrotron emission that will terminate at the cyclotron energy of the local magnetic field , which could be anywhere from @xmath5 10 kev at the neutron star surface down to @xmath5 0.1 kev several stellar radii above the surface . a measurement of the pulsed emission below 50 kev would determine how far the spectrum of the non - thermal component extends . if a cutoff in the spectrum is observed , it might be a measure of the local field strength in the emission region and thus , a limit of the proximity of a polar cap cascade to the neutron star surface . the seven gamma - ray pulsars detected by the compton gamma - ray observatory have high spin down luminosities and , because of their association with egret gamma - ray sources , probably have pulsed x - ray fluxes . in the case of psr b1706 - 44 there is not enough spectral information to classify the emission as thermal vs non thermal nor determine whether any fraction is pulsed . rosat has made pointed observations of this pulsar on three different occasions with both the pspc ( feb 1992 and sept 1992 see becker et al 1995 ) and the hri ( mar 1995 see finley et al 1998 ) for 11.8 ks and 28 ks respectively . the former showed a net count rate of 0.022 @xmath6 0.002 cps for the energy range 0.1 - 2.4 kev . the net count rate reported by hri was 0.0087 cps which was considered consistent with the pspc count rate . the hri - determined j2000 source location was in good agreement with the previous radio ( interferometric and timing ) positions of frail et al ( 1994 ) and johnston et al ( 1992 ) . in addition , asca sis and gis instruments imaged the region around psr b1706 - 44 for net exposures of 20 and 23 ks respectively on 10 sept 1994 with reported net count rates of 0.016 and 0.018 cps respectively . while the 2@xmath7 upper limit of the pulsed fraction is 18% in the rosat pspc band , the upper limit for the asca band gives a 22% limit in the 2.0 - 10.0 kev band at a 99% confidence level . the size of the synchrotron nebula ( to the flux levels searched by rosat hri ) was found to be @xmath8 in radius ( finley et al . 1998 ) . spectral analysis of rosat pspc data did not distinguish between black body and power law models , but the lack of pulsations and the similarity of spin parameters of psr b1706 - 44 with those of vela were used by becker et al . ( 1995 ) to favor a power law model with @xmath9 which suggested synchrotron emission from a pulsar powered nebula as the likely origin of detected ( steady ) x - rays . the corresponding neutral hydrogen column depth @xmath10 was required to be 5.4 @xmath11 from spectral fits . the flux within the 0.1 - 2.4 kev range and the bolometric luminosity ( isotropic ) at a distance of 1.8 kpc are about 3.2 @xmath12 and 1.3 @xmath13 respectively . the energy spectra obtained with asca sis and gis are well fit by a single power law model which gives a photon index @xmath14 and @xmath15 with l(2 - 10 kev ) = 5.3 @xmath16 . the bright low mass x - ray binary 4u 1705 - 44 is about @xmath17 away from psr b1706 - 44 ( see fig 1 ) . in addition there is a faint source in soft x - ray band at @xmath18 dec=-44@xmath19 33@xmath20 28@xmath21 . these two sources cause confusion and background problems for rxte since they are all within the fwhm response circle of the rxte pca detector . additionally , the large scattering wings of the asca mirrors mean that flux from these sources contributes to the asca background for psr b1706 - 44 . the egret instrument shows a photon spectral index of 1.57 @xmath6 0.05 in the gamma - ray energy range of 100 mev to 2 gev for this pulsar and steepens at higher energies ( fierro 1995 ) . in addition there are upper limits to the optical ( r - band ) magnitude of this pulsar which are discussed together with multiband observations in the context of theoretical models in section 4.3 . psr b1706 - 44 was observed as an rxte target of opportunity ( too ) when the nearby bright lmxb 4u 1705 - 44 , went into a low state ( corresponding to 5 ct / s in the all sky monitor ) . this lmxb , which is about @xmath17 away from psr b1706 - 44 , remains in the fov of the pca instrument and is therefore a source of contamination which reduces the available pulsed signal to noise . hence it is advantageous to perform the experiment to detect pulsations when the lmxb is in a low state . a gray - scale image of the region of the sky downloaded from the high energy astrophysics archive ( heasarc ) with a superimposed circle denoting the xte pca hwhm radius is shown in figure 1 . psr b1706 - 44 was allocated 147 ksec of too observation . it was observed on two different occasions between november 9 - 12 , 1996 and may 16 - 19 , 1997 . the dates , instrument configurations , exposures etc . , are listed in table 1 . after excising the times of saa passage and earth occultations etc . , the on - source live time was 132 ks . during this time all 5 proportional counter units ( pcu ) were kept on . on certain occasions , the nearby lmxb 4u1705 - 44 underwent type i x - ray bursts , enhancing its luminosity by several orders of magnitude for a few tens to a hundred seconds even though generally , the lmxb was in a quasiperiodic low state ( as seen by the all sky monitor as referred to above ) for @xmath5 20 days . several hundred seconds of data , comprising the bursts and a roughly equal amount of time immediately after each burst were excluded from our timing analysis for psr b1706 - 44 x - ray pulse detection by visually examining the standard2 light curves and then using the ftools software package _ grosstimefilt_. a preliminary investigation of the data revealed that the average count rate in channels 5 - 22 ( in goodxenon-16s mode ) was substantially higher than that in channels 23 - 49 . it is also well known that the instrumental background in channels 49 through 255 usually is too high to make low signal to noise detections of pulsed emission feasible . hence we divided the data into bands containing channels 5 through 22 and 23 through 49 and used mainly the latter set of channels for timing analysis towards pulse detection because the contaminating emission from the lmxb was lower in these channels . for the span of data which was of the mode ea-125us-64m-0 - 1s in november 1996 , we chose the appropriate channel boundaries ( cpix4 ) which corresponded to the same range of energy as the chosen channels ( 23 - 49 ) for the goodxenon mode , for a combined analysis . in order to ameliorate contamination from the lmxb , we considered offset pointing from the pulsar away from the lmxb . however , we calculated , and later confirmed with raster scans of the region , that this would have resulted in a net decrease in the signal to noise ratio . the photon events from the two separate experiment data systems ( pca / eds ) configurations were separately analyzed and fed into the `` fasebin '' ftool . this package handles the solar system barycentric corrections of the arrival times of the photons ( via the jpl de200 solar system and the rxte orbital ephemerides ) , and then epoch - folds the data according to a supplied pulsar ephemeris . the radio pulsar ephemerides used in the epoch folding program , kindly supplied by v. kaspi et al , are displayed in table 2 . the radio observations ruled out any sudden period jumps of the pulsar spin period ( glitches ) near or in between the two epochs of rxte observations , which are separated by nearly six months . other ftools packages allow manipulation of the epoch folded phaseogram according to spectral channels , binning in adequate number of phase bins etc . the epoch folded phaseogram of these events is displayed in figure 2 . the origin of the phaseogram corresponds to the absolute phase of the radio pulse . there is no compelling evidence for pulsed emission at the pulsar period . note that the reduced @xmath22 per degree of freedom ( 19 degrees of freedom ) for the histogram displayed in figure 2 is 1.99 . this means that a random set of fluctuations consistent with a uniform background of the zero level shown ( corresponding to 94.16 cps ) , i.e. the absence of a pulsation at the radio folding period , is ruled out at a 99.25 % confidence level . currently we can only obtain the following 2 @xmath7 upper limit to the _ pulsed _ flux , based on the total observation of 132 ks , in the energy interval 9 kev @xmath23 e@xmath24 @xmath23 18.5 kev band ( ulmer et al 1991 ) : : @xmath25 where we have taken the pulsar duty cycle @xmath26 = 1/2 and the rxte effective area a@xmath27 to be 6000 @xmath28 . this upper limit is shown together with the detections and upper limits of psr b1706 - 44 in other energy bands in figure 3 . the photon spectrum shown in figure 3 from the rosat ( kev ) band to the tev band has implications for theoretical models of pulsar radiation mechanisms . high energy electromagnetic radiation from pulsars is due to relativistic electrons and positrons which are accelerated in the magnetosphere surrounding the neutron star . the models for magnetospheric accelerators fall into two main classes depending upon their location with respect to the neutron star . in the polar cap model invoked originally to power radio emission from a pulsar ( ruderman and sutherland 1975 ; daugherty and harding 1982 ) the acceleration zone for the primary electrons ( positrons ) is relatively close to the neutron star surface . the outer magnetospheric accelerators ( holloway 1973 ; the outer gap " ) , proposed for energetic photons ( cheng , ho , & ruderman 1986a , b , hereafter chra and chrb ) , accelerate particles at a distance which is a good fraction of the light cylinder radius from the neutron star . as extensive e@xmath29 flows are generated by both polar cap and outer - gaps , the pairs from one would be expected to quench the other , and it is difficult to expect that both could survive on the same field lines , unless the field lines threading the two acceleration zones have differing curvatures and signs of net charge density etc . ( see wang et al 1998 for a discussion ) . currently , there are several different versions of the outer gap model in the literature which differ in significant detail as to the mechanisms of gap discharge through e@xmath29 pair production and the process of electromagnetic radiation emitted by these relativistic particles . the classic version of the chr model is now extended by zhang and cheng ( 1997 ) , cheng and ding ( 1994 ) as well as by wang et al ( 1998 ) . in addition there are variants of the classic outer gap model by romani(1996 ) and romani and yadigaroglu ( 1995 ) who consider only the outwardly flowing e@xmath29 pairs and compute curvature , synchrotron and inverse compton spectra from primary and secondary particles . in polar cap models , primary particles are accelerated near the surface of the neutron star and produce @xmath30-rays through curvature radiation and inverse - compton emission . these @xmath30-rays convert to electron - positron pairs in the strong magnetic field , initiating electromagnetic cascades as the pairs radiate synchrotron @xmath30-rays which produce further pair generations . the x - ray and @xmath30-ray spectrum from such a cascade initiated by curvature radiation was computed by daugherty & harding ( 1996 ) for the parameters of the vela pulsar . as already discussed , the rotation powered pulsars emit x - rays which are a combination of differing amounts of three spectral components : 1 ) power - law emission , resulting from non - thermal radiation of particles accelerated in the pulsar magnetosphere , 2 ) soft blackbody emission from surface cooling of the neutron star , and 3 ) a hard thermal component from heated polar caps . in addition , spectra may include a background of unpulsed emission from a synchrotron nebula . clearly , component 1 ) relates most directly to the pulsar particle acceleration models . however , components 2 ) and 3 ) can also be indirectly related to the magnetospherically accelerated particles under certain circumstances , especially if they are seen as a pulsed component . the nebular background is due to energetic particles injected in the surrounding nebula outside the light cylinder . while the study of its properties is interesting in its own right ( for example , to determine the injected particle spectrum and the strength of the nebular magnetic field - see harding and de jager 1997 ) , for the purposes of detecting pulsed emission , it is a background component . figure 3 shows the pulsed photon spectrum of psr b1706 - 44 from egret detections in the hard gamma - ray bands as well as numerous upper limits in the soft and hard x - ray bands and tev gamma - rays . since @xmath31 and @xmath32 for psr b1706 - 44 are very close to those of vela , the cascade spectrum of the two pulsars in the polar cap model is expected to be similar . we have plotted the vela cascade spectrum of the pc model ( dh96 and daugherty and harding , in preparation ) in figure 3 , extended down to 1 kev , and normalized to the psr b1706 - 44 egret data points . in the vela cascade model , the primary electrons are assumed to begin their acceleration above the surface at a height of 2.0 stellar radii . the acceleration zone extends to a height where the electrons have achieved an energy of 10 tev and the cascade begins near the top of the acceleration zone . acceleration starting at some height above the surface is likely , due to the screening of the parallel electric field near the surface by the cascades from returning positrons ( see section 4.2 and harding & muslimov 1998 ) . the cascade spectrum is primarily synchrotron radiation from pairs , and has a high energy turnover around 3 gev due to pair production attenuation in the local magnetic field . this turnover is quite sharp , because the pair production attenuation coefficient is an exponential function of photon energy , and thus no pulsed emission is expected at tev energies . the spectrum has a low - energy break or turnover at the local cyclotron energy in the particle rest frame , blue - shifted by the particle energy . in this case , the local cyclotron energy in the particles rest frame falls at 1.2 kev ( the local field is @xmath33 g ) and the pairs have an average lorentz factor of 50 , giving a turnover energy of about 50 kev . below the turnover , the spectrum flattens into a power - law formed by the broadening of the cyclotron fundamental by the pair energy distribution . this model spectrum falls just below the rxte upper limit , indicating that the position of the cyclotron energy turnover is near its minimum acceptable value . the rxte limit thus imposes a stringent upper limit on the emission radius in the polar cap model . for psr b1706 - 44 , the emission must occur within 2 stellar radii of the neutron star surface . in the chra , b model the pair production and radiation mechanisms in the outer gap for crab and vela type pulsars are different although it was shown subsequently that different angles of inclination between the spin and magnetic axes can alter the picture substantially ( chen and ruderman 1993 ) . in the chr model the synchrotron radiation from secondary e@xmath29 pairs generated in the outer gap ( by the collision of the primary photons and the tertiary ir photons ) is responsible for the photon spectrum in the 100 kev to 3 gev range . the pulsed photon spectrum for vela type pulsars ( with p @xmath34 s ) in the gamma - ray region is given by cheng and ding ( 1994 ) ( see also ruderman and cheng 1988 ) as : @xmath35 where e@xmath36 and e@xmath37 are given by : @xmath38 and e@xmath37 is given by : @xmath39 for psr b1706 - 44 . here sin @xmath40 is the mean pitch angle for the synchrotron emitting secondary particles and @xmath41 corresponds to the maximum energy of the secondary pairs , while @xmath42 corresponds to the secondary energy at which the radiative loss equals the magnetospheric crossing time and @xmath43 corresponds to the local cyclotron frequency . the above outer gap spectrum is also plotted as a dashed line in fig 2 and falls well below the rxte limit when normalized to the egret points . specifically , cheng and ding quote the e@xmath37 for vela and psr b1706 - 44 as ( 11mev , 1.6 mev ) and e@xmath36 as ( 7.5 gev , and 0.7 tev ) respectively . although the e@xmath37 of the two pulsars which have very similar periods and magnetic fields are within an order of magnitude , no clear explanation is given as to why the e@xmath36 of the two pulsars differ by nearly two orders of magnitude , especially with the same or similar pitch angle distributions of the secondary particles . physically , the upper spectral break ( @xmath44 ) in the chr models comes from the maximum energy of the outer gap @xmath30-rays which create the energetic @xmath45 pairs whose synchrotron radiation is in the @xmath30-ray / x - ray regime . the lower spectral break is related to the finite residence time ( ruderman and cheng 1988 ) of the synchrotron radiating pairs on the open field lines of an outer magnetosphere such that they leave the magnetosphere ( light cylinder ) radius before they have a chance to radiate away most of their energy . since the e@xmath29 leaving the light cylinder can not retain the memory of neutron star rotation , the radiation beyond the light cylinder will not be pulsed , hence the 3/2 power law pulsed synchrotron spectrum in this model is expected to be valid only above e@xmath46 which is typically in the mev energy range . ( chrb obtained the spectral break between 3/2 and 2/3 power around 100 kev , corresponding to a local magnetic field strength of 5 @xmath47 g ) . note that ruderman and cheng ( 1988 ) give the upper and lower bounds as the following functions of the period of vela - type pulsars : @xmath48 and @xmath49 respectively . presumably the higher value of e@xmath36 (= 0.7 tev ) was chosen by cheng and ding ( 1994 ) because psr b1706 - 44 does not appear to have a sharp high energy cutoff in the egret range . however , this choice of this high e@xmath36 also violates the upper limit of pulsed flux at somewhat lower energies around 250 gev due to chadwick et al ( 1997 ) . we note that , if the egret gamma - ray photon spectrum is to be explained by a 3/2 power law below 3 gev due to synchrotron radiation from the secondary e@xmath29 pairs in the outer magnetosphere , as in chr or cheng and ding , then the observed break in the power - law spectrum of psr b1706 - 44 in egret data around 3 gev might indicate an e@xmath36 close to that of the vela pulsar as reported in cheng and ding ( 7.5 gev ) . the break in the photon spectrum at 3 gev implies that ( given a canonical maximum energy of the primary photon of 10@xmath50 mev ) , the innermost region of the magnetosphere where the production of the synchro - curvature " photons from the secondaries are taking place ( see zhang and cheng 1997 ) is typically at a local magnetic field of several times @xmath51 gauss , i.e at about 100 stellar radii ( or 1/5 of the light cylinder distance ) for psr b1706 - 44 . for moderately inclined pulsars , this inner boundary of the outer gap would imply an inclination angle @xmath52 ( @xmath53 ) derived from the expression due to halpern and ruderman ( 1993 ) : @xmath54 , to be around 60 degrees . if the egret detected fluxes are indeed described by a 3/2 power law below 3 gev , then it appears that to accommodate the rxte upper limit obtained in this paper , one requires a break in the photon spectrum at energies higher than about 50 kev ( see fig 3 ) . as already mentioned , the pc model spectrum , if fitted to the egret pulsed fluxes , can not extend to the kev regions unless there is a break as mentioned here . the outer gap model s prediction due to cheng and ding ( 1994 ) ( which has a slope of 2/3 with a spectral break at 1 mev ) as shown by a dashed line in fig 3 , passes well below the rxte upper limit and is not constrained by it . in a modification of the outer gap model ( wang et al 1998 ) the inward directed @xmath30-ray curvature photons ( of energy @xmath5 100 mev , instead of the much higher energy photons of energy @xmath55 mev as in the chr version ) radiated by the inward directed primary particles produce e@xmath29 pairs in the stronger magnetic field ( 10@xmath56 g ) near the neutron star surface . the radius of the magnetic pair production ( at about 5 stellar radii ) is large compared to what is invoked in polar cap models . it is to be noted that the region of pair production and subsequent x - ray and low energy @xmath30-ray emission is much closer to the neutron star ( and is directed starward ) than the high energy emission in the earlier chr version of the outer gap model . wang et al give a photon spectral index of 3/2 with lower and upper cutoff energies at 0.1 kev and 5 mev respectively . we note that such a component is unlikely to be a continuation of the 3/2 power law photon spectrum from the egret energy bands , as such an extension would violate the rxte upper limit as noted above and the lower cutoff of 0.1 kev can be ruled out for an extrapolated egret spectrum . on the other hand , such a power - law component can be present with a lower strength , in addition to the same power law component ( due to outwardly directed particle flux ) as previously discussed by ruderman and cheng ( 1988 ) ( see also cheng and ding 1994 ) if the latter is taken to describe the egret spectrum below @xmath57 gev . since the former component is generated by a starward directed beam of gamma - rays , to be observable this x - ray component would come from the side of the magnetosphere opposite to the observer direction . the two components ( outwardly directed x - ray and @xmath30-ray beams and starward directed beams ) will be separated in pulsar spin phase , possibly with different strengths in the x - ray bands and with different spectra below @xmath5 1 mev . these could be seen in phase resolved spectra should such spectra be available in the future . in fig 4 we have plotted an expanded view of the x - ray spectrum expected from the non - thermal emission from the pc and og models , as discussed above , and the thermal radiation from the heated polar cap due to returned energetic particles from the acceleration regions . also shown are the upper limits to the pulsed fluxes from rosat pspc , rosat hri , asca , xte and osse ( see thompson et al 1996 for the references to these works ; the rosat hri limit is due to finley et al , 1998 ) . the rosat pspc upper limit corresponds to the 18% modulation of the total flux reported by becker et al ( 1995 ) ( with a reported spectral index of @xmath58 in the 0.1 - 2.4 kev band ) while the rosat hri upper limit corresponds to the upper limit of the pulsed flux as reported in finley et al ( 1998 ) in the same band ( we assume the same photon spectral index here as in becker et al . ( 1995 ) ) . the asca upper limits are due to the fluxes in the 0.7 - 2.0 kev and 2.0 - 10.0 kev bands and their photon index of the combined sis and gis data ( @xmath59 = 1.7 ) as reported in finley et al ( 1998 ) . the rxte upper limit in the 9.0 to 18.5 kev regime reported here is centered on a photon energy of 12.8 kev . the osse upper limit is due to schroeder et al ( 1995 ) . the polar cap and the outer gap models have different predictions of the returned e@xmath29 particle flux reaching the surface of the neutron star . if the particle fluxes indeed translate proportionally to the _ observable _ x - ray luminosities , then these models could potentially be distinguished if the associated hot thermal pulsed component in the x - rays from the heated polar caps are detected and are distinguishable from the the soft thermal component due to the cooling neutron star . in the polar cap model , an accurate determination of the fraction of backflowing positrons generated in the pair formation fronts requires a self - consistent calculation of the screening of the electric field component parallel to the local magnetic field due to the cascade pairs and its feedback on the pair spatial distribution . the generated positrons which are turned around slightly suppress the voltage all the way down to the bottom of the polar magnetic flux tube , so that the accelerating electric field ( @xmath60 ) component vanishes at a height above the stellar surface rather than at the actual surface . this is effectively an enhancement of the maximum number of electrons to be injected into the acceleration region . there is a maximum possible power in the backflowing positrons that will not shut off the parallel electric field . this will lead to the polar cap heating luminosity estimated by harding and muslimov ( 1998 ; hm98 ) as : @xmath61 where l@xmath62 is the spin down power of the pulsar and @xmath63 , and @xmath64 , i , m , r being the moment of inertia , mass and radius of the neutron star and @xmath65 . if the x - ray emission is beamed into a solid angle of 4@xmath66 steradian , the x - ray emission due to returning particle flux gives rise to a hot polar cap temperature of : @xmath67^{1/4}\ ] ] where r is the radius of the heated polar cap . taking r = @xmath68 , the standard polar cap radius for the period of psr b1706 - 44 , and @xmath69 , the backflowing positrons in the polar cap model are expected to produce a luminosity of @xmath70 beamed into a solid angle of 4@xmath66 steradian with a polar cap temperature of 5 @xmath71 k. in fig 4 , we have plotted the blackbody curve corresponding to @xmath72 for the heated polar cap ( hard ) thermal radiation - the maximum flux expected from psr b1706 - 44 on the basis of returned positron current as in hm98 . this curve lies significantly above the rosat hri and asca limits , implying a returning particle luminosity fraction much less than the maximum theoretically allowable . wang et al ( 1998 ) have argued that the x - ray emission properties of @xmath30-ray pulsars differ from those that are not observed to be strong @xmath30-ray emitters . for the gamma - ray pulsars , with accelerators presumably at very many stellar radii ( outer gap " ) above the neutron star surface , curvature radiation gamma - rays moving starward produce a dense blanket of e@xmath29 pairs on closed field lines around the star . the pair blanket affects the x - ray emission properties by a ) producing a nonthermal synchrotron spectrum with photon power law index of 3/2 between 0.1 kev @xmath73 5 mev , as discussed above , whose intensity ( typically @xmath5 10@xmath74 ) depends upon the backflow current and b ) forming a cyclotron resonance blanket around the star , which shields the direct observation of hard x - ray thermal emission from the heated polar cap except through two narrow holes along the open field lines threading the polar caps . less strongly modulated , softer x - rays can also escape by diffusing through the cyclotron resonant blanket . the ratio of the hard radiation flux from the heated polar cap coming through the holes ( f@xmath75 ) to the flux through the blanket itself is given by wang et al ( 1998 ) as : @xmath76 , where the expected optical depth @xmath77 is roughly 10 - 200 and the blanket area is roughly @xmath78 times greater than the hole size . therefore the radiation flux leaking through the holes may be 10@xmath79 to 0.2 times the more uniformly distributed softer flux escaping from the blanket . with constant bombardment of the relativistically inflowing particles at a rate of 10@xmath80 ( the goldreich - julian current ) which radiate away much of their energy before they reach the polar cap , wang et al estimated a total x - ray luminosity of @xmath81 , where the factor f is the fraction of goldreich - julian current @xmath82 that flows back through the outer gap accelerator to the polar cap ( taken to be 1/2 for @xmath30-ray pulsars not too far from their death lines ) . here , @xmath83^{-1/3 } \sim 5.5 \rm \ ; erg $ ] , is the residual energy of the charged particles impacting the polar cap . using this luminosity and the above range of the fractional flux ratio @xmath84 for the open field lines bundle for psr b1706 - 44 , the resultant hard x - ray ( thermal ) component , should the beam intercept the observer , would translate to a photon flux of @xmath85 -1 } \ ] ] @xmath86 here , the heated polar cap temperature for psr b1706 - 44 is taken as the same as that predicted for psr0656 + 14 , by wang et al ( 1998 ) , namely 2@xmath71 k. the range of the heated polar cap fluxes corresponding to the above range of @xmath87 is shown as the shaded area in fig 4 . note that the available observational upper limits are also constraining the extent of the heated polar cap emission in both pc and og models by limiting @xmath88 or @xmath87 . the upper range of @xmath87 is ruled out by the rosat hri limit requiring that all but about 1 % of the returned particle luminosity must be hidden at the hard x - ray and redistributed at lower bands . the rosat and asca limits restrict @xmath88 ( @xmath89 ) in the pc model to a value less than @xmath90 , the observed limit for the vela pulsar . zhang and cheng ( 1997 ) using the classic thin outer gap models of chr give an implied ir luminosity of the tertiary photons for geminga . using the same line of argument , we can estimate that psr b1706 - 44 should show a total ir band luminosity of @xmath91 @xmath92 should the tertiary photons indeed play a crucial role in sustaining the e@xmath29 discharges through ir-@xmath30-ray photons collision ( the latter in turn produced by inverse compton scattering of the same ir photons by highly energetic primary electrons ) . chakrabarty and kaspi ( 1998 ) have found an upper limit to the r - band magnitude for a region coincident with the pulsar of 18 mag , corresponding to a flux of 2.4 @xmath93 at 0.7 @xmath94 m , which translates to an r - band total luminosity of 9 @xmath95 at the pulsar . if the spectrum at the ir wavelengths is adequately described by a rayleigh jeans spectrum , then the corresponding flux limit from the r - band upper limit is well below the theoretical ir luminosity predicted from the thin outer gap model . inconsistencies between the ir luminosity required by outer gap models and observed ir limits ( iras ) for other pulsars has been pointed out by usov ( 1994 ) , zhang and cheng(1997 ) . the r - band upper limit corresponding to 18 mag , however is not sufficiently constraining compared to the scaling of the optical vs gamma - ray luminosities expected in outer gap models ( chr ) as given by usov ( 1994 , see his equation ( 24 ) ) . given a pulsed gamma - ray flux of 3@xmath96 jyhz at 10@xmath97 hz ( thompson et al 1997 ) , usov predicts an optical flux of @xmath98 for chr outer gap models . however , the more stringent ( possible ) upper limit of 21 mag quoted by chakrabarty and kaspi ( 1998 ) would violate the expected optical flux in the outer gap model . the main outcome of our work has been to provide an upper limit of the pulsed x - ray flux in the 9 - 18 kev band which constrains any non - thermal emission spectrum extending from egret gamma - ray bands towards the soft x - ray bands of rosat . in particular , the cyclotron turnover required for the polar cap model downwards of 50 kev , restricts the height of polar cap acceleration to less than two stellar radii above the ns surface . on the other hand , if a break in the egret spectrum around 3 gev is to be consistent with the spectra at lower energies , then the boundary of the outer gap acceleration zone nearest to the neutron star would be nearly 100 stellar radii away . in particular , our rxte limit would tend to rule out a particular component of the outer gap model emission ( starwardly directed beam of curvature gamma - rays materializing in relatively strong magnetic field to produce secondary e@xmath29 that synchrotron radiate to produce the power law x - ray gamma - ray spectra ) requiring a lower cutoff of 0.1 kev , unless it is a weaker component that is below the extrapolation from the egret spectrum to lower energy bands . future phase resolved spectra of this pulsar might therefore have information to reveal with regard to the nature and strength of these possibly different components tied to different mechanisms . we thank v. kaspi , m. bailes , n. wang and r.n . manchester for providing the radio ephemerides of psr b1706 - 44 measured from atnf . we thank d.j . thompson and arnold rots for discussions and for the compilation of data on this pulsar at other energy bands and the staff of xte guest observer facility for help with data analysis . thanks joe daugherty for assistance in producing the polar cap model cascade spectrum . acknowledges support through a senior research associateship of the national research council at nasa / goddard and the rxte guest observer program . arons , j. 1981 , apj 248 , 1099 . becker , w. , and trmper , j. 1997 , a & a. 326 , 682 . becker , w. , brazier , k.t.s . and trmper , j. 1995 , a & a. 298 , 528 . chadwick , p.m. et al 1997 proc . 4th compton symp . , williamsburg , ( ed c. dermer , m.s . strickman and j.d . kurfess ; aip proc 410 ) . chakrabarty , d. and kaspi , v. m. 1998 , lanl e - print astro - ph/9801017 v2 . chen , k .- y . and ruderman , m.a . 1993 , apj 408 , 179 . cheng , k.s . and ding , w. k .- y . , 1994 , apj 431 , 724 . cheng , k.s . , ho , c. and ruderman , m.a . 1986 , apj 300 , 500 ( chra ) . cheng , k.s . , ho , c. and ruderman , m.a . 1986 , apj 300 , 522 ( chrb ) . daugherty , j.k . and harding , a.k . 1982 , apj 252,337 . daugherty , j.k . and harding , a.k . 1996 , apj 458 , 278 . fierro , j. 1995 , ph.d . thesis , stanford university , palo alto . frail , d.a . , goss , w.m . , and whiteoak , j.b.z . 1994 , apj 437 , 781 . finley , j.p . , srinivasan , r. , saito , y. , hiriyama , m. , kamae , t. , and yoshida , k. , 1998 apj 493 , 884 . halpern , j. and ruderman , m.a . 1993 apj 415 , 286 . harding , a.k . , and muslimov , a. 1998 , ap j , submitted . harding , a.k . and de jager , o. 1997 , proc of kruger national park workshop on tev astronomy . ho , c. 1989 , apj 342 , 396 . holloway , n.j . 1973 , nature , 246,6 . johnston , s. , lyne , a.g . , manchester , r.n . , kniffen , d.a . , damico , n. , lim , j. and ashworth , m. 1992 , mnras , 225 , 401 . nel , h. et al 1996 apj , 465 , 898 . ray , a. , harding , a.k . and strickman , m.s . 1997 in proc of 4th compton symp , williamsburg , ( ed c. dermer et al , aip proc 410 ) . ray , a. and benford , g. 1981 , phys rev d , 23 , 2142 . romani , r.w . 1996 , apj 470 , 460 . romani , r.w . and yadigaroglu , i .- a . 1995 , apj 438 , 314 . ruderman , m.a . and cheng , k.s . 1988 , 335 , 305 . ruderman , m.a . and sutherland , p.g . 1975 , apj 196 , 51 . sturner , s.j . and dermer , c.d . 1994 , apj 420 , l79 . thompson , d.j . 1996 , in pulsars : problems & progress ( eds s. johnston et al ) astron soc pacific ser . 105 , 307 . thompson , d.j . et al 1997 , in proc of 4th compton symp . williamsburg , ( ed c. dermer et al , aip proc 410 ) . schroeder , p.c . , ulmer , m.p . , matz , s.m . , grabelsky , d.a . , purcell , w.r . , grove , j.e . , johnson , w.n . , kinzer , r.l . , kurfess , j.d . and strickman , m.s . 1995 , apj , 450 , 784 ulmer , m.p . , purcell , w.r . , wheaton , w.a . and mahoney , w.a . 1991 , apj 369 , 485 . usov , v.v . 1994 , apj 427 , 394 . wang , f.y .- h . , ruderman , m. , halpern , j.p . , and zhu , t. 1998 , columbia university preprint . zhang , l. and cheng , k.s . 1997 , apj 487 , 370 .

as the number of online resources and web documents continues to increase , the need for better organizational structures that guide readers towards the information they seek increases . hierarchies and taxonomies are invaluable tools for this purpose . taxonomies are widely used in libraries via the library of congress system or the dewey decimal system , and hierarchies were a fixture of the early world wide web ; perhaps the most famous example being the yahoo search engine , which originally was a taxonomic - collection of hyperlinks organized by topic . these systems were developed because their effectiveness at topical organization and their logarithmic depth allowed users to quickly find the relevant documents for which they were searching . unfortunately , taxonomy curation of documents , articles , books , etc . is mostly a manual process , which is only possible when the number of curated documents is relatively small . this process becomes increasingly impractical as the number of documents grows to web - scale , and has motivated research towards the automatic inference of taxonomies @xcite . most document repositories contain linkages between the documents creating a _ document - graph_. these links provide , among other things , proper context to the terms and topics in each document . document - graphs are especially common in nonfiction and scientific literature , where citations are viewed as inter - document links . similarly , the world wide web ( www ) can be considered to be a single , very large document - graph , where web pages represent documents and hyperlinks link documents . web sites on the www could also be considered document graphs because a web site is simply a subgraph of the www . web site subgraphs , in particular , are a collection of documents with a specific and purposeful organizational structure , which are often carefully designed to guide the user from the entry page , _ i.e. _ , a homepage , to progressively more specific web pages . similarly , scientific literature can be categorized into a hierarchy of increasingly specific scientific topics by their citation links , and encyclopedia articles can be categorized into a hierarchy of increasingly specific articles by their cross references . thus , we assert that most document - graphs , or , more generally , information networks , contain hidden node hierarchies . in this paper we draw specific distinctions between a _ hierarchy _ and a _ taxonomy_. a taxonomy is defined to be a classification of objects into increasingly finer granularities , where each non - leaf node is a conceptual combination of its children . a biological taxonomy is a canonical example of this definition because a classified species , say ( _ i.e. _ , humans ) , can only be placed at a leaf in the taxonomy ; the inner nodes , _ e.g. _ , , , , do not declare new species , rather they are conceptual agglomerations of species . furthermore , each species is described by its path through the taxonomy . for example , s , can be more generally described as s , s and as s ( among others ) . a hierarchy , on the other hand , is an arrangement of objects where some objects are considered to be _ above _ , _ below _ or _ at the same level as _ others . this necessarily means that objects of a hierarchy live at the internal nodes . for example , a business , government or military chain of command is a hierarchy because the is a specific object that is above the s , which are above the s , and so on . in this case , high - level nodes like are not agglomerations of s just as the is not an aggregation of s or s. in the similar studies , `` document hierarchies '' were not actually hierarchies of documents in the literal sense . for example , hierarchical lda ( hlda ) @xcite , topicblock @xcite , and the tree structured stick breaking ( tssb ) model @xcite learn a conceptual taxonomy in which the non - leaf topics are a combination of words and does not represent a real document in the corpus ; the hierarchical pachinko allocation model ( hpam ) @xcite constructs a tree - like conceptual taxonomy like hlda , but where each topic can have multiple parents . in these related models , only the leaves contained the actual , literal documents . contrary to our perspective , the internal nodes of existing models contain ephemeral word - topic distributions , rather than actual documents . see figure [ fig : inh_relatedwork ] for a brief comparison of model outputs . the hdtm model introduced in this paper requires that inner nodes , which in previous work are made of ephemeral distributions , be literal documents . this requirement asserts that _ some documents are more general than others_. here we explore this assertion through examples and a review of similar assertions made in previous research . a web site @xmath0 can be viewed as a directed graph with web pages as vertices @xmath1 and hyperlinks as directed edges @xmath2 between web pages @xmath3 excluding inter - site hyperlinks . in most cases , designating the web site entry page as the root @xmath4 allows for a web site to be viewed as a rooted directed graph . web site creators and curators purposefully organize the hyperlinks between documents in a topically meaningful manner . as a result , web documents further away from the root document typically contain more specific topics than web documents graphically close to the root document . for example , the web site at the university of notre dame , shown in figure [ fig : inh_webtree ] contains a root web document ( the entry page ) , and dozens of children web documents . even with a very small subset of documents and edges , the corresponding web graph can be quite complicated and messy . a breadth first traversal of the web graph starting with the root node is a simple way to distill a document hierarchy from the web graph . unfortunately , a fixed breadth - first hierarchy can not account for many of the intricacies of real world web graphs . for explanation purposes , let s assume there are four types of hyperlink edges in a web site : ( 1 ) parent - to - child links , ( 2 ) upward links , ( 3 ) shortcuts , and ( 4 ) cross - topic links . parent - to - child links direct the user from one web page to a more topically specific web page ; _ e.g. _ , a hyperlink from engineering.nd.edu to cse.nd.edu is a parent - to - child hyperlink because computer science is topically more specific than engineering . upward links are hyperlinks that reference a more general document ; _ e.g. _ , there may exist a hyperlink from engineering.nd.edu to /the - arts because the engineering college would like to reference artistic happenings at the university . shortcut links are hyperlinks that skip from very general web documents to very specific web documents as a way of featuring some specific topic ; _ e.g. _ , if a computer science professor wins a prestigious award or grant , the professor s web page may be linked to from the news section of the root web page . cross topic links are hyperlinks that move across topical subtrees ; _ e.g. _ , the college of science may reference some working relationship with the college of engineering by creating a hyperlink between the two web pages . because the goal is to infer the document hierarchy , the hdtm model is , in a sense , trying to find parent - to - child links . in the event that there is more than one parent - to - child link to a particular web page , the goal is to find the best topically - relevant parent for each web document in the inferred hierarchy ( except the root ) . web researchers and practitioners have used hyperlink structures to organize web documents for many years . the pagerank and hits algorithms are two of the most famous examples of information propagation through links . pagerank , for instance , uses the model of a random web surfer ( _ i.e. _ random walker ) , who randomly follows hyperlinks over the web ; a current measure of a web page s authority corresponds to the probability that a random surfer lands upon that web page the pagerank score . in the hdtm model , pagerank s notion of authority loosely corresponds to topical generality , that is , web pages with a high random surfer probability are likely to be topically more general than others . the document - graph structure is also used to enrich the document description by adding features to improve retrieval performance . some of the intuition behind these previous works are helpful in framing the generative model . a limitation of the random walker model is that it only looks at the graphical structure of the network . the word distributions found in each document are clearly an important factor to consider when generating document hierarchies . previous work by song , _ et al . _ @xcite and qin , _ et al . _ @xcite show that a given web page can be enriched by propagating information from its children . their relevance propagation model modifies the language distribution of a web page to be a mixture of itself and its children according to the formula : @xmath5 where @xmath6 is the frequency of term @xmath7 in a document @xmath8 before propagation , @xmath9 is the frequency of term @xmath7 in document @xmath8 after propagation , @xmath10 is a child page of @xmath8 in the sitemap @xmath11 , and @xmath12 is a parameter to control the mixing factor of the children . this propagation algorithm assumes that the sitemap , @xmath11 , is constructed ahead of time . for the purposes of web information retrieval language models are often used to normalize and smooth word distributions . for illustration purposes , we apply a dirichlet prior smoothing function @xcite to smooth the term distribution where the @xmath9 from above is used in place of the usual @xmath13 from the original dirichlet prior smoothing function yielding : @xmath14 where @xmath15 is the distribution over all terms in @xmath1 , @xmath16 is the smoothing parameter , and the length is modified by the propagation algorithm to be @xmath17 . as a result of the upward propagation @xmath18 function , the root document ( web site entry page ) will contain all of the words from all of the web pages in the web site with different non - zero probabilities . the most probable words are those that occur most frequently and most generally across all documents , and are thus propagated the most . as a small , preliminary example , table [ tab : inh_lm ] shows the top six most probable words in the top document ( via text propagation ) and in root topics of hlda and hdtm of the computer science department s web site at the university of notre dame . this small example reinforces the intuition that certain web sites have a hidden hierarchical topical structure . in the previous term propagation work , web site sitemaps were constructed ahead of time using url heuristics or manually . the goal of hdtm is to learn the document hierarchy automatically and in conjunction with the topical hierarchy . documents from many different collections exist in hidden hierarchies . while technically a web site , wikipedia documents and categories form a unique document graph . wikipedia categories are especially interesting because they provide a type of ontology wherein categories have more specific sub - categories and more general parent - categories . most wikipedia articles are are represented by at least one category description ; this allows for users to drill down to relevant articles in a very few number of clicks by browsing the category graph . a partial example of the wikipedia category graph is shown in figure [ fig : inh_wikicat ] . this figure illustrates how a document graph can be construed into a document hierarchy . specifically , the wikipedia community has hand - crafted a category hierarchy , represented by colored circles and edges , on top of the article - graph , represented by grey squares and grey edges . although the category graph is imperfect and incomplete , wikipedia browsers can get a sense of the granularity , topic and fit simply by viewing the article s placement in the category hierarchy . with the wikipedia construction in mind , our goal can be loosely interpreted as automatically inferring the wikipedia category hierarchy from the wikipedia article graph ; this inference is tried and evaluated in section [ sec : inh_experiments ] . bibliographical networks may also be hierarchically structured . in a bibliographic network , papers or authors ( wherein each author could be a collection of documents ) are represented by nodes and each citation is represented by an edge in the graph . apart from web and citation graphs , bioinformatics networks , for example protein networks , can also be hierarchically organized for protein fold prediction @xcite . the nodes in such network are proteins and two proteins are connected if they have structural or evolutionary relationships @xcite . the goal of this work is to construct node hierarchies from an information network using the node features ( _ e.g. _ , text ) and inter - node edges . for these purposes a node which appears _ above _ , _ below _ or _ at the same level as _ another node refers to the conceptual granularity of the nodes . in other words , given an information network with an explicitly identified root , such as a web site homepage , we aim to learn a node - hierarchy that best captures the conceptual hierarchy of the document - graph . this problem poses three technical challenges : 1 . * document - topic inference*. in document hierarchies the parent documents consist of topics that are more general than their children . this requires that the parent documents are viewed as a mixture of the topics contained within its children , and children documents should topically fit underneath their selected parent . in this paper we introduce the _ hierarchical document - topic model _ ( hdtm ) which generates a course - to - fine representation of the text information , wherein high - level documents live near the top of the hierarchy , and low - level , more specific documents live at or near the leaves . 2 . * selecting document placement*. placement of a document within the hierarchy drives the topic mixing during inference . because links between edges hint at the context of the relationship between documents , the document placement in the inferred hierarchy is constrained by their edges within the original document - graph . in other words , if an edge exists in the final , inferred hierarchy , then it must also exist in original document - graph ( not vice versa ) . unlike existing models , such as hlda @xcite , that select topic paths using the nested chinese restaurant process ( ncrp ) , hdtm performs document placement based on a stochastic process resembling random walks with restart ( rwr ) over the original document - graph . the use of a stochastic process over the document - graph frees the algorithm from rigid parameters ; perhaps most importantly the adoption of the rwr stochastic process instead of ncrp allows documents to live at non - leaf nodes , and frees the algorithm from the depth parameter of hlda . * analysis at web site - scale*. in many document - graph collections , the number of edges grows quadratically with the number of nodes . this limits the scalability of many topic diffusion algorithms @xcite . an important side - effect of the rwr process is the ability to adapt the hdtm model inference algorithm into a distributed graph processing system that is capable of processing billion - node graphs . the remainder of this paper is organized as follows : after reviewing the related work we introduce the hdtm model and show how inference is performed . next we show how the inference can be adapted to a large scale graph processing system in order to run on web - scale data sets . the experiments section describes a myriad of tests that were run on various data sets and the quantitative and qualitative evaluations that were performed . we conclude with a examples of real - world use cases , discuss avenues for future research , and provide a link to hdtm s source code . initial efforts in hierarchical clustering used greedy heuristics such as single - link or complete - link agglomoration rules to infer dendrograms @xcite , in which the root node is split into a series of branches that terminate with a single document at each leaf . ho , _ et al . _ , point out that manually - curated web hierarchies like the open directory project are typically flatter and contain fewer inner nodes than agglomerative clustering techniques produce @xcite . other hierarchical clustering algorithms include top - down processes which iteratively partition the data @xcite , incremental methods like cobweb @xcite , classit @xcite , and other algorithms optimized for hierarchical text clustering . the processes that typically defines most hierarchical clustering algorithms can be made to fit in a probabilistic setting that build bottom - up hierarchies based on bayesian hypothesis testing @xcite . on the other hand , a lot of recent work uses bayesian generative models to find the most likely explanation of observed text and links . the first of these hierarchical generative models was hierarchical latent dirichlet allocation ( hlda ) . in hlda each document sits at a leaf in a tree of fixed depth @xmath19 as illustrated in figure [ fig : rwhlda ] . note that all non - leave nodes in figure [ fig : rwhlda ] are conceptual topics containing word distribution instead of a document . each document is represented by a mixture of multinomials along the path through the taxonomy from the document to the root . documents are placed at their respective leaf nodes stochasically using the nested chinese restaurant process ( ncrp ) along side an lda - style word sampling process . ncrp is a recursive version of the standard chinese restaurant process ( crp ) , which progresses according to the following analogy : an empty chinese restaurant has an infinite number of tables , and each table has an infinite number of chairs . when the first customer arrives he sits in the first chair at the first table with probability of 1 . the second customer can then chose to sit at an occupied table with probability of @xmath20 or sit at a new , unoccupied table with probability of @xmath21 , where @xmath22 is the current customer , @xmath23 is the number of customers currently sitting at table @xmath24 , and @xmath25 is a parameter that defines the affinity to sit at a previously occupied table . the nested version of the crp extends the original analogy as follows : at each table in the chinese restaurant are cards with the name of another chinese restaurant . when a customer sits at a given table , he reads the card , gets up and goes to that restaurant , where he is reseated according to the crp . each customer visits @xmath19 restaurants until he is finally seated and is able to eat . this process creates a tree with a depth of @xmath19 and a width determined by the @xmath25 parameter . this process has also been called the chinese restaurant franchise because of this analogy @xcite . adams , _ et al . _ proposed a hierarchical topic model called tree structured stick breaking ( tssb ) , illustrated in figure [ fig : rwtssb ] , wherein documents can live at internal nodes , rather than exclusively at leaf nodes@xcite . however , this process involves chaining together conjugate priors which makes inference more complicated , and it also does not make use of link data . other work along this line include hierarchical labeled lda ( hllda ) by petinot _ et al . _ @xcite hllda , as well as fixed structure lda ( fslda ) by reisinger and pasca @xcite which modify hlda by fixing the hierarchical structure and learning hierarchical topic distributions . the hierarchical pachinko allocation model(hpam ) , shown in figure [ fig : rwhpam ] , produces a directed acyclic graph ( dag ) of a fixed depth allowing for each internal ( non - document ) node to be represented a mixture of more abstract , _ i.e. _ , higher level , topics @xcite . in network - only data , community discovery is the process of finding self - similar group , or clusters . the shrink algorithm creates hierarchical clusters by identifying tightly - knit communities and by finding disparate clusters by looking for hubs and other heuristics @xcite . clauset , _ et al _ , discover dendrograms by monte carlo sampling @xcite ; however , dendrograms poorly represent the manually curated hierarchies and taxonomies that we are pursuing . stochastic block models ( sbm ) are an alternative line of network clustering research that partitions nodes into communities in order to generatively infer link probabilities @xcite . several extensions to the original sbm have since been proposed ( for a survey see @xcite ) . one downside to block - model processes is that they assign probabilities to every possible edge requiring @xmath26 complexity in every sampling iteration . furthermore , sbm methods typically are not concerned with topical / conceptual properties of the nodes . because hdtm merges document text and inter - document links into a single model , we assume that the words and their latent topics fit within the link structure of the graph , and that the graph structure explains topical relationships between interlinked documents . topic modeling with network structure ( tmn ) is similar in this regard because it regularizes a statistical topic model with a harmonic regularizer based on the graph structure in the data ; the result is that topic proportions of linked documents are similar to each other @xcite . however , hierarchical information is not discovered nor can be easily inferred from this model . topic - sensitive pagerank combines document topics with the pagerank algorithm , by arguing that the pagerank score of a document ought to be influenced by its topical connection to the referring document @xcite . like tmn model , topic - senstive pagerank does not construct any type of information network hierarchy . other work on generative models that combine text and links include : a probabilistic model for document connectivity @xcite , the link - plsa - lda and pairwise - link - lda methods @xcite , the latent topic model for hypertext ( lthm ) method @xcite , role discovery in social networks @xcite , the author - topic - model @xcite , and others . the above models operate by encoding link probability as a discrete random variable or a bernoulli trial that is parameterized by the topics of the documents . the relational topic model ( rtm ) builds links between topics , where observed links are given a very high likelihood @xcite . the topicblock model combines the non - parametric hlda and stochastic block models @xcite to generate document taxonomies from text and links @xcite ; however , topicblock , like hlda , does not permit documents to reside at non - leaf nodes of the resulting tree . to apply topic modeling algorithms on web - scale data , several parallel algorithms have been introduced . newman , _ et al _ proposed an exact distributed gibbs sampling algorithm as well as an approximate distributed gibbs sampling algorithm that uses local gibbs sampling and global synchronization @xcite . smyth , _ et al _ introduced an asynchronous distributed algorithm that was capable of learning lda - style topics @xcite ; and ahmed , _ et al _ recently released yahoo lda , which is a scalable approximate inference framework on large - scale streaming data @xcite . these parallel approaches use the conventional , procedural programming paradigms , and as a result can not guarantee statistically sound samples from their gibbs iterations . although conventional parallel and distributed algorithms indeed divide the document set into smaller groups , the maximum number of subgroups is subject to the number of processors . in contrast , the vertex - programming paradigm , _ e.g. _ , pregel , graphlab and graphx @xcite , can distribute sampling operations at a much finer granularity by treating each graph - node as an independent computing unit . although the distributed variants of topic inference have made significant contributions in large scale topic models on the one hand , and large scale graph processing on the other hand , we are unaware of any parallel algorithm capable of joining these two subjects to infer topical hierarchies on large scale information networks . in contrast to the previous work , hdtm builds a hierarchy of documents from text and inter - document links . in this model , each node in the hierarchy contains a single document , and the hierarchy s width and depth is not fixed . the distributed version of proposed algorithm has the ability to handle graph with millions of nodes and billions of tokens . the problem of inferring the document hierarchy is a learning problem akin to finding the single , best parent for each document - node . unlike previous algorithms , which discover latent topic taxonomies , the hierarchical document - topic model ( hdtm ) finds hidden hierarchies by selecting edges in the document graph . this section presents a detailed description of the model . beginning with a document graph @xmath27 of documents @xmath1 and edges @xmath2 . each document is a collection of words , where a word @xmath7 is an item in a vocabulary . the basic assumption of hdtm and similar models is that each document can be generated by probabilistically mixing words from among topics . distributions over topics are represented by @xmath28 , which is a multinomial variable with an associated set of distributions over words @xmath29 , where @xmath30 is a dirichlet hyper - parameter . document - specific mixing proportions are denoted by the vector @xmath31 . parametric - bayes topic models also include a @xmath32 parameter that denotes the number of topics , wherein @xmath28 is one of @xmath32 possible values and @xmath31 is a @xmath32-d vector . hdtm , and other non - parametric bayesian models , do not require a @xmath32 parameter as input . instead , in hdtm there exist @xmath33 topics , one for each graph node , and each document is a mixture of the topics on the path between itself and the root document . in the original lda model , a single document mixture distribution is @xmath34 . the process for generating a document is ( 1 ) choose a @xmath31 of topic proportions from a distribution @xmath35 , where @xmath35 is a dirichlet distribution ; ( 2 ) sample words from the mixture distribution @xmath36 for the @xmath31 chosen in step 1 . hlda is an extension of lda in which the topics are situated in a taxonomy @xmath37 of fixed depth @xmath19 . the hierarchy is generated by the nested chinese restaurant process ( ncrp ) which represents @xmath31 as an @xmath19-dimensional vector , defining an @xmath19-level path through @xmath37 from root to document . because of the ncrp process , every document lives at a leaf and the words in each document are a mixture of the topic - words on the path from it to the root . the ncrp stochastic process could not be used to infer document hierarchies because the ncrp process forces documents to the leaves in the tree . hdtm replaces ncrp with random walk with restart ( rwr ) ( which is also known as personalized pagerank ( ppr ) ) @xcite . in contrast , random walk with teleportation ( aka pagerank ) random walks by selecting a random starting point , and , with probability @xmath38 , the walker randomly walks to a new , connected location or chooses to jump to a random location with probability @xmath25 , where @xmath25 is called the jumping probability , however , this would be ambiguous with the dirichlet hyper - parameter @xmath12 . ] . in hdtm , the root node is fixed , either as the entry page of a web site , by some other heuristic or manually . therefore , for the purposes of hierarchy inference , the random walker is forced to start and restart at the root node . [ alg : rwr ] ( ) @xmath39 ( ) @xmath40 @xmath41 say we wish to find the rwr - probability between some node @xmath42 and some target node @xmath43 . we model this by a random walker visiting document @xmath42 at time @xmath44 . in the next time step , the walker chooses a document @xmath45 from among @xmath42 s outgoing neighbors @xmath46 in the hierarchy @xmath37 uniformly at random . in other words , at time @xmath47 , the walker lands at node @xmath48 with probability @xmath49 , where @xmath50 is the outdegree of some document @xmath51 . if at any time , there exists an edge to @xmath52 , _ i.e _ , an edge between the current node @xmath42 and the target node @xmath43 in the original graph @xmath0 , then we record the probability of that new path possibility for later sampling . alg . [ alg : rwr ] describes this process algorithmically . this procedure allows for new paths from the root @xmath53 to be probabilistically generated based on the current hierarchy effectively allowing for documents to migrate up , down and through the hierarchy during sampling . because a document hierarchy is a tree , each document - node can only have one parent . selecting a path for a document @xmath8 in the graph @xmath0 is akin to selecting a parent @xmath54 ( and grandparents , etc . ) from @xmath55 in the document graph @xmath0 . hdtm creates and samples from a probability distribution over each documents parent , where the probability of document @xmath42 being the parent of @xmath8 is defined as : @xmath56 where @xmath57 is the walkers current position at time @xmath44 , @xmath58 is the depth of @xmath8 in @xmath37 , and @xmath59 is the outdegree of @xmath57 in the hierarchy @xmath37 . in other words , the probability of landing at @xmath8 is the product of the emission probabilities from each document in the path through @xmath37 from @xmath4 to @xmath8 . this random walker function assigns higher probabilities to parents that are at a shallower depth than those at deeper positions . this is in line with the intuition that flatter hierarchies are easier for human understanding than deep hierarchies @xcite . simply put , the restart probability @xmath25 controls how much resistance there is to placing a document at successive depths . algorithmically , hdtm infers document hierarchies by drawing paths @xmath60 from the @xmath4 to the document @xmath8 . thus , the documents are drawn from the following generative process : 1 . each document @xmath61 is assigned a topic @xmath62 dir(@xmath63 ) : 2 . for each document @xmath61 : 1 . draw a path @xmath64 rwr(@xmath25 ) 2 . draw an @xmath19-dim topic proportion vector @xmath31 from dir(@xmath12 ) , where @xmath65len@xmath66 . 3 . for each word @xmath67 : 1 . choose topic @xmath68 mult(@xmath69 ) . choose word @xmath70 mult(@xmath71 ) , where @xmath72 is the topic in the @xmath28th position in @xmath60 . in this generative process hierarchical nodes represent documents _ and _ topics , where internal nodes contain the shared terminology of its descendants . the illustration in figure [ fig : topicdocuments ] shows the two potential outputs from a three - node graph of the wikipedia articles : , and . clearly , data mining and machine learning could both be regarded as children of computer science , but there is indeed a wiki - link from to in the graph . there are two key ideas conveyed in this illustration . the first is that the hierarchy on the left picks edges so as to ignore the to wiki - link ( denoted as a thin - grey line ) , whereas the hierarchy on the right picks edges so as to ignore the to wiki - link . thus , in this toy example , there are two possible hierarchy outcomes presented : the one on the left and the one on the right in figure [ fig : topicdocuments ] . the second idea conveyed in this illustration is that the word - topic distributions ( in colored circles ) are stored along with documents words ( in black squares ) at the same position in the hierarchy . the word distribution within each topic is constrained such that a document may only propagate terms upwards through the currently selected / sampled hierarchical topology . thus , it is impossible , in the hierarchy on the left , for the topic distribution at to contain a word unique to or . it is similarly impossible , in the hierarchy on the right , for the topic distribution in to contain a word unique to even though this is indeed possible in the hierarchical topology on the left . like in earlier models , there is statistical pressure to have more general terms in topics towards the root of the hierarchy . this is because every path in the hierarchy includes the root node and there are more paths through nodes at higher levels than through nodes at lower levels . moving down the tree , the topics , and therefore the documents , become more specific . hyperparameters also play an important role in the shape and character of the hierarchy . the @xmath12 parameter affects the smoothing on topic distributions , and the @xmath63 parameter affects the smoothing on word distributions . the @xmath25 parameter is perhaps the most important parameter because it affects the depth of the hierarchy . specifically , if @xmath25 is set to be large ( _ e.g. _ , @xmath73 ) then resulting hierarchy is shallow . low values ( _ e.g. _ , @xmath74 ) may result in deep hierarchies , because there is a smaller probabilistic penalty for each step that the random walker takes . exact inference on this model is intractable , so an approximation technique for posterior inference is used . the gibbs sampling algorithm is ideal in this situation because it simultaneously allows exploration of topic distributions and potential graphical hierarchies . the variables needed by the gibbs sampler are : @xmath75 , the @xmath22th word in document @xmath8 ; @xmath76 , the assignment of the @xmath22th word in document @xmath8 ; and @xmath77 , the topic corresponding to document at the @xmath28th level . the @xmath31 and @xmath30 variables are integrated out forming a collapsed gibbs sampler . the sampling is performed in two parts : ( 1 ) given the current level allocations of each word @xmath76 sample the path @xmath77 , ( 2 ) given the current state of the hierarchy , sample @xmath76 . in other words , we use the topic distributions to inform the path selections that make up the hierarchy , and the hierarchy topology to inform the topic distributions . the first gibbs sampling step is to draw a path from each document to the root through the graph . the sampling distribution for a path @xmath78 is @xmath79 where @xmath80 is the count of terms in document @xmath8 , and @xmath81 are the words without document @xmath8 . this equation is an expression of bayes theorem where the first term represents the probability of data given some choice of path from the root , and the second term represents the probability of selecting some path . specifically , the second term represents the probability of drawing the path @xmath82 to document @xmath8 at depth @xmath43 from the rwr process . recall that each node has an emission probability of @xmath83 , and a restart probability of @xmath25 . the probability is defined recursively : @xmath84 in other words , the probability of reaching @xmath8 is equal to the probability of a random walker with restart probability @xmath25 being at document @xmath8 at time @xmath43 . the first term in eq . [ eq : pathsample ] is the probability of a given word based on the current path @xmath85 and topic assignment @xmath86 : @xmath87 + w\eta } ) } { \prod_w { \gamma ( \#[\mathbf{c}_{-d , k } = c_{d , k } , \mathbf{w}_{-d } = w ] + \eta } ) } } \times \\ & \quad\hspace{1.1 cm } \frac { \prod_w{\gamma ( \#[\mathbf{z } = k , \mathbf{c}_{k } = c_{d , k } , \mathbf{w } = w ] + \eta ) } } { \gamma ( \sum { \#[\mathbf{z } = k , \mathbf{c}_{k } = c_{d , k } , \mathbf{w } = w ] } + w\eta ) } , \label{eq : pwgivenc } \end{split}\ ] ] where @xmath88 $ ] counts the elements of an array that satisfy the given condition , and @xmath89 is the maximum depth of the current hierarchy state . the expression @xmath90 $ ] counts : ( ii ) @xmath91 , _ i.e. _ , the number of words @xmath7 that do not appear in @xmath8 , for each ( i ) @xmath92 , _ i.e. _ , the number of paths to the current document @xmath8 except those where the path length is @xmath43 . the expression @xmath93 counts : ( iii ) @xmath94 , _ i.e. _ , the number of words @xmath7 such that , ( ii ) @xmath95 , _ i.e. _ , the words appear in document @xmath8 and are situated at the end of a path of length @xmath43 , where ( i ) @xmath96 , _ i.e. _ , @xmath43 is one of the topics in @xmath86 . @xmath97 is the size of the vocabulary . [ eq : pwgivenc ] is adapted from the standard ratio of normalizing constants for the dirichlet distribution @xcite . given the current state of all the variables , the word sampler must first pick an assignment @xmath28 for word @xmath22 in document @xmath8 . the sampling distribution of @xmath76 is @xmath98 where @xmath99 and @xmath100 . the first term is a distribution over word assignments : @xmath101 + \eta \end{split}\ ] ] which is the @xmath63-smoothed frequency of seeing word @xmath75 in the topic at level @xmath76 in the path @xmath78 . the second term is the distribution over levels @xmath102}{\#[\mathbf{z}_{d ,- n } \ge j ] } } \right ) \times \\ & \frac{1-\gamma}{\mathrm{deg}_t(d_{k-1 } ) } \frac{\#[\mathbf{z}_{d ,- n } = k]}{\#[\mathbf{z}_{d ,- n } \ge k ] } , \end{split } \label{eq : abuse}\ ] ] where @xmath88 $ ] is the number of elements in the vector which satisfy the given condition . [ eq : abuse ] abuses notation so that the product from @xmath103 to @xmath104 combines terms representing nodes at the @xmath105th level in the path @xmath85 down to the parent of @xmath106 , and the second set of terms represents document @xmath106 at level @xmath43 . the @xmath107 symbol in eq . [ eq : abuse ] refers to terms representing all ancestors of a particular node , and @xmath108 refers to the ancestors of a node including itself . a common complaint among data science practitioners is that graphical models , especially , non - parametric bayesian graphical models , do not perform well at scale . with this in mind , we also implemented the hdtm inference algorithm in the scalable , distributed vertex - programming paradigm . the mechanism behind gibbs sampling , and other markov chain monte carlo methods , requires sequential sampling steps , and execution of each step depends on the results of the previous step making gibbs samplers , and mcmc method in general , difficult to parallelize . approximate distributed lda ( ad - lda ) is one attempt to find approximate , distributed solutions to the serial inference problem by dividing documents into @xmath109 parts where @xmath109 is the number of processors and initializes the topic distribution @xmath28 globally . then , for every gibbs iteration , each processor samples @xmath110 of the dataset using the @xmath111 from last gibbs sampling iteration . when all processors are finished , a global synchronization is performed and @xmath28 is updated @xcite . following the distribution lessons from ad - lda we sought to also implement a scalable , distributed version of hdtm . however , a major difference between the lda / hlda and hdtm is that hlda uses the ncrp stochastic process to assign terms to topics , while hdtm samples paths from a graph of documents using graph - based random walk with restart method . the process of random walks over the network topology combined with the term sampling process described above is a good candidate for the vertex - programming paradigm using frameworks like pregel @xcite or graphlab @xcite . although mapreduce is a widely used , general purpose parallel scheme that can easily deal with scalable data , it is not optimized for _ iterative _ computational tasks such as statistical inference or logistic regression @xcite . this is because mapreduce materializes all intermediate results and to disk in order to tolerate task failures . mapreduce , therefore , has relatively high i / o costs compared to other designs that keep data in memory across iterations @xcite . apart from mapreduce , another scalable solution is to build a custom distributed system using message passing interface ( mpi ) . custom approaches are usually closer to optimal because developers can tune the code based on their own needs and minimize the unnecessary overhead . however , the drawbacks are also significant : because mpi is a barebone communication specification developers need to write their own code for job dispatching , load balancing , and dealing with node failure @xcite . after evaluating aforementioned approaches , we decided to use an emerging computational paradigm called vertex programming . vertex programming aims to improve the performance of graph / network computing by automatically distributing in - memory computation and with vertex - centric scheduling . unlike the mapreduce paradigm , which writes every intermediate result to disk , vertex programming keeps data up - to - date , in - memory and reduces i / o overhead by only materializing data to disk through periodic checkpoints . vertex - centric scheduling views every graph - vertex as an elementary computing unit and uses mpi ( or some other message passing system ) to transfer data over graph - edges . typically , vertex - programs are rather easy to implement , can be distributed easily , and are much more computationally efficient than conventional , procedural programming when working with iterative computational tasks @xcite . several vertex programming frameworks , including spark - graphx , hama , graphlab , graphlab create and graphlab 2.0 ( powergraph ) , were evaluated , and the powergraph framework was ultimately chosen . the decision was based , in part , from insurmountable troubles experienced during several months of implementation attempts on spark s graphx framework . ultimately , we concluded that c++ and mpi based powergraph was faster , more scalable and had a much smaller memory footprint than other frameworks . [ alg : dist_rwr ] ( ) vertex @xmath112 [ alg : dist_upate ] ( ) vertex @xmath112 the distributed hdtm inference algorithm is similar to procedural hdtm . we do not detail the entire distributed hdtm inference algorithm in this paper ; however , the source code is referenced in the section [ sec : conclusions ] . to fit hdtm to the vertex - programming model , changes in sampling sequence and attention to global synchronization were required . firstly , random walk with restart must be executed during each gibbs iteration so that every visited node can have a random walk probability . next , every node gathers @xmath81 , @xmath86 , and @xmath85 separately and decides their new path @xmath85 according to eq [ eq : pathsample ] and shown in alg [ alg : dist_rwr ] . after a path is sampled , each node will pick sample assignments @xmath28 for each word @xmath22 across all documents / nodes @xmath8 in parallel according to eq [ eq : wordsample ] . a global synchronization step is required so that each document / node can update the number of words @xmath22 and the topic assignments @xmath28 globally ; fortunately , because the topics and words are sampled according to the information provided in the _ path _ from root @xmath4 to the each document @xmath106 ( document @xmath8 at level @xmath43 ) , it suffices to update the nodes on the path from @xmath4 to @xmath106 instead of an actual global update to all nodes in @xmath1 . this vertex - programming based path - global update function is shown in alg [ alg : dist_upate ] . furthermore , this update is executed during the _ synchronization barrier _ , which is built - in to most vertex - programming frameworks , is highly optimized , and does not lock the global system any more than the synchronization barrier already does . this section describes the method and results for evaluating the hdtm model . a quantitative and qualitative analysis of the hierarchical document - topic model s ability to learn accurate and interpretable hierarchies of document graphs is shown . the main evaluations explore the empirical likelihood of the data and a very large case study wherein human judges are asked to evaluate the constructed hierarchies . hdtm is evaluated on four corpora : the wikipedia category graph , the wikipedia document graph , a computer science web site from the university of illinois , and a bibliographic network of all cikm and sigir conferences . the wikipedia data set has been used several times in the past for topic modeling purpose . however , because of the computational resources needed to infer topic models , prior studies have severely constricted the dataset size . , crawled 105 pages starting with the article on the nips conference finding 799 links @xcite . ho _ et al . _ performed a larger evaluation of their topicblock model using 14,675 document with 152,674 links ; however , they truncated each article to only the first 100 terms and limited the vocabulary to the 10,000 most popular words @xcite . the wikipedia category data set is a crawl of the _ category _ graph of wikipedia , beginning at the category as shown in figure [ fig : inh_wikicat ] . in wikipedia each category has a collection of articles and a set of links to other categories ; however , categories do nt typically have text associated with them , so the text of each article associated with a particular category is associated to the category s text . for example , the category includes articles : , etc . in total the crawled category graph consisted of 609 categories with text from 6,745 articles . the category graph is rather sparse with only 2,014 edges between categories , but has vocabulary size of 146,624 with 5,570,868 total tokens . we did not perform any text preprocessing procedure , including stop word removal and stemming , in any of the experiments due to empirical findings that these models are robust to the presents of stop words , etc @xcite . such settings can also explore hdtm s robustness in presence of a large and noisy corpus . a computer science department web site from the university of illinois urbana - champaign was chosen as the second data set because it represents a rooted web graph with familiar topics . by inferring the document hierarchy , the goal is to find the organizational structure of the computer science department . the intuition is that web sites reflect the business organization of the underlying entity ; thus it is expected that subtrees consisting of courses , faculty , news , research areas , etc . are found at high levels , and specific web pages are found at lower levels in the hierarchy . web site was crawled starting at the entry page and captured 1,078 web pages and 63,052 hyperlinks . in total there were 15,101 unique terms from 771,309 tokens . the bibliographic network consists of documents and titles from 4,713 articles from the sigir and cikm conferences . there exist 3,908 terms across 43,345 tokens in the document collection . in this collection , links include citations between papers within the cikm and sigir conferences . due to different citation styles , vendor abbreviations , and ambiguous names , without human interference most bibliographical meta - data extraction algorithms can only extract a portion of correct data @xcite . here we choose to use the most complete citation data set available from by the authors of the arnetminer project @xcite , however the citation graph is not guaranteed to be complete . a sigir 1998 paper by ponte and croft @xcite was chosen to be the root document because , in the data set , it had the most in - collection citations . to construct a larger data set with finer granularity , the english wikipedia article graph is used . in wikipedia , every wiki - article has two different link types : 1 ) category links , which point from the article to its categories , and 2 ) wiki - links , which point from the article to other wiki - articles . the category links are important because they hint at some encoded hierarchical structure that was codified on top of wikipedia after the article - graph was largely developed . for example , the wiki - article has a category link to and wiki - links to and among others . in later experiments we infer a document hierarchy from wikipedia and compare it to the crowd - encoded wikipedia category hierarchy . our wikipedia snapshot has 4,606,884 documents with 1,878,158,318 tokens , 44,739,242 links , and a vocabulary of 6,064,216 words . because the document graph is not a single connected component , we picked the largest component in that graph as data set . this giant component contains 1,957,268 documents and 44,673,134 document to document links ; 42.49% of nodes and 70% of edges are in this component . additionally , 1,316,879,537 terms with 4,225,765 unique tokens appear in this component , which is approximately 70% of the total number of terms . as a graph - preprocessing step , we replaced all edges pointing to redirection pages with edges that point directly to the actual wiki - article . using either the traditional inference model from sec . [ sec : inference ] or the high - throughput distributed sampling algorithm from sec . [ sec : distinference ] we basically sample a full hierarchy @xmath113 by sampling paths for each node @xmath78 except the root and the words in the document @xmath76 . given the state of the sampler at some time @xmath44 , _ i.e. _ , @xmath114 and @xmath115 , we iteratively sample each variable conditioned on the others as illustrated in [ sec : generating ] . the conditional distribution of the latent variables in hdtm model the given document - network . after running the markov chain a sufficient number of times we approach the stationary distribution ; the process of approaching the stationary distribution is called the `` burn - in '' period . after the burn - in we collect samples at a selected interval , _ i.e. _ , the sampling lag . the collected `` samples '' from the markov chain are full hierarchies constructed from the selection of a path for each node @xmath78 and a word for each document @xmath76 . therefore each sampled hierarchy contains one estimation about position of each document in the hierarchy and the position of each word in a document . for a given sampled hierarchy , we can assess the goodness of the hierarchy by measuring the log probability of that hierarchy and the observed words conditioned on the hyperparameters : @xmath116 using this log likelihood function it is possible to pick the most sampled hierarchy that maximizes the log likelihood as the final result . later , we will relax this assumption and use a more robust measurement that computes the mode of the sampled hierarchies and determines the _ certainty _ of the result . hdtm has some distinct qualities that make apples to apples comparison difficult . because hdtm is the first model to generate document hierarchies based on graphs , there is nothing to _ directly _ compare against . however , some of the models in the related work perform similar tasks , and so comparisons were performed when applicable . the related models typically perform quantitative evaluation by measuring the log likelihood on held out data or by performing some other task like link prediction . log likelihood analysis looks at the goodness of fit on held out data . unfortunately , the creation of a `` hold out '' data set is not possible , because each document , especially documents on the first or second level document , is very important to the resulting hierarchy . removing certain documents might even cause the graph to separate , which would make hierarchy inference impossible . instead , for the first quantitative evaluation , we compare the highest log likelihood generated from each model s sampler . quantitative experiments were performed on many of the aforementioned algorithms including : hlda @xcite , topicblock @xcite , tssb @xcite , and fslda @xcite . the fixed structure in fslda is determined by a breadth first iteration over the document graph because url heuristics used in the original paper do not exist in most of our data sets . the depth of hlda and topicblock is 4 . in all cases , a gibbs sampler was run for 5,000 iterations ; with a burn - in of 2,000 and a sampling lag of 20 . figure [ fig : lliterations ] shows the log likelihood for each sample on the computer science web site ; the other data sets showed similarly - shaped results . interestingly , figure [ fig : llavgdepth ] shows that higher likelihood values are strongly correlated with hierarchies of deeper average depth . lower @xmath25 values result in deeper hierarchies . ] the gibbs sampling algorithm was run on hdtm for various values of @xmath25 , and figure [ fig : bestcumll ] shows the best cumulative log likelihood for each of the tested values of @xmath25 . we observe that hdtm with @xmath74 achieved the best likelihood score . likelihood scores decreased steadily for increasing @xmath25 values , and hdtm with @xmath73 achieved the worst likelihood score . table [ tab : llresults ] shows the results of the different algorithms on the three smaller data sets . the topicblock and tssb clearly infer models with the best likelihood . the remaining algorithms , including hdtm , have mixed results . although these log likelihood results initially seem as if hdtm performs poorly , we demonstrate in section [ subsec : discussion ] that these likelihood results are expected and preferred . recall that the goal of hdtm is to generate document hierarchies from some original document - graph . thus , a positive result should somehow demonstrate that hdtm demonstrates hierarchies that are similar to the hierarchy that a human ( or many humans ) would create given the same document graph . with this in mind we performed a large scale experiment on the full english wikipedia article graph . also recall that the wikipedia category graph establishes a hierarchy that was human - built on top of wikipedia that maintains a pseudo - hierarchical classification of almost every document . the structure of categories is described to be a _ tree _ rooted by or ; however , we find that 87.01% of categories have an in - degree larger than 1 indicating the actual category is more complex . we find that although the wikipedia category graph _ mostly _ resembles a top - down ( coarse to fine ) hierarchy , it is technically an undirected loopy graph . as for wiki - articles , they almost always belong to more than one category because the idea or object described in each article can rightfully have multiple classification perspectives . for example , the wiki - article belongs to a category of the same name , but it also belongs to and among others . for a large scale evaluation our task is to infer a hierarchy from the full wikipedia article graph and see how well it matches the category graph . unfortunately , it is difficult to measure similarity between category graph and hdtm - hierarchies using conventional evaluation methods like the log likelihood . consider , for example , the category , which contains wiki - articles of faculty members from vastly different topical fields . in such cases , the bond is described primarily by the links rather than the content ; nevertheless , the log likelihood score judges models based on the goodness of fit of words . thus it is unreasonable to measure hierarchy quality by likelihood scores alone ; indeed , the hierarchy with the best possible likelihood score may be very different from actual , human created hierarchy . in order to perform a more thorough evaluation we should also compare the the graph structure . graph topologies can be compared in many different ways . among many options we chose deltacon @xcite to measure the similarity between hdtm s hierarchy and the wikipedia category hierarchy three reasons : 1 ) deltacon can calculate the similarity between two graphs with nodes that partially overlap , as observed in the messy wikipedia category graph , 2 ) the underlying metric in deltacon is the affinity score calculated by fast belief propagation ( fabp ) , which models the connectivity of nodes , and 3 ) deltacon can process millions of nodes in only a few hours . the deltacon metric returns a similarity score between the two compared graphs ranging between 0 and 1 inclusive where 0 is completely dissimilar and 1 is completely similar . although comparing with hlda and its variants is important , all existing hierarchical topic models are incapable of processing on data sets at this scale . we attempted to use the hlda implementation in the mallet package @xcite , but 130 hours was needed in order to compute a single gibbs sampling iteration . fortunately , we were able to use graphlab s implementation of lda which is based on models developed by ahmed _ et al . _ @xcite and smyth _ et al . _ although lda does not directly generate a graph , we are able to compare it with the category hierarchy by connecting the documents within a single lda - cluster to a made - up root node to simulate a hierarchy . the large scale models were inferred using 400 gibbs sampling iterations with a burn - in of 100 and a lag of 10 . in addition to hdtm and lda , we also use deltacon to compare the original wikipedia article graph and a randomly generated article - hierarchy as baselines . in the case of wikipedia article graph , we expect the wikipedia article graph ( the full graph , not a hierarchy ) to perform reasonably well because the category hierarchy is largely built on top of the article graph topology and they currently coexist and coevolve . in the case of the randomly generated article hierarchy , we randomly select a single parent for each node in the original article graph . .comparison of the wikipedia category graph to other generated hierarchies . deltacon scoring means higher is better . the random article hierarchy is generated by randomly picking one parent among all possible parents , _ i.e. _ , references , for each article . the differences in deltacon scores show that hdtm can preserve crucial connectivity information when constructing a new hierarchical structure from the original graph . [ cols=">,^ " , ] table [ tab : deltacon ] shows that the hierarchy inferred by hdtm is indeed most similar to the wikipedia category graph , followed by the random article hierarchy . recall that deltacon looks at graph topology similarity , thus the hdtm hierarchy is expected to more topologically similar to another hierarchy ( even if random ) than the article graph . these results demonstrate that hdtm can identify and preserve critical topological features when inferring a hierarchy from the graph . as usual , an increase in the number of lda topics increases the goodness of fit score ( in this case measured by deltacon instead of likelihood ) . interestingly , the @xmath25 parameter and the selection of different roots did not significantly influence deltacon score . to understand why this is , recall that different root - nodes and @xmath25 values result in different classification perspectives , but the different perspectives are still subsets of the same category graph with similar topological properties . hence they should have similar scores using the deltacon metric . recall that the hdtm , in its most basic form , generates a hierarchy by picking the best parent for each node / document ( except the root ) . in the iterative gibbs sampling process it is an almost certainty that a given node will pick different parents during different iterations . for example , say node @xmath8 has two parents @xmath117 and @xmath118 , it is possible that during iterations 15 node @xmath8 samples node @xmath117 as its parent , but then in iterations 620 node @xmath8 samples node @xmath118 to be its parent . two questions come to mind : 1 ) which parent should be ultimately picked for node @xmath8 in the final output graph ? and 2 ) what can the distribution of samples say about the certainty of our inferred graph ? in classic gibbs sampling , maximum a posteriori estimation is determined by the the mode of the samples . for some node @xmath8 , this translates to choosing the parent that appeared in the most samples as the final parent ; in the above case node @xmath8 would chose node @xmath118 as its parent because node @xmath118 was sampled 15 times compared to only 5 samples of node @xmath117 . to answer the second question , and therefore model the certainty of an inferred hierarchy , we calculate the number of times some node @xmath8 picks its final parent over the total number of samples normalized by the number of possible choices , _ i.e. _ , parents , node @xmath8 has . this results in a certainty score for node @xmath8 : @xmath119 where @xmath22 represents the total number of samples , @xmath120 is the number of times the final parent @xmath121 is sampled , and @xmath122 is the indegree of node @xmath8 representing the total number of parents that node @xmath8 could pick from . in the certainty score the outside fraction is used to measure the normalized difference between the raw probability and probability of random guess . applying this function to the above example we would calculate that the certainty score of node @xmath8 would be @xmath123 . figure [ fig : wiki_certainty_density ] shows the probability density functions for the certainties in hdtms parent sampling process . in the figures on both left and right we find that the probability densities appear to be polynomial distributions with interesting plateaus that end just before the 50% , 66% , and 75% certainty scores corresponding to the near - certain ( but not perfectly - certain ) scores for nodes with indegree values of 2 , 3 , and 4 respectively . in these results , and others not displayed in this paper , we find that when the root is a general document like , then changing @xmath25 does not affect the certainty distribution . given a relatively specific root like , larger @xmath25 values increase certainty overall . as discussed earlier , measuring topical similarity in hierarchies can be precarious because , in many cases , documents that are correctly situated under a topically unrelated parent may still have a strong contextual association with the parent document that outweighs the topical / language - oriented similarity as determined by the @xmath25 parameter . as an example , consider the wiki - articles and : even though these two articles are topically dissimilar , hdtm parameterized with a high @xmath25 value is likely to place both articles as children of the hierarchy s root because the high @xmath25 values weigh the topological link as more important that the documents inferred topicality . if , on the other hand , hdtm was parameterized with a very low @xmath25 value , then is more likely to be situated with other state senators , and the is more likely to be situated with other legislation . to quantitatively evaluate the topical and topological fitness of hierarchy together , we compare the similarily among the sets of parents and children in the inferred hierarchy against sets of wiki - articles and their wikipedia categories . because we are comparing sets of items we use the jaccard coefficient for this task . specifically , the jaccard coefficient is calculated by @xmath124 for each document @xmath8 that belongs to a set of categories @xmath125 , and where @xmath126 is the union of the corresponding categories of the nodes in @xmath8 s ancestors @xmath60 . the jaccard coefficient is useful here because it gives an quantitative measure describing how well the inferred hierarchical structure matches with the human - annotated categorical structure of wikipedia . figure [ fig : wiki_jaccard ] shows the relationship between jaccard coefficients of the hierarchies . for illustration clarity , in this figure all jaccard coefficients are binned into 10 equal intervals . we observe a clear correlation between the node s certainty function and its jaccard coefficient . this means that decisions that hdtm is most certain about tend to be the most similar to the human - created hierarchies . recall that hdtm uses the @xmath25 parameter to balance the weight of topological links and topical similarities . hence when specific nodes like are chosen as the root , higher @xmath25 will reduce jaccard coefficient for high certainty nodes . this is because higher @xmath25 values favor topology rather than content during the path and parent selection process . these results also underscore the importance of the selection of the root node . when provided a root node , which can be any node in the document graph , the inferred hierarchy is created from the perspective of that root . the hierarchy rooted at would therefore organize , say , congressional legislation differently than a hierarchy generated by the wiki - article on . yet , as we ve seen , the graph similarity metrics views the drastically different document - topic hierarchies as topologically similar . in order to properly understand the results captured in table [ tab : llresults ] from above , recall that log likelihood is a metric on the _ fit _ of the observations on the configuration of the model . the original work on lda @xcite found that likelihood increases as the number of topics increases . along those lines , chang , _ et al . _ demonstrated that more fine grained topics , which appear in models with a larger number of topics have a lower interpretability , despite having higher likelihood scores @xcite . simply put , there exists a negative correlation between likelihood scores and human interpretability in lda and similar topic models . applying those lessons to our experiments , recall that hdtm has as many topics as there are documents , and non - root document topics are mixtures of the topics on the path to the root . also recall that hlda , topicblock and tssb all generate a large number of latent topics . in hlda and topicblock , there are infinitely many topics / tables in the ncrp ; and practically speaking , the number of topics in the final model is much larger than the number of documents ( conditioned on the @xmath25 parameter ) . in tssb , the topic generation is said to be an interleaving of two stick breaking processes ; practically , this generates even larger topic hierarchies . the fslda algorithm has as many topics as there are in hlda , however , the fslda hierarchy is not redrawn during gibbs iterations to fit the word distributions resulting in a lower likelihood . simply put , the number of topics in hdtm and fslda @xmath127 @xmath128 hpam , hlda and topicblock @xmath128 tssb . similarly , figure [ fig : llavgdepth ] shows that deeper hierarchies have higher likelihood scores . this is because long document - to - root paths , found in deep hierarchies , are able to provide a more fine grained fit for the words in the document resulting in a higher likelihood . therefore , the better likelihood values of hlda , topicblock and tssb are due to the larger number of topics that these models infer and not necessarily due to better model outputs . in this case , a better way to evaluate model accuracy is by some external task or by manually judging the coherence of the topics . to measure the coherence of the inferred groupings , the _ word intrusion _ task developed by chang _ et al _ @xcite is slightly modified to create the _ document intrusion _ task . in this task , a human subject is presented with a randomly ordered set of eight document titles . the task for the human judge is to find the intruder , that is , the judge is asked to find which document is out of place or does not belong . if the set of documents without the intruder document all make sense together , then the human judge should easily be able to find the intruder . for example , given a set of computer science documents with titles \ { } , most people , even non - computer scientists , would pick as the intruder because the remaining words make sense together they are all computer science disciplines . for the set \ { } , identifying a single intruder is more difficult . human judges , when forced to make a choice , will choose an intruder at random , indicating that the grouping has poor coherence . to construct a set of document titles to present to the human judge , a grouping from the hierarchy is selected at random , and select seven documents are further selected at random from the grouping . if the there are fewer than 7 documents available in the selected grouping , then we select all of the documents available ; groupings of size less than 4 are thrown out . in addition to these documents , an intruder document is selected at random from among the entire collection of documents minus the documents in the test group . titles are then shuffled and presented to the human judges . extracting document groupings for evaluation is slightly different for each model . hdtm and fslda store a document at each node in the hierarchy . a grouping is selected by first picking a document at random , and then choosing its siblings . topicblock and hlda store documents at the leaves of the taxonomy , which often include several documents . a grouping is selected from these models by first picking a document at random , and then choosing the other documents in the leaf - topic . the hierarchies that the tssb model constructed allowed multiple documents to live at inner nodes . attempts to evaluate groupings on inner nodes with more than 4 documents were unsuccessful . nodes with 4 or more siblings were also difficult to find because the hierarchies that were generated were too sparse to find practical groupings . thus human judges with tssb groupings could not be found . each document - graph collection had different types of labels presented to the judges . the compsci web site collection was labeled by the web page title and url ; the wikipedia collection was labeled by the category title as shown in figure [ fig : intruder ] ; the bibliography network was labeled by the title of the paper . the intruder detection tasks described above were offered on amazon mechanical turk . no specialized training is expected of the judges . 50 tasks were created for each data set and model combination ; each user was presented with 5 tasks at a time at a cost of $ 0.07 per task . each task was evaluated by 15 separate judges . in order to measure the trustworthiness of a judge , 5 easy tasks were selected , _ i.e. _ , groupings with clear intruders , and gold - standard answers were created . judges who did not answer 80% of the gold - standard answers correctly are thrown out and not paid . in total the solicitation attracted 31,494 judgments , across 14 models of 50 tasks each . of these , 13,165 judgments were found to be from trustworthy judges . the _ model precision _ is measured based on how well the intruders were detected by the judges . specifically , if the intruder word @xmath129 is from model @xmath130 and task @xmath43 , and @xmath131 is the intruder selected by the human judge @xmath105 on task @xmath43 in model @xmath130 then @xmath132 where @xmath133 is the indicator function and @xmath134 is the number of judges . the model precision is basically the fraction of judges agreeing with the model . figure [ fig : turks ] shows boxplots of the precision for the four models on three corpora . in most cases , hdtm performs the best . as in @xcite , the likelihood scores do not necessarily correspond to human judgments . paired , two - tailed t - tests of statistical significants ( @xmath135 ) performed between hdtm @xmath136 and @xmath137 and the other models are represented by @xmath138 and @xmath139 in figure [ fig : turks ] respectively . . words at the root document represent the most probable words in the root topic . most probable words for other documents are not shown due to space constraints . ] the bibliography network data had relatively low precision scores . this is almost certainly because it was more difficult for the judges , who were probably not computer scientists , to differentiate between the topics in research paper titles . figure [ fig : dblphierarchy ] shows a small portion of the document hierarchy for the bibliographic network data set constructed with hdtm @xmath140 . the root document has 20 children in the hierarchy despite having 145 in - collection links . the remaining 120 documents live deeper in the hierarchy because hdtm has determined that they are too specific to warrant a first level position , and have a better fit in one of the subtrees . recall that each document is associated with the topics from itself to the root , where the root is a single , general topic . the seven most probable terms at the root level are also shown adjacent to the root s title in figure [ fig : dblphierarchy ] . these terms , like in hlda and topicblock , are terms that are general to the entire collection . similar sets of words exist at each node in the hierarchy , but are not shown in this illustration to maintain clarity . hdtm source code , analysis code and the scripts which generated the results found in this paper can be downloaded from https://github.com/nddsg/hdtm/releases/tag/kais . the wikipedia , web site and bibliographical data are all publicly available and free to download and are replicated in our own data repository . hierarchical document - topic model ( hdtm ) , is a bayesian generative model that creates document and topic hierarchies from rooted document graphs . the initial hypothesis was that document graphs , such as web sites , wikipedia and bibliographic networks contain a hidden hierarchy . unlike most previous work , hdtm allows documents to live at non - leaf nodes in the hierarchy , which requires the random walk with restart path sampling technique . an interesting side - effect of the random walker adaptation is that the path sampling step , eq . [ eq : pathsample ] , is much faster and easier to scale than the ncrp because rwr only creates a sampling distribution for the parents of a document , whereas the ncrp process creates a sampling distribution over all possible paths in the taxonomy . several quantitative experiments comparing hdtm with related models were performed . however , the results show that likelihood scores are a poor indicator of hierarchy interpretability , especially when the number of topics are different between comparison models . a large qualitative case study was also performed which showed that the cohesiveness of the document groupings generated by hdtm were statistically better than many of the comparison models despite the poor likelihood scores . 55ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty in http://dblp.uni-trier.de/db/conf/nips/nips2010.html#adamsgj10[__ ] ( ) pp . http://dl.acm.org/citation.cfm?id=944919.944937 [ * * , ( ) ] in link:\doibase 10.1145/2187836.2187936 [ _ _ ] ( , ) p. in http://dblp.uni-trier.de/db/conf/nips/nips2010.html#chambersss10[__ ] ( ) pp . http://dblp.uni-trier.de/db/journals/jmlr/jmlrp5.html#changb09 [ * * , ( ) ] in link:\doibase 10.1145/1273496.1273576 [ _ _ ] ( , ) pp . in @noop _ _ ( ) pp . @noop * * , ( ) in http://dblp.uni-trier.de/db/conf/trec/trec2004.html#songwsxlqzzxm04[__ ] ( ) in link:\doibase 10.1145/1076034.1076105 [ _ _ ] ( , ) p. link:\doibase 10.1145/984321.984322 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) in http://dblp.uni-trier.de/db/journals/jmlr/jmlrp15.html#nallapatimm11[__ ] , vol . ( ) pp . in http://dblp.uni-trier.de/db/conf/icwsm/icwsm2008.html#furukawamoui08[__ ] ( ) link:\doibase 10.1016/0306 - 4573(88)90027 - 1 [ * * , ( ) ] link:\doibase 10.1007/s10618 - 005 - 0361 - 3 [ * * , ( ) ] \doibase 10.1023/a:1022852608280 [ * * , ( ) ] link:\doibase 10.1016/0004 - 3702(89)90046 - 5 [ * * , ( ) ] in link:\doibase 10.1145/1102351.1102389 [ _ _ ] ( , ) pp . in http://dl.acm.org/citation.cfm?id=2002736.2002866[__ ] ( ) pp . http://dl.acm.org/citation.cfm?id=1690219.1690233 [ ( ) ] in link:\doibase 10.1145/1871437.1871469 [ _ _ ] ( , ) p. link:\doibase 10.1038/nature06830 [ * * , ( ) ] link:\doibase 10.1016/0378 - 8733(83)90021 - 7 [ * * , ( ) ] http://arxiv.org/abs/0912.5410 [ , * * , ( ) ] in link:\doibase 10.1145/1367497.1367512 [ _ _ ] ( , ) pp . in link:\doibase 10.1145/511446.511513 [ _ _ ] ( , ) pp . in http://dblp.uni-trier.de/db/conf/nips/nips2000.html#cohnh00[__ ] ( ) pp . in link:\doibase 10.1145/1401890.1401957 [ _ _ ] ( , ) pp . in http://dblp.uni-trier.de/db/conf/uai/uai2008.html#gruberrw08[__ ] ( ) pp . in http://dblp.uni-trier.de/db/conf/ijcai/ijcai2005.html#mccallumcw05[__ ] ( ) pp . in http://dblp.uni-trier.de/db/conf/uai/uai2004.html#rosen-zvigss04[__ ] ( ) pp . in @noop _ _ ( ) pp . in @noop _ _ ( ) pp . in @noop _ _ ( , ) pp . in @noop _ _ ( , ) @noop ( ) http://dblp.uni-trier.de/db/journals/pvldb/pvldb4.html#bahmanicg10 [ * * , ( ) ] in @noop _ _ ( , ) pp . in @noop _ _ , vol . ( ) p. link:\doibase 10.1145/2094114.2094118 [ * * , ( ) ] @noop ( ) @noop ( ) in @noop _ _ ( ) pp . in @noop _ _ ( , ) pp . in @noop _ _ ( , ) pp . in link:\doibase 10.1145/1401890.1402008 [ _ _ ] ( , ) p. in link:\doibase 10.1145/290941.291008 [ _ _ ] ( , ) pp . @noop ( ) @noop ( ) in @noop _ _ ( ) pp .

in this paper we construct a generalized universal cover for a very large class of uniform spaces called _ coverable _ spaces , which includes all geodesic metric spaces ( corollary [ inner ] ) , connected and locally pathwise connected compact topological spaces ( corollary [ 0628new ] ) , and in particular peano continua ( corollary [ peano ] ) . coverable spaces also include some more pathological spaces like the topologist s sine curve ( see below ) and totally disconnected spaces ( @xcite ) . associated with this generalized universal cover is a functorial uniform space invariant called the _ deck group _ , which is related to the classical fundamental group by a natural homomorphism . three types of uniform spaces are of greatest importance : topological groups , compact topologial spaces and metric spaces . topological groups were considered in @xcite and @xcite ; the relationship to the present paper may be found in section [ topsec ] , along with a correction to bptg . for compact topological spaces , which have a unique uniform structure compatible with the topology , the deck group is a topological invariant that coincides with the fundamental group in the setting of what we will call poincar spaces ( i.e. , connected , locally pathwise connected , semilocally simply connected spaces ) . the special case of metric spaces is of particular interest . in addition to more classical examples such as the hawaiian earring and related spaces ( @xcite , @xcite , de , @xcite , @xcite , @xcite ) , generalized universal covers and fundamental groups have recently been studied in connection with gromov - hausdorff limits of riemannian manifolds ( @xcite and @xcite ) . such limits are always geodesic spaces , and hence coverable . a recent example of menguy shows that limits of riemannian manifolds with positive ricci curvature can have bad local topology ( @xcite)precisely the sort of metric spaces at which the present work is aimed . we will consider metric spaces in more detail in an upcoming paper . one of the main impediments to generalizing the classical construction of the universal cover is the traditional definition of covering map , the most important property of which is the ability to lift curves and homotopies . however , this lifting property is traditionally gained at the expense of requiring that a space and its cover be locally homeomorphic in a fairly strong way , and consequently traditional universal ( or simply connected ) covers exist _ only _ for poincar spaces . earlier work concerning systems coverings of uniform spaces ( cf . @xcite , @xcite ) was also limited by considering only traditional covers , and therefore a universal object is impossible to obtain even for basic examples such as the countable product of circles or the hawaiian earring . one may take a hint from topological groups for how to proceed . in this category , a traditional covering map is a quotient homomophism with discrete kernel . the action by the kernel is not only properly discontinuous , it is uniformly so . moreover , as we showed in @xcite , one can exploit this more uniform kind of action to define generalized covers as quotients with central kernels that are ( complete and ) prodiscrete , i.e. are inverse limits of discrete groups . these generalized covers have the lifting properties of traditional covers , but are not in general local homeomorphisms . this allows one to abandon assumptions concerning ( semi-)local simple connectivity . the kernel of the generalized universal cover is a kind of generalized fundamental group . the key to generalizing the results of @xcite is to build a group action right into the definition of generalized cover . actually doing so is , unfortunately , somewhat technical , and this was carried out by the second author in @xcite . the basic idea is to consider a kind of uniformly properly discontinuous action called a _ discrete action _ on a uniform space . inverse systems of discrete actions define _ prodiscrete actions _ that generalize the notion of the action on a topological group via a prodiscrete , central subgroup . the action by a subgroup also preserves the uniform structure ( as long as one matches the left or right action to the left or right uniformity ) ; this property is generalized by something called an isomorphic action on a uniform structure , which also broadens the notion of an isometric action on a metric space . a generalized _ cover _ of uniform spaces is defined in @xcite to be a quotient via a prodiscrete , isomorphic action . at the beginning of @xcite is a review of basic definitions and properties of uniform spaces that are used for the construction ; we will use the same notation in this paper ( see @xcite for a more in - depth discussion ) . what follows is a sketch of our construction and main results . all of these constructions involve a choice of basepoint(s ) , but basepoint choice has an impact in this setting similar to that in the traditional setting for pathwise connected spaces , and for simplicity we will save detailed discussion of basepoints for the body of the paper . for now , all functions are simply assumed to preserve some chosen basepoints ( this is particularly important for the uniqueness statements , which are only true up to choice of basepoint ) . for each uniform space @xmath0 there is an inverse system @xmath1 indexed on the collection of entourages of @xmath0 partially ordered by reverse inclusion , called the _ fundamental inverse system _ of @xmath0 . @xmath2 consists of equivalence classes of finite @xmath3-chains starting at a fixed basepoint . an @xmath3-_chain _ between points @xmath4 is an ordered set of points @xmath5 such that for all @xmath6 , @xmath7 . an @xmath3__-loop _ _ is an @xmath3-chain that starts and ends at the same point . two @xmath3-chains from @xmath8 to @xmath9 are equivalent if one can be obtained from the other through finitely many steps , each of which involves removing or adding a point , always leaving the endpoints fixed and keeping an @xmath3-chain at each stage . the collection of all equivalence classes of @xmath3-loops at the basepoint forms a group @xmath10 with respect to concatenation . this group is finitely generated when @xmath0 is compact ( theorem [ compact ] ) . when @xmath2 is provided with a natural lifted uniformity , @xmath10 acts discretely on @xmath11 by concatenating an @xmath3-loop at the beginning of an @xmath3-chain . in a sense @xmath12 detects holes in @xmath0 that are , roughly speaking , larger than @xmath3 . the restriction @xmath13 of the bonding map @xmath14 to @xmath15 preserves concatenation and produces another inverse system @xmath16 of groups and homomorphisms ; in fact these two systems form an inverse system of actions as defined in @xcite . the inverse limit of the group system consists of a group @xmath17 , called the _ deck group _ of @xmath0 , which acts prodiscretely and isomorphically on @xmath18 . observe that @xmath19 is actually a ( prodiscrete ) topological group and the homomorphisms induced on the deck group are continuous homomorphisms . however , we do not know of examples of spaces having deck groups that are abstractly , but not continuously , isomorphic . in general the projection @xmath20 may not be surjective , and hence not a cover . in fact @xmath21 may be only a single point even when @xmath0 is not . the next definition , which is central to the paper , deals with this issue . [ coverdef]let @xmath0 be a uniform space . an entourage @xmath3 such that the projection @xmath22 is surjective is called a covering entourage . a uniform space @xmath0 is called coverable if there is a uniformity base of covering entourages including @xmath23 . the collection of all such entourages is called the covering base @xmath24 . if @xmath0 is coverable then each @xmath25 is a covering map ( in the sense of @xcite ) and we refer to the projection @xmath20 as the _ universal cover of _ @xmath0 . the next theorem combines results of theorems [ 73 ] and [ 101 ] . ( induced mapping ) let @xmath26 be coverable spaces with universal covers @xmath27 and @xmath28 . if @xmath29 is uniformly continuous then there is a unique uniformly continuous function @xmath30 such that @xmath31 . moreover , 1 . if @xmath32 is a cover then @xmath33 is a uniform homeomorphism . if @xmath34 is a coverable space and @xmath35 is uniformly continuous then @xmath36 . the restriction of @xmath33 to @xmath19 in the above theorem is a homomorphism @xmath37 ( theorem [ 73 ] ) . therefore the deck group is a functorial invariant of uniform structures . in the case of compact spaces the deck group is a topological invariant ; if @xmath0 is a compact poincar space then the deck group is naturally isomorphic to the fundamental group ( corollary 0518equiv ) . for non - compact spaces the deck group need not be a topological invariant ; for example the surface of revolution @xmath38 obtained by rotating the graph of @xmath39 about the @xmath40-axis has trivial deck group even though it is homeomorphic to a standard cylinder the deck group of which is @xmath41 . the problem is that a generator of @xmath42 may be represented by a path that extends down the cusp arbitrarily far , wraps around the small cusp , and then travels back to the basepoint . this generator will therefore not be detected by any of the groups @xmath43 , where @xmath44 is the metric entourage of size @xmath45 . the deck group does indicate that the standard cylinder and @xmath38 are not _ uniformly _ homeomorphic and that @xmath38 is not uniformly semilocally simply connected . the next theorem follows from corollary [ partialstep ] and theorem [ 101 ] . ( universal property ) if @xmath0 and @xmath46 are coverable and @xmath47 is a cover then there is a unique cover @xmath48 such that @xmath49 , where @xmath50 is the universal cover of @xmath46 . with regard to the universal property it should be pointed out that , in contrast to the situation for coverable topological groups , we do not know whether the composition of covers between coverable spaces ( or uniform spaces in general ) is a cover . ( the situation for topological groups is significantly simpler because the deck group is actually a central subgroup of the universal cover and covers are simply quotient homomorphisms , making their composition easier to understand . ) recall that the composition of traditional covering maps between connected topological spaces need not be a traditional cover ( @xcite ) . we resolve this problem in the same way that it is resolved for topological spaces ( see @xcite ) : define a category whose objects are covers @xmath51 between coverable uniform spaces and whose morphisms are commutative diagrams @xmath52where @xmath53 and @xmath54 are covers and @xmath32 is uniformly continuous . it is an immediate consequence of the preceding theorem that the universal cover @xmath55 is a universal object in this category . [ universaldef]a uniform space @xmath0 is called universal if there is a base @xmath56 for the uniformity such that for any @xmath57 , @xmath58 is a uniform homeomorphism . the collection of all such uniformities is called the universal base of @xmath0 ( which always contains @xmath23 ) . if @xmath0 is coverable then @xmath21 is universal ( theorem [ xic ] ) and every universal space is coverable ( corollary [ uic ] ) . moreover , a coverable space @xmath46 is universal if and only if @xmath19 is trivial , or equivalently every @xmath3-loop is @xmath3-homotopic to the trivial loop for entourages @xmath3 in a particular basis ( corollary [ uit ] ) , a condition that is reminiscent of simply connected . the next theorem follows from theorem [ 76 ] . ( lifting ) let @xmath0 be universal , @xmath46 be coverable and @xmath47 be uniformly continuous . then there exists a unique uniformly continuous function @xmath59 such that @xmath60 , where @xmath50 is the universal cover of @xmath46 . any connected , uniformly locally pathwise connected ( see definition unidef ) , simply connected uniform space is universal in particular any compact , connected , locally pathwise connected , simply connected topological space is universal ( theorem [ 0627simply ] and corollary [ compactu ] ) . therefore we may apply the lifting theorem to paths and homotopies of paths . this allows one to define a natural mapping @xmath61 , for any coverable space @xmath0 , by lifting a loop that represents an element of the fundamental group and taking the deck transformation that takes the basepoint to the endpoint of the loop . even though the action is not discrete this mapping is well defined , and a homomorphism . there is a satisfying relationship between the two most basic algebraic properties of the map @xmath62 and topological properties of @xmath63 when @xmath0 is pathwise connected : ( 1 ) @xmath62 is injective if and only if @xmath21 is simply connected ( proposition [ inj ] ) , and ( 2 ) @xmath64 is surjective if and only if @xmath21 is pathwise connected . more precisely , for arbitrary coverable @xmath0 , the image of @xmath64 is the stabilizer in the deck group @xmath19 of the pathwise connected component of @xmath21 that contains the basepoint ( theorem [ nat ] ) . hence if @xmath21 is both pathwise connected and simply connected then the deck group and the fundamental group are isomorphic . this is true as mentioned earlier when @xmath0 is a compact poincar space and also when @xmath0 is a locally compact , pathwise connected topological group ( cf . if @xmath0 is connected and uniformly locally pathwise connected then the pathwise connected component of @xmath65 is dense in @xmath21 ( proposition [ dense ] ) and therefore @xmath21 is connected . in this case it is also true that @xmath66 is dense in @xmath19 . note that , in the case of the surface @xmath38 described above , @xmath38 is itself universal . @xmath38 is pathwise connected but not simply connected and correspondingly @xmath67 is surjective but not injective . as another example , consider the ( closed ) topologist s sine curve @xmath68 , illustrated in figure [ tsc ] with its universal cover . the deck group of @xmath69 is @xmath41 ; the action shifts the universal cover in a way similar to the action of @xmath41 on @xmath70 . note that arbitrarily fine chains may wrap around @xmath68 , while no path does . hence the deck group indicates topology where the fundamental group , which is trivial in this case , does not . finally , @xmath71 is simply connected but not pathwise connected , and correspondingly @xmath72 is injective but not surjective ( and the image of @xmath62 need not be dense in the deck group because @xmath68 is not locally pathwise connected ) . we show that if @xmath0 is coverable with uniform dimension @xmath73 in a sense due to isbell ( @xcite ) then @xmath21 has the same property ( theorem [ dimthm ] ) , and we conjecture that the the uniform dimensions of @xmath0 and @xmath65 are the same . when @xmath0 has dimension @xmath74 , we show that @xmath65 is simply connected and contains no topological circles . as a consequence the function @xmath62 is injective . we prove some results concerning dimension and coverable spaces , although it seems much more may be said . further progress may require additional work extending theorems concerning covering dimension to theorems about uniform dimension . however , this partial result is particularly useful when @xmath75 because then @xmath21 is forced to be at most one dimensional and if @xmath0 contains any non - trivial curve then the uniform dimension of @xmath76 must be exactly @xmath74 . one - dimensional metric spaces are of interest since they include , for example , planar fractals and metrized cayley graphs , as well as many familiar pathalogical examples . it is known that the fundamental group of a compact , connected , @xmath74-dimensional metrizable space embeds in an inverse limit of free groups . given this fact , it seems very likely that for appropriate choice of @xmath3 , the groups @xmath77 are finitely generated free groups when @xmath0 is compact , connected , and @xmath74-dimensional . we have verified this in some special cases ( see section [ pssec ] ) . it would be very interesting to know whether for suitable choice of @xmath3 , @xmath78 is free when @xmath0 is coverable ( or just connected ? ) of uniform dimension @xmath74 . in this paper we show that , if @xmath0 is coverable of uniform dimension @xmath79 then @xmath21 has no topological circles ( proposition [ circle ] ) . it follows that @xmath21 is simply connected and hence ( assuming @xmath0 is pathwise connected ) @xmath80 is an embedding . in general , @xmath21 may not be pathwise connected and hence @xmath62 may not be surjective , but at least in the case when @xmath0 is uniformly locally pathwise connected we know that the ( isomorphic ) image of the fundamental group is dense in the deck group . in other words , @xmath0 is a quotient via a free action of its fundamental group on a connected , simply connected , uniformly one dimensional space . we conclude with a general result about metric spaces . given a metric space @xmath81 , @xmath21 is metrizable ( proposition [ metrizable ] ) , and if @xmath0 is coverable then by definition of cover the action of @xmath19 on @xmath65 is isomorphic in the sense that there is a uniformity base for @xmath21 that is invariant with respect to @xmath19 . it is natural to ask whether there is a metric on @xmath21 with respect to which the action of @xmath19 is isometric and such that the metric on @xmath21 is the quotient metric . ( since @xmath19 is complete and acts prodiscretely , the orbits of @xmath19 are closed and hence if @xmath21 has an invariant metric there is a well defined quotient metric on @xmath0 , namely the distance between the corresponding orbits in @xmath21see @xcite , 4.4 for details on quotient metrics . ) we do not know the answer to this question in general . however , if one examines the proof in @xcite that a uniform space with a countable base possesses a compatible pseudometric , it is clear that the explicitly constructed pseudometric is invariant with respect to the action of a group @xmath82 if the entourages used in the construction are invariant . in other words , if @xmath0 is a coverable metric space then we may put a metric on @xmath21 that is invariant with respect to the action of @xmath19 . although the quotient metric on @xmath0 may not be the original , it is still uniformly equivalent to the original . this proves : if @xmath0 is a coverable metric space then @xmath0 is uniformly homeomorphic to the metric quotient @xmath83 of a universal metric space @xmath46 with respect to the isometric prodiscrete action of a group @xmath82 isomorphic to @xmath19 . note that for any entourage @xmath3,@xmath84that is , @xmath85 is the set of all pairs of points joined by an @xmath3-chain of length @xmath86 . a uniform space @xmath0 is called _ chain connected _ ( sometimes called uniformly connected ) if every uniformly continuous function from @xmath0 into a discrete uniform space is constant . [ consume]the following are equivalent : 1 . @xmath0 is chain connected 2 . for any entourage @xmath3 , @xmath87 . 3 . for any entourage @xmath3 , every pair of points in @xmath0 is joined by an @xmath3-chain . 4 . for any entourage @xmath3 and @xmath88 , @xmath89 . the equivalence of the first two conditions is proved in @xcite , 9.34the statement of 9.34 is wrong but the proof is right ! the last three are equivalent by formula ( [ ton ] ) . if @xmath0 is chain connected and @xmath47 is a uniformly continuous surjection then @xmath46 is chain connected . for a topological group @xmath82 , chain connected is equivalent to @xmath82 being locally generated ( generated by every neighborhood of the identity ) , and @xmath3-chains are the same as what we called @xmath90-chains in @xcite . if one fixes a single point @xmath8 and entourage @xmath3 in an arbitrary uniform space @xmath0 then if @xmath91 is the set of all points that can be joined to @xmath8 by a @xmath3-chain , it is easy to check that @xmath92 is both open and closed . therefore every connected uniform space is chain connected . on the other hand , the rational numbers are chain connected and totally disconnected ( see @xcite for related topics ) . [ 315system]let @xmath93 be an inverse system of sets such that each of the projections @xmath94 is surjective . then for any subsets @xmath95 of @xmath96 , @xmath97 . in particular for any @xmath86 , @xmath98 . we have @xmath99 if and only if there exists some @xmath100 such that @xmath101 and @xmath102 . since @xmath103 is surjective this is equivalent to : for some @xmath104 , @xmath105 and @xmath106 . but this is equivalent to @xmath107 . [ 41]let @xmath108 be an inverse system of chain connected uniform spaces . if each of the projections @xmath109 is surjective then @xmath110 is chain connected . for any entourage @xmath111 in @xmath0 and @xmath112 we have @xmath113 for some @xmath86 . then @xmath114 . an @xmath3-extension of a @xmath3-chain @xmath115 is an @xmath3-chain @xmath116 , where @xmath117 . two @xmath3-chains from @xmath118 to @xmath119 are said to be @xmath3-related if one is a @xmath3-extension of the other . an @xmath3-homotopy between @xmath3-chains @xmath120 and @xmath121 is a sequence @xmath122 of @xmath3-chains such that @xmath123 is @xmath3-related to @xmath124 for all @xmath125 . the number @xmath126 is called the _ length _ of the homotopy . we denote the @xmath3-homotopy class of an @xmath3-chain @xmath127 by @xmath128_{e}$ ] . now fix a basepoint @xmath129 . let @xmath2 denote the set of all @xmath3-homotopy classes of @xmath3-chains in @xmath0 starting at @xmath130 and define a function ( the endpoint map ) @xmath58 by @xmath131)=x_{n}$ ] . the proof of the next lemma is immediate . [ lemsur5]if @xmath0 is a uniform space and @xmath3 is an entourage then @xmath132 is surjective if and only if every pair in @xmath0 is joined by an @xmath3-chain ; in particular @xmath133 is surjective if @xmath0 is chain connected . the next lemma will often be used without reference . the proof is straightforward but tedious . [ lemsur2]let @xmath134 and @xmath135 be @xmath3-chains for some entourage @xmath3 in a uniform space @xmath0 . then @xmath136 and @xmath137 are @xmath3-homotopic if and only if the @xmath3-loop @xmath138 is @xmath3-homotopic to the trivial chain @xmath139 . let @xmath0 be a uniform space with entourage @xmath3 . for any entourage @xmath140 , define @xmath141 as follows : let @xmath142_{e},[\beta ] _ { e})\in d^{\ast } $ ] if and only if @xmath143_{e},[\beta ] _ { e})=([\ast = x_{0}, ... ,x_{n},y]_{e},[\ast = x_{0}, ... ,x_{n},z]_{e})\text { with } ( y , z)\in d\text{.}\ ] ] [ technical]let @xmath0 be a uniform space with entourage @xmath3 . for any entourages @xmath144 , @xmath142_{e},[\beta ] _ { e})\in d^{\ast } f^{\ast } $ ] if and only if for some @xmath145_{e}=[\ast = x_{0}, ... ,x_{n}]_{e}$ ] we have @xmath146_{e}=[\ast = x_{0}, ... ,x_{n},y]_{e}\]]@xmath147_{e}=[\ast = x_{0}, ... ,x_{n},z]_{e}\]]for some @xmath148 such that @xmath149 and @xmath150 . in particular , @xmath151_{e},[\beta ] _ { e}\in b([\gamma ] _ { e},\left ( d^{\ast } \right ) ^{2})$ ] if and only if these conditions hold with @xmath152 . the reverse implication is obvious . conversely , if @xmath142_{e},[\beta ] _ { e})\in d^{\ast } f^{\ast } $ ] then we may take @xmath153 , @xmath154 , @xmath155 and @xmath156 where @xmath157 is @xmath3-homotopic to @xmath127 and @xmath150 and @xmath149 . now @xmath137 is @xmath3-related to @xmath158 and the latter is @xmath3-homotopic to @xmath159 using the @xmath3-homotopy from @xmath127 to @xmath160 . finally , @xmath136 is @xmath3-related to @xmath161 . [ xee]let @xmath0 be a chain connected uniform space with entourage @xmath3 , and @xmath162 be the endpoint map . then 1 . @xmath133 is injective when restricted to any @xmath163-ball . 2 . for any entourage @xmath140 and @xmath151_{e}\in x_{e}$ ] , @xmath164_{e},d^{\ast } ) ) = b(\phi _ { xe}([\alpha ] _ { e}),d)\]]and @xmath165 . 3 . the collection of all @xmath141 such that @xmath140 is a base for a uniformity of @xmath2 . 4 . @xmath133 is a bi - uniformly continuous surjection with respect to this uniformity . in particular the restriction of @xmath133 to any @xmath166-ball is a uniform homeomorphism onto the corresponding @xmath167-ball . if @xmath151_{e}$ ] and @xmath168_{e}$ ] are in @xmath169_{e},e^{\ast } ) $ ] and @xmath170_{e})=\phi _ { xe}([\beta ] _ { e})$ ] then by definition of @xmath163 , @xmath136 and @xmath137 are @xmath3-homotopic . this proves the first part . for part ( 2 ) let @xmath171 ; it is obvious from the definition of @xmath133 and @xmath141 that @xmath164_{e},d^{\ast } ) ) \subset b(\phi _ { xe}([\alpha ] _ { e}),d)=b(x , d)\text{.}\]]now suppose @xmath172 and let @xmath173 . since @xmath136 is @xmath3-homotopic to @xmath174 , we have that @xmath142_{e},[\beta ] _ { e})\in d^{\ast } $ ] and @xmath175_{e},[\beta ] _ { e}))=(x , y)\text{.}\]]this implies that @xmath165 . we will now check the conditions for a uniformity base . clearly , for entourages @xmath3 and @xmath176 we have @xmath177 ( and in fact they are equal but it is not necessary to prove this ) . since @xmath167 is symmetric , so is @xmath141 . next , let @xmath176 be an entourage such that @xmath178 . suppose that @xmath142_{e},[\beta ] _ { e})\in ( f^{\ast } ) ^{2}$ ] . then for some @xmath145_{e}$ ] , @xmath179_{e},[\beta ] _ { e}\in b([\gamma ] _ { e},f^{\ast } ) $ ] . applying lemma technical we may write @xmath151_{e}=[\ast = x_{0}, ... ,x_{n},y]_{e}$ ] and @xmath180_{e}=[\ast = x_{0}, ... ,x_{n},z]_{e}$ ] with @xmath181 and hence @xmath182 . by definition @xmath142_{e},[\beta ] _ { e})\in d^{\ast } recall from @xcite that @xmath133 is bi - uniformly continuous by definition if the image or inverse image of any entourage with respect to @xmath133 is again an entourage . part ( 4 ) is now an immediate consequence of parts ( 2 ) and ( 3 ) . we can now see precisely how the choice of basepoints affect things . let @xmath183 and @xmath184 be @xmath185-chains for some entourage @xmath3 in a uniform space @xmath0 . define the concatenation of @xmath136 to @xmath137 by@xmath186 [ basepoint]let @xmath0 be a chain connected uniform space , @xmath187 , and @xmath3 be an entourage . let @xmath188 denote the space of @xmath3-homotopy classes of @xmath3-chains based at @xmath189 . let @xmath127 be any @xmath3-chain from @xmath54 to @xmath53 . there is a natural map from @xmath190 to @xmath191 defined by taking @xmath151_{e}$ ] to @xmath192_{e}$ ] , where @xmath193 is the @xmath3-chain obtained by concatenating @xmath194 to the end of @xmath127 . it is easy to check that this function is a bijection with inverse @xmath195 . for any entourage @xmath176 in @xmath0 , there are corresponding entourages @xmath196 in @xmath197 and @xmath198 in @xmath199 . now @xmath200 if and only if @xmath201 and in particular , @xmath190 and @xmath199 are uniformly homeomophic . therefore , as in the case of traditional covering space theory , the choice of basepoint @xmath130 plays only a minor and predictable role . given a basepoint @xmath130 in @xmath0 , we will always take @xmath202_{e}$ ] for the basepoint in @xmath2 . in general , for any function @xmath47 we will always suppose that @xmath32 is pointedin the sense that @xmath203 . for each chain @xmath204 in @xmath0 we denote the chain @xmath205 by @xmath206 , with similar notation for any finite sequence of chains . we will always take the uniform structure on @xmath2 to be the one given by proposition [ xee ] . if @xmath95 are entourages of @xmath26 , respectively , such that @xmath207 and @xmath127 is an @xmath3-chain then @xmath206 is an @xmath176-chain in @xmath46 . note that if @xmath208 is an @xmath3-homotopy between @xmath3-chains @xmath136 and @xmath127 then @xmath209 is an @xmath176-homotopy between @xmath210 and @xmath206 . in particular the function in the following definition is well defined ( and pointed ) . [ induced]given @xmath47 is a function between uniform spaces such that for entourages @xmath95 of @xmath26 , respectively , @xmath207 , we define @xmath211 by @xmath212_{e})=[f(\gamma ) ] _ { f}$ ] . for any entourages @xmath140 in a uniform space @xmath0 , define @xmath213 by @xmath214_{d})=[\alpha ] _ { e}$ ] . note that by definition @xmath215 , where @xmath216 is the identity , and @xmath217 for any entourages @xmath140 in a uniform space @xmath0 , @xmath218 is uniformly continuous . let @xmath219 be an entourage in @xmath2 and @xmath220 be an entourage in @xmath0 . we will need to distinguish between @xmath221 , which we will refer to as @xmath222 and @xmath223 , which we will refer to as @xmath224 . if @xmath225_{d},[\beta ] _ { d})\in g_{d}^{\ast } $ ] then by definition we can take @xmath226 and @xmath227 with @xmath228 . now @xmath229_{d},[\beta ] _ { d}))=([\alpha ] _ { e},[\beta ] _ { e})\]]and by definition @xmath142_{e},[\beta ] _ { e})\in g_{e}^{\ast } that is , @xmath230 . [ identify]let @xmath0 be a uniform space and @xmath140 be entourages . then @xmath231is a uniform homeomorphism such that the following diagram commutes:@xmath232moreover , for any entourage @xmath233 in @xmath0 , we have @xmath234 . for simplicity denote @xmath235 by @xmath236 . let @xmath237_{e}=[\alpha _ ... ,[\alpha _ { n}]_{e}\}$ ] be a @xmath166-chain in @xmath2 , where each @xmath238 ends at a point @xmath239 . then @xmath240_{d^{\ast } } ) = [ \phi _ { xe}([\ast ] _ { e}), ... ,\phi _ { xe}([\alpha _ { n}]_{e})]_{d}=[z_{0}=\ast , z_{1}, ... ,z_{n}]_{d}\text{. } \label{identifyfor}\ ] ] since @xmath241_{e},[\alpha _ { 1}]_{e})\in d^{\ast } $ ] , @xmath242_{e}=[z_{0}=\ast , z_{1}]_{e}\text{.}\]]proceeding inductively , for all @xmath6 we have that @xmath243_{e}=[z_{0}=\ast , z_{1}, ... ,z_{i}]_{e } \label{0621sequence}\]]which implies that @xmath236 is injective . given any @xmath167-chain @xmath244 we can let @xmath245 for all @xmath246 . since @xmath247 is a @xmath167-chain , @xmath248_{d}, ... ,[\alpha _ { n}]_{d}\}$ ] is a @xmath141-chain and @xmath249_{e}=[\alpha _ { 0}]_{e}, ... ,[\alpha _ { n}]_{e}]_{d^{\ast } } ) = [ z_{0}=\ast , z_{1}, ... ,z_{n}]_{d}\]]which shows that @xmath236 is surjective . by definition , @xmath250_{d^{\ast } } ) \right ) = [ z_{0}=\ast , z_{1}, ... ,z_{n}]_{e}\text{.}\]]that is , the upper triangle of the diagram commutes , and we already know that the bottom triangle does . finally , suppose that @xmath251_{e}=[\alpha _ { 0}]_{e}, ... ,[\alpha _ { n}]_{e}]_{d^{\ast } } , [ [ \ast ] _ { e}=[\beta _ { 0}]_{e}, ... ,[\beta _ { m}]_{e}]_{d^{\ast } } ) \in ( f^{\ast } ) ^{\ast } \]]where , according to formula [ 0621sequence ] we can suppose that for some @xmath252 , @xmath253 and @xmath254 for all @xmath6 . by definition this means that we can suppose that @xmath255 and @xmath256_{e},[\beta _ { n}]_{e})\in f^{\ast } $ ] . this in turn is equivalent to @xmath257 , which is equivalent to @xmath258_{e}=[\alpha _ { 0}]_{e}, ... ,[\alpha _ { n}]_{e}]_{d^{\ast } } , [ [ \ast ] _ { e}=[\beta _ { 0}]_{e}, ... ,[\beta _ { m}]_{e}]_{d^{\ast } } ) ) \in f^{\ast } \text{.}\ ] ] the preceding somewhat technical - looking proposition in fact has a very nice interpretation . essentially it identifies @xmath259 with @xmath260 by taking a @xmath141-chain of @xmath3-chains to the @xmath167-chain of their endpoints . in other words , @xmath2 and @xmath0 are locally the same ; @xmath141 and @xmath163 are really just copies of @xmath167 and @xmath3 inside @xmath261 . dealing with @xmath260 rather than @xmath262 means we are dealing with chains rather than chains of chains . at the same time , proposition [ identify ] identifies the mapping @xmath263 with the more easily understood mapping @xmath264 . with such identifications we can express the proposition as @xmath265 . this proposition is very useful because it will allow us to immediately apply results proved for @xmath133 to the more general functions @xmath266 . the next lemma illustrates this . [ chainhom]suppose that @xmath0 is a uniform space and @xmath3 is an entourage in @xmath0 such that @xmath2 is chain connected . then for any entourage @xmath176 and @xmath185-chain @xmath127 in @xmath0 , @xmath127 is @xmath3-homotopic to an @xmath176-chain . without loss of generality we can suppose that @xmath267 . since @xmath2 is chain connected , @xmath268 is surjective by lemma [ lemsur5 ] . but this means that if @xmath127 is an @xmath3-chain in @xmath0 there is some @xmath176-chain @xmath136 in @xmath0 such that @xmath151_{e}=\phi _ { ef}([\alpha ] _ { f})=[\gamma ] _ { e}$ ] . given entourages @xmath269 in @xmath0 , we have functions @xmath213 and @xmath270 with @xmath271_{d}))=[x_{0}=\ast , ... , x_{n}]_{f}=\phi _ { fd}([x_{0}=\ast , ... , x_{n}]_{d})\text{.}\]]in other words , @xmath272 and @xmath273 forms an inverse system of uniformly continuous functions having as its indexing set the set of all entourages of @xmath0 partially ordered by reverse inclusion . we will call the inverse system @xmath274 the fundamental inverse system of @xmath0 and let @xmath21 denote the inverse limit of this inverse system with the inverse limit uniformity . we will let @xmath275 be the projection ; we simply denote @xmath276 by @xmath50 . we will always choose for our basepoint @xmath130 in @xmath65 the element having as each of its coordinates the basepoint @xmath277_{e}$ ] in @xmath2 . since a uniform space is metrizable if and only if it has a countable base for its uniformity ( @xcite ) , if @xmath0 is metrizable then each @xmath2 is metrizable . we may index the fundamental system with this countable base and conclude : [ metrizable]if @xmath0 is metrizable then @xmath21 is metrizable . [ 60]let @xmath0 and @xmath46 be uniform spaces , @xmath47 be uniformly continuous , and @xmath95 be entourages in @xmath26 , respectively , such that @xmath278 . then @xmath279 is uniformly continuous and satisfies @xmath280moreover , if @xmath2 is chain connected then @xmath279 is the unique uniformly continuous function satisfying ( [ commute1 ] ) such that @xmath281_{e})=[f(\ast ) ] _ { f}$ ] . to prove @xmath279 is uniformly continuous we need only consider entourages of the form @xmath141 in @xmath282 , where @xmath283 is an entourage in @xmath46 . since @xmath32 is uniformly continuous there exists some entourage @xmath284 such that @xmath285 . let @xmath286 and @xmath287 be such that @xmath143_{e},[\beta ] _ { e})\in g^{\ast } \text{,}\]]which means by definition @xmath228 . then @xmath288and therefore @xmath289_{e},[\beta ] _ { f})=([\ast = f(x_{0}), ... ,f(x_{n}),f(y)]_{f},[\ast = f(x_{0}), ... ,f(x_{n}),f(z)]_{f})\in d^{\ast } \text{.}\]]this shows that @xmath279 is uniformly continuous . that @xmath290 is an immediate consequence of the definition . to prove the last statement suppose @xmath2 is chain connected and @xmath291 is a uniformly continuous function such that @xmath292 and @xmath293_{e})=[f(\ast ) ] _ { f}$ ] . let @xmath82 be an entourage in @xmath0 such that @xmath294 . by way of lemma chainhom it is sufficient to show that if @xmath295 is a @xmath296-chain then @xmath297_{e})=f_{ef}([\beta ] _ { f})$ ] . we will prove it by induction on @xmath86 . the case @xmath298 is given ; suppose the statement is true for @xmath299 and consider @xmath300 with @xmath301 . suppose that @xmath302_{e}):=[f(\ast ) = z_{0}, ... ,z_{m}]_{f}$ ] ; then @xmath303_{e})=\phi _ { yf}\circ f^{\prime } ( [ \ast = x_{0}, ... ,x_{n+1}]_{e})=z_{m}\text{.}\]]by the inductive hypothesis , @xmath304_{e})=f_{ef}([\alpha ] _ { e})$ ] and by definition of @xmath296 we have @xmath305_{e}),f^{\prime } ( [ \beta ] _ { e}))=(f^{\prime } ( [ \alpha ] _ { e}),f^{\prime } ( [ \beta ] _ { e}))\in f^{\ast } \]]and @xmath305_{e}),f_{ef}([\beta ] _ { e}))\in f^{\ast } \text{.}\]]in other words , both @xmath306_{e})$ ] and @xmath302_{e}))$ ] lie in @xmath307_{e},f^{\ast } ) $ ] and by proposition [ xee ] @xmath308 is injective on this ball . the fact that @xmath309_{e})=f^{\prime } ( [ \beta ] _ { e})$ ] now follows from @xmath310_{e})=f(\phi _ { xe}([\beta ] _ { e}))=\phi _ { yf}(f^{\prime } ( [ \beta ] _ { e}))\text{.}\ ] ] if @xmath0 is a uniform space and @xmath267 are entourages in @xmath0 such that @xmath311 is chain connected then @xmath14 is the unique uniformly continuous function such that @xmath312_{f})=[\ast ] _ { e}$ ] and @xmath313 . the proof of the next lemma is virtually identical to the proof of lemma 65 in @xcite ; one need only replace statements like @xmath314 with @xmath315 . let @xmath26 be uniform spaces , @xmath47 be a uniformly continuous surjection , @xmath176 be an entourage in @xmath46 and @xmath316 . let @xmath317 be an @xmath3-chain in @xmath0 and @xmath208 be an @xmath176-homotopy from the @xmath176-chain @xmath318 to another @xmath176-chain @xmath319 . then @xmath208 lifts to an @xmath3-homotopy from @xmath320 to an @xmath3-chain @xmath321 . that is , there exist an @xmath3-chain @xmath322 and an @xmath3-homotopy @xmath323 between @xmath317 and @xmath321 such that @xmath324 . [ 66]let @xmath26 be uniform spaces , @xmath47 be a uniformly continuous surjection , @xmath176 be an entourage in @xmath46 and @xmath316 . if @xmath58 is surjective and there exists a uniformly continuous function @xmath325 such that the following diagram commutes @xmath326then @xmath133 is a uniform homeomorphism . we need only show that @xmath133 is injective . equivalently we need only show that if @xmath327 is an @xmath3-loop in @xmath0 then @xmath317 is @xmath3-homotopic to the trivial loop @xmath139 . let @xmath318 , which is an @xmath176-loop in @xmath46 and @xmath328_{e}):=[h]_{f}$ ] ; by the commutativity of the diagram @xmath329_{f})=f(\phi _ { xe}([c]_{e})=f(\ast ) $ ] . in particular , @xmath330 is a loop in @xmath46 . moreover , @xmath331_{f}=\psi ( \phi _ { xe}([c]_{e}))=\psi ( \ast ) $ ] . on the other hand , @xmath332_{f}=f_{ef}([\ast ] _ { e})=\psi ( \phi _ { xe}([\ast ] _ { e}))=\psi ( \ast ) \text{. } \]]that is , @xmath331_{f}=[f(\ast ) ] _ { f}$ ] and therefore @xmath330 is @xmath176-homotopic to the trivial loop @xmath333 . now by definition , @xmath334_{f}=f_{ef}([c]_{e})=[f(x_{0}), ... ,f(x_{n})]_{f}=[f(c)]_{f}=[d]_{f}\]]and therefore @xmath335 is also @xmath176-homotopic to @xmath336_{f}$ ] . the chain lifting lemma now finishes the proof . [ 67]let @xmath0 and @xmath46 be uniform spaces with @xmath0 chain connected , @xmath29 be a surjective uniformly continuous map , @xmath176 be an entourage in @xmath46 and @xmath316 . if @xmath337 is bijective then @xmath162 is a uniform homeomorphism . note that since @xmath0 is chain connected , @xmath133 is surjective . let @xmath338 , which is uniformly continuous since @xmath339 is bi - uniformly continuous . certainly @xmath340 and @xmath341since @xmath342 is bijective we conclude that @xmath343 and the conditions of proposition [ 66 ] are satisfied , completing the proof . [ coverchain]if @xmath0 is a coverable uniform space and @xmath3 is a covering entourage in @xmath0 then @xmath2 is coverable and hence chain connected . if @xmath3 is a covering entourage then we may index the fundamental inverse system of @xmath2 using the set of all entourages @xmath219 where @xmath344 is a covering entourage in @xmath0 . then @xmath345 is surjective for all such @xmath176 . but @xmath346 is naturally identified with @xmath347 by proposition [ identify ] and therefore @xmath348 is also surjective . therefore the collection of all such @xmath219 is a covering basis for @xmath2 . it is straightforward but tedious to check that if @xmath151_{e}=[\alpha ^{\prime } ] _ { e}$ ] and @xmath168_{e}=[\beta ^{\prime } ] _ { e}$ ] then @xmath349_{e}=[\alpha ^{\prime } \ast \beta ^{\prime } ] _ { e}$ ] . for an entourage @xmath3 in a uniform space @xmath0 , we define @xmath77 to be the group of all @xmath3-homotopy classes of @xmath3-loops at @xmath130 with the group operation induced by concatenation . we will call this the @xmath3-deck group of @xmath0 . that is , given @xmath3-loops @xmath136 and @xmath137 based at @xmath130 , we let @xmath146_{e}\ast \lbrack \beta ] _ { e}:=[\alpha \ast \beta ] _ { e}\text{.% } \]]note that the identity chain is @xmath202_{e}$ ] and if @xmath350 then @xmath151_{e}^{-1}=[\alpha ^{-1}]_{e}$ ] where @xmath351 . it is easy to check that @xmath78 is in fact a group . [ compact]if @xmath0 is a compact uniform space and @xmath3 is an entourage in @xmath81 such that @xmath2 is chain connected then @xmath77 is finitely generated . let @xmath176 be an entourage in @xmath0 such that @xmath352 . since @xmath0 is compact there exists some finite @xmath176-dense set , i.e. , a set @xmath353 such that for every @xmath88 there exists some @xmath354 such that @xmath355 . we will first show that if @xmath356 is any @xmath3-chain with @xmath357 then @xmath358 is @xmath3-homotopic to an @xmath3-chain @xmath359 such that @xmath360 for all @xmath6 . by lemma [ chainhom ] we may assume that @xmath136 is in fact an @xmath176-chain . now for each @xmath361 , @xmath362 , there is some @xmath239 such that @xmath363 . we may now proceed iteratively , removing each @xmath361 and then replacing it by @xmath239 . for example , since @xmath364 , we may remove @xmath365 and still have an @xmath3-chain . since @xmath366 and @xmath367 , @xmath368 . similarly @xmath369 , so we may add @xmath370 . likewise , @xmath371 , so we may remove @xmath372 , and then @xmath373 and @xmath374 , so we may add @xmath375 . after finitely many steps we have the desired @xmath3-chain . we may suppose that @xmath376 . we now claim that @xmath77 is generated by elements of the form @xmath377_{e}\]]where @xmath378 for all @xmath6 , and for @xmath379 we have @xmath380 . we will call elements of this form minimal elements . that is , a minimal element is represented by a loop that consists of a chain made of distinct points of @xmath381 , followed by a loop ( which may be empty ) of additional distinct points of @xmath381 , followed by the initial chain in reverse order . if we prove this claim then the proof of the theorem is finished because @xmath381 , and hence the set of minimal elements , is finite . let @xmath382 be an arbitrary @xmath3-loop . we will show by induction on @xmath383 that @xmath145_{e}$ ] is a product of minimal elements . for @xmath384 the proof is trivial ; suppose it is true for @xmath385 . if all of the points @xmath239 @xmath386 are distinct then @xmath145_{e}$ ] is already a minimal element . otherwise , let @xmath387 be the smallest index such that for some @xmath388 , @xmath389 . let @xmath390we may apply the inductive hypothesis to conclude that @xmath391 _ { e}$ ] is the product of minimal elements . on the other hand , let @xmath392by construction , @xmath136 is minimal and since @xmath393 _ { e}=% \left [ \alpha \ast \beta \right ] _ { e}$ ] , the proof is finished . the following mapping is certainly well - defined . let @xmath0 be a uniform space and @xmath3 be an entourage . for each @xmath394_{e}\in \delta _ { e}(x)$ ] , define a mapping @xmath395 by @xmath396_{e})=[\lambda \ast \alpha ] _ { e}$ ] . before proceeding we will recall some notation and definitions from @xcite . let @xmath0 be a uniform space . we denote by @xmath397 the topological group of uniform homeomorphisms of @xmath0 with composition as the operation . suppose @xmath82 is a subgroup of @xmath397 . an entourage @xmath3 is called @xmath82__-invariant _ _ if @xmath398 for every @xmath399 . if @xmath0 has a uniformity base consisting of @xmath400-invariant entourages we say that @xmath82 acts isomorphically . the action of @xmath82 is said to be _ discrete _ provided there exists some entourage @xmath3 such that if @xmath401 for some @xmath88 and @xmath402 then @xmath403 must be the identity . if @xmath82 acts discretely and isomorphically on @xmath0 then the natural mapping @xmath404 is called a discrete cover . [ discrete]let @xmath0 be a uniform space and @xmath3 be an entourage . 1 . for any @xmath405 we have @xmath406 and @xmath78 is naturally isomorphic to a subgroup of @xmath407 . 2 . @xmath77 acts discretely and isomorphically on @xmath2 . 3 . if @xmath133 is surjective ( in particular if @xmath0 is chain connected ) then @xmath162 is a discrete cover with covering group @xmath77 . we have @xmath408_{e})=[\lambda _ { 1}\ast \lambda _ { 2}\ast \alpha ] _ { e}=\overline{\lambda _ { 1}\ast \lambda _ { 2}}([\alpha ] _ { e})\text{.}\]]this implies that each @xmath409 is a bijection and therefore is an element of @xmath407 , while the inclusion @xmath410 is a homomorphism . if @xmath411_{e})=[\alpha ] _ { e}$ ] for some @xmath136 , then @xmath412_{e}=[\lambda \ast \alpha ] _ { e}\ast \lbrack \alpha _ { e}]^{-1}=[\lambda \ast \alpha \ast \alpha ^{-1}]_{e}=[\lambda ] _ { e}\text{.}\]]this implies both that @xmath413 is injective and the action is free . the action of @xmath77 is free . by the very definition , for any entourage @xmath267 and all @xmath414 , @xmath415 and therefore the action is isomorphic . now suppose that @xmath416 for some @xmath417 . then @xmath418 and since @xmath133 is injective on every @xmath163-ball by proposition [ xee ] , @xmath419 . since we have shown the action is free , it follows that @xmath409 is the identity and we have shown that the action is discrete . for the third part note that by theorem 11 ( see also remark 13 ) , @xcite , we need only check first that @xmath133 is bi - uniformly continuous ( which we already know ) and second that if @xmath88 and @xmath420 are such that @xmath421 , then the preimage @xmath422 is precisely the orbit @xmath423 . for any @xmath88 , @xmath424_{e}\}\text{.}\]]now there is some @xmath3-chain @xmath425 since @xmath426 is surjective . moreover , for any @xmath62 , the endpoint of @xmath427_{e})$ ] is still @xmath40 and hence @xmath428_{e})\in \phi _ { xe}^{-1}(x)$ ] . that is , the orbit @xmath429_{e}$ ] is contained in @xmath422 . on the other hand , if @xmath137 is any other @xmath3-chain to @xmath40 and we let @xmath430 then @xmath431_{e}\in \delta _ { e}(x)$ ] and @xmath428_{e})=[\beta ] _ { e}$ ] and therefore @xmath422 is contained in the orbit @xmath432_{e}$ ] . from proposition [ identify ] we have : [ 0530cov]if @xmath0 is a uniform space then for any entourages @xmath95 in @xmath0 with @xmath267 such that @xmath433 is surjective ( in particular when @xmath2 is chain connected ) , @xmath14 is a discrete cover with covering group @xmath434 . in particular @xmath435 is a uniform homeomorphism if and only if @xmath434 is trivial . let @xmath0 be a uniform space and @xmath436 denote the restriction of @xmath14 to @xmath437 . the collection @xmath438 forms an inverse system . we denote @xmath439 by @xmath19 and call it the deck group of @xmath0 . [ chainhom2]if @xmath0 is a uniform space and @xmath3 is an entourage such that @xmath2 is chain connected then for any entourage @xmath267 , @xmath440 is surjective . according to lemma [ chainhom ] , if @xmath127 is an @xmath3-loop in @xmath0 , @xmath441 is @xmath3-homotopic to an @xmath176-chain @xmath136 . but since @xmath3-homotopy preserves endpoints , @xmath136 must also be an @xmath176-loop and by construction @xmath442_{f})=[\alpha ] _ { e}=[\gamma ] _ { e}$ ] . [ coin]let @xmath0 be a uniform space . then 1 . for any entourage @xmath3 , the group @xmath77 is discrete with respect to both the topology of uniform convergence and the topology induced by the inclusion of @xmath77 in @xmath2 . 2 . the group @xmath19 is prodiscrete with respect to the inverse limit topology , which is the same as the topology induced by the inclusion in @xmath21 . since @xmath77 acts discretely ( theorem [ discrete ] ) it follows from corollary 32 of @xcite that @xmath77 is discrete with respect to the topology of uniform convergence . on the other hand , since @xmath133 is injective on @xmath443 and @xmath444 , @xmath445 , which shows that @xmath77 is a discrete subset of @xmath133 . the second part now follows from the definitions . before we consider coverable spaces in more detail we need to revisit the issue of basepoints . note that the construction of @xmath21 is dependent on the initial choice of basepoint , and in fact @xmath21 itself may depend on the choice of the basepoint . for example , if one takes @xmath446 $ ] with the subspace metric then one can check that @xmath65 based at @xmath447 will consist of a single point , while @xmath65 based at any point in @xmath448 $ ] will be @xmath448 $ ] . the following lemma clears up this issue : [ baseok]suppose that @xmath0 is a uniform space such that for some choice of basepoint @xmath130 , the projection @xmath20 is surjective . then @xmath0 is chain connected . moreover , if @xmath0 satisfies the definition of coverable for @xmath130 then 1 . for any two basepoints there is a natural system of uniform homeomorphisms between the fundamental inverse systems of @xmath0 with respect to the two basepoints . in particular the spaces @xmath21 constructed with each basepoint are naturally uniformly homeomorphic . 2 . @xmath0 satisfies the definition of coverable for any choice of basepoint . if @xmath20 is surjective then since @xmath449 , @xmath133 is surjective . it now follows from lemma [ lemsur5 ] that @xmath0 is chain connected . the first statement now follows from the observations in remark [ basepoint ] . the second statement follows from the first ; essentially the two inverse systems are the same and so the projections are surjective in one if and only if they are surjective in the other . [ maincov]let @xmath0 be coverable and @xmath450 . then @xmath451 and @xmath452 comprise an inverse system of discrete covers and @xmath55 is a cover with covering group @xmath453 . since @xmath0 is coverable we may take @xmath454 for our indexing set . if @xmath64 is an @xmath176-loop and @xmath136 is an @xmath176-chain then @xmath455_{f}([\alpha ] _ { f}))=\phi _ { ef}([\lambda \ast \alpha ] _ { f})=[\lambda \ast \alpha ] _ { e}\]]@xmath456_{e}([\alpha ] _ { e})=\theta _ { ef}([\lambda ] _ { f})(\phi _ { ef}([\alpha ] _ { f})\text{.}\]]that is , the system is compatible in the sense of @xcite . since each @xmath457 is surjective , so is each @xmath14 and according to corollary 0530cov , each @xmath14 is a discrete cover . the proof is now finished by @xcite , theorem 44 . when @xmath0 is coverable we will refer to the projection @xmath458 as the universal covering map of @xmath0 . we will always take @xmath19 to have the ( prodiscrete ) inverse limit topology , with respect to which @xmath19 is complete and hausdorff . @xmath459 is also prodiscrete with respect to the topology of uniform convergence since @xmath50 is a cover ( see @xcite ) . it may be that the topology of uniform convergence is the the same as the induced topology , but we have no need for such a statement in this paper . [ 0607uniform]let @xmath0 be a uniform space and @xmath56 be a uniformity base . the following are equivalent for a fixed basepoint @xmath130 : 1 . @xmath56 is universal . 2 . @xmath460 and for any @xmath267 with @xmath461 , @xmath14 is a uniform homeomorphism . 3 . @xmath460 and for any @xmath57 , @xmath462 is a uniform homeomorphism . suppose that @xmath56 is universal and @xmath267 are entourages in @xmath463 . then @xmath433 satisfies @xmath464 and is therefore a uniform homeomorphism . since @xmath23 lies in any universal base , 1)@xmath4652 ) . if 2 ) holds then we may use @xmath56 as our indexing set for the fundamental inverse system of @xmath0 and since each of the bonding maps is a uniform homeomorphism the inverse system is in fact trivial . therefore each @xmath466 is a uniform homeomorphism and 3 ) follows . if 3 ) is given then note that for any @xmath57 we have that @xmath467 is a uniform homeomorphism and we see that @xmath56 is universal . definition [ universaldef ] implicitly involves the choice of a basepoint for the construction of the spaces @xmath2 . however , as the next corollary ( the proof of which is immediate from the proposition [ 0607uniform ] and lemma [ baseok ] ) shows that the definition is independent of basepoint . [ uic]any universal space is coverable , hence chain connected . in particular , if @xmath0 is universal with respect to one basepoint then @xmath0 is universal with respect to any basepoint . [ 0518u]if @xmath0 is universal and @xmath3 is an entourage such that @xmath2 is chain connected then @xmath162 is a uniform homeomorphism . let @xmath267 be an entourage in the universal base . since @xmath2 is chain connected , @xmath433 is surjective . since @xmath468 is bijective and @xmath469 , @xmath14 must also be injective . since @xmath2 is chain connected , lemma [ lemsur5 ] implies that @xmath14 is surjective , hence a uniform homeomorphism because @xmath14 is bi - uniformly continuous . therefore @xmath470 is a uniform homeomorphism . [ xic]if @xmath0 is coverable then @xmath21 is universal . applying proposition [ coverchain ] and lemma [ 41 ] we have that @xmath65 is chain connected . a basis element for the uniformity of @xmath65 is of the form @xmath471 , where @xmath267 are entourages in @xmath0 . let @xmath167 be a covering entourage contained in @xmath472 . then @xmath473 is surjective and @xmath474@xmath475 in other words , @xmath476 and we may take for our basis elements of @xmath21 entourages of the form @xmath477 . we will now apply corollary [ 67 ] , taking @xmath32 to be the surjective map @xmath478 . we have the following diagram : @xmath479according to proposition [ identify ] the function @xmath480 is a uniform homeomorphism and we may conclude that @xmath481 is also a uniform homeomorphism . the collection of all such @xmath82 is therefore a universal base for @xmath76 . [ uit]the following are equivalent for a coverable space @xmath0 . 1 . @xmath0 is universal . 2 . for every entourage @xmath3 such that @xmath2 is chain connected , @xmath482 is trivial ( i.e. , every @xmath3-loop based at @xmath130 is @xmath3-homotopic to the trivial loop ) . @xmath19 is trivial . the implication ( 1)@xmath465(2 ) follows from corollary [ 0518u ] . if ( 2 ) holds then for any coverable entourage @xmath3 , @xmath483 is a uniformly contionuous surjection and hence @xmath2 is chain connected . therefore each @xmath77 is trivial and the inverse limit @xmath19 is trivial . finally , if @xmath19 is trivial then the universal covering map is a uniform homeomorphism and since @xmath21 is universal , so is @xmath0 . we will see later ( theorem [ 0627simply ] ) that @xmath70 is universal . consider the entourage @xmath3 consisting of all @xmath484 such that @xmath485 , where @xmath486 . it is not hard to see that @xmath487 is an @xmath3-loop based at @xmath488 that is not @xmath3-homotopic to the trivial loop . at the same time it is true that @xmath489 is uniformly homeomorphic to @xmath490 , where @xmath41 has the discrete uniformity , and hence @xmath489 is not chain connected ( see example 48 in @xcite for more details ) this shows that one can not expect @xmath78 to be trivial for every choice of @xmath3 when @xmath0 is universal . if @xmath47 is a uniformly continuous bijection between coverable spaces @xmath0 and @xmath46 and @xmath46 is universal then @xmath0 is universal . suppose @xmath46 is universal and let @xmath176 be an entourage in the universal base for @xmath46 . then @xmath337 is a uniform homeomorphism and by corollary [ 67 ] @xmath162 is a uniform homeomorphism , where @xmath316 . [ disccover]let @xmath491 be a discrete cover , where @xmath0 is chain connected . for any sufficiently small @xmath82-invariant entourage @xmath3 and @xmath492 , the function @xmath211 is a uniform homeomorphism . since @xmath32 is a discrete cover , there exists an entourage @xmath167 such that if @xmath493 satisfies @xmath494 for some @xmath88 then @xmath495 . suppose that @xmath3 is any invariant entourage such that @xmath496 . since @xmath32 is a quotient mapping , @xmath497 is an entourage in @xmath46 . we will first show that @xmath498 is injective . by lemma [ lemsur2 ] we may equivalently prove that if @xmath499 is an @xmath3-chain in @xmath0 such that @xmath500 for some @xmath402 and the @xmath176-loop @xmath206 is @xmath176-homotopic to the trivial loop @xmath501 then @xmath495 ( so @xmath127 is an @xmath3-loop ) and @xmath127 is @xmath3-homotopic to the trivial loop @xmath502 . let @xmath503 . we will prove the statement by induction on the minimal length @xmath126 of an @xmath176-homotopy between @xmath206 and @xmath504 . if @xmath505 then @xmath206 is already trivial and so is @xmath441 , and the proof is finished . suppose we have proved it for some @xmath506 , and there is some @xmath176-homotopy of @xmath206 to @xmath504 of length @xmath126 . suppose that the first step in the @xmath176-homotopy is to add a point : @xmath206 is @xmath176-related to @xmath507 . that is , @xmath508 . now there exists @xmath509 such that @xmath510 and @xmath511 , and therefore some @xmath402 such that @xmath512 . since @xmath3 is invariant , we have that if @xmath513 then @xmath514 and @xmath515 . likewise there is some @xmath516 such that @xmath517 and @xmath518 . since @xmath519 , @xmath520 , and @xmath518 , it follows that @xmath521 . but since @xmath522 , @xmath523 for some @xmath524 and by choice of @xmath167 , it must be that @xmath525 and @xmath526 . the inductive hypothesis now finishes the proof . now suppose that the first step in the homotopy is to remove a point : @xmath206 is @xmath176-related to @xmath527 ; that is , @xmath528 . as in the preceding argument there is some @xmath493 such that @xmath529 . but @xmath530 , so @xmath531 , and we conclude that @xmath532 . therefore @xmath533 and we may again apply the inductive hypothesis . to see why @xmath279 is surjective , let @xmath534 be an @xmath176-chain in @xmath46 . we will prove the statement by induction in @xmath86 . if @xmath298 then the proof is obvious since @xmath535 by assumption . suppose we have proved it for @xmath536 . then we can find some @xmath3-chain @xmath537 such that @xmath538 for all @xmath6 . since @xmath539 , and @xmath540 , we may again use the invariance of @xmath3 to see that there exists some @xmath541 such that @xmath542 and @xmath543 . then @xmath115 is an @xmath3-chain such that @xmath544 . to finish the proof of the proposition , let @xmath141 be an entourage in @xmath545 , where @xmath140 is an invariant entourage in @xmath0 ( @xmath167 exists since @xmath82 acts isomorphically ) . since @xmath0 is chain connected , @xmath546 by proposition [ xee ] , and @xmath547 is an entourage in @xmath548 . the proof will be complete if we show that @xmath549 , which makes @xmath550 an entourage . let @xmath142_{f},[\beta ] _ { f})\in k^{\ast } $ ] which means @xmath143_{f},[\beta ] _ { f})=([\ast = y_{0}, ... ,y_{n},y]_{f},[\ast = y_{0}, ... ,y_{n},z]_{f})\]]with @xmath551 . using the invariance of @xmath3 and @xmath167 and proceeding inductively as we have done above , we can find @xmath3-chains @xmath552 and @xmath553 such that @xmath554 and @xmath555 , and @xmath556 . by definition , @xmath557_{e},[\beta ^{\prime } ] _ { e})\in d^{\ast } $ ] and @xmath558_{e}),f_{ef}([\alpha ^{\prime } ] _ { e})\right ) = ( [ \alpha ] _ { f},[\beta ] _ { f})$ ] . [ base2]we give one final comment about basepoints and lifts of functions following lemma [ baseok ] . the lifting theorems below are true for any choice of basepoints such that the functions involved are basepoint - preserving . for example , in the proposition below we may start with a basepoint @xmath130 in @xmath0 , choose @xmath535 in @xmath46 , use any basepoint @xmath559 to construct @xmath560 and then choose another basepoint @xmath130 in @xmath560 so that @xmath561 . [ liftlem]let @xmath0 be universal , @xmath46 be uniform and @xmath47 be uniformly continuous . then 1 . for any entourage @xmath3 in @xmath46 there is a unique uniformly continuous function @xmath562 such that @xmath563 and @xmath564 . there is a unique uniformly continuous function @xmath565 such that @xmath566 and @xmath567 , where @xmath568 is the projection . define @xmath562 as follows . let @xmath176 be an entourage in the universal base of @xmath0 such that @xmath569 and @xmath570 be the unique uniformly continuous function given by theorem [ 60 ] ( @xmath571 is a uniform homeomorphism and therefore @xmath311 is chain connected ) . define @xmath572 . if @xmath573 were any such function then it follows from the uniqueness of @xmath574 that @xmath575 and hence that @xmath576 . note that by uniqueness , if @xmath577 are entourages in @xmath46 we have that @xmath578 and by the universal property of the inverse limit there is a unique function @xmath59 such that @xmath579 for every entourage @xmath3 and @xmath580 . suppose that @xmath581 is uniformly continuous such that @xmath582 and @xmath583 . note that for any entourage @xmath3 in @xmath46 we have @xmath584and therefore by uniqueness in part ( 1 ) , @xmath585 . since @xmath586 is also induced by the functions @xmath587 , @xmath588 . the functions @xmath589 and @xmath590 will both be referred to as lifts of @xmath32 [ 76]let @xmath0 and @xmath46 be uniform , @xmath47 be a cover , @xmath34 be universal , and @xmath591 be uniformly continuous . then there exists a unique uniformly continuous function @xmath592 such that @xmath593 and @xmath594 . suppose first that @xmath32 is a discrete cover . according to proposition disccover there is an entourage @xmath3 in @xmath0 such that if @xmath492 then @xmath595 is a uniform homeomorphism and @xmath290 . define @xmath592 by @xmath596 ( where @xmath597 is the lift of @xmath403 given by proposition [ liftlem ] ) . then @xmath594 and @xmath598@xmath599to prove uniqueness , suppose that @xmath600 is any such function . consider the lift @xmath601 . we have @xmath602by uniqueness of lifts ( proposition [ liftlem ] ) , @xmath603 . but then @xmath604 and @xmath605 now suppose that @xmath32 is an arbitrary cover . by theorem 48 in @xcite there exists an inverse system @xmath606 such that @xmath607 is a discrete cover and @xmath608 and @xmath609 for some minimal element @xmath74 . according to what we proved above , for each @xmath136 there is a unique uniformly continuous function @xmath610 such that @xmath611 and @xmath612 . if @xmath613 we have that @xmath614and by uniqueness @xmath615 . by the universal property of inverse limits there is a unique uniformly continuous function @xmath616 such that for all @xmath136 , @xmath617 and @xmath594 . now suppose @xmath618 is any uniformly continuous function such that @xmath619 and @xmath620 . define @xmath621 . we have that @xmath622 and @xmath623by the uniqueness of @xmath624 , @xmath625 and therefore @xmath626 . if @xmath46 is coverable then by theorem [ xic ] , @xmath560 is universal and we obtain : [ partialstep]let @xmath47 be a cover where @xmath0 is uniform and @xmath46 is coverable , and @xmath568 be the projection . then there exists a unique uniformly continuous function @xmath627 such that @xmath628 and @xmath629 . [ quick]if @xmath47 is a bi - uniformly continuous surjection where @xmath0 is universal and @xmath46 is uniform then @xmath46 is coverable . let @xmath56 be a universal base for @xmath0 . since @xmath32 is a bi - uniformly continuous surjection , the set @xmath630 of all @xmath631 such that @xmath632 is a base for the uniformity of @xmath46 . we may index the fundamental system for @xmath46 using @xmath630 , and for each @xmath633 in @xmath634 we may use @xmath176 in the construction of the lift @xmath589 . since @xmath635 we will be finished by the definition of coverable if we can show that each @xmath589 is surjective . let @xmath636 be an @xmath3-chain in @xmath46 . we will show by induction on @xmath86 that there is an @xmath176-chain @xmath637 such that @xmath638 and , by definition of @xmath589 , this completes the proof . for @xmath298 the proof is obvious . now suppose we have an @xmath176-chain @xmath639 such that @xmath640 . since @xmath641 there is some ordered pair @xmath642 such that @xmath643 . the @xmath176-chain @xmath644 now satisfies @xmath645 . [ bi]if @xmath0 is coverable and @xmath47 is a bi - uniformly continuous surjection then @xmath46 is coverable . in particular , any quotient by an equiuniform action , hence by an isomorphic action , on a coverable uniform space is coverable . since the universal cover @xmath20 is a bi - uniformly continuous surjection , so is @xmath646 . the proof is finished by corollary [ quick ] and theorem [ xic ] . [ 73]let @xmath26 be coverable spaces , @xmath47 be uniformly continuous , and @xmath20 and @xmath647 be the projections . then there is a unique uniformly continuous function @xmath30 , such that @xmath648 and @xmath649 . moreover , 1 . for any @xmath650 and @xmath651 , @xmath652 . 2 . the restriction @xmath653 of @xmath33 to @xmath19 is a homomorphism into @xmath654 . if @xmath32 is a discrete cover then @xmath33 is a uniform homeomorphism . 4 . if @xmath34 is uniform and @xmath35 is uniformly continuous then @xmath655 ( and in particular @xmath656 ) . for the main statement , define @xmath657 . if @xmath658 is a uniformly continuous function with @xmath659 then by definition @xmath403 is a lift of @xmath660 and so @xmath661 . for part ( 1 ) , note that if @xmath662 then @xmath663 and @xmath664therefore @xmath665 . for any entourage @xmath176 in @xmath46 let @xmath3 be an entourage in @xmath0 such that @xmath207 . now @xmath666 and @xmath667 are both lifts of @xmath660 to @xmath282 and therefore @xmath668 ( see proposition [ liftlem ] ) . now let @xmath669 and @xmath650 with @xmath670 _ { e}$ ] and @xmath671 _ { e}$ ] . by definition of the action of the inverse limit group @xmath82 on the inverse limit space @xmath21,@xmath672 _ { e}\right ) \text{.}\]]the latter quantity , by definition of @xmath279 , is equal to@xmath673 _ { f}=\left [ f(\gamma _ { e})\ast f(\alpha _ { e})\right ] _ { f}=\left [ f(\gamma _ { e})\right ] _ { f}\ast \left [ f(\alpha _ { e})\right ] _ { f}\text{.}\]]as a special case when the @xmath403 is the identity , @xmath674_{f}\right ) \text{.}\]]if @xmath675 then @xmath676 _ { f}\text{.}\]]combining these we obtain @xmath677when @xmath662 , we have@xmath678which gives the second statement . if @xmath491 is a discrete cover then @xmath32 is bi - uniformly continuous and we may index the fundamental system of @xmath46 using entourages of the form @xmath679 where @xmath3 is an invariant entourage in @xmath0 . then by uniqueness , @xmath33 is induced by the functions @xmath279 . proposition [ disccover ] implies that the functions @xmath279 are all uniform homeomorphisms and hence @xmath33 must be a uniform homeomorphism . the last part follows from uniqueness of @xmath680 . [ 101]if @xmath0 and @xmath46 are coverable spaces and @xmath47 is a cover then @xmath30 is a uniform homeomorphism . if @xmath0 is coverable then @xmath681 is a cover with covering group @xmath682 . let @xmath20 and @xmath28 be the projections . we will show that the lift @xmath683 given by corollary [ partialstep ] and proposition [ liftlem ] is an inverse to @xmath33 . first note that @xmath684that is , @xmath685 is the unique lift of the identity on @xmath46 and hence must be the identity on @xmath560 . now according to theorem [ 76 ] , @xmath50 is the unique lift of the function @xmath660 ; that is , the unique uniformly continuous function @xmath686 such that such that @xmath687 and @xmath688 . but we also have that @xmath689 in other words , @xmath690 . we now have @xmath691and @xmath692 must be the unique lift of the identity on @xmath0 , hence the identity on @xmath21 . we have shown that both @xmath693 and @xmath33 are uniform homeomorphisms and inverses of one another . for the second part note that since @xmath0 is coverable , @xmath50 is bi - uniformly continuous and since @xmath694 , @xmath695 is bi - uniformly continuous . now @xmath696 for @xmath697 if and only @xmath698 , or equivalently , letting @xmath699 and @xmath700 there is some @xmath651 such that @xmath701 . but this is equivalent to @xmath702(the second equality comes from theorem [ 73 ] ( 1 ) ) . that is , the orbits of @xmath703 are precisely the preimages of points with respect to @xmath695 . according to @xcite this means that @xmath695 is the quotient with respect to the action of @xmath703 . finally , since @xmath704 , which acts prodiscretely and isomorphically , so does @xmath703 and @xmath705 is a cover with covering group @xmath703 ( see remark 13 in @xcite ) . since a universal space is uniformly homeomorphic to its own universal cover by proposition [ 0607uniform ] , we obtain : the universal cover of a coverable space is unique up to uniform homeomorphism . more precisely , if @xmath0 is coverable , @xmath46 is universal , @xmath706 is a cover and @xmath20 is the universal cover then @xmath707 is a uniform homeomorphism . we will say that a topological space @xmath0 is simply connected if every loop in @xmath0 is null - homotopic , regardless of whether @xmath0 is pathwise connected . recall that @xmath0 is called locally connected ( resp . locally pathwise connected ) if for every @xmath88 and open set @xmath90 containing @xmath40 there is a connected ( resp . pathwise connected ) open set @xmath708 with @xmath709 . @xmath0 is semilocally simply connected if each @xmath88 is contained in an open set @xmath90 such that every loop in @xmath90 based at @xmath40 is null - homotopic in @xmath0 ( see @xcite ) . in a uniform space it is natural to consider the situation when these local conditions are true uniformly . [ unidef]a uniform space @xmath0 is called uniformly locally connected ( resp . uniformly locally pathwise connected ) if for each entourage @xmath3 there is an entourage @xmath267 such that all @xmath176-balls are open and connected ( resp . open and pathwise connected ) . @xmath0 is called uniformly semilocally simply connected if there exists an entourage @xmath3 such that any loop in @xmath710 based at @xmath88 is null - homotopic in @xmath0 . note that given any @xmath3 as in the above definition , any entourage @xmath711 has the same property . a uniform space @xmath0 is uniformly locally connected ( resp . uniformly locally pathwise connected ) if and only if for every entourage @xmath3 there exists an entourage @xmath176 with open balls such that for every @xmath88 there exists some open set @xmath712 such that @xmath713 and @xmath712 is connected ( resp . @xmath90 is pathwise connected ) . necessity is trivial ; take an entourage @xmath176 with open balls and let @xmath714 . to prove the converse , let @xmath3 be any entourage in @xmath0 and @xmath715 be an entourage such that @xmath716 . let @xmath167 be an entourage such that for each @xmath88 there exists an open set @xmath712 that is connected ( resp . pathwise connected ) , such that @xmath717 . define @xmath718 by @xmath719@xmath176 is symmetric by definition and certainly contains the diagonal . now @xmath720which , being the union of connected ( resp . pathwise connected ) open sets @xmath721 containing @xmath40 , is open and connected ( resp . pathwise connected ) . @xmath176 is an entourage since by definition @xmath283 . finally , if @xmath722 then @xmath723 and @xmath724 for some @xmath725 containing @xmath40 . since @xmath726 , @xmath727 . [ 428unif]let @xmath0 be a locally connected ( resp . locally pathwise connected ) compact topological space . then @xmath0 , with the unique uniformity compatible with its topology , is uniformly locally connected ( resp . uniformly locally pathwise connected ) . if @xmath0 is both locally pathwise connected and semilocally simply connected then @xmath0 is uniformly semilocally simply connected . recall that the unique compatible uniformity on @xmath0 has as a basis all symmetric open subsets of @xmath23 that contain the diagonal . let @xmath3 be any entourage in @xmath0 and for any @xmath88 let @xmath712 be an open and connected ( resp . pathwise connected ) neighborhood of @xmath40 such that @xmath728 . let @xmath267 be the union of all sets @xmath729 . then @xmath176 is open and symmetric , hence an entourage . if @xmath730 then @xmath731 for some @xmath732 , and therefore @xmath733 and @xmath734 both lie in @xmath735 . that is , @xmath736 is the union of all sets @xmath737 such that @xmath738 . since each @xmath735 is connected ( resp . pathwise connected ) , so is @xmath736 . if @xmath0 is both locally pathwise connected and semilocally simply connected then by what we proved above , for every @xmath88 there is an arbitrarily small entourage @xmath739 such that the ball @xmath740 is open and pathwise connected . by choosing @xmath739 small enough we may assume that every loop in @xmath740 based at @xmath40 is null - homotopic in @xmath0 . for each @xmath40 let @xmath741 be an entourage with open balls such that @xmath742 . let @xmath743 be a finite open cover of @xmath0 , and let @xmath744 . for any @xmath88 there is some @xmath745 such that @xmath746 . if @xmath723 then @xmath747 and @xmath748 . that is , @xmath749 . given any loop @xmath127 in @xmath736 based at @xmath40 , join it to @xmath745 by a path in @xmath750 from @xmath40 to @xmath745 . the resulting loop based at @xmath745 is null - homotopic in @xmath0 , and hence so is @xmath127 . [ 315conn]let @xmath0 be a chain connected uniform space . if there is an entourage @xmath3 of @xmath0 such that the @xmath3-balls of @xmath0 are connected ( resp . pathwise connected ) then @xmath0 is connected ( resp . pathwise connected ) . it will follow from proposition [ consume ] and a standard theorem from topology if we show by induction that all @xmath85-balls are connected . the @xmath751 case is given . suppose that @xmath85-balls and @xmath3-balls are all conneced for some @xmath86 . let @xmath752 . by definition of @xmath753 there is some @xmath732 such that @xmath754 and @xmath755 . each of these balls is connected and they intersect in @xmath732 , hence their union is connected set . we have shown that every element of @xmath756 is contained in a connected subset of @xmath756 containing @xmath40 , and so @xmath753 is connected . if the @xmath3-balls are all pathwise connected then by what we have just proved @xmath0 is connected and locally pathwise connected , hence pathwise connected . if @xmath0 is chain connected and uniformly locally connected ( resp . uniformly locally pathwise connected ) then @xmath0 is connected ( resp . pathwise connected ) . [ 52]let @xmath0 be a uniform space and @xmath3 be an entourage in @xmath0 such that the @xmath3-balls have one of the following properties : chain connected , connected , pathwise connected . then @xmath2 has the same property . suppose that the @xmath3-balls of @xmath0 are chain connected . let @xmath267 be an entourage and @xmath637 be an @xmath3-chain . since the @xmath3-balls are chain connected we may suppose up to @xmath3-homotopy that @xmath194 is an @xmath176-chain . for example , we may join @xmath118 and @xmath757 by an @xmath176-chain @xmath758 that lies entirely in @xmath759 . we may remove the points @xmath760 one at a time in order to obtain an @xmath3-homotopy from @xmath761 to @xmath136 . then letting @xmath762 we have that @xmath763_{e},[\alpha _ { i+1}]_{e})\in f^{\ast } $ ] for all @xmath6 . in particular , @xmath764_{e}, ... ,[\alpha _ { n}]_{e}=[\alpha ] _ { e}\}$ ] is an @xmath219-chain to @xmath202_{e}$ ] . this shows that @xmath2 is chain connected . now if @xmath185 has connected ( resp . pathwise connected ) balls then these balls are chain connected and therefore @xmath2 is chain connected by what we just proved . according to proposition [ xee ] , the @xmath163-balls in @xmath2 are connected ( resp . pathwise connected ) , and lemma [ 315conn ] now shows that @xmath2 is connected ( resp . pathwise connected ) . [ 0627simply]if @xmath0 is a uniformly locally pathwise connected , connected and simply connected uniform space then @xmath0 is universal . let @xmath3 be an entourage with pathwise connected open balls . then @xmath2 is pathwise connected by proposition [ 52 ] . moreover , by corollary 34 of @xcite , the action of @xmath77 on @xmath2 is properly discontinuous . since @xmath0 is a poincar space , @xmath765 is a traditional cover ( cf . @xcite ) and since @xmath0 is simply connected , @xmath133 must be a trivial cover , hence bijective . this means that the bi - uniformly continuous mapping @xmath133 is a uniform homeomorphism . [ compactu]every compact , connected , locally pathwise connected , simply connected topological space is universal . proposition [ liftlem ] now implies : [ pathlift]if @xmath766\rightarrow x$ ] is a ( continuous ) path , where @xmath0 is uniform , with @xmath767 , then there is a unique lift @xmath768\rightarrow \widetilde{x}$ ] such that @xmath769 and @xmath770 , where @xmath20 is the projection . a similar statement holds for homotopies . if @xmath766\rightarrow x$ ] is a path from @xmath130 to @xmath88 then @xmath771 . we have : [ pathsur]if @xmath0 is a uniform space with projection @xmath458 then the pathwise connected component of @xmath0 at @xmath130 is contained in @xmath772 , where @xmath773 is the pathwise connected component of @xmath774 in @xmath21 . [ 0628new]if @xmath0 is a connected , uniformly locally pathwise connected uniform space then @xmath0 is coverable . in particular , every compact , connected , locally pathwise connected topological space is coverable . we may index the fundamental system of @xmath0 using entourages having pathwise connected balls . but for each such @xmath3 , @xmath2 is pathwise connected by proposition [ 52 ] and therefore each @xmath775 is surjective by the previous corollary . [ 0518simp]suppose @xmath0 is a uniformly locally pathwise connected uniform space having an entourage @xmath3 such that for all @xmath88 every loop in @xmath776 is null - homotopic in @xmath0 and an entourage @xmath176 having pathwise connected balls such that @xmath777 . then @xmath347 is pathwise connected , uniformly locally pathwise connected and simply connected , hence universal . according to proposition [ 52 ] , @xmath347 is pathwise connected , and the @xmath778-balls , which are uniformly homeomorphic to the @xmath176-balls , are pathwise connected . by proposition [ identify ] the function @xmath779 is a uniform homeomorphism . this implies that every @xmath219-loop in @xmath347 is @xmath219-homotopic to the trivial loop . now let @xmath766\rightarrow x_{f}$ ] be a loop based at @xmath130 . since @xmath317 is uniformly continuous there is some @xmath219-loop @xmath780 such that each @xmath781 for some @xmath6 and @xmath782)\subset b(x_{i},f^{\ast } ) $ ] for all @xmath6 . we may use any @xmath219-homotopy from @xmath441 to @xmath130 to construct a null - homotopy from @xmath317 to the trivial loop as follows . suppose that @xmath783 for some @xmath784 . we may join @xmath745 to @xmath734 and @xmath734 to @xmath785 by paths @xmath786 and @xmath787 in @xmath788 . let @xmath789 be the restriction of @xmath317 to @xmath790 $ ] . then we have a loop @xmath791 , each segment of which lies in the @xmath219-ball centered at its endpoint . now let @xmath792 , which is a loop consisting of three segments , each of which lies in the @xmath176-ball centered at its endpoint . since @xmath777 , @xmath793 lies entirely in the @xmath3-ball centered at a point , and hence is null - homotopic in @xmath0 . according to theorem [ 76 ] any null - homotopy of @xmath794 lifts to a null - homotopy of @xmath137 . in other words , the path @xmath317 is homotopic to a path @xmath795 obtained by replacing @xmath789 by @xmath786 concatenated by @xmath787 . we may carry out a similar process when removing any point in @xmath127 to form a new @xmath219-chain as part of an @xmath219-homotopy of @xmath127 . finitely many such steps show that @xmath317 is null - homotopic . since the @xmath176-balls of @xmath0 , and hence the @xmath219-balls of @xmath347 , are pathwise connected , @xmath347 is uniformly locally pathwise connected and pathwise connected , hence universal by theorem 0627simply . let @xmath766\rightarrow x$ ] be a path in a coverable space @xmath0 such that @xmath797 . since @xmath798 , @xmath799 , where @xmath50 is the universal covering map of @xmath0 and @xmath800 is the lift given by proposition [ liftlem ] that is , @xmath801 . define a function @xmath80 by @xmath802)=c_{l}(1)$ ] , where @xmath803 $ ] is the homotopy equivalence class of a loop @xmath317 based at @xmath130 . this function is well defined . in fact , suppose @xmath317 and @xmath335 are loops based at @xmath130 that are homotopic via a homotopy @xmath804\times \lbrack 0,1]\rightarrow x$ ] , with @xmath805 and @xmath806 . then @xmath807 is a path joining @xmath808 to @xmath809 , both of which lie in @xmath810 . but @xmath811 is a prodiscrete group ( hence totally disconnected ) with respect to the subspace topology . therefore @xmath812 . [ lifter]let @xmath0 be coverable , @xmath803\in \pi _ { 1}(x)$ ] , where @xmath813\rightarrow x$ ] is based at @xmath130 , and @xmath3 be an entourage in @xmath0 . if @xmath814 are such that for every @xmath6 , @xmath815 for all @xmath816 $ ] then @xmath817))=[\ast = c(t_{0}), ... ,c(t_{n})]_{e}$ ] . ( such values @xmath818 always exist since @xmath317 is uniformly continuous . ) note that by definition,@xmath819))=\theta _ { e}(c_{l}(1))=c_{e}(1)\]]where @xmath820 is the lift given by proposition [ liftlem ] . let @xmath821 , where the points @xmath818 satisfy the conditions of the proposition . we will show by induction on @xmath6 that @xmath822_{e}$ ] for all @xmath246 . the case @xmath823 is trivial . suppose that we have proved the statement for some @xmath6 , @xmath117 . by definition , @xmath824_{e},[c(t_{0}),c(t_{1}), ... ,c(t_{i}),c(t_{i+1})]_{e})\in e^{\ast } \]]and @xmath825_{e})=c(t_{i+1})$ ] . recall from proposition [ xee ] that @xmath133 is a uniform homeomorphism when restricted to any @xmath163-ball of @xmath2 , and since we also have @xmath826 we will be finished if we can show @xmath824_{e},c_{e}(t_{i+1}))\in e^{\ast } \text{.}\]]but the unique lift @xmath827 of @xmath828}$ ] to @xmath2 starting at @xmath829 must be @xmath830}$ ] , where @xmath208 is the inverse of @xmath133 restricted to @xmath831_{e},e^{\ast } ) $ ] . since @xmath832}$ ] also satisfies the conditions for this lift , we must have that @xmath833_{e},e^{\ast } ) \text{.}\ ] ] [ nat]if @xmath0 is coverable then the natural mapping @xmath834 is a homomorphism , the image of which is the normal subgroup @xmath835 of @xmath19 that leaves invariant the pathwise connected component of @xmath21 containing @xmath130 . in particular , @xmath62 is surjective if and only if @xmath21 is pathwise connected . to see why @xmath62 is a homomorphism note that @xmath836)\circ d_{l}\right ) ) = c\ast d$ ] ( here @xmath837)\circ d_{l}$ ] is the translate via the uniform homeomorphism @xmath837)$ ] of @xmath838 to the endpoint of @xmath800 and @xmath839 is the universal covering map of @xmath0 ) and therefore by uniqueness , @xmath840)\circ d_{l}\right ) = \left ( c\ast d\right ) _ { l}$ ] . but the endpoint of @xmath841)\circ d_{l}$ ] is @xmath842)\lambda ( [ d])$ ] . if @xmath651 stabilizes the pathwise connected component of @xmath21 containing @xmath774 then @xmath130 and @xmath843 are joined by an path @xmath136 , and @xmath844 is a loop based at @xmath130 with @xmath845)=g$ ] . on the other hand , if @xmath403 lies in the image of @xmath62 then @xmath130 and @xmath846 are joined by an path ( namely the lift of a loop , the image of whose equivalence class is @xmath403 ) . now suppose @xmath40 lies in the pathwise connected component of @xmath21 containing @xmath130 . we may join @xmath40 to @xmath130 by an path @xmath317 . but then @xmath847 joins @xmath848 and @xmath843 , and we have already observed that the latter is joined to @xmath130 by an path . the kernel of @xmath62 , by definition , consists of those @xmath803 $ ] such that @xmath849 lifts to a loop in @xmath21 , or equivalently the lift of @xmath317 to each @xmath2 is a loop . it is of obvious interest when @xmath850 is trivial . this can be checked in some special cases . for example , in the case of the hawaiian earring , a problem that essentially amounts to the injectivity of @xmath62 occupied several papers ( @xcite , @xcite , d ) and was simply stated in @xcite ( along with an incorrect statement that amounts to surjectivity of @xmath851 . [ inj]let @xmath0 be coverable . then 1 . if @xmath21 is simply connected then @xmath834 is injective . 2 . if @xmath0 is pathwise connected and @xmath61 is injective then @xmath21 is simply connected . if @xmath21 is simply connected and @xmath803\in \ker \lambda $ ] then @xmath317 lifts to a loop @xmath800 in @xmath21 based at @xmath130 , which is then null - homotopic . the image of any null - homotopy of @xmath800 via the universal covering map @xmath50 is a null - homotopy of @xmath317 . now suppose that @xmath0 is pathwise connected and @xmath834 is injective . let @xmath766\rightarrow \widetilde{x}$ ] be a loop based at some point @xmath40 . then @xmath852 is a loop in @xmath0 , where @xmath50 is the universal covering map of @xmath0 . let @xmath853\rightarrow x$ ] be the concatenation of a path @xmath136 from @xmath130 to @xmath854 , followed by @xmath855 , parameterized so that the restriction to @xmath856 $ ] is @xmath852 . so @xmath335 represents an element of @xmath857 . let @xmath651 be such that @xmath858 ( such a @xmath403 exists since @xmath859 ) . now the composition of @xmath403 with @xmath860}$ ] is simply @xmath317 and therefore @xmath861}$ ] is a loop . but then @xmath862)=0 $ ] and since @xmath62 is injective , @xmath335 must be null - homotopic . but then any null - homotopy of @xmath335 lifts to one of @xmath838 , and the composition of the lifted homotopy with @xmath403 gives rise to a null - homotopy of @xmath317 . [ dense]if @xmath0 is a uniformly locally pathwise connected , connected uniform space then the pathwise connected component of @xmath21 is dense in @xmath21 and @xmath863 is dense in @xmath453 . these two statements follow from the following general result concerning induced functions on inverse limits : let @xmath864 be an inverse system of topological spaces with continuous bonding maps and @xmath865 . let @xmath866 be a collection of continuous surjections from a topological space @xmath46 such that @xmath867 for all @xmath613 . then the induced mapping @xmath868 has dense image in @xmath0 . we do not have a reference for this exact statement but the proof is straightforward and similar to the proof of iii.7.3 proposition 2 in @xcite . now if @xmath3 has pathwise connected balls then @xmath2 is pathwise connected and by corollary [ pathsur ] the restriction @xmath793 of @xmath25 to the pathwise connected component @xmath773 of @xmath21 is surjective onto @xmath2 and the proof of the first part is finished by the above general statement . on the other hand , if @xmath869 then there is some path @xmath136 from @xmath130 to @xmath40 , and @xmath870 is a loop @xmath127 in @xmath0 based at @xmath130 such that the unique lift @xmath871 has @xmath40 as its endpoint . but then @xmath872)=x$ ] . in other words , @xmath873 is surjective , and the proof is finished by the above general statement . note that the above proof really only requires that there be a basis for the uniformity of @xmath0 such that for each @xmath3 in the basis , @xmath2 is pathwise connected . since the closure of a connected set is connected we have : if @xmath0 is a uniformly locally pathwise connected , connected uniform space then @xmath21 is connected . [ newquotient]if @xmath0 is a locally pathwise connected and connected space and @xmath773 is the pathwise connected component of @xmath21 containing @xmath130 then @xmath0 is the quotient of @xmath773 by the free isomorphic action of @xmath874 . we already know that @xmath863 acts freely and isomorphically on @xmath773 since @xmath19 does . we need to check that the restriction @xmath794 of @xmath50 to @xmath773 is bi - uniformly continuous and that for any @xmath875 , @xmath876 is precisely the orbit @xmath877 of @xmath40 ( see remark 13 of @xcite ) . we know that @xmath793 is uniformly continuous . let @xmath878 be an entourage in @xmath773 ; we may assume that the @xmath3-balls are pathwise connected , hence @xmath2 is pathwise connected , hence the restriction @xmath879 of @xmath25 to @xmath773 is surjective by corollary [ pathsur ] . since @xmath879 is surjective , @xmath880now @xmath881which is an entourage since @xmath133 is bi - uniformly continuous . this shows that @xmath793 is bi - uniformly continuous . the statement about the orbits simply follows from the fact that the orbits of @xmath793 are precisely the orbits of @xmath50 intersected with @xmath773 and that @xmath882 for any @xmath883 . note that when @xmath62 is injective , the action in the above proposition is in fact an action by @xmath884 . on the other hand , when @xmath62 is not surjective one has the disadvantage that @xmath773 is not complete with the uniformity induced by @xmath21 . a uniform space @xmath0 is called strongly coverable if @xmath0 is chain connected and for some entourage @xmath3 , @xmath2 is universal , hence the universal cover of @xmath0 . it is clear from the discussion in the introduction that the topologist s sine curve is strongly coverable . [ strong]if @xmath0 is strongly coverable and uniformly locally pathwise connected then @xmath80 is surjective . let @xmath3 be an entourage such that @xmath2 is universal and @xmath267 be an entourage having pathwise connected balls . then @xmath347 is pathwise connected and @xmath219 is in the universal base of @xmath2 by corollary [ 0518u ] . hence @xmath2 and @xmath347 are uniformly homeomorphic and therefore @xmath885 is also pathwise connected . the proof is finished by theorem [ nat ] . from theorem [ 0518simp ] , lemma [ strong ] , and proposition [ inj ] we obtain : every connected , uniformly locally pathwise connected and uniformly semi - locally simply connected uniform space @xmath0 is strongly coverable with pathwise connected , simply connected universal cover and @xmath886 . [ 0518equiv]if @xmath0 is a compact poincar space then @xmath21 is the traditional universal cover of @xmath0 and @xmath834 is an isomorphism . we refer the reader to @xcite for more background on dimension and uniform spaces . suppose that @xmath0 has uniform dimension @xmath73 , which we denote by @xmath887 . this means that any uniform open cover @xmath888 of @xmath0 has a refinement by a uniform open cover of order @xmath889 ( a uniform open cover is an open cover that is refined by the cover of @xmath0 by @xmath176-balls for some fixed entourage @xmath176 ) . note that this particular notion of dimension is called large dimension in @xcite , and is denoted by @xmath890 . we will use theorems from @xcite concerning another dimension , called uniform dimension in @xcite and denoted by @xmath891 . this particular notion of dimension uses finite covers and is somewhat more difficult to work with in the present situation . however , it is always true that @xmath892 ( theorem v.5 in @xcite in fact the dimensions are equal if @xmath890 is finite ) and therefore if we know that @xmath893 then we may use theorems from @xcite that require @xmath894 . if @xmath0 is compact then both of these dimensions are equal to covering dimension , which we denote by @xmath895 . [ dimprop]if @xmath0 is a uniform space with @xmath893 then for any entourage @xmath3 , @xmath896 . every uniform open cover of @xmath2 is refined by the cover of @xmath2 by @xmath778-balls for some entourage @xmath176 such that @xmath777 and therefore we need only consider the open cover of @xmath2 by @xmath219-balls for such @xmath176 . by definition of uniform dimension , the cover of @xmath0 by @xmath176-balls has a refinement by a uniform open cover @xmath888 such that every @xmath88 is contained in at most @xmath889 sets in @xmath888 . let @xmath897 . then @xmath898 for some @xmath88 . let @xmath899 be the collection of all sets of the form @xmath900 where @xmath901 and @xmath902 . first , @xmath903 is a cover . in fact , if @xmath904 , @xmath905 for some @xmath906 , with @xmath898 . but @xmath133 restricted to @xmath907 is a uniform homeomorphism onto @xmath908 , which contains @xmath909 since @xmath777 . therefore there is some @xmath910 . but @xmath911 and @xmath912 ; therefore @xmath913 . that is , @xmath914 . since @xmath888 is a uniform cover , so is @xmath903 . in fact , if the @xmath167-ball cover of @xmath0 refines @xmath915 for some @xmath283 then the @xmath141-ball cover refines @xmath916 . by definition , @xmath903 refines the @xmath219-ball cover of @xmath2 . finally , to check that the order of @xmath903 is at most @xmath889 , we need only check that if @xmath917 are distinct and @xmath918 then @xmath919 \cap \left [ \phi _ { xe}^{-1}(a)\cap b(z , f^{\ast } ) \right ] = \varnothing \text{.}\]]but @xmath788 and @xmath920 are already disjoint , because otherwise for any @xmath921 we would have @xmath922 , which contradicts the fact that @xmath133 is injective on @xmath163-balls . in @xcite , theorem iv.32 it is shown that the inverse limit of spaces with uniform dimension at most @xmath86 must have uniform dimension at most @xmath86 . therefore by proposition [ dimprop ] : [ dimthm]if @xmath0 is a coverable uniform space with @xmath893 then @xmath923 . note that essentially the same argument shows that if @xmath0 and @xmath46 are uniform spaces , @xmath47 is a cover , and @xmath924 then @xmath887 . we conjecture that in this situation @xmath0 and @xmath46 have exactly the same dimension . [ circle ] if @xmath0 is coverable uniform space with @xmath925 then @xmath926 contains no simple closed curve ( i.e. topological circle ) . suppose that there is a topological embedding @xmath927 with the inverse homeomorphism @xmath928 . by theorem [ dimthm ] we have @xmath929 and therefore by theorem v.13 of @xcite there is an extension of @xmath403 to a uniformly continuous function @xmath930 . choose any point @xmath130 as the basepoint in both @xmath38 and @xmath21 , and choose @xmath843 as the basepoint in @xmath931 ( see remark [ base2 ] ) . let @xmath932 be the traditional universal cover of @xmath931 also with some choice of basepoint @xmath130 in @xmath933 which is also the universal cover in the sense of the present paper by corollary 0518equiv . since @xmath934 is universal by theorem [ xic ] , proposition [ liftlem ] ( 2 ) implies that there is a unique lift @xmath935 such that @xmath936 and @xmath937 thus we get that @xmath938this implies that @xmath939 is a topological embedding of @xmath931 into @xmath940 , which is impossible . [ laminj]if @xmath0 is coverable uniform space with @xmath941 ( in particular if @xmath0 is compact with covering dimension @xmath942 ) , then @xmath65 is simply connected and if @xmath0 is pathwise connected the homomorphism @xmath943 is injective . by proposition [ inj ] , it is enough to prove that @xmath934 is simply connected . suppose that @xmath766\rightarrow \tilde{x}$ ] is a loop in @xmath944 . then its image @xmath945 is a peano continuum that contains no simple closed curves by proposition [ circle ] . by the hahn - muzurkiewicz theorem , @xmath945 is locally connected , hence a dendrite ( see section 51 , vi in @xcite ) . then @xmath946 is contractible by corollary 7 in section 54 , vii of @xcite . therefore @xmath945 has trivial fundamental group . this means that any loop at any basepoint in @xmath945 ( including the loop @xmath317 ) is null - homotopic in @xmath945 hence in @xmath65 . combining the above theorem with proposition [ newquotient ] we have the following : let @xmath0 be a pathwise connected , uniformly locally pathwise connected uniform space with @xmath947 . then @xmath0 is the quotient of a one - dimensional pathwise connected , simply connected uniform space via a free isomorphic action of @xmath884 . [ tcc]let @xmath0 be a uniform space . we define an entourage @xmath3 to be chain connected if every @xmath3-ball in @xmath0 is chain connected . we say @xmath0 is totally chain connected if @xmath0 has a uniformity base that includes @xmath948 such that each entourage in the base is chain connected . from proposition [ 52 ] we immediately have : [ 315totally]if @xmath0 is a totally chain connected uniform space and @xmath3 is a chain connected entourage then @xmath2 is totally chain connected . if moreover @xmath0 is uniformly locally connected ( resp . uniformly locally pathwise connected ) then @xmath2 has the same property . in the proof of theorem 9 , @xcite , it was shown that the character group of @xmath949 is not coverable . but this group is known to be connected and locally connected , hence totally chain connected . every totally chain connected pseudometric space @xmath0 is coverable . since @xmath0 is a pseudometric space we can find a countable sequence @xmath950 of chain connected entourages forming a base for the uniformity of @xmath0 ( hence cofinal in the set of all entourages of @xmath0 ) . therefore @xmath951 . each of the spaces @xmath952 is chain connected by lemma [ 315totally ] and when @xmath953 we may identify @xmath954 with @xmath955 by proposition [ identify ] . therefore , lemma [ lemsur5 ] implies that each bonding map @xmath956 is surjective . since the inverse system is countable it follows that the projections @xmath957 are all surjective . since connected sets are chain connected , we have the following two corollaries : [ pseu]every connected , uniformly locally connected pseudometric space is coverable . recall that a peano continuum is a hausdorff topological space that is the continuous image of an interval . equivalently ( by the hahn - mazurkiewicz theorem ) , a peano continuum is a compact , connected , locally ( pathwise ) connected metrizable space . [ peano]every peano continuum @xmath0 is coverable . moreover , there is a compact subset @xmath38 of the pathwise connected component of @xmath21 such that the restriction of the universal covering map @xmath50 to @xmath38 is surjective . the first statement follows from corollary [ 0628new ] . let @xmath813\rightarrow x$ ] be a continuous surjection . then @xmath958)$ ] has the desired properties . [ metnote]to simplify matters , when @xmath0 is a metric space we refer to an @xmath44-loop , where @xmath44 is the metric entourage having open @xmath45-balls as @xmath44-balls , as an @xmath959-loop ( @xmath960 ) , @xmath44-homotopies as @xmath959-homotopies , etc . we will denote @xmath961 by @xmath962 , @xmath963 by @xmath964 , @xmath965 by @xmath966 , and @xmath967 by @xmath968 . note that for a metric space @xmath0 and @xmath960 , the cover @xmath969 is a broadening of the notion of @xmath45-cover , something that goes back at least to spanier s book ( @xcite ) , and has been used in @xcite and @xcite to study universal covers of limits of riemannian manifolds . however , the construction of @xmath45-covers uses paths and standard homotopies rather than chains . it is not hard to construct examples of uniformly locally pathwise connected metric spaces having metric balls that are not necessarily pathwise connected , and so that the metric entourages are not covering entourages . recall that an _ inner metric space _ is a metric space such that the distance between any two points is the infimum of lengths of curves joining them . a _ geodesic _ space further has the property that the distance is realized as the length of some curve , called a minimal geodesic . the metric balls of any inner metric space are pathwise connected and therefore the metric entourages are covering entourages . we have : [ inner]every inner metric space @xmath0 is coverable . moreover , @xmath970 , where @xmath971 is any sequence of positive values decreasing to @xmath488 . from lemma [ 315totally ] and corollary [ uit ] we have : an inner metric space @xmath0 is universal if and only if for any @xmath972 , every @xmath45-loop based at some point is @xmath45-homotopic to the trivial loop . note that the topologist s sine curve , discussed in the introduction , is coverable but not totally chain connected because at some points every small neighborhood contains pairs of points that can not be joined by an arbitrarily fine chain that stays inside the neighborhood . we next give a unified calculation of the deck groups @xmath19 of the hawaiian earring @xmath973 , the sierpinski gasket @xmath974 , and sierpinski carpet @xmath975 . this unified calculation is possible because each of these spaces admits a description as a countable intersection @xmath976 let us begin with the case @xmath977 define @xmath978 as the closed disc @xmath979 of diameter 1 , lying in the upper euclidean half - plane @xmath980 and tangent to the @xmath40-axis at the origin @xmath981 . let @xmath982 be its boundary circle . denote by @xmath983 the @xmath136-homothety of euclidean plane @xmath984 with the center @xmath985 we omit @xmath317 in the notation if @xmath986 now define @xmath987 , @xmath988 iteratively by the formulas @xmath989then @xmath973 is defined by formulas ( [ ind1 ] ) and ( [ inters ] ) . for @xmath974 let @xmath990 be the isosceles triangle of diameter 1 in the euclidean quarter - plane @xmath991 , with one side on the @xmath40-axis and one vertex @xmath992 ; denote the other vertices by @xmath993 . now define @xmath987 iteratively by the formulas @xmath994then @xmath974 is defined by formulas ( [ ind2 ] ) and ( [ inters ] ) . for @xmath975 let @xmath995 be the square of diameter 1 in the euclidean quarter - plane @xmath991 with one side on each of the coordinate axes . denote by @xmath996 its vertices and @xmath997 the midpoints of its sides . now define @xmath987 iteratively by the formulas @xmath998then @xmath975 is defined by formulas ( [ ind3 ] ) and ( [ inters ] ) . denote by @xmath999 the number @xmath1000 in the cases @xmath1001 and @xmath974 and the number @xmath1002 in the case @xmath975 . moreover , by our choice of @xmath1003 , @xmath1004 ( see notation [ metnote ] ) is the universal cover of @xmath987 , which means that the projection @xmath1005 is an isomorphism . one can easily see that in all cases @xmath987 is a compact poincar space and @xmath1006 is isomorphic to @xmath1007 via the map @xmath1008 defined in definition [ induced ] , where @xmath1009 is the inclusion . according to corollary [ 0518equiv ] , @xmath1010 is an isomorphism . therefore for any fixed @xmath86 , the function @xmath1011 is an isomorphism . according to proposition [ lifter ] , the image with respect to @xmath1012 of an equivalence class of a loop @xmath317 based at @xmath130 is the @xmath3-homotopy class of any sufficiently fine @xmath3-chain of the form @xmath1013 . therefore , if @xmath1014 we have the following commutative diagram:@xmath1015where @xmath1016 is inclusion and @xmath1017 with @xmath1018 being the inclusion - induced mapping given by definition [ induced ] . but @xmath987 is homotopic to a wedge product of @xmath1019 circles , where @xmath1020 for @xmath973 , @xmath1021 for @xmath974 , and @xmath1022 for @xmath1023 so , in any case , @xmath1024 where @xmath1025 is the free group with @xmath9 generators . note that @xmath1026 and for all three examples the inverse sequence @xmath1027 is a cofinal sequence in the inverse system @xmath1028 where @xmath1029 is the unique surjective homomorphism that kills one extra generator . although all three spaces have the same deck group , the spaces @xmath1030 and @xmath1031 are mutually non - homeomorphic one - dimensional peano continua . in fact , the last two spaces are not semi - locally simply connected at any point , while @xmath1032 has unique point ( @xmath1033 ) , at which it is not semi - locally simply connected . the spaces @xmath1034 and @xmath1031 are not homeomorphic , because the first space have a countable subset @xmath1035 such that @xmath1036 is null - homotopic in itself , while the second one has no such subset . we can take @xmath1037where @xmath126 is the midpoint of the segment @xmath1038,$ ] and @xmath1039for every subset @xmath1040 since @xmath1034 and @xmath1031 are non - homeomorphic one - dimensional peano continua that are not semi - locally simply connected at any point , then by a result of @xcite , @xmath1041 and @xmath1042 are not isomorphic as abstract groups . thus it follows from this and the above result @xmath1043 that the homomorphism @xmath1044 can not be a bijection for all three examples . finally , note that the conditions of proposition dense are satisfied and hence the image of the fundamental group with respect to @xmath62 is dense in @xmath19 in each case . a similar discussion may be applied to the sierpinski sponge @xmath1045 ( also known as the universal sierpinski curve ) which is formed by successively removing middle cores from the unit cube , implying that @xmath1046 is also the inverse limit of finitely generated free groups . it is well - known that every @xmath74-dimensional locally connected metrizable continuum can be topologically embedded in @xmath1047 . by a result of curtis and fort ( @xcite ) , these embeddings induce inclusions of the fundamental group , and it follows that the fundamental group of any such continuum embeds in an inverse limit of finitely generated free groups . this fact has been previously established by various authors , most recently @xcite . the construction , in @xcite , of the group @xmath1048 for a ( hausdorff ) topological group @xmath82 is the same as the construction that we use in the present paper . however , the construction in the prior paper includes a compatible group structure on each group in the fundamental inverse system , which induces a group structure on @xmath1048 so that the natural homomorphism @xmath1049 is a quotient map with closed , central , prodiscrete kernel . ( there are also several results in @xcite that have no analogs for uniform spaces in general , such as results concerning extensions of local homomorphisms . ) for a topological group @xmath82 , the condition that we call universal in this paper is equivalent to what was called locally defined in @xcite , by proposition 61 of @xcite . a coverable group was defined to be the quotient of a locally defined group via a closed normal subgroup . according to corollary [ bi ] in the present paper , such a group is a coverable uniform space . the converse of this statement involves theorem 90 in @xcite , which requrires a correction . in fact , professor helge glckner of t. u. darmstadt has pointed out that the proof of lemma 42 in @xcite has a gap , and we do not know whether this lemma is true . the only direct reference to lemma 42 is theorem 90 , and in light of this gap part ( 2 ) of theorem 90 should be restated as the following stronger condition : ( 2@xmath1050 ) @xmath82 has a basis for its topology at @xmath1051 consisting of locally generated symmetric neighborhoods , and @xmath1052 is surjective for all @xmath90 in this basis . the proof may be modified to show ( 1 ) implies ( 2@xmath1053 ) as follows . in the first part of the proof it is shown that @xmath50 is surjective and that @xmath1054 is coverable . the same argument may then be applied to show that @xmath1052 is surjective , proving ( 2@xmath1053 ) . in the proof of ( 2)@xmath465(3 ) , lemma 42 is only used to prove that all of the homomorphisms @xmath1052 are surjective , and so ( 2@xmath1053 ) eliminates the need for lemma 42 . note that ( 2@xmath1053 ) is actually a stronger condition than the definition of coverable in the present paper , which now completes the proof that a topological group is coverable in the sense of @xcite if and only if it is coverable in the present sense . we know of no example of a uniform space @xmath0 ( let alone topological group ) that is not coverable and @xmath20 is surjective , hence it is still possible that theorem 90 ( or its generalization to uniform spaces ) is true as stated . however , the revised version is sufficient for all applications in @xcite and @xcite except for three . the first is corollary 91 of @xcite , which is not used elsewhere . the second exception is that the alternate hypothesis in theorem 15 or @xmath1049 is surjective needs to be taken out ( or replaced by a stronger assumption related to the new condition ( 2@xmath1053 ) ) . the third exception is corollary 107 , which is nt used elsewhere . in the interest of completeness we will now address every reference to theorem 90 both in @xcite and @xcite to show that the new version is sufficient . since ( 2@xmath1053 ) is stronger than ( 2 ) , there is no problem with any statement that does nt involve ( 2)@xmath465(1 ) or ( 2)@xmath1055(3 ) . theorem 90 is used this way in proofs of the following statements in @xcite : theorem 4 , theorem 5 , proposition 10 , example 99 , theorem 92 ( necessity ) , theorem 101 , and theorem 15 ( second reference ) . now consider the remaining references : in theorem 92 ( sufficiency ) of bptg , all of the projections are surjective and therefore ( 2@xmath1053 ) holds . in @xcite , theorem 24 is simply a restatement of theorem 90 ( with an additional statement about metric spaces added ) and therefore ( 2 ) must be replaced by ( 2@xmath1053 ) . theorem 24 is only used in the proof of theorem 7 , and the only problematic usage is that theorem 24 , ( 2)@xmath1055(3 ) , is used to prove theorem 7 , ( 6)@xmath465(1 ) . however , it is well - known that an pathwise connected , locally compact group is locally pathwise connected . therefore for any pathwise connected symmetric open set @xmath90 containing the origin , @xmath1054 is pathwise connected and hence @xmath1056 is surjective . that is , the conditions for ( 2@xmath1053 ) are satisfied . the relationship between covers in the present sense and covers in the sense of @xcite is considered in @xcite . note that our paper partially answers problem 152 in @xcite , which asks whether the generalized fundamental group in that paper , which is the same as the deck group in the present paper , is a topological invariant . from the present paper we know that the deck group is in fact an invariant of uniform structures and for compact groups a topological invariant . there are several other theorems from @xcite that likely can be generalized to the more setting of uniform spaces , and these will be considered in a future paper . some of the questions of that paper have analogs for uniform spaces , most notably :

modern computer and communications networks constitute an arena of economic interactions among multiple autonomous self - interested entities ( network access providers , end - users , electronic commerce enterprises , content storage and distribution enterprises ) . the internet is perhaps the most massive and global field where economic networking interactions take place . the recently established study of _ network formation games _ @xcite(chapter 19 ) aims at understanding how the competitive ( or coalitional ) activity of multiple such _ selfish agents _ affects the network s characteristics and its efficiency . in this paper we consider the setting of @xmath0 _ content distribution network _ enterprises interacting over a network . enterprise @xmath7 takes decisions on where to store replicas of digital content over the network , so as to satisfy access demand of local customers , situated at node @xmath8 . demand is expressed by a non - negative weight @xmath9 for enterprise @xmath7 . every enterprise chooses strategically a location @xmath10 on the network for installation of content replicas , so as to minimize its individual expenses for _ ( a ) _ storage / management of the content , and _ ( b ) _ weighted connection ( bandwidth / delay ) costs to the chosen location . we study a _ facility location _ game played among selfish agents ( enterprises ) , that naturally models this situation . we use a fair cost allocation rule - known as shapley cost - sharing @xcite - for facility installation costs : for a location @xmath10 installation cost @xmath11 is shared in a fair manner among agents that receive service from @xmath10 , so that agent @xmath7 pays @xmath12 , where @xmath13 denotes the total demand forwarded to @xmath10 . the weighted connection cost is given by @xmath14 , where @xmath15 denotes connection cost per unit of demand . our work focuses on bounding the _ social cost _ of cost - efficient and _ stable _ network infrastructures ; stable networks will be represented by _ pure _ strategy nash equilibria and _ strong _ equilibria of the corresponding facility location game . the social cost will be the sum of individual costs experienced by the agents - players of the game . strong equilibria are an extension of pure nash equilibria , dicussed in @xcite , and essentially introduced by aumann in @xcite . strong equilibria are resilient to coalitional _ pure _ deviations : a strategy profile is a strong equilibrium if no subset of agents can deviate ( by jointly adopting a different _ pure _ strategy ) , so that all of its members are better off . thus this notion captures the possibility of coalitional behavior among agents . occurence of such behavior is naturally expected in rapidly evolving modern markets , especially the ones involving the exploitation and exchange of digital goods and services . we derive bounds on the _ price of stability _ ( pos ) of pure equilibria , defined as the cost of the least expensive equilibrium relative to the socially optimum cost @xcite . for strong equilibria we derive bounds on their ( strong ) _ price of anarchy _ ( spoa ) , i.e. the cost of the most expensive strong equilibrium relative to the socially optimum cost . let us note that the _ price of anarchy _ of pure nash equilibria was introduced in @xcite as the cost of the most expensive pure nash equilibrium relative to the socially optimum cost , and measures essentially social cost losses incurred due to selfishness of agents , _ and _ lack of coordination among them . the pos on the other hand measures social cost losses only due to selfishness : if all players coordinate ( even by following some externally provided instructions , or by interacting on the basis of a common protocol ) they may reach the least expensive equilibrium . the notion of strong equilibria inherently permits coordination among subsets of agents . therefore we can intuitively expect the spoa to almost match the pos . our results confirm this intuition . the study of the price of stability in network design games with fair allocation of network link costs was initiated in @xcite . one challenging problem remaining open since then is improving upon an upper bound of @xmath16 and a lower bound ( recently shown in @xcite ) of @xmath17 , for unweighted players on undirected networks . a series of recent works @xcite provides results ( polylogarithmic upper bounds ) with respect to the social cost of pure nash and ( approximate ) strong equilibria in the network design game model of @xcite , also for the case of weighted players . we review these results in section [ section : related - work ] . we also explain how the facility location game that we study is a special case of the more general model introduced in @xcite , that also includes _ `` delay '' _ costs in using network links ( refered to as connection costs in the case of facility location ) , apart from fairly allocated _ `` installation '' _ costs . for the case of metric connection costs we were able to prove greatly improved and almost tight bounds for the pos and the spoa ; this makes the _ metric _ facility location game and exceptional special case of the model introduced in @xcite . [ [ summary - of - results ] ] summary of results + + + + + + + + + + + + + + + + + + we analyze the pos of the unweighted metric facility location game ( section [ section : unweighted - metric - pos ] ) , and prove constant upper and lower bounds . note that an @xmath1 general upper bound implied by the work of @xcite is tight for non - metric networks ( we discuss this in section [ section : definitions ] ) . our technique relies on direct analysis of social cost evolution during an iterative best response performed by the agents , until they reach equilibrium . we show in particular that given any initial configuration ( including the social optimum ) , the social cost of the reached equilibrium is at most @xmath18 times the initial social cost and at least @xmath19 times this cost in the worst case . metric facility location is the first case of the model proposed in @xcite , in which an additional structural network property ( triangle inequality ) yields a constant almost tight price of stability . it is also interesting that incorporation of link delays ( connection costs ) makes the previously derived @xmath20 upper bounds tight for the non - metric unweighted game . subsequently we study strong equilibria ( section [ section : apx - strong - equilibria ] ) . although they do not always exist , we prove that @xmath21-approximate strong equilibria exist ( no subset deviation causes factor @xmath21 improvement to all of its members ) , for @xmath22 in general networks and weighted agents . for the price of anarchy of approximate strong equilibria ( spoa ) we show an @xmath4 general upper bound which becomes @xmath6 for unweighted agents ( section [ section : strong - poa ] ) . for the metric unweighted case we prove a constant upper bound ( section [ subsection : metric - unweighted - spoa ] ) . except for the spoa analysis in paragraph [ subsection : metric - unweighted - spoa ] ( theorem [ theorem : metric - spoa - ub ] ) , the rest of the described results have appeared in @xcite , in a shorter version -approximate strong equilibria that appeared in a previous version of this report was erroneous and has been removed . ] . we note that the analysis of the spoa of approximate strong equilibria for the metric unweighted case in paragraph [ subsection : metric - unweighted - spoa ] essentially generalizes the analysis of the corresponding result appearing in @xcite ( theorem 2 ) ; this result concerned the spoa of exact strong equilibria only _ when _ they exist . .summary : facility location games with fair allocation of facility costs . [ cols="^,^,^,^ " , ] anshelevich _ et al . _ first studied the price of stability for network design games with general @xcite and fair cost allocation @xcite . in the latter case @xmath0 agents wish to connect node subsets over a network , by strategically selecting links to use . every network link is associated to two cost components , an _ installation _ cost and a _ delay _ cost . link installation cost is shared fairly among agents using the same link in @xcite , while every agent experiences a delay cost given by a polynomial function of the number of agents using the link . the authors showed that _ for unweighted agents _ these network design games belong to the class of _ potential _ games introduced by monderer and shapley in @xcite , hence they have pure strategy nash equilibria . in particular , a potential function defined in @xcite is associated to these games in the following manner : an arbitrarily initialized iterative best response procedure carried out by the players reaches a local minimum of the potential function , which corresponds to a pure strategy nash equilibrium of the game . in @xcite the authors developed elegant arguments using the potential function , so as to upper bound the pos for unweighted agents . in case of polynomial delay costs of degree at most @xmath23 the pos was shown to be at most @xmath24 . for the case of directed networks this bound was shown to be tight ( for k=0 and zero delays ) , whereas for undirected networks a lower bound of @xmath25 was given . this was recently improved to @xmath17 by fiat et al . @xcite , who also showed an upper bound of @xmath26 for @xmath27 , when exactly one agent resides on each node of the network and all agents need to connect to a common sink node ( single - sink case ) . for weighted agents a potential function proving existence of pure equilibria was developed for only @xmath28 agents in @xcite . chen et al . @xcite showed that pure strategy nash equilibria do not always exist for weighted network design games with fair allocation of link installation costs . they developed a trade - off for the social cost of approximate equilibria versus the approximation factor , as a function of the maximum weight taken over all agents . an extensive study of _ strong _ equilibria for network design games with fair cost allocation ( but without delays ) appeared recently by albers in @xcite , for unweighted and weighted agents . strong equilibria do not always exist , but she showed that @xmath20 and @xmath4-approximate strong equilibria do exist ( @xmath5 is the sum of the agents weights ) for unweighted and weighted games respectively . she proved @xmath20 and @xmath29 upper bounds for the spoa of approximate strong equilibria which are tight for directed graphs . for undirected graphs she showed @xmath30 and @xmath31 lower bounds . finally and most importantly , she showed an @xmath32 lower bound for the pos of weighted network design games . strong equilibria in the context of ( single - sink ) unweighted network design games with fair cost sharing were first studied in @xcite . the authors gave topological characterizations for the existence of strong equilibria on directed networks , and proved that @xmath33 . as of the recent literature , the gap for the identification of the pos of network design games with fair cost allocation remains open for unweighted agents , with the lower bound of @xmath17 being the best known so far , to the best of our knowledge . in effect , this means that the impact of an undirected network structure to the pos is not well understood , since the upper bounding potential function arguments developed in @xcite do not incorporate network structure . the facility location game is a special case of the model studied in @xcite , that is interesting on its own right : it finds numerous applications and exhibits intriguing characteristics . it emboddies non - shareable delays explicitly and in a sense specializes _ single - sink _ network design considered in @xcite : we can simply augment the network with a node @xmath34 and set links @xmath35 to have fairly shareable cost , equal to the facility opening cost at @xmath10 . the original network links have a delay cost only . then every agent needs to choose at most two edges from the node it resides on , to @xmath34 . [ [ facility - location - games ] ] facility location games + + + + + + + + + + + + + + + + + + + + + + + an unweighted metric facility location game with uniform facility costs was studied in the context of _ selfish caching _ in @xcite . the network model of that work is essentially equivalent to the one we discuss here , apart from the fact that fair cost - sharing of facility costs was _ not _ used : an agent could connect to a facility payed exclusively by another agent . another difference is that facility opening cost was uniform accross all nodes of the network , whereas we consider node - dependent costs . the poa and pos of this model were shown to be unbounded . the authors devised an extension of their game model , with payments exchanged among agents , in which the socially optimum configuration of the original game is rendered a pure strategy nash equilibrium ( hence @xmath36 for the extended game ) . vetta studied a class of games for _ competitive _ facility location @xcite , for which he proved existence of pure strategy nash equilibria and an upper bound of @xmath28 for the poa . in this competitive setting enterprises open facilities at certain nodes and try to attract customers to connect to them . he also illustrated applicability of his model in the context of the @xmath23-median problem model . in @xcite the authors considered _ single - sink _ network design with fair cost - sharing of all resources ( network links ) used by agents . they argued how this model can be viewed as a facility location model , but did not take connection costs ( non - shareable delays ) into account . the network will be a complete graph @xmath37 , each edge @xmath38 of which is associated to a non - negative cost @xmath39 . we consider a set @xmath40 of @xmath0 agents . each agent @xmath7 resides on a node @xmath41 , and is associated to a non - negative demand weight @xmath9 . the strategy space of agent @xmath7 is @xmath42 : @xmath7 chooses a location @xmath43 , where it will receive service from . opening a facility at node @xmath43 costs @xmath11 . a strategy profile ( configuration ) is denoted by @xmath44 , where @xmath45 . the total weight of agents receiving service from @xmath10 under @xmath46 is @xmath47 . the cost @xmath48 experienced by agent @xmath7 in @xmath46 is : @xmath49 every agent @xmath7 pays a fraction @xmath50 of the facility installation cost at @xmath51 . denote by @xmath52 the set of facility locations specified under @xmath46 . the _ social cost _ @xmath53 is then : @xmath54 we use @xmath55 for the sum of weights of agents in set @xmath56 . by @xmath57 we denote the total cost of agents in @xmath56 under @xmath46 . furthermore , given a facility @xmath58 , we use @xmath59 to denote @xmath60 . * pure nash equilibria * the _ unweighted _ facility location game specializes network design games first studied in @xcite , and is a potential game @xcite ; it is associated to a potential function , that can be locally minimized by iterative best response performed by players , and the reached local minimum corresponds to a pure nash equilibrium . in iterative best response we choose iteratively an arbitrary player @xmath7 , and let him / her decide a strategy that minimizes its individual cost with respect to the current configuration @xmath61 for the rest of the players . the process ends when no player @xmath7 can improve his / her individual cost under @xmath62 . then @xmath46 is a pure nash equilibrium . a simple example shows that the poa of pure equilibria can be @xmath0 in the worst case . take a network of two nodes @xmath63 . opening a facility on either node has a cost @xmath64 . we set @xmath65 , and let @xmath0 agents reside on @xmath66 . then if every agent @xmath7 plays @xmath67 , @xmath68 , the resulting configuration is a pure nash equilibrium of cost @xmath69 , because every agent pays exactly @xmath64 and has no incentive to open a facility at node @xmath66 by paying the same cost @xmath64 . the social optimum consists of a single facility opened on @xmath66 and all agents being locally serviced by it , at zero connection cost . thus the socially optimum cost is @xmath64 . potential function arguments developed in @xcite yield an upper bound of @xmath70 for the pos , where @xmath70 is the @xmath0-th harmonic number . this bound is tight for unweighted non - metric facility location . [ figure : general - pos - lb ] presents a non - metric lower bounding example for unweighted agents . assume uniform facility costs equal to 1 . @xmath71 unweighted agents reside on boldly drawn nodes , @xmath0 of them on @xmath72 . the social optimum consists of facilities on @xmath73 and @xmath72 , where @xmath72 serves only the @xmath0 agents residing on @xmath72 . the rest are served by @xmath73 . however they deviate to @xmath72 from @xmath73 one by one . it is then : @xmath74 . when it comes to metric networks however , the pos is constant as we show in the following section . this result essentially indicates the impact of the network s structure ( triangle inequality ) to the social cost of efficient pure nash equilibria . * ( strong equilibria ) @xcite * a strategy profile @xmath46 is a strong equilibrium if no subset of agents can deviate in coordination , so that each and every one of its members is better off . it is an @xmath21-approximate strong equilibrium if no subset of agents can deviate in coordination , so that each and every one of its members is better off by a factor strictly more than @xmath75 . we analyze evolution of an equilibrium through iterative best response , initialized at configuration @xmath76 . in fact , for _ any _ strategy profile @xmath46 that is reached through iterative best response initialized at @xmath76 , we will upper bound the ratio @xmath77 . thus we do not require that @xmath76 be the socially optimum configuration , nor that @xmath46 is an equilibrium strategy profile . we only require that @xmath46 is reached by iterative best response initialized at @xmath76 . then taking @xmath76 to be the socially optimum configuration and @xmath46 to be the reached equilibrium will yield an upper bound for the price of stability , as a corollary . to facilitate clarity , we give a brief description of the analysis plan . given any facility node @xmath78 denote by @xmath79 the subset of agents connected to @xmath10 in @xmath76 . then @xmath80 will be the total cost incurred by @xmath81 collectively . we will upper bound @xmath77 as follows : @xmath82 define @xmath83 to be the increase in cost caused by agent @xmath7 during iterative best response initialized at @xmath76 and reaching @xmath46 . for any subset @xmath84 of agents we also use @xmath85 . to evaluate the upper bounding expression ( [ equation : plan ] ) , we are going to use an upper bound on @xmath86 , valid _ for any _ @xmath78 . this is the increase in social cost caused during iterative best response by agents connected to @xmath10 in @xmath76 . then for any @xmath78 we will maximize @xmath87 , by lower bounding @xmath80 for any @xmath78 . in determining an upper bound on @xmath88 we prove the following lemma , which charges any specific agent @xmath7 a bounded amount of social cost increase during the execution of iterative best response . let @xmath81 be the subset of agents that are connected to @xmath10 in @xmath76 . for any @xmath89 define @xmath90 to be the subset of agents that have not yet deviated from @xmath10 , exactly before the first deviation of @xmath7 . then @xmath91 . [ lemma : ibr - charging ] for simplicity let @xmath92 . clearly @xmath93 . let us analyze contribution of @xmath7 to social cost increase during its first deviation . by deviating , @xmath7 reduces its individual cost from @xmath94 to @xmath95 , by joining another facility node @xmath96 . then : @xmath97 @xmath98 and @xmath99 is the connection cost payed by @xmath7 before and after its first deviation . @xmath100 is the number of agents sharing facility cost at @xmath96 , including @xmath7 . since @xmath101 , it is @xmath102 . let @xmath103 be the difference caused to the social cost by this single deviation of @xmath7 . we examine the following cases : 1 . @xmath104 , @xmath105 : then @xmath106 . @xmath107 , @xmath108 : then @xmath109 . @xmath104 , @xmath110 : then @xmath111 . @xmath107 , @xmath112 : then @xmath113 . clearly the above hold in general during execution of iterative best response , for any agent that performs a single deviation from a node @xmath10 to a node @xmath96 . now we implement a charging procedure along with iterative best response . give all agents an initial label @xmath114 , before executing iterative best response . the labels will be updated during iterative best response , so that increases caused by agent @xmath7 will be charged to an appropriate agent @xmath115 . initialize @xmath116 for all @xmath117 . we will make use of the auxiliary variables @xmath118 and @xmath100 , as they were described before . at any time @xmath119 is the number of agents ( including @xmath7 ) connected to @xmath10 , right after @xmath7 _ joined _ @xmath10 . we need to initialize @xmath119 to a value for every @xmath78 and @xmath89 , to implement the charging scheme . set @xmath120 to a distinct value from @xmath121 . @xmath122 will be always ( during iterative best response ) the number of agents connected to @xmath10 exactly before deviation of @xmath7 from @xmath10 ( before @xmath7 _ leaves _ @xmath10 ) . charging is then implemented by relabeling deviating agents in the following manner . 1 . if @xmath123 no relabeling is needed . 2 . otherwise there must be some @xmath124 connected to @xmath10 such that @xmath125 . in this case exchange labels of @xmath7 and @xmath126 . subsequently add the increase caused by deviation of @xmath7 to @xmath127 . finally , set @xmath128 equal to the number of agents connected to @xmath96 right after @xmath7 has joined @xmath96 . by the previous definitions it follows that if @xmath129 , then it is always @xmath130 , i.e. @xmath7 has joined @xmath10 before some agent @xmath126 with @xmath125 , but leaves @xmath10 `` out of order '' , i.e. before @xmath126 leaves . by exchanging labels of @xmath7,@xmath126 we add the increase caused by @xmath7 to the agent that previously labeled @xmath126 . possible increases in 1.,2.,3.,4 . , imply that any agent is charged by the end of iterative best response at most @xmath131 for some @xmath7 . if we `` guess '' the exact order of first deviation of all agents , we can initialize @xmath132 . then each @xmath7 itself will be charged at most @xmath131 . @xmath133 in what follows we are going to upper bound the ratio @xmath134 for any @xmath78 . if no agent @xmath89 ever deviates from playing @xmath10 , then clearly it will be @xmath135 . our analysis focuses on two remaining cases : either _ ( i ) _ there is a non - empty set of agents @xmath136 that never deviate from @xmath10 during iterative best response or _ ( ii ) _ all agents @xmath89 deviate from @xmath10 during iterative best response . case _ ( i ) _ is examined in proposition [ intermmediate - case ] , whereas the analysis is concluded by analysis of _ ( ii ) _ in theorem [ theorem : metric - pos - ub ] . let @xmath46 be a strategy profile reached by iterative best response initialized at strategy profile @xmath76 . for any facility @xmath78 define @xmath79 and let @xmath137 the subset of agents that never deviated from @xmath10 , during iterative best response . if @xmath138 , then @xmath139 . [ intermmediate - case ] as discussed previously , we will upper bound the ratio @xmath87 , by deriving an upper bound for @xmath88 and a lower bound for @xmath80 . by lemma [ lemma : ibr - charging ] follows immediately : @xmath140 \label{equation : delta - ub}\ ] ] because agents in @xmath141 never deviated , whereas the rest caused a cost increase equal to @xmath11 times a harmonic series term each . in what follows we show a lower bound on @xmath80 . fix any facility @xmath78 , and define @xmath142 . without loss of generality assume such an order of agents , that agents @xmath143 ( and for every @xmath78 ) `` best - respond '' consecutively ; e.g. assume that the algorithm scans the facilities in @xmath144 in an arbitrary fixed order and for each facility , the agents connected to it in an arbitrary fixed order . we focus only on the first deviation of each agent @xmath143 for some @xmath78 . let @xmath96 be the node that @xmath7 deviates to . for convenience we use @xmath145 and let @xmath146 . let @xmath147 be the total number of agents connected to @xmath96 _ right after _ deviation of @xmath7 . the new cost of @xmath7 right after its first deviation is : @xmath148 . for any other agent @xmath149 that , either deviates from @xmath10 to some node @xmath150 after the deviation of @xmath7 , or never deviated from @xmath10 ( in this case consider a trivial deviation with @xmath151 ) , we have respectively : @xmath152 substitute @xmath153 in ( [ equation : deviation - preference ] ) by triangle inequality : @xmath154 . let @xmath155 denote the number of agents connected to @xmath10 _ right before _ deviation of @xmath7 from @xmath10 . also note that @xmath156 , because @xmath7 decreases its individual cost by deviating . thus we obtain : @xmath157 because @xmath158 , and because the latter has to hold for every pair of distinct agents @xmath159 , we deduce : @xmath160 \label{equation : connection - lb-1}\ ] ] we use ( [ equation : connection - lb-1 ] ) for the connection cost of agents in @xmath161 , and @xmath162 for agents in @xmath163 ; in essence for every @xmath164 we simply set in ( [ equation : connection - lb-1 ] ) @xmath151 , @xmath165 , and take @xmath166 . the cost @xmath80 is then : lcl c_v(s^ * ) & & _ v + _ k=|a_s(v)|^|a_s^*(v)| ( - ) + & = & _ v+ ( -h(|a_s^*(v)|)+h(|a_s(v)| ) ) [ equation : v - cost - lb ] using the bounds ( [ equation : v - cost - lb ] ) and ( [ equation : delta - ub ] ) we deduce the following upper bound : lcl ( _ ia_s^*(v)c_i(s))/c_v(s^ * ) & & + & & [ equation : metric - pos - ub ] the latter inequality follows by bounds for the harmonic number , i.e. @xmath167 , where @xmath168 is the _ euler _ constant . by setting @xmath169 , we can numerically maximize the upper bound with respect to @xmath170 , to at most @xmath18 , for @xmath171.@xmath133 if @xmath46 is a strategy profile for the unweighted metric facility location game , reached by iterative best response initialized at a strategy profile @xmath76 , then @xmath172 . [ theorem : metric - pos - ub ] we only need to complement the result of proposition [ intermmediate - case ] by considering the case where @xmath173 . that is , @xmath174 , and all agents connected to @xmath10 in @xmath76 deviate during iterative best response . if for each @xmath143 that deviates from @xmath10 to @xmath96 ( @xmath96 is not necessarily the same for every @xmath7 ) it is @xmath175 , then : @xmath176 thus , assume there is at least one agent @xmath126 with @xmath177 . in general , let @xmath178 be the _ integer , so that @xmath179 for all @xmath180 . using the same arguments that led to ( [ equation : connection - lb-1 ] ) , and taking @xmath181 , we deduce : @xmath182 using ( [ equation : x - lb ] ) we end up with a similar lower bound to ( [ equation : v - cost - lb ] ) for @xmath183 : @xmath184 to finish the proof , we note that by ( [ equation : x - lb ] ) , deviation of agents with @xmath185 , yields a total cost increase ( payed as new connection cost ) of at most @xmath186 , which is exactly equal to the cost saved by closing the facility at @xmath10 ( because all agents deviate ) . thus we can simulate the situation with the case analyzed in proposition [ intermmediate - case ] , by setting @xmath187 and @xmath188 , so as to end up with ( [ equation : metric - pos - ub ] ) . this is technically equivalent to assuming that agents @xmath7 with @xmath185 never deviated.@xmath133 the price of stability for the unweighted metric facility location game with fairly allocated facility costs is at most @xmath18 . we analyze a worst - case example , that makes the pos for unweighted metric facility location at least @xmath19 asymptotically . experimental evidence shows a lower bound @xmath189 , which we believe is tight for pos . this is because the analysis of theorem [ theorem : metric - pos - ub ] embodies some losses due to bounding of harmonic numbers by logarithms . our construction appears in fig . [ figure : metric - pos - lb](a ) . take @xmath71 agents , @xmath0 of them residing on a singe node @xmath10 of the network . facility opening costs are assumed @xmath190 everywhere . the social optimum @xmath76 has @xmath191 facilities : @xmath10 and @xmath192 , @xmath193 , where @xmath194 . let the @xmath0 agents residing on @xmath10 be serviced by @xmath10 in @xmath76 , whereas the rest are equally partitioned to facilities @xmath192 , @xmath193 ; every facility @xmath192 services a batch of @xmath194 agents . in the least expensive equilibrium @xmath46 , all agents from each facility @xmath195 of @xmath76 will deviate and miss - connect to @xmath10 . we will analyze for _ any _ single facility @xmath192 - henceforth denoted by @xmath196 - the ratio @xmath197 . analysis for the rest of the facilities will be similar , and will yield asymptotically the same result . in particular , by choosing to analyze _ any _ batch of @xmath23 agents that deviate from @xmath196 to @xmath10 , the pos will be lower bounded as : @xmath198 for some _ constant _ @xmath199 we will have @xmath200 of the agents in @xmath81 increase their connection cost significantly by deviating , hence the social cost . for every @xmath89 let @xmath201 be the distance of @xmath7 from @xmath196 , and @xmath202 the increase in connection cost after deviation . we determine @xmath201 and @xmath202 by following best responses of agents in @xmath203 . let @xmath204 be the number of agents already connected to @xmath10 before agents of @xmath203 deviate to @xmath10 in order . in particular : * for @xmath205 set @xmath206 ; a small decrease @xmath207 causes @xmath7 to deviate to @xmath10 . * for each of the remaining @xmath208 agents set @xmath209 . by similar arguments as for the derivation of ( [ equation : deviation - preference]),([equation : connection - lb-0 ] ) and triangle inequality , we obtain @xmath210 , where @xmath211 . this yields @xmath212 for @xmath213 ( the remaining @xmath208 agents ) . note that the @xmath208 remaining agents will deviate to @xmath10 , because they prefer to pay strictly less than @xmath214 , instead of a cost share @xmath215 each , for the facility at @xmath196 . by deviation they `` shut - down '' the facility at @xmath196 and decrease the social cost by @xmath190 , but pay a total connection cost @xmath216 for @xmath10 , which tends to @xmath190 for large @xmath0 . let us now derive the cost @xmath217 . for convenience we use the notation @xmath218 to denote @xmath219 . for the first @xmath178 agents , summing up as in ( [ equation : v - cost - lb ] ) yields : @xmath220- \frac{1}{2}\frac{r}{\lambda+r}+\frac{1}{2}\delta h(\lambda+r,\lambda ) \label{equation : v - opt}\ ] ] the cost of agents in @xmath203 after deviation is : @xmath221 . by substituting @xmath222 , we obtain : @xmath223- \frac{1}{2}\left ( \frac{r}{\lambda+r}+\delta h(\lambda+r,\lambda ) \right ) \label{equation : eq - cost}\ ] ] we simplify ( [ equation : v - opt ] ) and ( [ equation : eq - cost ] ) and substitute to ( [ equation : metric - pos - lb ] ) appropriately . for @xmath224 we use the following bounds : @xmath225 where @xmath226 is the _ euler _ constant . also we use @xmath227 , because @xmath228 . then ( [ equation : metric - pos - lb ] ) becomes : @xmath229- \frac{1}{2}\frac{r}{\lambda+r}-\frac{1}{2}\delta h(\lambda+r,\lambda ) } { \frac{1}{\sqrt{n}}+ 1+\frac{1}{2 } \bigl[\frac{r}{k - r+1}-\delta h(k , k - r)\bigr]- \frac{1}{2}\frac{r}{\lambda+r}+\frac{1}{2}\delta h(\lambda+r,\lambda ) } \rightarrow \label{equation : pos - exp}\ ] ] @xmath230- \frac{1}{2}\frac{r}{\lambda+r}-\frac{1}{2}\delta h(\lambda+r,\lambda ) } { \frac{1}{\sqrt{n}}+ 1+\frac{1}{2 } \bigl[\frac{1-p}{p}+1-\gamma-\ln\frac{1}{p-1/k}\bigr]- \frac{1}{2}\frac{r}{\lambda+r}+\frac{1}{2}\delta h(\lambda+r,\lambda ) } \ ] ] notice that for large @xmath0 , @xmath231 , because @xmath232 , and @xmath204 . furthermore it is @xmath233 , because @xmath234 , with @xmath204 and @xmath235 . assuming that @xmath236 is a constant , the limits of numerator and denominator exist , and we can use them to calculate the limit of the fraction . given @xmath168 the pos lower bound is simplified to : @xmath237 numerical maximization over @xmath199 yields @xmath238 , for @xmath239 . we also searched computationally for @xmath178 maximizing ( [ equation : pos - exp ] ) , for increasing @xmath0 . for @xmath240 we found the values appearing in the table [ figure : metric - pos - lb](b ) , that indicate @xmath241 . one can verify that any configuration other than @xmath76 and @xmath46 is more expensive , by definition of distances of agents from @xmath10 and @xmath192 . strong equilibria do not generally exist in the facility location game , even for unweighted agents on metric networks with uniform facility costs . we illustrate this by an example , over the network shown in fig . [ figure : strong ] . consider the _ metric _ case in figure [ figure : strong ] . there are three unweighted agents @xmath242 situated on distinct nodes @xmath8 of the depicted 6-node cycle . for any equilibrium strategy profile @xmath46 there exist two agents @xmath243 , with @xmath244 with @xmath245 , @xmath246 . by agreeing to open a facility on vertex @xmath247 , @xmath7 and @xmath126 would change their costs to @xmath248 and @xmath249 respectively , each lowering their cost by at least @xmath250 . hence , this example does not have strong equilibria . existence of pure equilibria for weighted agents is also an open issue . however , we were able to reduce to a constant the logarithmic approximation factor @xmath21 known for general network design @xcite . in fact , our result is even more general , as it concerns _ strong _ equilibria . we make use of the following remark . if an instance of the facility location game does not have strong equilibria , then there is at least one cycle of deviations of particular coalitions that results in a circular sequence of configurations @xmath251 with @xmath252 . [ cyclic - remark ] given such a sequence @xmath251 , we denote the coalition that deviates from @xmath253 to form @xmath254 by @xmath255 . such a deviation causes a cost decrease of agents in @xmath256 and possibly a cost increase of agents in @xmath257 . recall that @xmath40 is the set of all agents . we define two quantities , the _ weighted improvement _ @xmath258 for agents in @xmath256 and the _ weighted damage _ @xmath259 caused by agents in @xmath255 respectively : @xmath260 we derive an approximation factor that eliminates cycles . [ lemma : damage ] let @xmath251 with @xmath252 be a cycle of configurations in a facility location game instance , caused by consecutive deviations of coalitions . the game instance has an @xmath21-approximate strong equilibrium if for all such sequences @xmath261 , where @xmath262 . if there is no @xmath21-approximate strong equilibrium we know that there is at least one cycle @xmath251 such that @xmath263 . because @xmath252 we have that @xmath264 for every agent @xmath7 . then : @xmath265 it follows that @xmath266 . hence the lemma follows by contradiction.@xmath133 we derive an approximation factor as an upper bound of @xmath267 for any cycle . for every @xmath268 there exist @xmath21-approximate strong equilibria in the facility location game with fairly allocated facility costs , even for weighted agents and general networks . we prove that @xmath269 for every cycle @xmath251 of configurations and the result follows from lemma [ lemma : damage ] . let @xmath270 be the set of agents going to @xmath10 in @xmath253 , but not in @xmath254 , and @xmath271 be the set of agents going to @xmath10 in both @xmath253 and @xmath254 . note that @xmath272 , therefore : @xmath273 @xmath274 hence , we need only consider what happens at the worst case node . for an agent @xmath7 in @xmath271 we get that : @xmath275 it follows that : @xmath276 @xmath133 approximate strong equilibria are also approximate pure nash equilibria , thus : the facility location game with fairly allocated facility costs and weighted agents has @xmath21-approximate pure strategy nash equilibria for every @xmath268 . we derive next an upper bound on the spoa of @xmath21-approximate strong equilibria , for the general facility location game . [ theorem : spoa - ub ] for any constant @xmath268 , the price of anarchy of @xmath21-approximate strong equilibria for the facility location game with fairly allocated facility costs , is upper bounded tightly by @xmath20 for unweighted and by @xmath4 for weighted agents , where @xmath5 is the sum of weights . let @xmath46 and @xmath76 be the most expensive strong equilibrium and the socially optimum configuration respectively . for any facility node @xmath78 let @xmath81 and @xmath161 be respectively the subsets of agents connected to @xmath10 in @xmath76 , and connected to @xmath10 in @xmath76 but not in @xmath46 . we will upper bound the spoa by @xmath277 $ ] . because @xmath46 is an @xmath21-approximate strong equilibrium , _ for every _ subset of @xmath161 there is at least one agent @xmath7 , that is not willing to deviate to @xmath10 in coordination with the rest agents of the subset . let @xmath278 and @xmath279 , where @xmath7 is not willing to deviate in coordination with @xmath280 . then : @xmath281 @xmath282 @xmath283 @xmath284 the result will follow from ( [ equation : weighted - harmonic ] ) , because @xmath285 , and agents in @xmath286 play @xmath10 in @xmath46 and @xmath76 . we analyze the spoa ratio . for @xmath287 it is @xmath288 , which yields @xmath289 . for weighted agents set @xmath290 and use @xmath291 for @xmath292 on ( [ equation : weighted - harmonic ] ) : @xmath293 @xmath133 * unweighted lower bound * we refer to fig . [ figure : apx - spoa - lb ] for a tight ( non - metric ) lower bounding example in case of unweighted agents . facility opening costs are @xmath190 . it is clear that all agents being connected to node @xmath294 is the most expensive @xmath21-approximate equilibrium , of social cost @xmath295 : no agent coalition has incentive to deviate to @xmath296 , which is practically the node that all agents reside on , and results in a social cost of virtually @xmath190 , for @xmath297 ( e.g. @xmath298 ) . -approximate strong equilibria ( @xmath268 ) . ] we prove the following for the spoa of the unweighted metric facility location game : the price of anarchy of @xmath2-approximate strong equilibria in the unweighted metric facility location game with fairly allocated facility costs is upper bounded by a constant . [ theorem : metric - spoa - ub ] the proof of theorem [ theorem : metric - spoa - ub ] consists of several partial results that we will synthesize . let @xmath46 denote any @xmath21-approximate strong equilibrium and @xmath76 the socially optimum cofiguration . we will upper bound the @xmath299 again by @xmath300 . for any facility node @xmath78 , define @xmath301 to be the subset of those agents that are connected to @xmath10 both in @xmath76 and @xmath46 . define @xmath302 to be the subset of agents that are connected to @xmath10 in @xmath76 but not in @xmath46 . see fig . [ figure : metric - spoa - setup ] for an illustration of the definitions . to simplify notation we use @xmath303 , for @xmath89 . at first we consider the following simple case : . ] let @xmath161 be a subset of misconnected agents under @xmath21-approximate strong equilibrium profile @xmath46 . let @xmath304 denote agents connected to @xmath10 under @xmath46 , with @xmath305 . if under @xmath306 no agent of @xmath161 has incentive to deviate to @xmath10 in coordination with @xmath161 , then @xmath307 . if under @xmath306 no agent @xmath143 has incentive to deviate in coordination with @xmath308 to @xmath10 , then for every @xmath143 it is @xmath309 , because @xmath46 is an @xmath21-approximate strong equilibrium . then : lcl _ ia_s^*(v)c_i(s ) & = & _ ia_s(v)c_i(s)+_ii_s^*(v)c_i(s ) + & & _ v+_ia_s(v)x_i^*+ ( _ v+_ii_s^*(v)x_i^ * ) + & & ( 1+)(_v+_ia_s^*(v)x_i^ * ) the latter equals @xmath310.@xmath133 for the rest of the analysis we treat the complementary case of that described in the previous lemma . assume there exists at least one agent @xmath143 willing to deviate to @xmath10 in coordination with the coalition @xmath161 under @xmath306 . define a _ minimal disagreeing subset _ @xmath311 to be a _ minimal subset of misconnected agents _ containing an agent @xmath7 that would actually deviate to @xmath10 with @xmath312 , under @xmath306 . then also define @xmath313 . fix @xmath314 to be from now on the agent ( or one of them if there are many ) that would deviate to @xmath10 in coordination with @xmath312 . we call @xmath7 the _ unstable _ agent of the minimal disagreeing subset @xmath312 . by definition , the following holds for an _ unstable _ agent @xmath7 of the minimal disagreeing subset : @xmath315 the rest of the analysis consists of bounds for agents in @xmath316 and @xmath317 separately . in particular , lemmas [ lemma : rest - ub ] and [ lemma : dev - ub ] describe upper bounds for these sets respectively . let @xmath161 be a subset of misconnected agents under an @xmath21-approximate strong equilibrium profile @xmath46 . let @xmath312 be a minimal disagreeing subset of @xmath161 and @xmath314 an unstable agent . then : @xmath318 [ lemma : rest - ub ] by minimality of @xmath312 , every agent @xmath319 is _ not willing _ to deviate to @xmath10 under @xmath306 with a coalition of size @xmath320 . thus for every @xmath319 : @xmath321 summing over @xmath312 yields : @xmath322 which is upper bounded by at most as in ( [ equation : rest - ub]).@xmath133 let @xmath161 be a subset of misconnected agents under an @xmath21-approximate strong equilibrium profile @xmath46 . let @xmath312 be a minimal disagreeing subset of @xmath161 and @xmath313 . then : @xmath323 \label{equation : dev - ub}\ ] ] [ lemma : dev - ub ] without loss of generality name agents @xmath324 by distinct indices @xmath325 and define a series of supersets of @xmath312 , as follows : @xmath326 . because @xmath46 is an @xmath21-approximate strong equilibrium , every set @xmath327 contains an agent that is not willing to deviate to @xmath10 in coordination with @xmath327 . this agent is found either in @xmath328 or in @xmath329 . we can assume without any loss of generality that for subset @xmath327 this agent is @xmath126 ; otherwise we only need to exchange @xmath126 with some agent from @xmath330 . one easily verifies that , by definition of a minimal disagreeing subset , such an exchange will not affect any of our previous results up to now . then we have : @xmath331 we omit @xmath332 and sum the inequality over @xmath324 . the result follows.@xmath133 the following lemma will provide a lower bound for @xmath333 appearing in ( [ equation : dev - ub ] ) . note that lemma [ lemma : rest - ub ] provides a concrete upper bound for the cost ( under strategy profile @xmath46 ) of agents in @xmath161 , as a function of their connection cost in the socially optimum configuration @xmath76 and the corresponding facility cost @xmath11 . this is not the case with lemma [ lemma : dev - ub ] for agents in @xmath317 ; the upper bound ( [ equation : dev - ub ] ) involves socially optimum connection cost plus @xmath11 multiplied by additional terms of harmonic numbers . thus we need to determine how low the socially optimum connection cost can be , in order to derive an upper bounding ratio for the @xmath299 . let @xmath161 be a subset of misconnected agents under @xmath21-approximate strong equilibrium profile @xmath46 . let @xmath312 be a minimal disagreeing subset of @xmath161 and @xmath313 . then : @xmath334 let @xmath7 be the fixed _ unstable _ agent of @xmath312 . note that , under strategy profile @xmath46 , @xmath7 does not have an incentive to join facility node @xmath335 for any @xmath324 . thus if @xmath126 pays for @xmath335 a share of @xmath336 ( that is , @xmath335 serves @xmath337 agents in total in @xmath46 ) : @xmath338 @xmath339 the latter inequality derives by usage of ( [ equation : j - ub ] ) for @xmath340 , and by safely omitting @xmath332 . using ( [ equation : everyone - ub ] ) and the lower bound for @xmath48 given in ( [ equation : i - lb ] ) , we can solve for @xmath341 . by the definition of @xmath327 in lemma [ lemma : dev - ub ] , it is @xmath342 , @xmath343 . then we obtain : @xmath344 finally we sum up the latter bound over all @xmath126 . notice that @xmath341 becomes non - negative only when @xmath345 . since @xmath346 is an integral value , it turns out that @xmath341 becomes non - negative for those values of @xmath126 for which it is @xmath347 . then , by setting @xmath348 , and by summing up over all @xmath126 we obtain the specified lower bound.@xmath133 [ [ proof - of - theoremtheoremmetric - spoa - ub ] ] proof of theorem [ theorem : metric - spoa - ub ] + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + now we put everything together . lower bound _ on @xmath80 is : @xmath349 accordingly , we obtain the following _ upper bound _ on @xmath350 by ( [ equation : rest - ub ] ) : lcl _ ia_s^*(v)c_i(s ) & & _ li_s^*^0(v)c_l(s)+ _ jj_s^*(v)c_j(s)+_ia_s(v)c_i(s ) + & & 2(_li_s^*^0(v)x_l^*+_v)+ _ jj_s^*(v)c_j(s)+_ia_s(v)c_i(s ) using the latter bounds for @xmath80 and @xmath350 , we deduce : lcr spoa & & 1 + ( 2_v+ 2_li_s^*^0(v)x_l^*+ _ jj_s^*(v)c_j(s ) ) / ( _ v+_li_s^*^0(v)x_l^*+_jj_s^*(v)x_j^ * ) + & & 1 + 2 + & = & 1 + 2 + 2 [ equation : metric - spoa - ub ] the latter inequality emerged firstly by substitution from ( [ equation : dev - ub ] ) , and secondly by removal of @xmath351 from numerator and denominator . we maximize ( [ equation : metric - spoa - ub ] ) by taking the lower bound of the denominator , given in ( [ equation : spoa - connection - lb ] ) and obtain : @xmath352 because @xmath353 , and by using logarithmic bounds for the harmonic numbers : @xmath354 by substituting @xmath355 for @xmath168 and @xmath356 we can maximize numerically the resulting upper bounding function of @xmath357 to at most a constant . note that , when strong equilibria exist , it is @xmath358 . substituting so in ( [ equation : analytic - metric - spoa - ub ] ) yields an upper bounding expression similar to the one of pos ( [ equation : metric - pos - ub ] ) , up to constant multiplicative and additive terms.@xmath133 further investigation of the weighted game is mostly challenging : it is not known whether this game possesses pure equilibria . designing a counter - example seems quite demanding , as the game specializes in some sense the single - sink weighted network design game studied in @xcite . for this case pure equilibria were shown not to exist generally . extending our analysis of the pos and spoa to the weighted metric case appears to be also non - trivial . finally , derivation of good lower bounds for the pos and the spoa of the ( non - metric ) weighted case is an interesting aspect of research : lower bounding techniques developed in @xcite do not readily apply for facility location . for usefull discussions we thank the participants of the _ open problems jam sessions _ of cagt : daniel andersson , gudmund skovbjerg frandsen , kristoffer arnsfelt hansen , peter bro miltersen , rocio santillan rodriguez , troels bjerre soerensen , nikolaos triandopoulos . e. anshelevich , a. dasgupta , j. m. kleinberg , e. tardos , t. wexler , and t. roughgarden . the price of stability for network design with fair cost allocation . in _ proceedings of the ieee symposium on foundations of computer science ( focs ) _ , pages 295304 , 2004 . c. chekuri , j. chuzhoy , l. lewin - eytan , j. naor , and a. orda . non - cooperative multicast and facility location games . in _ proceedings of the acm conference on electronic commerce ( ec ) _ , pages 7281 , 2006 . b. chun , k. chaudhuri , h. wee , m. barreno , c. h. papadimitriou , and j. kubiatowicz . selfish caching in distributed systems : a game - theoretic analysis . in _ proceedings of the acm symposium on principles of distributed computing ( podc ) _ , 2004 . a. fiat , h. kaplan , m. levy , s. olonetsky , and r. shabo . on the price of stability for designing undirected networks with fair cost allocations . in _ proceedings of the international colloquium on automata languages and programming ( icalp ) _ , pages 608618 , 2006 . t. d. hansen and o. a. telelis . on pure and ( approximate ) strong equilibria of facility location games . in _ proceedings of the international workshop on internet and network economics ( wine ) _ , pages 490497 , 2008 . a. vetta . nash equilibria in competitive societies , with applications to facility location , traffic routing and auctions . in _ proceedings of the ieee symposium on foundations of computer science ( focs ) _ , pages 416 , 2002 .

typical eukaryotic protein - coding genes contain introns , which are removed prior to translation . key constituents of the spliceosome , which is the rna - protein complex that performs the intron excision , can be traced back to the last common ancestor of extant eukaryotes ( leca ) . even deep - branching lineages have introns . it is thus almost certain that spliceosomal introns were present in leca . moreover , when comparing distant eukaryotes , intron positions often agree . the similarity is likely due more to conservation of early introns than to parallel gains . it is thus compelling to use genome - scale comparisons to study intron evolution in different lineages , and even to estimate the exon - intron organization in extinct ancestors . one of the first such studies , by , involved orthologous gene sets in eight eukaryotes . the same data set was reanalyzed by different authors , using novel methods developed for intron data . subsequent inquiries attest to a renewed interest in understanding the specifics of intron evolution within different eukaryotic lineages . this paper introduces novel computational techniques for the analysis of spliceosomal intron evolution , anticipating more large - scale studies to come . section [ sec : identification ] describes an alignment method for intron - annotated protein sequences , as well as a segmentation method for identifying conserved portions of a multiple alignment . section [ sec : likelihood ] describes a likelihood framework in which intron evolution can be analyzed in a theoretically sound manner . section [ sec : compress ] scrutinizes a compression technique that accelerates the evaluation of the likelihood function . the compression involves an @xmath4-time preprocessing step for @xmath5 sites and a phylogeny with @xmath6 species . we show that the subsequent evaluation takes sublinear , @xmath7 time almost surely in the yule - harding model of random phylogenies , even in the case of arbitrary , constant - size alphabets . fast evaluation is particularly important when the likelihood is maximized in a numerical procedure that computes the likelihood function with many different parameter settings . section [ sec : applications ] describes two applications of our methods . in one application , intron - aware alignment was used to validate some unexpected features of intron evolution before leca . in a second application , we analyzed intron evolution in 18 eukaryotic species . we found evidence of intron - rich early eukaryotes and a prevalence of intron loss over intron gain in recent times . orthologous introns can be identified by using whole genome alignments in the case of closely related genomes . for distant eukaryotes , however , intron orthology can only be established through protein alignments . the usual procedure is to project intron positions onto an alignment of multiple orthologous proteins . if introns in different species are projected to the same alignment position , then the introns are assumed to be orthologous . intron annotation can be included in protein alignments by defining intron match and mismatch scores . the alignment score is then computed as the sum of scores for amino acid matches , gaps , and intron matches . incorporating intron annotation should lead to better alignments at the amino acid level , and to a more reliable identification of orthologous introns . figure [ fig : iali ] shows an example of such improvement . intron scoring can be based on log - likelihood ratios in a probabilistic model . the model is defined by the joint distribution @xmath8 for the intron state in two sequences @xmath9 and @xmath10 , and the prior distributions @xmath11 and @xmath12 . aligned sites have states @xmath13 with probability @xmath14 if the two sites are unrelated , or with probability @xmath15 in case of homology . an @xmath13 alignment is scored with a value that is proportional to @xmath16 . we used the data set of to assess the strength of intron - match signals . since the data include no sites in which all species lack introns , but the model does allow for that , we added extra sites with no introns . the original data comprise 7236 intron sites in 684 genes , across 8 species . using a method described earlier , we added 35000 unobserved intron sites . using estimates for @xmath8 and @xmath17 , we computed the appropriate scores . the intron score is asymmetric and varies with evolutionary distance and intron conservation . matches for absent introns have an insignificant score , but shared introns have a high score , such as 93 ( human - plasmodium ) , 106 ( human - arabidopsis ) , 152 ( human-_s . pombe _ ) or 303 ( drosophila - anopheles ) on a 1/60-bit scale . shared introns thus give a signal comparable to amino acid matches : in the 1/60-bit scaled version of the vtml240 matrix , a tryptophan match scores 289 , and an arginine identity scores 113 . consider the case of aligning two protein sequences , @xmath9 and @xmath10 , which are annotated with the intron positions . every residue has two associated intron sites ( after the first and second nucleotides of their codons ) , and there is an intron site between consecutive amino acid positions ( phase-0 introns ) . intron sites may or may not be filled in by introns in either sequence . we use the notation for @xmath18}$ ] for the phase-0 site preceding the codon for amino acid @xmath19 , and @xmath20}$ ] , @xmath21}$ ] for phase-1 and -2 sites within the codon . intron presence is encoded by @xmath22 , and intron absence is encoded by @xmath23 throughout the paper . the intron annotation is specified by the variables @xmath24}\in\{{\texttt{1}},{\texttt{0}}\}$ ] . ( there can be no introns after the last amino acid . ) scores for aligned introns are specified by a @xmath25 scoring matrix @xmath26 . phase-1 and phase-2 intron sites are automatically placed by their associated amino acids . if @xmath27 is the amino acid scoring matrix , then the alignment of @xmath28 $ ] and @xmath29 $ ] entails a score of @xmath30}}}\\{{{t}[j]}}\end{smallmatrix}}\bigr]}+{{\lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{{t}[j]}}\end{smallmatrix}}\bigr]}$ ] with @xmath31}}}\\{{{t}[j]}}\end{smallmatrix}}\bigr ] } = { { \lambda}\bigl[{\begin{smallmatrix}{{{s}[i\colon1]}}\\{{{t}[j\colon1]}}\end{smallmatrix}}\bigr]}+{{\lambda}\bigl[{\begin{smallmatrix}{{{s}[i\colon2]}}\\{{{t}[j\colon2]}}\end{smallmatrix}}\bigr]}$ ] . similarly , aligning @xmath28 $ ] with an indel implies a score of @xmath30}}}\\{{\texttt{-}}}\end{smallmatrix}}\bigr]}+{{\lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{\texttt{0}}}\end{smallmatrix}}\bigr]}={{m}\big[{\begin{smallmatrix}{{{{s}[i]}}}\\{{\texttt{-}}}\end{smallmatrix}}\bigr]}+{{\lambda}\bigl[{\begin{smallmatrix}{{{s}[i\colon1]}}\\{{\texttt{0}}}\end{smallmatrix}}\bigr]}+{{\lambda}\bigl[{\begin{smallmatrix}{{{s}[i\colon2]}}\\{{\texttt{0}}}\end{smallmatrix}}\bigr]}$ ] , in addition to possible gap opening and closing penalties . standard alignment procedures need to be modified to deal with phase-0 introns , since the placement of phase-0 introns is not fixed with respect to gaps . it is not possible to simply add a new character to the alphabet to represent phase-0 introns because they affect gaps differently from amino acids . we added intron scoring into a multiple alignment framework , using a sum - of - pairs scoring policy with affine gap - scoring . it is np - hard to find the alignment of two multiple alignments under these optimization criteria . even the alignment of a single sequence to a multiple alignment necessitates sophisticated techniques . our solution therefore uses a gap - counting heuristic : namely , a gap - open penalty is triggered for an indel aligned with an amino acid if the indel is preceded by an amino acid or a phase-0 intron . gap opening thus corresponds to a pattern @xmath32 or @xmath33 . here , @xmath34 are intron states for the phase-0 site , and @xmath35 denotes either state . in addition , @xmath36 denotes an arbitrary amino acid , @xmath37 is the indel character , and @xmath38 is either of the latter two . we implemented affine gap - scoring by separate gap - open and -close penalties , so that gaps at the alignment extremities can be penalized less severely . gap closing is counted for the patterns @xmath39 and @xmath40 . table [ tbl : ali ] gives the recurrences for a dynamic programming algorithm that aligns an intron - annotated sequence @xmath9 to a multiple alignment @xmath41 of @xmath42 intron - annotated protein sequences . in order to simplify the presentation , we represent the sequences in such a way that every odd position of @xmath9 and @xmath41 is a regular residue or alignment column , annotated with information on the presence of phase-1 and phase-2 introns , whereas every even position is a phase-0 intron site . we use @xmath31}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr]}$ ] to denote the sum of intron - match scores for the intron sites associated with the positions @xmath28 $ ] , @xmath43 $ ] . we use also the shorthand @xmath44}$ ] to denote scoring for the alignment of an amino acid @xmath45 with an amino acid profile @xmath46 . the algorithm uses three types of variables , @xmath47}$ ] , @xmath48}$ ] and @xmath49}$ ] , which correspond to partial prefix alignments ending with aligned residues , gaps in @xmath9 , or gaps in @xmath41 , respectively . in case of @xmath48}$ ] , the last indel must be aligned with an amino acid column , and , thus @xmath50 must be odd ; for @xmath49}$ ] , @xmath19 must be odd . @xmath51 } & = \begin{aligned}[t ] & { { m}\big[{\begin{smallmatrix}{{{{s}[i]}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr]}+{{\lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr]}+{\gamma^{(-)}_{1}}(j ) + \max\bigl\ { { \mathsf{a}[i-2,j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{1}}(j ) + { \gamma^{(>)}_{1}}(j),\\ * & { \mathsf{gs}[i-2,j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{1}}(j ) + { \gamma^{(>)}_{2}}(j-2 ) , { \mathsf{gs}[i-1,j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{1}}(j ) + { \gamma^{(>)}_{2}}(j-2),\\ * & { \mathsf{gp}[i-2,j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{2}}(j ) + { \gamma^{(>)}_{3}}(j ) , { \mathsf{gp}[i-2,j-1 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(>)}_{2}}(j ) \bigr\}\end{aligned } & \text{odd $ i$ , odd $ j$}\\ { \mathsf{gs}[i , j ] } & = \begin{aligned}[t ] & { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(-)}_{2}}(j ) + \max\bigl\ { { \mathsf{a}[i , j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j ) + { \gamma^{(>)}_{1}}(j),\\ * & { \mathsf{gs}[i , j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } , \\ * & { \mathsf{gp}[i , j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j)+{\gamma^{(>)}_{3}}(j ) , { \mathsf{gp}[i , j-1 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j ) + { \gamma^{(>)}_{2}}(j ) \bigr\ } \end{aligned } & \text{odd $ i$ , odd $ j$}\\ { \mathsf{gp}[i , j ] } & = \begin{aligned}[t ] & { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(-)}_{1}}(j ) + \max\bigl\ { { \mathsf{a}[i-2,j ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j),\\ * & { \mathsf{gs}[i-2,j]}+ { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j ) + { \gamma^{(>)}_{2}}(j ) , { \mathsf{gs}[i-1,j]}+ { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j ) + { \gamma^{(>)}_{2}}(j),\\ * & { \mathsf{gp}[i-2,j]}+ { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } \bigr\ } \end{aligned } & \text{odd $ i$ , odd $ j$ } \\ { \mathsf{gs}[i , j ] } & = \begin{aligned}[t ] & { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(-)}_{2}}(j ) + \max\bigl\ { { \mathsf{a}[i-1,j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j ) + { \gamma^{(>)}_{1}}(j ) , \\ * & { \mathsf{gs}[i-1,j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{4}}(i , j ) + { \gamma^{(>)}_{4}}(i , j ) , { \mathsf{gs}[i , j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } , \\ * & { \mathsf{gp}[i-1,j-2 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{{{p}[j-1]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j ) + { \gamma^{(>)}_{3}}(j ) , { \mathsf{gp}[i-1,j-1 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{3}}(j ) + { \gamma^{(>)}_{2}}(j ) \bigr\ } \end{aligned } & \text{even $ i$ , odd $ j$ } \\ { \mathsf{gp}[i , j ] } & = \begin{aligned}[t ] & { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr ] } + { \gamma^{(-)}_{1}}(j ) + \max\bigl\ { { \mathsf{a}[i-2,j-1 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{5}}(j ) + { \gamma^{(>)}_{5}}(j),\\ * & { \mathsf{gs}[i-2,j-1 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{5}}(j ) + { \gamma^{(>)}_{2}}(j-1 ) , { \mathsf{gs}[i-1,j-1 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{\texttt{0}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{5}}(j ) + { \gamma^{(>)}_{2}}(j-1),\\ * & { \mathsf{gp}[i-2,j-1 ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{{{p}[j]}}}\end{smallmatrix}}\bigr ] } + { \gamma^{(<)}_{6}}(j ) + { \gamma^{(>)}_{6}}(j ) , { \mathsf{gp}[i-2,j ] } + { { \lambda}\bigl[{\begin{smallmatrix}{{{{s}[i-1]}}}\\{{\texttt{0}}^h}\end{smallmatrix}}\bigr]}\bigr\ } \end{aligned } & \text{odd $ i$ , even $ j$}\end{aligned}\ ] ] gaps are scored by using affine penalties , with gap - open , -extend , and -close scores , denoted by @xmath52 . the gap - counting heuristic implies that gap scores in the equations of table [ tbl : ali ] are defined by the number of certain patterns in up to three consecutive alignment columns . for instance , @xmath53 equals the gap - open penalty multiplied by the number of such rows in @xmath41 where column @xmath50 contains an indel , and column @xmath54 has a phase-0 intron . the corresponding pattern is described as @xmath55 . table [ tbl : gaps ] lists the patterns for the gap - counting heuristic . @xmath56}}$ has intron , otherwise nothing } \\ * { \gamma^{(<)}_{5 } } & { \texttt{1}}\text { or } { \texttt{x}}{\texttt{0 } } & { \gamma^{(<)}_{6 } } & { \texttt{1}}\\ * { \gamma^{(>)}_{1 } } & { \texttt{-}}{\texttt{0}}{\texttt{x}}\text { or } { \texttt{-}}{\texttt{1}}{\texttt { * } } & { \gamma^{(>)}_{2 } } & { \texttt{x}}\\ * { \gamma^{(>)}_{3 } } & { \texttt{1}}{\texttt{*}}\text { or } { \texttt{0}}{\texttt{x } } & { \gamma^{(>)}_{4 } } & { \texttt{x}}\text { if $ { { { s}[i]}}$ has intron , otherwise nothing } \\ * { \gamma^{(>)}_{5 } } & { \texttt{-}}{\texttt{1 } } & { \gamma^{(>)}_{6 } } & { \texttt{1}}\\ * { \gamma^{(-)}_{1 } } & { \texttt{- } } & { \gamma^{(-)}_{2 } } & { \texttt{x}}\end{array}\ ] ] table [ tbl : ali ] does not show the initialization of the variables , nor the final gap - counting : they employ a logic analogous to the recurrences . at the end of the algorithm , the best of @xmath57}$ ] , @xmath58}$ ] , @xmath59}$ ] is selected , and the actual alignment is found by standard traceback techniques . we implemented the algorithm in java . the program iteratively realigns one sequence at a time to the rest of the sequences in a multiple alignment . instead of sequence - dependent intron match - mismatch scores , the implementation uses only two parameters : one for intron conservation and another for intron loss / gain . in order to reliably identify orthologs , we need to be able to distinguish regions of the alignment that are highly conserved from those that are less well - conserved . in poorly conserved regions , we can not confidently infer intron orthology . proposed post - processing pairwise sequence alignments into alternating blocks that score above a threshold parameter @xmath60 or below @xmath61 . we attained a similar goal by adapting algorithmic technniques from . the procedure separates a multiple alignment into alternating high- and low - scoring regions . using a complexity penalty @xmath60 , a segmentation with @xmath62 high - scoring regions has a segmentation score of @xmath63 where @xmath64 is the sum of scores for the aligned columns . column scores are computed without gap - open and -close penalties . the best segmentation of an alignment of length @xmath5 can be found in @xmath65 time after the column scores are computed . the result of this computation is that the total of column scores in each selected high - scoring region will be greater than @xmath60 . there may be small sub - regions of negative scores , but the total score of such a sub - region can not be less than @xmath61 . conversely , unselected regions score below @xmath61 and can not have sub - regions scoring above @xmath60 . we use a markov model for intron evolution , as in previous studies . for the sake of generality , we describe the markov model over an arbitrary alphabet @xmath66 of fixed size @xmath67 . the intron alphabet is @xmath68 . phylogeny _ over a set of species @xmath69 is defined by a rooted tree @xmath70 and a probabilistic model . the leaves are bijectively mapped to elements of @xmath69 . each tree node @xmath71 has an associated random variable @xmath72 , which is called its _ state _ or _ label _ , that takes values over @xmath66 . the tree @xmath70 with its parameters defines the joint distribution for the random variables @xmath72 . the distribution is determined by the _ root probabilities _ @xmath73 and _ edge transition probabilities _ @xmath74 assigned to every edge @xmath75 . the root probabilities give the distributon of the root state . edge transition probabilities define the conditional distributions @xmath76 . along every path away from the root , the node states form a markov chain . the leaf states form the _ character _ an input data set ( or _ sample _ ) consists of independent and identically distributed ( iid ) characters : @xmath78 . in case of intron evolution , introns are generated by a two - state continuous - time markov process with _ gain _ and _ loss rates _ @xmath79 along each edge @xmath75 . the edge length is denoted by @xmath80 . using standard results , the transition probabilities on edge @xmath75 with rates @xmath81 and length @xmath82 can be written as @xmath83 , @xmath84 , and @xmath85 , @xmath86 . in the absence of established edge lengths , we fix the edge length scaling in such a way that @xmath87 . independent model parameters are thus @xmath88 , @xmath89 and @xmath80 for all edges @xmath75 . it is important to allow for branch - dependent rates , since loss and gain rates vary considerably between lineages . in a maximum likelihood approach , model parameters are set by maximizing the likelihood of the input sample . let @xmath90 be the input data . every @xmath91 is a vector of @xmath6 states , and we write @xmath92 to denote the observed state of leaf @xmath71 . by independence , the likelihood of @xmath93 is the product @xmath94 . each character s likelihood can be computed in @xmath95 time , using a dynamic programming procedure introduced by . the procedure relies on a `` pruning '' technique , which consists of computing the conditional likelihoods @xmath96 for every node @xmath71 and letter @xmath97 . @xmath96 is the probability of observing the leaf labelings in the subtree of @xmath71 , given that @xmath71 is labeled with @xmath97 . the likelihood for the character @xmath45 equals @xmath98 . intron data are somewhat unusual in that an all-@xmath23 character is never observed : the input does not include sites in which introns are absent in all of the organisms . the uncorrected likelihood function is therefore misleading , as it underestimates the probability of intron loss . to resolve this difficulty , we employ a correction technique proposed by for restriction sites . ( describes an alternative technique based on expectation maximization . ) we compute the likelihood under the condition that the input does not include all-@xmath23 characters . we use therefore the corrected likelihood @xmath99 , and maximize @xmath100 . our goal is to give a reliable characterization of the dynamics of intron evolution . in particular , we aim to give estimates for intron density in ancestral species , and for intron loss and gain events on the edges . notice that the estimation method needs to account for ancestral introns even if they got eliminated in all descendant lineages . it is possible to do that with the help of conditional expectations , which fit naturally into a likelihood framework . for an observed character @xmath45 , we define _ upper conditional likelihoods _ @xmath101 so that @xmath102 . upper conditional likelihoods are computed with dynamic programming , from the root towards the leaves , in @xmath4 time , even for non - binary trees . similar computations are routinely used in dna and protein likelihood maximization programs . here we allow irreversible probabilistic models , which explains why @xmath101 must be computed in a top - down fashion in . the posterior probability for the state at node @xmath71 is @xmath103 the posterior probabilities for state changes on an edge @xmath104 are @xmath105 now , the number of ancestral introns is estimated as the conditional expectation @xmath106 . the formula takes into consideration unobserved intron sites , by estimating their number @xmath107 as the mean of a negative binomial random variable . the number of intron state changes is estimated as @xmath108 . in particular , @xmath109 is the number of introns lost , and @xmath110 is the number of introns gained on the edge leading to @xmath111 . in order to compute @xmath112 , we initialize @xmath113 . on every edge @xmath104 , there are many heuristics that accelerate likelihood - based phylogenetic reconstruction , which mostly concentrate on the exploration of the tree space . we propose an improvement to the evaluation of the likelihood function , which normally takes linear time in the input size . our evaluation method yields an @xmath115 speedup for typical trees . we use it to optimize the parameters of intron evolution on a known tree , but the method is generally applicable to phylogenetic likelihood problems where the likelihood is numerically optimized . the key idea is that it is enough to carry out the pruning algorithm once for every different labeling within a subtree . different labelings within each subtree can be computed in a preprocessing step . subsequent evaluations of the likelihood function with different model parameters are faster , and depend on the number of different labelings in the data set . here we describe how the preprocessing step can be carried out in @xmath4 time . secondly , we analyze the computational complexity of subsequent evaluations , and show that an @xmath115 speedup is achieved for almost all random trees in the yule - harding model . the latter analysis produces some novel concentration results on the number of subtrees with a fixed size @xmath62 . to our knowledge , the closest idea to ours was articulated by . specifically , they proposed identifying characters in which leaves in a subtree have identical labels . they reported that in benchmark experiments with nucleotide sequences , likelihood optimization was accelerated by 1215% through this technique . the technique relies entirely on the fact that alignments of closely related sequences exhibit high levels of identity , and can not be extended to non - identical labelings . the yule - harding distribution is encountered in random birth and death models of species and in coalescent models , and is thus one of the most adequate random models for phylogenies . in one of the equivalent formulations , a random tree is grown by adding leaves one by one in a random order . the leaves are first numbered by using a random uniform permutation of the integers @xmath116 . leaves are joined to the tree in an iterative procedure . in step 1 , the tree is just leaf 1 on its own . in step 2 , the tree is a `` cherry '' with leaves @xmath117 and @xmath118 . in each subsequent step @xmath119 , a random leaf @xmath120 is picked uniformly from the set @xmath121 . the new leaf @xmath19 is added to the tree as the sibling of @xmath120 , forming a cherry : a new node is placed on the edge leading to @xmath120 and @xmath19 is connected to it . the resulting random tree in iteration @xmath6 has the yule - harding distribution . 0.98 c algorithm compress * for * every leaf @xmath71 * do * initialize @xmath122{\leftarrow}0 $ ] // _ @xmath123 $ ] is the first occurrence of state @xmath50 _ * for * @xmath124 * do * * if * @xmath125 = 0 $ ] * then * @xmath125{\leftarrow}i$ ] set @xmath126 $ ] * for * every non - leaf node @xmath71 in a post - order traversal * do * initialize the map @xmath127 as empty [ line : compression.map.init ] let @xmath128 be the left and right children of @xmath71 * for * @xmath124 * do * * if * @xmath129 * then * @xmath130 [ line : compression.map.set ] set @xmath131 [ line : compression.map.get ] the likelihood can be computed faster by first identifying the different @xmath91 values , along with their multiplicity in the data . for large trees , the input typically consists of many different labelings , but for small trees with @xmath132 , the compression is useful , since the number of different @xmath91 values is bounded by @xmath133 . in order to exploit the benefits of compression , we extend it to every subtree . [ def : compression ] define the multiset of observed labelings for every node @xmath71 as follows . let @xmath134 denote the number of leaves in the subtree rooted at @xmath71 , and let @xmath135 denote those leaves . define @xmath136 for every labeling @xmath137 of the leaves in the subtree of @xmath71 as the number of times @xmath138 is observed in the data : @xmath139 where @xmath140 denotes the indicator function that takes the value 1 if its argument is true , otherwise it is 0 . define also the set of observed labelings @xmath141 the multisets of observed labelings are computed in the preprocessing step . the likelihood function is evaluated subsequently by computing the conditional likelihoods at each node @xmath71 for the labelings of @xmath142 only , in @xmath143 time . in order to compute @xmath144 , we use a recursive procedure . it is important to avoid working with the @xmath95-dimensional vectors @xmath138 of definition [ def : compression ] directly , otherwise the preprocessing may take superlinear time in @xmath6 . for that reason , every labeling @xmath145 is represented by the index @xmath19 for which @xmath91 is the first occurrence of @xmath138 . accordingly , we compute an auxiliary array @xmath146 , which stores the first occurrence of each labeling @xmath91 in @xmath71 s subtree . in particular , @xmath147 if @xmath148 is the smallest index such that @xmath149 for all @xmath62 where @xmath150 are the leaves in @xmath71 s subtree . figure [ fig : compression ] shows that the values @xmath146 can be computed in a post - order traversal . after @xmath146 are computed for all @xmath19 and @xmath71 , the multiplicities @xmath144 and observed labelings @xmath142 are straightforward to calculate in linear time . the map @xmath151 in lines [ line : compression.map.init][line : compression.map.get ] is sparse with at most @xmath5 entries , and can be implemented as a hash table so that accessing and updating it takes @xmath152 time . consequently , algorithm compress takes @xmath4 time . 0.98 l algorithm loglikelihood * for * every leaf @xmath71 do set @xmath153[a']{\leftarrow}\chi\{a = a'\}$ ] for all @xmath154 and @xmath155 * for * every non - leaf node @xmath71 in a post - order traversal * do * * for * every labeling @xmath145 * do * let @xmath156 be the children of @xmath71 , and let @xmath157 be their subtree labelings * for * every @xmath158 @xmath159[a ] { \leftarrow}\prod_{j=1,2}\sum_{a'\in{\mathcal{a } } } p_{uu_j}(a\rightarrow a ' ) l_{u_j}[y_j][a']$ ] set @xmath160 * for * every labeling @xmath161 * do * @xmath162[a]\bigr)$ ] * return * @xmath163 after the preprocessing step , the conditional likelihoods are computed only for the different labelings within each subtree . figure [ fig : likelihood ] shows the evaluation of the likelihood function . the running time for the algorithm is @xmath164 where @xmath165 is the total number of different labelings within all subtrees : @xmath166 if @xmath167 denotes the number of leaves in the subtree rooted at @xmath71 , then @xmath142 has at most @xmath168 elements . hence , @xmath165 is bounded as @xmath169 observe that by sheer number of arithmetic operations , it is always worth evaluating the likelihood function this way . the worst situation is that of a caterpillar tree ( where every inner node has a leaf child ) . in that case , there are only a few ( @xmath170 ) non - leaf nodes for which we can compress the data , and it is possible to construct an artificial data set in which there are @xmath5 different labelings for @xmath171 subtrees . caterpillar trees , however , are rare in phylogenetic analysis . typical phylogenies have fairly balanced subtrees . in what follows , we examine the bound of more closely in the yule - harding model . the analysis relies on a characterization of the random number of subtrees with a given size , as expressed in theorems [ tm : sizes.mean ] and [ tm : sizes.concentration ] . [ tm : sizes.mean ] consider random evolutionary trees with @xmath6 leaves in the yule - harding model . let @xmath172 denote the number of subtrees with @xmath173 leaves in a random tree . the expected value of @xmath172 is @xmath174 trivially , @xmath175 . derives equation by appealing to a plya urn model . an equivalent result is stated by . @xmath176 [ tm : sizes.concentration ] for all @xmath177 , [ eq : sizes.tail ] @xmath178 consider the random construction of the tree . let @xmath120 denote the random leaf picked in step @xmath19 to which leaf @xmath19 gets connected , for @xmath179 . each random variable @xmath120 is uniformly distributed over the set @xmath121 , and @xmath180 are independent . moreover , they completely determine the tree @xmath10 at the end of the procedure . consequently , @xmath172 is a function of @xmath181 : @xmath182 . the key observation for the concentration result is that if we change the value of only one of the @xmath120 in the series , then @xmath172 changes by at most two . in order to see this , consider what happens to the tree @xmath10 , if we change the value of exactly one of the @xmath120 from @xmath138 to @xmath183 . such a change corresponds to a `` subtree prune and regraft '' transformation . specifically , the subtree @xmath184 , defined as the child tree of the lowest common ancestor @xmath71 of @xmath138 and @xmath19 containing the leaf @xmath19 , is cut from @xmath10 , and is reattached to one of the edges on the path from the root to @xmath183 . now , such a transformation does not affect @xmath172 by much . notice that subtree sizes are strictly monotone decreasing from the root on every path . on the path from the root to @xmath71 , subtree sizes decrease by the size @xmath185 of @xmath184 , and @xmath71 disappears , contributing a change of @xmath186 , @xmath187 or @xmath188 to @xmath172 . ( at most one subtree of size @xmath189 that contains @xmath184 now has size @xmath62 , and at most one subtree of original size @xmath62 is not counted anymore : it may be @xmath71 s subtree itself , or a subtree above it . ) an analogous argument shows that regrafting contributes a change of @xmath186 , @xmath187 , or @xmath188 . hence , the function @xmath190 is such that by changing one of its arguments , its value changes by at most @xmath118 . as a consequence , mcdiarmid s inequality ( @xcite ) can be applied to bound the probabilities of large deviations for @xmath172 . in particular , for all @xmath177 , @xmath191 where @xmath192 and @xmath193 since @xmath194 for all @xmath19 , @xmath195 , implying eq . . an identical bound holds for the right - hand tail of the distribution . @xmath176 remark . the particular case of @xmath196 was considered by . they showed that the distribution of @xmath197 is asymptotically normal with mean @xmath198 and variance @xmath199 . the result suggests that the best constant factor in the exponent of eqs . is @xmath200 for @xmath201 , instead of @xmath202 . gave exact formulas for the variance of @xmath172 . he showed that the variance of @xmath172 is @xmath203 . the variance formulas were given earlier in a different context by . ( see also the discussion by . ) the result suggests that by analogy with the cherries , the best constant factor in the exponent is @xmath204 . it is thus plausible that the probability is properly bounded by @xmath205 in theorem [ tm : compute ] below . [ tm : compute ] with probability @xmath206 , the likelihood function can be evaluated for random trees in the yule - harding model in @xmath207 time after an initial preprocessing step that takes @xmath4 time . evaluating the likelihood function takes @xmath4 time in the worst case , and @xmath208 time on average . we need the following lemma for the proof of theorem [ tm : compute ] . [ lm : sum.q^i / i ] for all @xmath209 , @xmath210 . for all @xmath211 and @xmath212 , @xmath213 . the proof is straightforward by induction in @xmath214 . notice that the right order of magnitude is @xmath215 but we need a bound for all @xmath214 . @xmath176 .effect of the compression on three data sets . the fourth column quantifies the direct evaluation method , the fifth column quantifies the effect of the compression restricted to the root , the sixth column corresponds to the bound of eq . and the seventh column gives the exact value of @xmath165 . the first data set is from rogozin et al . ( 2003 ) , the second data set is the one analyzed here in [ ss : app.ancestors ] , and the third one is an unpublished data set we have worked on , where an ambiguity character is included in the alphabet . [ cols=">,>,^ , > , > , > , > " , ] genbank flatfiles and protein sequences were downloaded from refseq and ensembl . exon - intron annotation was extracted from the genbank flatfiles . the _ c. briggs _ protein sequences and genome annotation were downloaded from wormbase , and intron annotation was extracted from the gff annotation file . table [ tbl : data ] lists the data sources and the species abbreviations . we used the 684 ortholog sets , each corresponding to a cluster of orthologous groups , or kog , from the study of as `` seeds '' for compiling a set of putative orthologs for our species . for each seed ( consisting of homologous protein sequences for eight species ) , we performed a position - specific iterated blast search . in case of plasmodium species , we used three iterations against a database of all protozoan peptide sequences in refseq . we used an e - value cutoff of @xmath216 for retaining candidates . each candidate hit was then used as a query in a reversed position - specific blast ( rpsblast ) search against the kog database . candidates were retained after this point if they had the highest scoring hit ( by rpsblast ) against the same kog as the kog of the seed data , and they scored within 80% of the best such hit for the species . from the resulting set of paralogs , we selected a putative ortholog set in the following manner . for each kog , we selected all possible triples of human - arabidopsis - saccharomyces paralogs and kept the triple that had the highest score in alignments built by muscle . alignment score was computed using the vtml240 matrix . additional putative orthologs were added for one species at a time , by aligning each paralog individually to the current profile , and keeping the one that gave the largest alignment score . at this iterative addition , scoring was done with the vtml240 matrix , by summing the five highest pairwise scores between a candidate and already included sequences . the resulting sets were then realigned using muscle , and then realigned again using our intron - aware alignment with a gap penalty of 300 , gap - extend penalty of 11 , vtml240 amino acid scoring , intron - match scores of 300 and intron - mismatch penalties of 20 . ( these latter were established using different intron - match and -mismatch scores , and selecting the ones that gave the fewest number of intron sites while decreasing the score of the implied protein alignment by less than 0.1% . ) conserved portions were extracted using our segmentation program with a complexity penalty of @xmath217 ( larger values gave identical segmentation results , and lower values resulted in too many scattered blocks ) . we penalized indels with an infinitely large value to exclude gap columns . phase-0 introns falling on the boundaries of conserved blocks were excluded . intron presence and absence in the aligned data was then extracted to produce the data for the likelihood programs . figure [ fig : posteriors ] shows the estimated intron densities for ancestral species . it is notable that ancestral nodes such as the bilaterian ancestor ( node 4 ) , the ecdysozoan ancestor ( node 5 ) , the opisthokont ancestor ( node 15 ) , and the bikont ancestor ( node 16 ) all have very high intron densities , surpassing most previous estimates . the ecdysozoan ancestor has an even higher estimated intron density ( 80% of human density ) than the otherwise quite generous estimates ( about 70% ) of , which is mostly due to the inclusion of the relatively intron - rich honeybee genes . intron density in the bilaterian ancestor is estimated to be almost as high as in humans ( 94% ) , agreeing with estimates of . sequences of a handful of intron - rich genes from the marine annelid _ platynereis dumerilii _ have already indicated that the bilaterian ancestor s genome was at least two - thirds as intron - rich as the vertebrate genomes even by conservative estimates . introns are for the most part lost on the branches . figure [ fig : changes ] shows the estimated changes in a few lineages . intron evolution has been much slower in the vertebrate lineage than in most other lineages : in more than 380 million years since the divergence with fishes , only about 3% of our introns got lost . fungi , for example , massively trimmed their introns in many lineages . a notable exception is _ c. neoformans _ , which seems to have gained introns , but that assessment may change if another basidiomycete genome becomes available besides the relatively intron - poor _ u. maydis_. we presented a novel alignment technique for establishing intron orthology , and a likelihood framework in which intron evolutionary events can be quantified . we described a compression method for the evaluation of the likelihood function , which has been extremely valuable in practice . we also showed that the compression leads to sublinear running times for likelihood evaluation . we illustrated our methods for analyzing intron evolution with a large and diverse set of eukaryotic organisms . the data set is more comprehensive than any used in other studies published to this day . the data indicate that ancestral eukaryotic genomes were more intron - rich than previous studies suggested . many circumstances influence intron loss , and realistic likelihood models need to introduce rate variation . usual rate variation models entail multiple evaluations of the likelihood function , and , thus , underline the importance of computational efficiency . we believe that the proposed methods will help to produce and analyze large data sets even within complicated likelihood models . this research is supported by a grant from the natural sciences and engineering research council of canada . adachi , j. and m. hasegawa ( 1995 ) . version 2.3 : programs for molecular phylogenetics based on maximum likelihood . volume 28 of _ computer science monographs _ , pp . tokyo , japan : institute of statistical mathematics . altschul , s. f. , t. l. madden , a. a. schffer , j. zhang , z. zhang , w. miller , and d. j. lipman ( 1997 ) . gapped blast and psi - blast : a new generation of protein database search programs . _ 25_(17 ) , 33893402 . ma , b. , z. wang , and k. zhang ( 2003 ) . alignment between two multiple alignments . in r. a. baeza - yates , e. chvez , and m. crochemore ( eds . ) , _ proc . combinatorial pattern matching ( cpm ) _ , volume 2676 of _ lncs _ , pp . springer . raible , f. , k. tessmar - raible , k. osoegawa , p. wincker , c. jubin , g. balavoine , d. ferrier , v. benes , p. de jong , j. weissenbach , p. bork , and d. arendt ( 2005 ) . vertebrate - type intron - rich genes in the marine annelid platynereis dumerilii . _ 310 _ , 13251326 . rogozin , i. b. , y. i. wolf , a. v. sorokin , b. g. mirkin , and e. v. koonin ( 2003 ) . remarkable interkingdom conservation of intron positions and massive , lineage - specific intron loss and gain in eukaryotic evolution . _ 13 _ , 15121517 . stamatakis , a. p. , t. ludwig , h. meier , and m. j. wolf ( 2002 ) . : a fast program for sequential and parallel phylogenetic tree calculations based on the maximum likelihood method . in _ proc . ieee computer society bioinformatics conference ( csb ) _ , pp .

in the recent study on quantum field theories and quantum mechanics , topologically neutral soliton molecules , which are locally composed of ( fractional ) instantons and anti - instantons , have been attracting a great deal of attention in relation to the ir - renormalon@xcite . imaginary ambiguities arising in amplitudes of such topologically neutral configurations can cancel out those arising in non - borel - summable perturbative series ( ir - renormalon ) in quantum theories under certain conditions on the spacetime manifold @xcite . in field theories on compactified spacetime with a small compact dimension , these objects are termed as bions " @xcite . it is expected that full semi - classical expansion including perturbative and non - perturbative sectors as bions , which is called resurgent " expansion @xcite , leads to unambiguous and self - consistent definition of field theories in the same manner as the conjecture in quantum mechanics @xcite . the resurgence in theoretical physics was at first investigated in the matrix model and topological string theory @xcite . then , the study on the topic has been extended to abjm theory @xcite , string and supersymmetric gauge theories @xcite , and general quantum systems @xcite . bions and resurgence in non - susy field theories , especially in the low - dimensional models , have been extensively investigated for the @xmath0 model @xcite , the grassmann sigma model @xcite , the principal chiral model @xcite , and the @xmath2 model @xcite . according to these studies , the leading - order renormalon ambiguity @xmath3 arising in non - borel - summable perturbative series , which corresponds to the singularity closest to the origin on the borel plane , is compensated by the amplitude of neutral bions . on the other hand , it is expected but not verified that the ambiguities corresponding to singularities further from the origin ( @xmath4 , @xmath5 , ... ) are cancelled by amplitudes of bion molecules with more than four instanton constituents . in the case of quantum mechanics , not only the sector of zero instanton charge but also those of nonzero instanton charge contribute to physical observables such as the energy levels . the authors in refs . @xcite investigated quantum mechanics with several types of potential including the sine - gordon type . they showed that the leading instanton contributions are consistent with the perturbative calculation , and conjectured the explicit equation connecting perturbative and instanton contributions , which they call the generalized quantization condition . recently the authors in refs . @xcite adopted the uniform - wkb method based on the boundary condition , which is equivalent to the quantization condition in refs . @xcite , and pointed out the general relation between perturbative and non - perturbative contributions . explicit calculations of multi - instanton amplitudes at each configuration level are expected to clarify the structure of resurgence and to verify the conjectured relation between perturbative and non - perturbative contributions @xcite . in this paper , we focus on a quantum mechanical system with the sine - gordon potential , and we calculate the multi - instanton amplitude by explicitly integrating quasi moduli parameters corresponding to separations of instanton - constituents in a semi - classical limit , in comparison with the uniform wkb calculations @xcite . we adopt an extension of bogomolnyi zinn - justin prescription @xcite for multi - instanton configurations with an appropriate subtraction scheme for divergent parts . we calculate contributions to the energy eigenvalue of the lowest band from each multi - instanton configuration in a semi - classical limit ( @xmath6 ) . for the configurations with only instantons ( anti - instantons ) such as @xmath7 $ ] , @xmath8 $ ] and @xmath9 $ ] , we have unambiguous results without imaginary parts . here , we have denoted an instanton ( anti - instanton ) as @xmath10 ( @xmath11 ) . for configurations containing both instantons and anti - instantons such as @xmath12 $ ] , @xmath13 $ ] and @xmath14 $ ] , the results contain ambiguous imaginary parts , which depend on the path of analytic continuation . these imaginary parts correspond to the large - order behavior of perturbation series around the saddle point without the @xmath12 $ ] pair . for instance , we show explicitly that the imaginary part of the multi - instanton amplitude @xmath13 $ ] cancels the imaginary part of the borel resummation of the large - order perturbation series around the nontrivial background with a single instanton @xmath15 $ ] . by investigating the uniform - wkb calculations in detail , we verify that all of our results agree completely with those based on the uniform - wkb calculations up to a four - instanton order . while the sine - gordon quantum mechanics is worth to study on its own , another strong motivation lies in its close relationship to small circumference limit of the two - dimensional @xmath0 model on @xmath1 ( circumference @xmath16 ) with the @xmath17-symmetric twisted boundary condition @xcite . however , we observe that some of the field configurations of the @xmath0 model are not faithfully represented by means of the sine - gordon quantum mechanics . one can derive the sine - gordon quantum mechanics from the two - dimensional @xmath0 model by applying the scherk - schwarz dimensional reduction , which requires a particular dependence of the phases of fields on the coordinate @xmath18 of compactified dimension ( @xmath19 ) . it is important to realize that only parts of field configuration of @xmath0 model can be consistent with this @xmath18 dependence . for instance , the bps solution of two fractional instantons is not consistent with the scherk - schwarz reduction , and hence its small circumference limit can not be described by the sine - gordon quantum mechanics . on the other hand , two adjacent instantons in the sine - gordon quantum mechanics are mutually non - bps , although each individual instanton may be understood as a limit of bps fractional instanton ( with a different @xmath18 dependence ) . even in the instanton and anti - instanton configurations , @xmath0 model has a significant difference compared to the sine - gordon quantum mechanics : the phase moduli of the fractional instantons in the @xmath0 model are neglected in the sine - gordon quantum mechanics . for the configuration of a neutral bion composed of a fractional instanton and an anti - fractional instanton , we find that the interaction between them strongly depends on the relative phase of constituents . we calculate the neutral bion contribution in the @xmath0 model , based on the interaction potential with the quasi moduli parameter corresponding to the relative phase between the fractional instanton and anti - instanton . we find that this calculation gives a correction factor compared to the neutral bion amplitude obtained in the sine - gordon quantum mechanics @xcite . this paper is organized as follows . in sec . [ sec : sg ] , we review instantons and their interactions in the quantum mechanics with sine - gordon potential and the borel summation . in sec . [ sec : ic ] we calculate amplitudes of multi - instanton configurations in sine - gordon quantum mechanics by integrating out the moduli parameters . in sec . [ sec : wkb ] we discuss the results from the uniform wkb calculations , and show that they completely agree with the instanton moduli calculations . in sec . [ sec : cpn ] we discuss the neutral bion contributions in the compactified @xmath0 model based on the interaction potential including the relative phase parameter . section [ sec : sd ] is devoted to a summary and discussion . in appendix [ app1 ] we give some details of four - instanton calculations . in this article , we focus on the sine - gordon quantum mechanics described by the schrdinger equation @xmath20 where we follow the notation in refs . @xcite except @xmath21 is replaced by @xmath22 here . the euclidian lagrangian for the sine - gordon quantum mechanics is given by , and their euclidean lagrangian is @xmath23 . ] @xmath24 in the @xmath25 limit , it reduces to the schrdinger equation of the harmonic oscillator . the energy eigenvalues of periodic potentials split into bands of states . within each band , they are labeled by the bloch angle @xmath26 $ ] defined by @xmath27 in this article , we are interested in the lowest band , although excited bands can be treated similarly . the energy eigenvalue @xmath28 of the lowest band can be expressed in terms of the path - integral @xmath29 for weak coupling , the path - integral has contributions @xmath30 around the perturbative vacuum , as well as contributions @xmath31 from nonperturbative saddle points @xmath32 perturbation series in powers of coupling constant @xmath33 in quantum field theories or in quantum mechanics are extremely useful , but are usually factorially divergent @xmath34 it is useful to define the borel transform @xmath35 @xmath36 the borel resummation @xmath37 of the divergent series @xmath30 is defined as an integral of the borel transform along the positive real axis in the complex borel plane @xmath38 @xmath39 if the factorially divergent series is alternating , the borel transform has no singularities along the positive real @xmath38 axis , and the borel resummation becomes well - defined ( borel - summable ) . for the potential with degenerate minima such as the sine - gordon quantum mechanics , however , the perturbation series is non - alternating factorially divergent . in that case , the borel transform is convergent with the finite radius of convergence , but the borel resummation is ill - defined because of singularities in the complex borel plane . since the series become alternating and the borel resummation @xmath37 is unambiguous for @xmath40 , we can analytically continue it from @xmath40 to the physical region @xmath41 to obtain a real analytic function @xmath37 . if there is no complex singularities , we obtain a branch cut along the positive real axis of complex @xmath33 plane . the imaginary part @xmath42 at @xmath41 is related to the large - order behavior ( @xmath43 ) of perturbation series @xmath30 in eq.([eq : perturbation - series ] ) through the dispersion relation @xcite @xmath44}{(g^2)^{k+1}}. \label{eq : dispersion_rel}\ ] ] this large - order behavior corresponds to the singularities of the borel transform @xmath45 in the complex borel plane @xmath38 . of course this ambiguous ( path - dependent ) imaginary part is unacceptable , and should disappear , since the energy eigenvalue @xmath28 should be real , and ambiguity due to the choice of path is unphysical . in fact , it has been found that the leading term of the imaginary ambiguities is cancelled by the contributions from non - perturbative saddle points associated with neutral objects composed of instantons @xcite . this phenomenon is called the resurgence of the perturbation series . let us now consider non - perturbative saddle points . by rescaling the variable @xmath46 the euclidean lagrangian in eq.([eq : lagrangian ] ) can be rewritten as @xmath47 and the instanton number as a topological charge may be defined by @xmath48 single instanton solution ( @xmath49 ) is given by . ] @xmath50 whereas single anti - instanton solution ( @xmath51 ) is given by @xmath52 with the euclidean action @xmath53 the moduli parameter @xmath54 is a zero mode ( moduli ) associated to the breakdown of translation , representing the location of the ( anti-)instanton . for even @xmath55 , the solutions ( [ eq : instanton ] ) and ( [ eq : anti - instanton ] ) satisfy the following bps equation of the ( anti-)bps equation ( [ eq : bps - bound ] ) and ( [ eq : anti - bps - bound ] ) is twice as large as the periodicity @xmath56 of the lagrangian in eq.([eq : lagrangian ] ) . ] saturating the bps bound for @xmath57 @xmath58 for odd @xmath55 , they satisfy the anti - bps equation saturating the anti - bps bound for @xmath57 @xmath59 by integrating over the translational zero mode @xmath54 , one finds the contribution @xmath60 of single instanton @xmath15 $ ] to the energy as @xmath61= - \left({e^{-s_{i}}\over{\sqrt{\pi g^{2}}}}\right)\ , e^{i\theta}\,.\ ] ] suppose , for instance , we have a bps instanton in eq.([eq : instanton ] ) with @xmath62 and wish to place another instanton or anti - instanton to its right , we are forced to take either instanton with @xmath63 in eq.([eq : instanton ] ) or anti - instanton with @xmath63 in eq.([eq : anti - instanton ] ) , both of which are anti - bps configurations . therefore two successive ( anti-)instantons are inevitably non - bps . the energy of the non - bps configuration of two successive instantons should be more than the sum of individual instanton energies . they are found to repel each other with the potential @xcite for large separations @xmath64 @xmath65\ , . \label{eq : interaction - i - i}\ ] ] the non - bps configuration of successive instanton and anti - instantons are found to attract each other with the potential for large separations @xmath64 @xmath66\ , . \label{eq : interaction - i - antii}\ ] ] for later convenience , we introduce the uniform - wkb ansatz by following ref . @xcite . with the coordinate variable @xmath67 in eq.([eq : rescaled_variable ] ) , eq.([seq1 ] ) can be rewritten as @xmath68 we define the potential as @xmath69 . by using the parabolic cylinder function @xmath70 satisfying the differential equation @xmath71 we introduce an ansatz for the wave function@xcite @xmath72 where the parameter @xmath73 is the shift of energy eigenvalue @xmath28 from the ground state energy of the harmonic oscillator ( @xmath74 limit ) . then the schrdinger equation ( [ seq2 ] ) becomes @xmath75 with @xmath76 . in the @xmath77 limit , eq.([eq : nonlinear_eq_u ] ) just reduces to @xmath78 , whose solution @xmath79 is @xmath80 which gives the zeroth - order argument of the parabolic cylinder function in eq.([dansatz ] ) and solves the schrdinger equation of the harmonic oscillator . in this section we calculate multi - instanton amplitudes in sine - gordon quantum mechanics . we need to integrate over the distances @xmath81 between instantons and ( anti-)instantons as quasi - moduli . since the interaction ( [ eq : interaction - i - i ] ) and ( [ eq : interaction - i - antii ] ) between instantons and ( anti-)instantons vanish at large distances , we need to regulate the integral by introducing a factor @xmath82 into the effective potential @xmath83 \,=\ , \pm { 2\over{g^{2}}}\exp(-r ) \,+\ , \epsilon r\ , , \label{pot2}\ ] ] where @xmath84 is for the instanton - instanton repulsive interaction and @xmath85 for the instanton anti - instanton attractive interaction . the regularization parameter @xmath82 can be identified as the number @xmath86 of fictitious fermions@xcite . after subtracting divergences , we need to take the limit @xmath87 . even after eliminating the divergence arising from large separations ( @xmath88 ) , we have another source of divergence for the case of the attractive instanton anti - instanton interaction : the integrand @xmath89 $ ] becomes divergent as @xmath90 , contrary to the repulsive case . therefore the moduli - integral gets divergent contributions from small @xmath81 regions , and is ill - defined in the semi - classical limit ( @xmath6 ) . this is why we need to introduce the bogomolnyi zinn - justin ( bzj ) prescription @xcite : we first regard @xmath91 as real positive ( @xmath92 ) to make the integral well - defined in the semi - classical limit , and then we analytically continue @xmath92 back to @xmath93 in the complex @xmath22 plane at the end of the calculation . the energy eigenvalue of the lowest band has contributions @xmath94 from the amplitude @xmath95_{\rm all}$ ] of @xmath55-instanton and @xmath96-anti - instanton configuration as @xmath97_{\rm all}\,,\ ] ] @xmath98_{\rm all}\,=\ , \left({e^{-s_{\mathcal i}}\over{\sqrt{\pi g^{2}}}}\right)^{n+m}\ , e^{i(n - m)\theta}\ , \int dr_{1}dr_{2} ... dr_{n+m-1 } \ , e^{-v[r_{1}]-v[r_{2}]-\cdot\cdot\cdot v[r_{n+m-1 } ] } \ , , \ ] ] where @xmath99_{\rm all}$ ] stands for the sum of configurations which can be composed of @xmath55 instantons and @xmath96 anti - instantons in all possible orderings . as shown in ref . @xcite , the contribution contains @xmath100 with @xmath101 being the instanton charge since the bloch angle @xmath102 shows up in a topological term @xmath103 in the euclidian action in eq.([eq : path - integral ] ) . we perform the quasi - moduli integral taking only interactions between neighboring instantons among the ( @xmath104 ) instantons . we should perform this multi - integral in the semi - classical region @xmath6 , and subtract the divergent parts appropriately at each level of the multi - integral . we will evaluate them explicitly from the next subsection . each horizontal line stands for the vacuum in the sine - gordon potential . , scaledwidth=40.0% ] the amplitude of two instantons shown in fig . [ fig : ii ] is obtained as @xmath105e^{-2i\theta}\xi^{-2}\,=\ , \int_{0}^{\infty}dr \exp\left(-{2\over{g^2}}e^{- r}-\epsilon r\right ) & \,=\ , \left({g^{2}\over{2}}\right)^{\epsilon } \int_{0}^{2/g^{2}}ds\ , e^{-s } s^{\epsilon-1 } \nonumber\\[5pt ] & \,\overset{|g^{2}|\ll 1}{\longrightarrow}\ , \left({g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \nonumber\\[5pt ] & \,=\ , -\left(\gamma\,+\,\log{2\over{g^{2}}}\right)\,+\ , o\left(\frac{1}{\epsilon}\right)\,+\ , o(\epsilon)\ , , \label{iia}\end{aligned}\ ] ] where @xmath106 is the euler constant and @xmath107 is an instanton factor defined by @xmath108 here we have neglected terms of order @xmath109 or higher . to simplify the formula , we divide the amplitude by @xmath110 and @xmath111 . precisely speaking , the interaction energy between instantons at small separation @xmath112 may not be precisely represented by the potential in eq.([pot2 ] ) . however , our result is unchanged as long as @xmath6 is satisfied . we need to subtract the divergent term @xmath113 while the @xmath114 term disappears in the @xmath115 limit . the contribution from this amplitude to the energy eigenvalue of the lowest band is then given by @xmath116 where the superscript @xmath117 stands for two - instanton and zero anti - instanton amplitude . we note that the contribution from the two anti - instanton amplitude is obtained by replacing @xmath111 by @xmath118 . the amplitude of one instanton and one anti - instanton amplitude is composed of two configurations @xmath12 $ ] and @xmath119 $ ] , as shown in fig . [ fig : iibar ] . in these cases , the interaction between the two constituents is attractive , and the quasi moduli integral is ill - defined . therefore we introduce the bogomolnyi zinn - justin ( bzj ) prescription @xcite : we first evaluate the integral by taking @xmath92 , and then we analytically continue the result from @xmath92 back to @xmath93 in the complex @xmath22 plane . this procedure provides the imaginary ambiguity depending on the path of the analytic continuation as @xmath120 . , [ \bar{\mathcal{i}}\mathcal{i}]$ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential . , title="fig:",scaledwidth=40.0% ] , [ \bar{\mathcal{i}}\mathcal{i}]$ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential . , title="fig:",scaledwidth=40.0% ] the amplitude of one - instanton and one anti - instanton configuration @xmath12 $ ] corresponding to the left of fig . [ fig : iibar ] is obtained as @xmath121\xi^{-2}&\,=\ , \int_{0}^{\infty}dr\ , \exp\left(-{2\over{-g^2}}e^{- r}-\epsilon r\right ) \overset{|g^{2}|\ll 1}{\longrightarrow}\ , \left({-g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \nonumber\\[5pt ] & \,\overset{-g^{2}= e^{\mp i\pi}g^{2}}{\longrightarrow}\ , -\left(\gamma\,+\,\log{2\over{e^{\mp i\pi}g^{2}}}\right ) \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\ , o(\epsilon ) \nonumber\\[5pt ] & \,=\ , -\left(\gamma\,+\,\log{2\over{g^{2}}}\right ) \mp i\pi \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\ , o(\epsilon)\ , , \label{eq : i - bari-1}\end{aligned}\ ] ] where we perform the integral in the first line by considering @xmath92 , and in the second line analytically continue @xmath92 back to @xmath93 in the complex @xmath22 plane @xcite . the third line shows a two - fold ambiguous expression of @xmath91 depending on the path of analytic continuation as @xmath120 . as with the two - instanton case , we have subtracted the divergent part @xmath113 while the @xmath114 term disappears in the @xmath115 limit . another amplitude of one - instanton and one anti - instanton configurations @xmath119 $ ] corresponding to the right of fig . [ fig : iibar ] turns out to give identical contribution as that in eq.([eq : i - bari-1 ] ) . the total contribution is given by the sum of them with dropping @xmath113 and @xmath114 terms , @xmath122\,+\,[\bar{\mathcal{i}}\mathcal{i } ] ) \xi^{-2}\,=\ , -2\left(\gamma\,+\,\log{2\over{g^{2}}}\right)\,\mp\ , 2i\pi\,.\ ] ] its contribution to the energy eigenvalue of the lowest band is then given by @xmath123\ , . \label{iibar}\ ] ] if the resurgence idea is valid , this imaginary ambiguity should cancel the imaginary ambiguity of the non - borel summable divergent series of the perturbative contribution . @xmath124 + { \rm im}[e_{\rm pert}]=0 . \label{imag_cancel}\ ] ] if we insert eqs.([iibar ] ) and ( [ imag_cancel ] ) into the dispersion relation in eq.([eq : dispersion_rel ] ) , we should be able to reproduce the large - order behavior of the perturbation series @xmath125\over{(g^2)^{k+1 } } } \,=\ , { 1\over{\pi}}\int_{0}^{\infty } \,d g^2 \ , { { \rm i m } [ \triangle e^{(1,1)}]\over{(g^2)^{k+1 } } } \nonumber\\ & \,=\ , -{1\over{\pi}}\int_{0}^{\infty } \,d(g^2 ) \ , { 2 e^{-1/g^2}\over{(g^2)^{k+2 } } } \,=\ , -{2\over{\pi } } \,k!\ , \quad\quad\quad\quad\quad\ , ( k\geq 2)\,,\end{aligned}\ ] ] in accordance with the leading large - order behavior of the perturbation series @xcite . thus we find that the imaginary ambiguity of the instanton anti - instanton amplitude correctly cancels the imaginary ambiguity of the borel resummation of the ( non - borel summable ) perturbation series . each horizontal line stands for the vacuum in the sine - gordon potential.,scaledwidth=40.0% ] for the three - instanton amplitude shown in fig . [ fig : iii ] , we have two quasi modulus @xmath126 corresponding to the separations between adjacent instantons . for multiple moduli integral of each given configuration , we need to specify subtraction scheme explicitly , and propose the following : 1 . enumerate possible ordering of quasi moduli integrations , such as @xmath127 and @xmath128 for the three instanton case . 2 . subtract possible poles like @xmath129 for the first integration , and then perform the next integration successively , and retain the finite piece . 3 . average the results of all possible orderings . incorporating the repulsive potentials between adjacent instantons , we obtain the three instanton amplitude as ) , but our subtraction scheme specify the successive subtraction which gives a different result from the product of the result of single moduli integral in eq.([iia ] ) . ] @xmath130e^{-3i\theta}\xi^{-3}&\,=\ , \int_{0}^{\infty}dr_{1}\int_{0}^{\infty}dr_{2 } \exp\left[-{2\over{g^2}}(e^{- r_{1}}+e^{- r_{2}})-\epsilon(r_{1}+r_{2})\right ] \nonumber\\[5pt ] & \,\overset{|g^{2}|\ll 1}{\longrightarrow}\ , \left({g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \left[\left({g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon)-{1\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,=\ , { 3\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 \,+ \ , { \pi^2\over{12 } } \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\ , o(\epsilon)\,.\end{aligned}\ ] ] in the three instanton case , we have two possible orderings of the multi integral , each of which gives the identical contribution . then the three instanton contribution to the energy eigenvalue of the lowest band is given by @xmath131\ , . \label{iii}\ ] ] we note that the contribution from the three anti - instanton amplitude is obtained by replacing @xmath132 by @xmath133 . the two - instanton and one anti - instanton amplitudes consist of three types of configurations , as shown in fig . [ fig : iiibar ] . , [ \mathcal{i } \mathcal{i}\bar{\mathcal{i } } ] , [ \bar{\mathcal{i}}\mathcal{i } \mathcal{i}]$ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=40.0% ] + , [ \mathcal{i } \mathcal{i}\bar{\mathcal{i } } ] , [ \bar{\mathcal{i}}\mathcal{i } \mathcal{i}]$ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=40.0% ] , [ \mathcal{i } \mathcal{i}\bar{\mathcal{i } } ] , [ \bar{\mathcal{i}}\mathcal{i } \mathcal{i}]$ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=40.0% ] the first one is @xmath134 $ ] , where the anti - instanton is sandwiched between two instantons . for this type of configuration , adjacent constituent ( anti-)instantons attract each other . therefore we first take @xmath40 in order to apply the bzj prescription to the integral , and follow our subtraction prescription @xmath135e^{-i\theta}\xi^{-3}&\,=\ , \int_{0}^{\infty}dr_{1}dr_{2 } \exp\left[-{2\over{-g^2}}(e^{- r_{1}}+e^{- r_{2 } } ) -\epsilon(r_{1}+r_{2})\right ] \nonumber\\[5pt ] & \,\overset{|g^{2}|\ll 1}{\longrightarrow}\ , \left({-g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \left[\left({-g^{2}\over{2}}\right)^{\epsilon } \gamma(\epsilon)-{1\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-g^{2}=g^{2 } e^{\mp i\pi } } { \longrightarrow}\ , { 3\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 \ , - \,{17\pi^2\over{12 } } \pm 3i\pi\left(\gamma+\log{2\over{g^{2}}}\right ) \,+\,o\left(\frac{1}{\epsilon}\right)\,+\ , o(\epsilon)\,,\end{aligned}\ ] ] where we again subtract @xmath129 in the first integral , before integrating the second quasi - moduli . the second type of configuration is @xmath136 $ ] , where one pair of constituents is repulsive and the other attractive . in order to apply the bzj prescription only to the attractive part of the interaction , we temporarily distinguish two coupling constants @xmath22 for the repulsive interaction and @xmath137 for the attractive interaction , and apply the bzj prescription to @xmath137 , but not to @xmath22 . after analytic continuation , we identify @xmath137 as the original @xmath22 at the end of the calculation . the moduli - integral of @xmath136 $ ] is given by @xmath138e^{-i\theta}\xi^{-3}\ , = \ , \int_{0}^{\infty}dr_{1}dr_{2}\ , \exp\left[-{2\over{g^2}}e^{- r_{1}}-{2\over{-\tilde{g}^2 } } e^{- r_{2}}-\epsilon(r_{1}+r_{2})\right]\,.\ ] ] for this configuration @xmath136 $ ] , we have two possible orderings of moduli integral . the first ordering is to integrate over @xmath139 and to subtract the @xmath129 pole , before integrating over @xmath140 . we call this ordering as @xmath141 @xmath142 \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi } } { \longrightarrow}\ , { 3\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 \ , - \,{5\pi^2\over{12 } } \pm 2i\pi\left(\gamma+\log{2\over{g^{2}}}\right ) \,+\ , o\left(\frac{1}{\epsilon}\right ) + o(\epsilon ) \,.\end{aligned}\ ] ] we call the result of another ordering as @xmath143 , which is given by integrating over @xmath140 first and then over @xmath139 @xmath144 \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi}}{\longrightarrow}\ , { 3\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 \,+ \,{\pi^2\over{12 } } \pm i\pi\left(\gamma+\log{2\over{g^{2}}}\right ) \,+\ , o\left(\frac{1}{\epsilon}\right ) \,+\ , o(\epsilon ) .\end{aligned}\ ] ] in the final expressions for @xmath145 , we implicitly put back @xmath137 to @xmath33 after analytic continuation . we obtain the amplitude of @xmath146 $ ] as an average of @xmath147 and @xmath148 with dropping @xmath113 and @xmath114 terms @xmath149e^{-i\theta}\xi^{-3 } & \,=\ , ( f_{1}\,+\,f_{2})/2 \nonumber\\ & \,=\ , { 3\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 \ , - \,{\pi^2\over{6 } } \pm { 3\over{2}}i\pi\left(\gamma+\log{2\over{g^{2}}}\right)\ , . \label{eq : iibari}\end{aligned}\ ] ] the third type of configuration is @xmath150 $ ] , which gives the identical result as the second type in eq.([eq : iibari ] ) : @xmath151= [ \mathcal{i}\mathcal{i } \bar{\mathcal{i}}]$ ] . by taking the sum of all three types of configurations , we end up with @xmath152\ , + \,[\mathcal{i } \mathcal{i } \bar{\mathcal{i } } ] \,+\,[\bar{\mathcal{i}}\mathcal{i } \mathcal{i}])e^{-i\theta}\xi^{-3 } \,=\ , { 9\over{2}}\left[\left(\gamma+\log{2\over{g^{2}}}\right)^2 \ , - \,{7\pi^2\over{18 } } \pm { 4\over{3}}i\pi\left(\gamma+\log{2\over{g^{2}}}\right)\right]\,.\end{aligned}\ ] ] its contribution to the energy eigenvalue of the lowest band is then given by @xmath153\ , . \label{iiibar}\ ] ] we note that the two anti - instanton and one - instanton contribution is obtained by replacing @xmath154 by @xmath155 . resurgence implies that the two anti - instanton and one - instanton amplitude should correspond to the large - order behavior of the perturbation series around the one - instanton saddle point . using the cancellation between imaginary ambiguities of borel resummed perturbation series around one - instanton saddle point and of the two anti - instanton and one - instanton amplitude , the large - order behavior of the perturbation series around the one instanton saddle point can be estimated from the imaginary part of ( [ iiibar ] ) by means of the dispersion relation as @xmath156 \over{(g^2)^{k+1 } } } \,=\ , { 6\over{\pi}}\int_{0}^{\infty } \,dg^2 \ , { e^{-1/g^2}(\gamma+\log(2/g^2))\over{(g^2)^{k+2 } } } \nonumber\\ & \,=\ , { 6\over{\pi } } k!\,\left(\log 2\,+\,{s(k+1,2)\over{k ! } } \right ) \quad\quad\quad\quad\quad\ , ( k\geq 2)\ , , \label{ak3}\end{aligned}\ ] ] where @xmath157 is the stirling number of the first kind , which is the solution of the recurrence relation @xmath158 . the first few numbers are given as @xmath159 for @xmath160 . @xcite , the large - order perturbative series around one instanton is numerically calculated as @xmath161 which is consistent with eq . ( [ ak3 ] ) for large @xmath162 . each horizontal line stands for the vacuum in the sine - gordon potential.,scaledwidth=40.0% ] for the four - instanton amplitude , the interaction between each pair of instantons is repulsive , as shown in fig . [ fig : iiii ] . since all the possible orderings of multi - moduli integral has the identical contribution , we easily obtain the amplitude using the polygamma function defined as @xmath163 @xmath164e^{-4i\theta } \xi^{-4}&\,=\,\int_{0}^{\infty}dr_{1}dr_{2}dr_{3}\ , \exp\left[-{2\over{g^2}}(e^{- r_{1}}+e^{- r_{2}}+e^{-r_{3 } } ) -\epsilon(r_{1}+r_{2}+r_{3})\right ] \nonumber\\[5pt ] & \,\overset{|g^{2}| \ll 1}{\longrightarrow}\ , \left({g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \left[\left({g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \left\{\left({g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) -{1\over{\epsilon}}\right\ } + { \gamma+\log(2/g^2)\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,=\ , -{8\over{3 } } \left [ \left(\gamma+\log{2\over{g^{2}}}\right)^3 \ , + \,{\pi^2\over{8}}\left ( \gamma+\log{2\over{g^{2}}}\right ) -{1\over{16}}\psi^{(2)}(1 ) \right ] \,+\,o\left(\frac{1}{\epsilon}\right)\,+\ , o(\epsilon)\,,\end{aligned}\ ] ] where we have first subtracted @xmath129 in the first integral @xmath165 , and subtracted @xmath166 in the next integral @xmath167 , and finally dropped the @xmath113 term in the integral @xmath168 . its contribution to the eigenvalue of the lowest band is then given by @xmath169\ , . \label{iiii}\ ] ] the four anti - instanton contribution is obtained by replacing @xmath170 by @xmath171 . the three - instanton and one anti - instanton amplitudes consist of four types of configurations @xmath172 $ ] , @xmath173 $ ] , @xmath174 $ ] and @xmath175 $ ] , as shown in fig . [ fig : iiiibar ] . $ ] , @xmath173 $ ] , @xmath174 $ ] and @xmath175 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=40.0% ] $ ] , @xmath173 $ ] , @xmath174 $ ] and @xmath175 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=40.0% ] $ ] , @xmath173 $ ] , @xmath174 $ ] and @xmath175 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=40.0% ] $ ] , @xmath173 $ ] , @xmath174 $ ] and @xmath175 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=40.0% ] in the first configuration @xmath172 $ ] , two pairs of adjacent instantons have repulsive interactions , and the other is attractive . we again apply the bzj prescription only to the integral for the pair with the attractive interaction , where we denote the coupling as @xmath137 , evaluate the integral at @xmath176 , and analytically continue to @xmath177 , while keeping @xmath93 for the repulsive interactions . the multi - moduli integral is given by @xmath178 \over{e^{2i\theta}\xi^{4}}}\,=\ , \int_{0}^{\infty}dr_{1}dr_{2}dr_{3 } \,\exp\left[-{2\over{g^2}}e^{- r_{1}}-{2\over{g^2 } } e^{- r_{2}}-{2\over{-\tilde{g}^2}}e^{-r_{3 } } -\epsilon(r_{1}+r_{2}+r_{3})\right]\ , . \label{g123}\ ] ] here we have three orderings of the multi - integral , distinguished by the ordering of @xmath165 ( attractive interaction ) relative to @xmath179 . we denote the results @xmath180 , @xmath181 , and @xmath182 as the first , the second , and the third integral , respectively . referring the calculations given in appendix [ app1 ] , we just show the result @xmath183 g_{2}(g^2)&\,=\,-{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{1\over{6 } } \pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1)\,\mp\,i\pi\big[{5\over{2 } } \left(\gamma+\log{2\over{g^{2}}}\right)^2 + { \pi^2\over{12}}\big]\ , , \nonumber\\[5pt ] g_{3}(g^2)&\,=\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,- \,{1\over{3 } } \pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1)\,\mp\,i\pi\big[{3\over{2 } } \left(\gamma+\log{2\over{g^{2}}}\right)^2 + { \pi^2\over{12}}\big]\,.\end{aligned}\ ] ] we obtain the amplitude of @xmath184 $ ] as the average of @xmath185 , @xmath181 and @xmath182 , @xmath186\over{e^{2i\theta}\xi^{4 } } } & \,=\ , ( g_{1}(g^2)\,+\,g_{2}(g^2)\,+\ , g_{3}(g^2))/3 \nonumber\\ & \,=\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \ , + \,{1\over{3}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1)\,\mp\,i\pi\big[{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^2 + { \pi^2\over{18}}\big]\ , . \label{eq : iiibari}\end{aligned}\ ] ] the amplitude of the second configuration @xmath173 $ ] is found to give identical result as the above configuration : @xmath187 = [ \mathcal{i}\mathcal{i}\mathcal{i } \bar{\mathcal{i}}]$ ] . in the third configuration @xmath174 $ ] , the two pairs of adjacent instantons have attractive interactions , while the other is repulsive . we again apply the bzj prescription only to the attractive pairs , by denoting the couplings as @xmath137 . the multi - integral is given as @xmath188\over { e^{2i\theta}\xi^{4 } } } \,=\,\int_{0}^{\infty}dr_{1}dr_{2}dr_{3}\ , \exp\left[-{2\over{g^2}}e^{- r_{1}}-{2\over{-\tilde{g}^2 } } e^{- r_{2}}-{2\over{-\tilde{g}^2}}e^{-r_{3 } } -\epsilon(r_{1}+r_{2}+r_{3})\right]\ , . \label{g456}\ ] ] we have three orderings of the multi - integral , distinguished by the ordering of @xmath168 ( repulsive interaction ) relative to @xmath189 . we denote the results @xmath190 , @xmath191 , and @xmath192 as the first , the second , and the third integral , respectively . referring the calculations given in appendix [ app1 ] , we just show the result @xmath193\ , , \nonumber\\[5pt ] g_{5}&\,=\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \ , + \,{19\over{6}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1)\,\mp\,i\pi\big[{11\over{2 } } \left(\gamma+\log{2\over{g^{2}}}\right)^2 -{5\pi^2\over{12}}\big]\ , , \nonumber\\[5pt ] g_{6}&\,=\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{14\over{3 } } \pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \,\mp\,i\pi\big[{13\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 -{11\pi^2\over{12}}\big]\,.\end{aligned}\ ] ] we obtain the amplitude of @xmath194 $ ] as the average of @xmath195 , @xmath196 and @xmath197 , @xmath198\over{e^{2i\theta}\xi^{4 } } } & \,=\ , ( g_{4}(g^{2})\,+\,g_{5}(g^{2})\,+\ , g_{6}(g^{2}))/3 \nonumber\\ & \,=\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,3\pi^{2 } \left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1)\,\mp\,i\pi\big[{16\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^2 -{7\pi^2\over{18}}\big]\ , . \label{eq : ibariii}\end{aligned}\ ] ] the amplitude of the fourth configuration @xmath175 $ ] turns out to be identical to the above third configuration : @xmath175 = [ \mathcal{i}\bar{\mathcal{i}}\mathcal{i}\mathcal{i } ] $ ] . the sum of all the four configurations gives @xmath199 + [ \bar{\mathcal{i}}\mathcal{i}\mathcal{i}\mathcal{i}]+ [ \mathcal{i}\bar{\mathcal{i}}\mathcal{i}\mathcal{i}]+ [ \mathcal{i}\mathcal{i}\bar{\mathcal{i}}\mathcal{i}])e^{-2i\theta}\xi^{-4 } \nonumber\\ & \,=\ , -{32\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{20\over{3 } } \pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 2\over{3}}\psi^{(2)}(1)\,\mp\,i\pi\big[16\left(\gamma + \log{2\over{g^{2}}}\right)^2 -{2\pi^2\over{3}}\big]\,.\end{aligned}\ ] ] therefore the contribution of three instantons and one anti - instanton to the energy eigenvalue of the lowest band is given by @xmath200 \right]\ , . \label{iiiibar}\end{aligned}\ ] ] we note that the three anti - instanton and one - instanton contribution is obtained by replacing @xmath111 by @xmath118 . resurgence implies that the three - instanton and one anti - instanton amplitude should correspond to the large - order behavior of the perturbation series around the two - instanton saddle point . using the cancellation between imaginary ambiguities of borel resummed perturbation series around two - instanton saddle point and of the three - instanton and one anti - instanton amplitude , the large - order behavior of the perturbation series around the two instanton saddle point can be estimated from the imaginary part of ( [ iiiibar ] ) by means of the dispersion relation as @xmath201 \over{(g^2)^{k+1 } } } \nonumber\\ & \,=\ , -{16\over{\pi}}\int_{0}^{\infty } \,dg^2 \ , { e^{-1/g^2}\over{(g^2)^{k+2 } } } \left ( ( \gamma+\log 2)^{2}-{\pi^{2}\over{24 } } \ , - \ , 2(\gamma+\log 2)\log(g^2)\,+\ , \log^{2}(g^2 ) \right)\,.\end{aligned}\ ] ] this integral can be performed numerically , and the first few results are @xmath202 for @xmath203 . one should be able to check this large - order behavior , by performing the perturbation around the two - instanton saddle point to high orders . the two - instanton and two - anti - instanton amplitudes consist of six types of configurations @xmath204 $ ] , @xmath205 $ ] , @xmath14 $ ] , @xmath206 $ ] , @xmath207 $ ] and @xmath208 $ ] , as shown in fig . [ fig : iiibaribar ] . $ ] , @xmath205 $ ] , @xmath14 $ ] , @xmath206 $ ] , @xmath207 $ ] and @xmath208 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=32.0% ] $ ] , @xmath205 $ ] , @xmath14 $ ] , @xmath206 $ ] , @xmath207 $ ] and @xmath208 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=32.0% ] + $ ] , @xmath205 $ ] , @xmath14 $ ] , @xmath206 $ ] , @xmath207 $ ] and @xmath208 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=32.0% ] $ ] , @xmath205 $ ] , @xmath14 $ ] , @xmath206 $ ] , @xmath207 $ ] and @xmath208 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=32.0% ] + $ ] , @xmath205 $ ] , @xmath14 $ ] , @xmath206 $ ] , @xmath207 $ ] and @xmath208 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=32.0% ] $ ] , @xmath205 $ ] , @xmath14 $ ] , @xmath206 $ ] , @xmath207 $ ] and @xmath208 $ ] ) . each horizontal line stands for the vacuum in the sine - gordon potential.,title="fig:",scaledwidth=32.0% ] the first one is @xmath204 $ ] , where all three adjacent pairs of constituents have attractive interactions . thus we apply the simple bzj prescription to all the integration . since all the orderings of the multi integral have the same contributions , the amplitude can be easily calculated as @xmath209\xi^{-4 } & \,=\ , \int_{0}^{\infty}dr_{1}dr_{2}dr_{3 } \exp\left[-{2\over{-g^2}}(e^{- r_{1}}+e^{- r_{2}}+e^{-r_{3 } } ) -\epsilon(r_{1}+r_{2}+r_{3})\right ] \nonumber\\[5pt ] & \,\overset{|g^{2}|\ll 1}{\longrightarrow}\ , \left({-g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \left[\left({-g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) \left\{\left({-g^{2}\over{2}}\right)^{\epsilon}\gamma(\epsilon ) -{1\over{\epsilon}}\right\ } + { \gamma+\log(2/g^2)\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-g^{2}\to g^{2}e^{\mp i\pi}}{\longrightarrow}\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{23\over{3 } } \pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \nonumber\\ & \qquad\qquad\qquad\mp\,i\pi\big[8\left(\gamma + \log{2\over{g^{2}}}\right)^2 -{7\pi^2\over{3}}\big ] \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\,o(\epsilon)\,.\end{aligned}\ ] ] the second configuration @xmath205 $ ] gives identical result as the first one : @xmath205= [ \mathcal{i}\bar{\mathcal{i}}\mathcal{i}\bar{\mathcal{i}}]$ ] . for the third @xmath14 $ ] and fourth configurations @xmath206 $ ] , the two pairs of interactions are repulsive , but the other pair is attractive . since these moduli - integrals are the same as that of @xmath172 $ ] in eq.([eq : iiibari ] ) , we obtain @xmath210 \xi^{-4 } = [ \bar{\mathcal{i}}\bar{\mathcal{i}}\mathcal{i}\mathcal{i}]\xi^{-4 } \nonumber\\ & \,=\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{1\over{3 } } \pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1)\,\mp\,i\pi\big[{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^2 + { \pi^2\over{18}}\big]\,.\end{aligned}\ ] ] for the fifth @xmath207 $ ] and the sixth configurations @xmath208 $ ] , the two pairs of interactions are attractive , but the other pair is repulsive . since these moduli - integrals are the same as @xmath174 $ ] in eq.([eq : ibariii ] ) , we obtain @xmath211\xi^{-4 } = [ \bar{\mathcal{i}}\mathcal{i}\mathcal{i}\bar{\mathcal{i}}]\xi^{-4 } \nonumber\\ & \,=\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,3\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1)\,\mp\,i\pi\big[{16\over{3}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 -{7\pi^2\over{18}}\big]\,.\end{aligned}\ ] ] the sum of all the six configurations gives @xmath212 + [ \bar{\mathcal{i}}\mathcal{i}\bar{\mathcal{i}}\mathcal{i}]+ [ \mathcal{i}\mathcal{i}\bar{\mathcal{i}}\bar{\mathcal{i}}]+ [ \bar{\mathcal{i}}\bar{\mathcal{i}}\mathcal{i}\mathcal{i}]+ [ \mathcal{i}\bar{\mathcal{i}}\bar{\mathcal{i}}\mathcal{i}]+ [ \bar{\mathcal{i}}\mathcal{i}\mathcal{i}\bar{\mathcal{i}}])\xi^{-4 } \nonumber\\ \,=\ , & -16 \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,22\pi^{2 } \left ( \gamma+\log{2\over{g^{2}}}\right ) + \psi^{(2)}(1)\,\mp\,i\pi\big[32\left(\gamma+\log{2\over{g^{2}}}\right)^2 -{16\pi^2\over{3}}\big]\,.\end{aligned}\ ] ] therefore the contribution of two instantons and two anti - instantons to the energy eigenvalue of the lowest band is given by @xmath213 \right]\ , . \label{iiibaribar}\end{aligned}\ ] ] according to the resurgence , the imaginary part of the two - instanton and two anti - instanton amplitudes should cancel the imaginary part of the borel resummation of the divergent perturbation series . using the dispersion relation , we can obtain the large - order behavior of the perturbation series corresponding to the two - instanton and two anti - instanton amplitude in eq . ( [ iiibaribar ] ) @xmath214 \over{(g^2)^{k+1 } } } \nonumber\\ & \,=\ , -{32\over{\pi^2}}\int_{0}^{\infty } \,dg^2 \ , { e^{-2/g^2}\over{(g^2)^{k+3 } } } \left ( ( \gamma+\log 2)^{2}-{\pi^{2}\over{6 } } \ , - \ , 2(\gamma+\log 2)\log(g^2)\,+\ , \log^{2}(g^2 ) \right)\ , . \label{4iak}\end{aligned}\ ] ] this integral can be performed numerically , and the first few results are @xmath215 for @xmath203 . because of the factor @xmath216 associated with the four - instanton action @xmath217 , the large - order ( @xmath218 ) behavior of @xmath219 in eq . ( [ 4iak ] ) can be estimated with a constant @xmath220 as @xmath221 \,,\ ] ] exhibiting the @xmath222 factor besides the factorial growth @xmath223 . this large - order behavior corresponds to the singularity of the borel transform @xmath45 at @xmath224 in the borel plane @xmath225 this is consistent with the known and expected results in quantum mechanics . we comment that we can also calculate the large - order perturbative behavior around the one - instanton and one anti - instanton vacuum as @xmath226/(g^{2})^{k+1}$ ] . for general cases with @xmath55 instantons and @xmath96 anti - instantons , we have @xmath227 configurations depending on the arrangement of constituents . we can classify these @xmath227 configurations into classes based on how many attractive and repulsive interactions are among the @xmath104 nearest - neighbor interactions . ( we can classify them into @xmath228 classes for @xmath229 , and @xmath230 classes for @xmath231 with @xmath232 . ) we use the expression @xmath233 as the number of configurations including @xmath162 attractive and @xmath234 repulsive interactions satisfying @xmath235 , and show examples of classification for @xmath236 , @xmath237 , @xmath238 and @xmath239 as following , @xmath240 when calculating the amplitude of each in the @xmath233 configurations , we have @xmath241 possible orderings of the multi moduli integrals in relation to the subtraction scheme which is described in sec . [ sec : general ] and [ sec:3instanton ] . after calculating these @xmath241 integrals , we average the results and obtain a contribution from one of @xmath233 configurations . as such , we calculate each contribution , and by summing up all the integral results , we end up with the semi - classical amplitude of the configuration with @xmath55 instantons and @xmath96 anti - instantons . a systematic method called the uniform - wkb method has been applied extensively to study quantum mechanics for potentials with degenerate minima , such as the sine - gordon quantum mechanics @xcite . in this method , one combines the ansatz in eq.([dansatz ] ) with a global boundary condition to take account of the other minima than the one to do perturbative computation . this boundary condition enables one to go beyond the ordinary perturbative computation , and obtain all nonperturbative contributions leading to the resurgence . defining the even @xmath242 and odd @xmath243 functions , and the wronskian @xmath244 , @xmath245\ , , \\ f_{2}(y ) \,&=\ , { 1\over{\sqrt{u'(y)}}}\left [ d_{\nu}(u(y)/g)\ , -\ , d_{\nu}(-u(y)/g ) \right]\ , , \\ \mathcal{w}\,&=\ , f_{1}f'_{2}-f'_{1}f_{2}\ , = \,-\sqrt{8\pi\over{g^{2}}}{1\over{\gamma(-\nu)}}\,,\end{aligned}\ ] ] the following uniform - wkb boundary condition @xcite is imposed at the midpoint @xmath246 of the adjacent minima of the sine - gordon potential @xmath247 where parameter @xmath102 is the bloch angle in eq.([eq : bloch - angle ] ) to label the state within the energy band . this boundary condition can be rewritten as @xmath248 where @xmath107 is the instanton factor with the instanton action @xmath249 in eq.([eq : instanton_action ] ) @xmath250\,,\ ] ] the function @xmath251 describing the perturbative fluctuations around the instanton is defined in terms of hypergeometric functions in ref . @xcite . we just show the necessary expression at the zeroth order of @xmath22 ( @xmath252 ) : @xmath253 this boundary condition of the uniform - wkb approximation is also derived as the quantization condition in ref . @xcite . since @xmath73 is the shift of energy eigenvalue @xmath28 from the harmonic oscillator ground state energy , the expansion of eq . ( [ uwkbc ] ) up to the @xmath254 order with respect to @xmath255 for the energy eigenvalue of the lowest band is @xmath256 \nonumber\\ & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\pm\,{i\pi\xi^2 \over{2 } } \left [ h_{0}^2 + \nu(2h_{0}h_{0}'+h_{0}^{2}q_{1}^{\pm } ) + \nu^{2}(2(h_{0}')^2 + 2h_{0}h_{0}''-h_{0}^{2}q_{2}^{\pm } ) \right ] \nonumber\\ & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,=\xi\cos\theta \left[h_{0 } + \nu h_{0 } ' + \nu^{2}h_{0}''+\nu^{3}h_{0 } ' '' \right]\ , , \label{expansion}\end{aligned}\ ] ] with @xmath257 with @xmath258 . for @xmath259 , @xmath260 and @xmath261 , we just replace @xmath262 by @xmath263 . we iteratively solve the expansion in eq . ( [ expansion ] ) in the semi - classical limit ( @xmath264 ) , then obtain the expression of @xmath255 as @xmath265 \nonumber\\ & \,+\,\xi^{3}h_{0}^{3}\left [ - ( 2q_{1}^2 + q_{2})\cos^{3}\theta \,+ \left ( { \pi^{2}\over{2 } } \mp i\pi{3\over{2}}q_{1 } \right ) \cos \theta \right ] \nonumber\\ & \,+\ , \xi^{4}h_{0}^{4}\big [ \left(5q_{1}^{3}+5q_{1}q_{2}+q_{3 } \right)\cos^{4}\theta \nonumber\\ & \,\,\,\,\,\,\,\,+\ , \left ( -2\pi^{2}q_{1 } \pm i\pi({9\over{2}}q_{1}^{2}+q_{2}-{\pi^{2}\over{4 } } ) \right ) \cos^{2}\theta \,-\ , \left({\pi^{2}\over{2}}q_{1}\pm i\pi{\pi^{2}\over{4 } } \right ) \big]\ , . \label{delnu}\end{aligned}\ ] ] for @xmath266 , the lowest energy level @xmath267 is expressed as @xmath268 from eq . ( [ delnu ] ) , we can calculate the contributions from each order of @xmath107 and @xmath154 . the order of @xmath107 is interpreted as the number of instanton and anti - instanton constituents in the corresponding configurations while the order of @xmath154 is interpreted as the instanton number . in the followings , we show that the coefficients at each order of @xmath107 and @xmath154 are completely consistent with the contributions derived from the instanton calculations in the previous section . we first start with the @xmath110 order , where we have two instanton constituents . in the @xmath110 order , the terms proportional to @xmath111 are @xmath269\,.\ ] ] it agrees with the contribution from the two - instanton amplitude in eq . ( [ ii ] ) . on the other hand , the terms proportional to @xmath270 are given by @xmath271\,.\ ] ] these terms composed of the real and imaginary parts precisely agree with the result of the one - instanton and one anti - instanton amplitude in eq . ( [ iibar ] ) . we next look into the @xmath272 order . in the @xmath272 order , the terms with @xmath132 are @xmath273\,,\ ] ] which agree with the contributions from the three - instanton amplitude in eq . ( [ iii ] ) . the terms proportional to @xmath154 are given by @xmath274\,.\ ] ] the real and imaginary parts in this result are identical to the results from the two - instanton and one anti - instanton amplitudes in eq . ( [ iiibar ] ) . we finally investigate the @xmath275 order . in this order , the terms proportional to @xmath170 are given by @xmath276\,,\ ] ] which agrees with the four - instanton contribution in eq . ( [ iiii ] ) . next , the terms with @xmath111 are @xmath277 \right]\,.\ ] ] these terms are the same ones as the contribution obtained from the three - instanton and one anti - instanton amplitudes in eq . ( [ iiiibar ] ) . finally , the terms proportional to @xmath270 are given by @xmath278 \right]\,,\ ] ] which precisely agree with the contribution from two - instanton and two anti - instanton amplitudes shown in eq . ( [ iiibaribar ] ) . in this section , we discuss fractional instantons and neutral bions in the @xmath0 model on @xmath1 with the @xmath279 twisted boundary conditions in comparison to the sine - gordom quantum mechanics . the @xmath0 model is described in terms of the @xmath280-component complex field @xmath281 satisfying the constraint @xmath282 with a constant @xmath283 , whose euclidean lagrangian is given by @xmath284 , \quad { \cal d}_\mu=\partial_\mu+i a_\mu , \quad a_\mu = \frac{i}{v^2 } { \partial}_\mu h h^{\dagger}. \label{eq : cpn_lag}\ ] ] the parameter @xmath285 serves as a coupling constant manifold is denoted as @xmath286 , and is related to the notation of dunne and unsal notation @xcite as @xmath287 , @xmath288 . ] of the @xmath0 model , which is asymptotically free . the bps solutions are obtained if a complex @xmath280-component vector @xmath289 called the moduli matrix is holomorphic ( independent of @xmath290 ) @xcite @xmath291 whereas the moduli matrix @xmath289 for anti - bps solutions should be anti - holomorphic ( @xmath292 independent of @xmath293 ) . here we have defined the complex coordinate @xmath294 with @xmath295 . the @xmath17 symmetric twisted boundary condition can be written as @xmath296 the moduli matrix for a fractional instanton as a bps solution is given by @xmath297 for which we find the topological charge to be fractional @xmath298 @xcite ( see also subsequent works @xcite ) . this solution can be regarded as a kink connecting two neighboring vacua @xcite , where real parameters @xmath299 are their modulus , representing the relative phase @xmath300 of the neighboring vacua and the position of the kink at @xmath301 . the action and topological charge densities of the single fractional instanton solution has no dependence on the coordinate @xmath18 of the compactified dimension , thus at the first glance the situation seems quite similar to that of the ( euclidean ) quantum mechanics with the periodic potential , namely the sine - gordon quantum mechanics . to clarify the similarities and differences with the sine - gordon quantum mechanics , we consider a small compactification length limit ( @xmath302 ) of the two - dimensional @xmath0 model . let us first take the @xmath303 model for simplicity . by a stereographic projection , we can parametrize @xmath304 target space in terms of two fields @xmath305 and @xmath306 corresponding to the zenith and azimuth angles of @xmath307 @xmath308 one should note that only the ratio of the first and second components @xmath309 is needed to parametrize @xmath307 as the inhomogeneous coordinate of the @xmath303 field space @xmath310 choosing @xmath311 in eq.([eq : cpn_lag ] ) , the lagrangian of the two - dimensional @xmath303 model can be rewritten in terms of @xmath305 and @xmath306 @xmath312 . \label{eq : cp1_lag}\ ] ] the scherk - schwarz dimensional reduction assumes the following ansatz of a particular @xmath18 dependence for the fields @xmath313 with constants @xmath314 . one should note that we have restricted @xmath315 to @xmath18 independent field and ignored the fluctuation of field @xmath316 . by inserting the ansatz ( [ eq : ss_ansatz ] ) and integrating over @xmath18 , we obtain the euclidean action @xmath57 as an integral of the euclidean lagrangian @xmath16 over the euclidean time @xmath38 as @xmath317 = \int dt l. \label{eq : ssred_cp1_lag}\ ] ] we notice that the euclidean lagrangian @xmath16 is identical to that of the sine - gordon quantum mechanics in eq.([eq : y_lagrangian ] ) with the identification @xmath318 , @xmath319 , and the sine - gordon coupling @xmath21 as @xmath320 since the @xmath279-twisted boundary condition ( @xmath311 for @xmath303 model ) requires @xmath321 we finally find the coupling of the sine - gordon quantum mechanics @xmath21 in terms of the parameters of the @xmath0 model @xmath322 which is independent of the compactification scale @xmath16 . in fig . [ fig : sg - cp1 ] we show how the sine - gordon model is embedded into the @xmath323 model . the target space of the @xmath323 model is @xmath324 in which the @xmath325 as a great circle of @xmath307 is the target space of the sine - gordon model . the two fixed points of the action of the twisted boundary conditions , the north and south poles denoted by n and s , respectively , corresponding to @xmath326 and @xmath327 ( modulo @xmath328 ) in the lagrangian in eq . ( [ eq : ssred_cp1_lag ] ) , are two vacua in the reduced sine - gordon model ( see fig . [ fig : sg - cp1 ] ) : @xmath329 model . the great circle @xmath325 denotes the target space of the sine - gordon model . n and s are two vacua @xmath330 and @xmath331 modulo @xmath328 of the sine - gordon model . , scaledwidth=20.0% ] to see the correspondence between instantons in the @xmath323 model and reduced sine - gordon model , let us first examine the bps one - instanton solution in eq.([eq:1instanton ] ) , whose inhomogeneous coordinate is given as @xmath332 @xmath333 we see that this solution is consistent with the assumption of the scherk - schwarz reduction in eq.([eq : ss_ansatz ] ) with eq.([eq : ssphase ] ) . this solution gives the vertical path starting from n at @xmath334 and reaching s at @xmath335 with @xmath336 when @xmath337 . with varying from @xmath337 to @xmath338 , this vertical path revolves in @xmath316 to end up at @xmath338 with another vertical path connecting n ( at @xmath334 ) to s ( at @xmath335 ) with @xmath339 . thus the paths sweep precisely half of @xmath307 , as illustrated in fig . [ sg123 ] . the solution @xmath315 in eq . ( [ eq : inhomo_1instanton ] ) is identical to the single instanton solution of the sine - gordon quantum mechanics in eq.([eq : instanton ] ) . therefore the fractional instanton solution of the @xmath303 model is captured correctly by the sine - gordon quantum mechanics as the instanton solution . the second homotopy group @xmath340 for the @xmath323 fractional instanton is 1/2 , while the first homotopy group @xmath341 for the reduced sine - gordon model is also 1/2 because it corresponds to a half orbit of the great circle . in the reduced quantum mechanics is depicted . it corresponds to a single line from the north to the south pole which is rotated over the half of @xmath342 homogeneously . the figure depicts the rotation of the line around the half sphere . the red arrows denote paths depending on @xmath18 with a constant @xmath343 , while the blue arrows denote the @xmath343 dependence of such paths . , scaledwidth=70.0% ] in fig . [ fig : fractional ] , we classify how all possible ( anti-)instanton configurations in the reduced sine - gordon model correspond to the fractional instantons in the original @xmath323 model . as shown in eqs . ( [ eq : bps - bound ] ) and ( [ eq : anti - bps - bound ] ) , a sine - gordon soliton connecting from a vacuum labeled by even @xmath55 to a vacuum labeled by odd @xmath55 is bps while that from odd @xmath55 to even @xmath55 is anti - bps . when the sine - gordon quantum mechanics is embedded into the @xmath323 model , a soliton from n to s on @xmath307 is bps while that from s to n is anti - bps , consistent with bps and anti - bps kinks in the @xmath323 model . in other words , configurations with positive values of the second homotopy class @xmath344 are bps while those with negative values are anti - bps both in the @xmath323 model and reduced sine - gordon model . thus , ( a ) and ( d ) in fig . [ fig : fractional ] are bps while ( b ) and ( c ) are anti - bps . [ cols="^,^,^,^,^ " , ] here we point out that the bps solution of two fractional - instantons in the @xmath303 model can not be described in the reduced sine - gordon model . the bps solution of two fractional - instantons , which contains the ordinary one - instanton bps solution ( @xmath49 ) in the limit of small separation of two fractional instantons is given by @xmath345 this is a composite of ( a ) and ( d ) in fig.[fig : fractional ] . the inhomogeneous coordinate of @xmath303 now reads @xmath346 which can not satisfy the assumption ( [ eq : ss_ansatz ] ) of the scherk - schwarz reduction . this is because in the reduced sine - gordon model a configuration starting from n and ending at s [ ( a ) in fig . [ fig : fractional ] ] can not be connected to another configuration starting from n and ending at s [ ( d ) in fig . [ fig : fractional ] ] . the former can be connected only to a configuration starting from s and ending at n. therefore the bps two fractional instanton solution can not be described by the sine - gordon quantum mechanics even in the limit of small @xmath16 . more generally , all the bps multi - fractional - instanton solutions are inconsistent with the scherk - schwarz reduction and hence the sine - gordon quantum mechanics fails to capture them . this is consistent with the fact that configurations containing @xmath55-instantons ( @xmath347 ) are always non - bps in the sine - gordon quantum mechanics . of neutral bion configurations for @xmath348 and @xmath349 in the @xmath303 model on @xmath1 . the same action density is depicted in two ways , as a function of @xmath350 ( left ) and @xmath343 ( right ) . there is no @xmath18 dependence in the action density , with @xmath18 being a coordinate of the compactied dimension.,title="fig:",scaledwidth=55.0% ] of neutral bion configurations for @xmath348 and @xmath349 in the @xmath303 model on @xmath1 . the same action density is depicted in two ways , as a function of @xmath350 ( left ) and @xmath343 ( right ) . there is no @xmath18 dependence in the action density , with @xmath18 being a coordinate of the compactied dimension.,title="fig:",scaledwidth=40.0% ] next let us consider the non - bps configuration of neutral bion , which is a composite of a fractional instanton and a fractional anti - instanton as depicted in fig . [ 025pi ] . we can write down an ansatz for the moduli matrix @xmath289 @xmath351 which is guaranteed to become an exact solution of the field equation in the limit of large separation ( @xmath352 ) between constituents . the inhomogeneous coordinate of @xmath303 now reads @xmath353 which satisfy the assumption ( [ eq : ss_ansatz ] ) of the scherk - schwarz reduction if and only if @xmath354 . in that case , we obtain the angular coordinate fields of @xmath307 as @xmath355 this configuration starts from n at @xmath334 . for the upper sign , it goes through s at @xmath356 and reaches to n with @xmath357 at @xmath335 , namely it winds once around the great circle . the configuration represents the double instanton configuration of the sine - gordon quantum mechanics as shown in fig.[fig : ii ] . for the lower sign , the configuration returns back to n with @xmath330 at @xmath335 approaching but never reaching s at any point in @xmath358 . this clearly represents the instanton and anti - instanton configuration @xmath12 $ ] of the sine - gordon quantum mechanics , as shown in the left panel of fig . [ fig : iibar ] . the sine - gordon quantum mechanics captures only field configurations that can cover the ( part of ) @xmath307 in the following specific fashion : when @xmath343 is varied with fixed @xmath18 , @xmath315 goes along the great circle ( namely fixed @xmath316 ) , whereas @xmath18 variation with fixed @xmath343 makes a rotation of @xmath316 with the constant velocity by an amount @xmath327 at fixed @xmath315 . the first homotopy group @xmath341 for the sine - gordon model is one for the upper sign and zero for the lower sign , but the second homotopy group @xmath340 for the @xmath323 model is zero for the both cases . in fig . [ sgii ] , we show the instanton anti - instanton and instanton - instanton configurations in the sine - gordon quantum mechanics corresponding to @xmath354 in eq . ( [ eq : ratio_2instanton ] ) , and how the corresponding configuration of the @xmath303 model in eq.([eq : inhomo_ibari ] ) cover the sphere @xmath307 . here , each of fractional instanton again corresponds to the line between the north and south poles sweeping around the half sphere . on the other hand , generic configurations of the neutral bion of the @xmath303 model in eq . ( [ eq : ratio_2instanton ] ) exhibit complicated ways of covering ( part of ) @xmath307 in terms of @xmath359 . in terms of the @xmath360 field , however , they are quite similar to those special configurations with @xmath354 describable by the sine - gordon quantum mechanics except for an important new feature : they have the relative phase @xmath361 between fractional instanton and anti - instanton as an additional moduli of the neutral bion configuration . this relative phase moduli introduces a striking physical effect into interactions between fractional instanton and anti - instanton constituents even in the limit of @xmath302 , as we see immediately . model are depicted.,scaledwidth=80.0% ] the absence of the relative phase moduli between the instanton constituents in the bion configuration is a crucial drawback of the sine - gordon quantam mechanical description . to clarify how crucial it is , we discuss the neutral bion configuration in the @xmath0 model on @xmath1 with the @xmath279 twisted boundary conditions . generalizing eq.([eq : bion_cp1 ] ) in the @xmath303 model , the neutral bion configuration with the relative phase moduli between the two constituents is given by @xcite @xmath362 where @xmath363 are real parameters corresponding to positions of constituent fractional instanton and fractional anti - instanton , and @xmath364 is the relative phase . here , the separation between the fractional instanton and anti - instanton is given by @xmath365 . we study the interaction potential as a function of separation @xmath81 of constituents for the fixed relative phase @xmath364 . as we vary this relative phase @xmath364 from @xmath366 to @xmath328 , we find that the interaction between the instanton and anti - instanton constituents is attractive for @xmath367 and @xmath368 , whereas it is repulsive for @xmath369 . we compare the energy density for various relative phases in fig . [ thetadep ] . the interaction potential between fractional instanton and fractional anti - instanton is derived by both numerically @xcite and analytically as @xmath83 \,=\ , -{4\kappa l v^2\cos \phi } \ , \exp(-\kappa r)\ , , \label{potnb}\ ] ] where we note @xmath370 and the coupling of @xmath0 model is denoted as @xmath285 . we here follow the definition in our previous work @xcite : our coupling @xmath285 is related to the coupling @xmath371 adopted in @xcite as @xmath372 . we note that the corresponding configuration in the sine - gordon quantum mechanics has no relative phase moduli and the interaction potential between the instanton and anti - instanton constituents is given by eq . ( [ pot2 ] ) with the minus sign ( attraction ) . therefore the neutral bion contribution in the @xmath0 model can not be calculated in the reduced quantum mechanics correctly . of neutral bion configurations for @xmath373 are shown for several values of relative phase moduli @xmath364 : @xmath374 ( left ) , @xmath375 ( center ) and @xmath327 ( right ) . the total action increases as the relative phase @xmath364 is varied from @xmath366 to @xmath327 , while configurations in a far - separated situation are almost the same for the three cases . the attractive force decreases in the region from @xmath376 to @xmath377 , then becomes repulsive for @xmath378.,title="fig:",scaledwidth=32.0% ] of neutral bion configurations for @xmath373 are shown for several values of relative phase moduli @xmath364 : @xmath374 ( left ) , @xmath375 ( center ) and @xmath327 ( right ) . the total action increases as the relative phase @xmath364 is varied from @xmath366 to @xmath327 , while configurations in a far - separated situation are almost the same for the three cases . the attractive force decreases in the region from @xmath376 to @xmath377 , then becomes repulsive for @xmath378.,title="fig:",scaledwidth=32.0% ] of neutral bion configurations for @xmath373 are shown for several values of relative phase moduli @xmath364 : @xmath374 ( left ) , @xmath375 ( center ) and @xmath327 ( right ) . the total action increases as the relative phase @xmath364 is varied from @xmath366 to @xmath327 , while configurations in a far - separated situation are almost the same for the three cases . the attractive force decreases in the region from @xmath376 to @xmath377 , then becomes repulsive for @xmath378.,title="fig:",scaledwidth=32.0% ] the neutral bion amplitude has been computed previously by ignoring the relative phase moduli ( @xmath376 ) . the result can be given in terms of the dimensionless separation variable @xmath379 @xcite @xmath380^{\phi=0 } & = & c e^{-2s_{i}/n}i(v^2 ) , \nonumber \\ i(v^2)&=&\int_0^\infty d\tau \exp\left(4\kappa l v^2e^{-\tau } -\epsilon \tau\right ) = \int_0^\infty d\tau \exp\left(\frac{8\pi v^2}{n}e^{-\tau } -\epsilon \tau\right ) \nonumber \\ & = & -\left(\gamma + \log \left(\frac{8\pi v^2}{n}\right)\right ) \mp i\pi , \label{eq : bion_no_phase}\end{aligned}\ ] ] where @xmath381 is the instanton action and @xmath220 denotes the numerical coefficient when the relative phase moduli is ignored ( @xmath376 ) . based on the interaction potential we obtained in eq . ( [ potnb ] ) , we find that the corrected contributions of the neutral bion is given by taking account of the relative phase moduli as @xmath382 = c e^{-2s_{i}/n}\int_0^{2\pi } d\phi \,i(v^2 \cos\phi).\ ] ] since the moduli integrals for @xmath383 $ ] and @xmath384 $ ] are identical , we will double the result for @xmath383 $ ] . we apply the bzj prescription to the integral in the case of the attractive interaction for @xmath385 $ ] , while we simply perform the integral for the repulsive case @xmath386 $ ] . using eqs.([iia ] ) and ( [ eq : i - bari-1 ] ) , we obtain @xmath387 \nonumber \\ & = -2\pi\left(\gamma + \log \left(\frac{4\pi v^2}{n}\right ) \right ) \,\mp\ , i\pi^{2}. \end{aligned}\ ] ] thus we obtain the neutral bion contribution as @xmath388 \,=\ , ce^{-\frac{2s_{i}}{n } } \left[-2\pi\left(\gamma + \log \left(\frac{4\pi v^2}{n}\right ) \right ) \,\mp\ , i\pi^{2 } \right ] \ , , \label{e0nb}\ ] ] in contrast to the result in eq.([eq : bion_no_phase ] ) where the relative phase is ignored . we have obtained quantitative corrections to that in ref . @xcite , by taking account of the effects of the integral of the relative phase moduli . to establish the absolute magnitude of the instanton contribution definitely , it is desirable to examine the one - loop determinant around the fractional instanton and anti - instanton background for the two - dimensional @xmath0 model which has an explicit ( weak ) dependence on @xmath18 . we consider these as future works . before closing this section , we note an alternative possibility to relate the sine - gordon quantum mechanics and the @xmath0 model without compactification . even if the compactification length is nonzero , the @xmath323 model can be deformed so as to produce the sine - gordon instanton solutions . to clarify this point , we move to the @xmath389 non - linear sigma model equivalent to the @xmath303 model . in this model , we have three real scalar fields @xmath390 subjected by the constraint @xmath391 . we introduce the two potential terms @xmath392 and @xmath393 , with a mass hierarchy @xmath394 . for the parameter region @xmath395 , the potential @xmath396 admits two discrete vacua @xmath397 and a @xmath323 domain wall solution interpolating these two vacua @xcite with the width @xmath398 . let us place it perpendicularly to the @xmath399 direction . with a small @xmath400 , the above vacua are shifted and the domain wall is deformed accordingly . in this case , the sine - gordon model is induced on the @xmath323 domain wall as the effective theory @xcite . then , an @xmath389 ( @xmath401 ) instanton is restricted to the @xmath323 domain wall and becomes a sine - gordon instanton with the width @xmath402 ( in the @xmath403 direction ) , in the domain wall effective theory @xcite . this setting gives a precise relation between the @xmath323 model and the sine - gordon quantum mechanics . by sending @xmath404 to zero , we may be able to investigate how the contribution of the relative complex phase disappears in the reduction process from the @xmath323 model to the sine - gordon quantum mechanics . in the case of the @xmath0 model , multiple @xmath405 parallel domain walls in it @xcite play the role to connect the @xmath0 model instantons and instantons in a sine - gordon - like model . in this paper we have calculated multi - instanton contributions in the quantum mechanics with the sine - gordon potential by integrating out separation moduli parameters between instantons and anti - instantons in the semi - classical limit ( @xmath6 ) . we have adopted an extended bogomolnyi zinn - justin prescription for multi - instanton configurations and the step - by - step subtraction scheme for divergent parts . we show that the imaginary parts of the multi - instanton amplitudes cancel those arising from the borel resummation of the large - order perturbation series . we verify that our results completely agree with those based on the uniform - wkb calculations @xcite up to a four - instanton order . we have also shown that the neutral bion amplitude in @xmath0 model based on the potential including the relative phase moduli parameter gives corrections to the results obtained in the sine - gordon quantum mechanics @xcite . our main results , that the multi - instanton amplitudes in the sine - gordon quantum mechanics are consistent with the large - order behavior of the perturbative calculations , and are completely reproduced by the uniform - wkb boundary conditions , strongly indicate the following facts : the uniform - wkb boundary condition provides the correct link between the perturbation series around the perturbative vacuum and non - perturbative instanton effects , and furthermore the perturbative calculation _ knows _ non - perturbative aspects of the quantum system in the first place . we can take the result as an evidence at least in the quantum mechanics , in favor of the resurgence conjecture , which gives an unambiguous and self - consistent definition of the quantum theory . as for the neutral bion in the @xmath0 model on @xmath1 , we have calculated , for the first time , the neutral bion amplitude beyond the sine - gordon quantum mechanics , with the moduli integral of the relative phase parameter , which does not exist in the sine - gordon quantum mechanics @xcite . as a future study , we will perform perturbative calculations and multi - instanton calculations in the @xmath0 model with keeping the compactified - direction dependence , then we may be able to show that its large - order behavior of the perturbations originates in the imaginary part of the neutral - bion and bion - molecule amplitude _ at the field theory level_. we here make a comment on the relation of the configuration eq . ( [ bion ] ) to the lefschetz thimble @xcite . in the recent attempt to perform the path integral in the extended theory @xcite , one complexifies the coupling constant ( @xmath406\not=0 $ ] ) and the field variables , then the standard path integral can be replaced by the integral along the steepest descent curves in the complex configuration space , or lefschetz thimbles with the imaginary part of the action being constant @xmath407 = { \rm const.}$ ] . in terms of this method , the configuration eq . ( [ bion ] ) corresponds to the special thimble for @xmath406 = 0 $ ] , called the stokes line , where the two critical points , in which fractional instantons are infinitely - separated or completely compressed , are directly connected by the configuration . although , for a general case @xmath406\not=0 $ ] , the two critical points should belong to two different lefschetz thimbles , our study on the stokes line @xmath406 = 0 $ ] could be a good starting point for investigating the general lefschetz thimbles . part of future work will be devoted to this study . we also note that the relation of lefschetz thimbles and the topologically neutral configurations has been investigated in terms of the hidden topological angles in ref . @xcite . we also comment that the cancellation of imaginary ambiguities in the sine - gordon quantum mechanics , or the resurgent structure in the system , can be understood as a special case of the generic resummation structure , called median resummation " @xcite . it is expected that this resummation is applicable to any quantum theories including quantum mechanics and field theory . thus , we consider that it is intriguing to study the application of the median resummation to the sine - gordon quantum mechanics and verify its structure as a future work . we are grateful to mithat nsal and gerald dunne for their interest and valuable comments and correspondences on their related work during the entire course of our study . t. m. and n. s. thank cern theory institute 2014 , resurgence and transseries in quantum , gauge and string theories " for the fruitful discussion and useful correspondence . this work is supported in part by the japan society for the promotion of science ( jsps ) grant - in - aid for scientific research ( kakenhi ) grant numbers ( 26800147 ( t. m. ) , 25400268 ( m. n. ) and 25400241 ( n. s. ) ) . in this appendix we show the details of calculation of the functions @xmath410 ( @xmath411 ) . the functions @xmath412 are based on the three subtraction patterns of the integral in eq . ( [ g123 ] ) . they are given by @xmath413 + { \gamma+\log(2/(-\tilde{g}^2))\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi}}{\longrightarrow}\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{7\over{6}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \nonumber\\ & \qquad\qquad\qquad\,\mp\,4i\pi\left(\gamma+\log{2\over{g^{2}}}\right)^2 \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\,o(\epsilon)\,,\end{aligned}\ ] ] @xmath414 + { \gamma+\log(2/g^2)\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi}}{\longrightarrow}\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{1\over{6}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \nonumber\\ & \qquad\qquad\qquad\,\mp\,i\pi\big[{5\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 + { \pi^2\over{12}}\big ] \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\,o(\epsilon)\,,\end{aligned}\ ] ] @xmath415 + { \gamma+\log(2/g^2)\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi}}{\longrightarrow}\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,- \,{1\over{3}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \nonumber\\ & \qquad\qquad\qquad\mp\,i\pi\big[{3\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 + { \pi^2\over{12}}\big ] \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\,o(\epsilon)\,.\end{aligned}\ ] ] @xmath417 + { \gamma+\log(2/g^2)\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi}}{\longrightarrow}\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{7\over{6}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \nonumber\\ & \qquad\qquad\qquad\,\mp\,i\pi\big[4\left(\gamma+\log{2\over{g^{2}}}\right)^2 + { \pi^2\over{6}}\big ] \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\,o(\epsilon)\,,\end{aligned}\ ] ] @xmath418 + { \gamma+\log(2/(-\tilde{g}^2))\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi}}{\longrightarrow}\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{19\over{6}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \nonumber\\ & \qquad\qquad\qquad\,\mp\,i\pi\big[{11\over{2 } } \left(\gamma+\log{2\over{g^{2}}}\right)^2 -{5\pi^2\over{12}}\big]\,+\ , o\left(\frac{1}{\epsilon}\right)\ , + \,o(\epsilon)\,,\end{aligned}\ ] ] @xmath419 + { \gamma+\log(2/(-\tilde{g}^2))\over{\epsilon}}\right ] \nonumber\\[5pt ] & \,\overset{-\tilde{g}^{2}= \tilde{g}^{2 } e^{\mp i\pi}}{\longrightarrow}\ , -{8\over{3 } } \left(\gamma+\log{2\over{g^{2}}}\right)^3 \,+ \,{14\over{3}}\pi^{2}\left ( \gamma+\log{2\over{g^{2}}}\right ) + { 1\over{6}}\psi^{(2)}(1 ) \nonumber\\ & \qquad\qquad\qquad\,\mp\,i\pi\big[{13\over{2}}\left(\gamma+\log{2\over{g^{2}}}\right)^2 -{11\pi^2\over{12}}\big ] \,+\ , o\left(\frac{1}{\epsilon}\right)\,+\,o(\epsilon)\,.\end{aligned}\ ] ] m. nsal , `` abelian duality , confinement , and chiral symmetry breaking in qcd(adj ) , '' phys . lett . * 100 * , 032005 ( 2008 ) [ arxiv:0708.1772 [ hep - th ] ] . m. nsal , `` magnetic bion condensation : a new mechanism of confinement and mass gap in four dimensions , '' phys . d * 80 * , 065001 ( 2009 ) [ arxiv:0709.3269 [ hep - th ] ] . m. shifman and m. nsal , `` qcd - like theories on r(3 ) x s(1 ) : a smooth journey from small to large r(s(1 ) ) with double - trace deformations , '' phys . d * 78 * , 065004 ( 2008 ) [ arxiv:0802.1232 [ hep - th ] ] . e. poppitz and m. nsal , `` conformality or confinement : ( ir)relevance of topological excitations , '' jhep * 0909 * , 050 ( 2009 ) [ arxiv:0906.5156 [ hep - th ] ] . m. m. anber and e. poppitz , `` microscopic structure of magnetic bions , '' jhep * 1106 * , 136 ( 2011 ) [ arxiv:1105.0940 [ hep - th ] ] . e. poppitz , t. schaefer and m. nsal , `` continuity , deconfinement , and ( super ) yang - mills theory , '' jhep * 1210 * , 115 ( 2012 ) [ arxiv:1205.0290 [ hep - th ] ] . p. argyres and m. nsal , `` a semiclassical realization of infrared renormalons , '' phys . lett . * 109 * , 121601 ( 2012 ) [ arxiv:1204.1661 [ hep - th ] ] . p. c. argyres and m. nsal , `` the semi - classical expansion and resurgence in gauge theories : new perturbative , instanton , bion , and renormalon effects , '' jhep * 1208 * , 063 ( 2012 ) [ arxiv:1206.1890 [ hep - th ] ] . g. v. dunne and m. nsal , `` resurgence and trans - series in quantum field theory : the cp(n-1 ) model , '' jhep * 1211 * , 170 ( 2012 ) [ arxiv:1210.2423 [ hep - th ] ] . g. v. dunne and m. nsal , `` continuity and resurgence : towards a continuum definition of the cp(n-1 ) model , '' phys . d * 87 * , 025015 ( 2013 ) [ arxiv:1210.3646 [ hep - th ] ] . r. dabrowski and g. v. dunne , `` fractionalized non - self - dual solutions in the cp(n-1 ) model , '' phys . d * 88 * , 025020 ( 2013 ) [ arxiv:1306.0921 [ hep - th ] ] . g. v. dunne and m. nsal , `` generating non - perturbative physics from perturbation theory , '' phys . d * 89 * , 041701 ( 2014 ) [ arxiv:1306.4405 [ hep - th ] ] . a. cherman , d. dorigoni , g. v. dunne and m. nsal , `` resurgence in qft : unitons , fractons and renormalons in the principal chiral model , '' phys . lett . * 112 * , 021601 ( 2014 ) [ arxiv:1308.0127 [ hep - th ] ] . g. basar , g. v. dunne and m. nsal , `` resurgence theory , ghost - instantons , and analytic continuation of path integrals , '' jhep * 1310 * , 041 ( 2013 ) [ arxiv:1308.1108 [ hep - th ] ] . g. v. dunne and m. nsal , `` uniform wkb , multi - instantons , and resurgent trans - series , '' phys . d * 89 * , 105009 ( 2014 ) [ arxiv:1401.5202 [ hep - th ] ] . a. cherman , d. dorigoni and m. nsal , `` decoding perturbation theory using resurgence : stokes phenomena , new saddle points and lefschetz thimbles , '' arxiv:1403.1277 [ hep - th ] . a. behtash , t. sulejmanpasic , t. schaefer and m. unsal , `` hidden topological angles and lefschetz thimbles , '' arxiv:1502.06624 [ hep - th ] . s. bolognesi and w. zakrzewski , `` clustering and decomposition for non bps solutions of the @xmath420 models , '' phys . d * 89 * , 065013 ( 2014 ) [ arxiv:1310.8247 [ hep - th ] ] . t. misumi , m. nitta and n. sakai , `` neutral bions in the @xmath0 model , '' jhep * 1406 * , 164 ( 2014 ) [ arxiv:1404.7225 [ hep - th ] ] ; t. misumi , m. nitta and n. sakai , `` neutral bions in the @xmath421 model for resurgence , '' j. phys . ser . * 597 * , no . 1 , 012060 ( 2015 ) [ arxiv:1412.0861 [ hep - th ] ] . t. misumi and t. kanazawa , `` adjoint qcd on @xmath422 with twisted fermionic boundary conditions , '' jhep * 1406 * , 181 ( 2014 ) [ arxiv:1405.3113 [ hep - ph ] ] . t. misumi , m. nitta and n. sakai , `` classifying bions in grassmann sigma models and non - abelian gauge theories by d - branes , '' ptep * 2015 * , 033b02 ( 2015 ) [ arxiv:1409.3444 [ hep - th ] ] . m. nitta , `` fractional instantons and bions in the principal chiral model on @xmath423 with twisted boundary conditions , '' arxiv:1503.06336 [ hep - th ] . m. nitta , `` fractional instantons and bions in the o@xmath424 model with twisted boundary conditions , '' jhep * 1503 * , 108 ( 2015 ) [ arxiv:1412.7681 [ hep - th ] ] . g. v. dunne and m. unsal , `` resurgence and dynamics of o(n ) and grassmannian sigma models , '' arxiv:1505.07803 [ hep - th ] . g. t hooft , `` can we make sense out of quantum chromodynamics ? , '' subnucl . ser . * 15 * , 943 ( 1979 ) . v. a. fateev , v. a. kazakov and p. b. wiegmann , `` principal chiral field at large n , '' nucl . b * 424 * , 505 ( 1994 ) [ hep - th/9403099 ] . v. a. fateev , p. b. wiegmann and v. a. kazakov , `` large n chiral field in two - dimensions , '' phys . * 73 * , 1750 ( 1994 ) . j. ecalle , les fonctions resurgentes " , i - iii ( publ . math . orsay , 1981 ) . m. marino , r. schiappa and m. weiss , `` nonperturbative effects and the large - order behavior of matrix models and topological strings , '' commun . * 2 * , 349 ( 2008 ) [ arxiv:0711.1954 [ hep - th ] ] . r. couso - santamaria , j. d. edelstein , r. schiappa and m. vonk , `` resurgent transseries and the holomorphic anomaly : nonperturbative closed strings in local @xmath425 , '' commun . phys . * 338 * , no . 1 , 285 ( 2015 ) [ arxiv:1407.4821 [ hep - th ] ] . e. b. bogomolny , `` calculation of instanton - anti - instanton contributions in quantum mechanics , '' phys . b * 91 * , 431 ( 1980 ) . j. zinn - justin , `` multi - instanton contributions in quantum mechanics , '' nucl . b * 192 * , 125 ( 1981 ) . j. zinn - justin , `` multi - instanton contributions in quantum mechanics . 2 . , '' nucl . b * 218 * ( 1983 ) 333 . j. zinn - justin and u. d. jentschura , `` multi - instantons and exact results i : conjectures , wkb expansions , and instanton interactions , '' annals phys . * 313 * , 197 ( 2004 ) [ quant - ph/0501136 ] . j. zinn - justin and u. d. jentschura , `` multi - instantons and exact results ii : specific cases , higher - order effects , and numerical calculations , '' annals phys . * 313 * , 269 ( 2004 ) [ quant - ph/0501137 ] . u. d. jentschura and j. zinn - justin , `` multi - instantons and exact results . iv : path integral formalism , '' annals phys . * 326 * , 2186 ( 2011 ) . m. a. escobar - ruiz , e. shuryak and a. v. turbiner , `` three - loop correction to the instanton density . i. the quartic double well potential , '' arxiv:1501.03993 [ hep - th ] ; `` three - loop correction to the instanton density . ii . the sine - gordon potential , '' arxiv:1505.05115 [ hep - th ] . m. eto , t. fujimori , y. isozumi , m. nitta , k. ohashi , k. ohta and n. sakai , `` non - abelian vortices on cylinder : duality between vortices and walls , '' phys . d * 73 * , 085008 ( 2006 ) [ hep - th/0601181 ] ; m. eto , t. fujimori , m. nitta , k. ohashi , k. ohta and n. sakai , `` statistical mechanics of vortices from d - branes and t - duality , '' nucl . b * 788 * , 120 ( 2008 ) [ hep - th/0703197 ] . m. eto , y. isozumi , m. nitta , k. ohashi and n. sakai , `` solitons in the higgs phase : the moduli matrix approach , '' j. phys . a * 39 * , r315 ( 2006 ) [ hep - th/0602170 ] . f. bruckmann , `` instanton constituents in the o(3 ) model at finite temperature , '' phys . lett . * 100 * , 051602 ( 2008 ) [ arxiv:0707.0775 [ hep - th ] ] ; w. brendel , f. bruckmann , l. janssen , a. wipf and c. wozar , `` instanton constituents and fermionic zero modes in twisted cp**n models , '' phys . b * 676 * , 116 ( 2009 ) [ arxiv:0902.2328 [ hep - th ] ] ; d. harland , `` kinks , chains , and loop groups in the cp**n sigma models , '' j. math . phys . * 50 * , 122902 ( 2009 ) [ arxiv:0902.2303 [ hep - th ] ] . f. bruckmann and t. sulejmanpasic , `` nonlinear sigma models at nonzero chemical potential : breaking up instantons and the phase diagram , '' arxiv:1408.2229 [ hep - th ] . e. r. c. abraham and p. k. townsend , `` q kinks , '' phys . b * 291 * , 85 ( 1992 ) ; e. r. c. abraham and p. k. townsend , `` more on q kinks : a ( 1 + 1)-dimensional analog of dyons , '' phys . lett . b * 295 * , 225 ( 1992 ) ; m. arai , m. naganuma , m. nitta and n. sakai , `` manifest supersymmetry for bps walls in n=2 nonlinear sigma models , '' nucl . b * 652 * , 35 ( 2003 ) [ hep - th/0211103 ] ; m. arai , m. naganuma , m. nitta and n. sakai , `` bps wall in n=2 susy nonlinear sigma model with eguchi - hanson manifold , '' in * arai , a. ( ed . ) et al . : a garden of quanta * 299 - 325 [ hep - th/0302028 ] . j. p. gauntlett , d. tong and p. k. townsend , `` multidomain walls in massive supersymmetric sigma models , '' phys . d * 64 * , 025010 ( 2001 ) [ hep - th/0012178 ] ; y. isozumi , m. nitta , k. ohashi and n. sakai , `` construction of non - abelian walls and their complete moduli space , '' phys . lett . * 93 * , 161601 ( 2004 ) [ hep - th/0404198 ] ; y. isozumi , m. nitta , k. ohashi and n. sakai , `` non - abelian walls in supersymmetric gauge theories , '' phys . rev . d * 70 * , 125014 ( 2004 ) [ hep - th/0405194 ] ; m. eto , y. isozumi , m. nitta , k. ohashi , k. ohta and n. sakai , `` d - brane construction for non - abelian walls , '' phys . d * 71 * , 125006 ( 2005 ) [ hep - th/0412024 ] ; m. eto , y. isozumi , m. nitta , k. ohashi , k. ohta , n. sakai and y. tachikawa , `` global structure of moduli space for bps walls , '' phys . d * 71 * , 105009 ( 2005 ) [ hep - th/0503033 ] . m. nitta , `` josephson vortices and the atiyah - manton construction , '' phys . d * 86 * , 125004 ( 2012 ) [ arxiv:1207.6958 [ hep - th ] ] ; m. kobayashi and m. nitta , `` sine - gordon kinks on a domain wall ring , '' phys . d * 87 * , no . 8 , 085003 ( 2013 ) [ arxiv:1302.0989 [ hep - th ] ] . y. tanizaki , `` lefschetz - thimble techniques for path integral of zero - dimensional @xmath426 sigma models , '' phys . d * 91 * , no . 3 , 036002 ( 2015 ) [ arxiv:1412.1891 [ hep - th ] ] ; t. kanazawa and y. tanizaki , `` structure of lefschetz thimbles in simple fermionic systems , '' jhep * 1503 * , 044 ( 2015 ) [ arxiv:1412.2802 [ hep - th ] ] ; y. tanizaki , h. nishimura and k. kashiwa , `` evading the sign problem in the mean - field approximation through lefschetz - thimble path integral , '' phys . d * 91 * , no . 10 , 101701 ( 2015 ) [ arxiv:1504.02979 [ hep - th ] ] .

* * \1 . data * 1.1 general considerations in regard to high dimensional and massive datasets * 1.2 characteristics of the data studied * 1.3 distribution of terms * \2 . objective of computational efficiency : using aggregated subsets to construct the factor space * 2.1 introduction * 2.2 projection into the factor space * 2.3 the practical benefit of a selected word set * 2.4 conclusion * * \3 . about the data ; followed by correspondence analysis of all recipes using a 247 word set * \4 . search using solr * 4.1 setting up data file and its description * 4.2 running the server and updating the index * 4.3 querying and other operations * 4.3.1 web browser user interface * 4.3.2 command line querying * \5 . consistency of solr searches * references * appendix 1 : sample recipe * appendix 2 : 247 ingredients used as attributes in the correspondence analysis * appendix 3 : alternative presentation of plot * appendix 4 : correspondence analysis for singular / plural association and potentially for disambiguation very high dimensional ( or equivalently , very low sample size or low @xmath0 ) data sets , by virtue of high relative dimensionality alone , have points mostly lying at the vertices of a regular simplex or polygon @xcite . in @xcite , the following issues are posed . firstly , what should we do , for our analytics , when we have thousands of documents ? correspondence analysis outputs ( and this is the same for other kindred methods , like lsa , latent semantic analysis ; plsa , probabilistic adaptation of lsa using a mixture of multinomial distributions ; factorial correspondence analysis ( ca ) ; kohonen maps ; or clustering methods ) are very huge ( much more than the original dataset ) . what we must therefore do is to display the output in an `` intelligent '' and friendly way . we explore the following . * we have 306 collections , labelled by `` recipe via meal - master ( tm ) v8.05 '' , each with , in principle , 500 recipes . * ( an additional set of 20 collections , relating to `` v8.06 '' , were considered as different , and set aside from this work . ) * there is one exception here : one of the collection files had two recipes less than the usual 500 complement . * hence we are dealing with 306 @xmath1 500 recipes ( less 2 ) , i.e. 152,998 . * appendix 1 shows a sample recipe . * this data , in text , contained : 5,948,739 lines of text ; 30,199,625 ( whitespace - demarcated ) words ; and 206,993,672 characters . * on the 306 recipe collections , comprising 152,998 recipes , we used our term extractor . this listed all terms of one character or more , having first removed punctuation and numeric character . we will use the term `` words '' for what was extracted . all upper case had been first set to lower case . abbreviations and acronyms showed up as words , some character combinations did so also following punctuation removal , and all forms of stemming were retained . + various misspellings are in the given data . in collection mm066001.mmf , there is : `` grate the zuccini into a calendar ( sp ) ... '' . what is intended is `` zucchini '' and `` colander '' . interestingly there are 57 times that `` zuccini '' appears , and once that `` zuccinis '' appears . + while spelt correctly , `` zucchini '' is present 7155 times , and `` zucchinis '' is present 56 times . there are various other incorrect words present , with one occurrence each : `` azucchini '' , `` brushingzucchini '' , `` lambzucchini '' , `` ofzucchini '' , `` pzucchini '' , `` zucchinia '' , `` zucchiniabout '' , `` zucchinii '' , `` zzzucchini '' . + since we take the data as such , notwithstanding such inaccuracies , it is to be noted that the correct spellings are predominant , and that the misspellings only show up if quite atypical features in the data ( such as rare occurrence ) are looked into . let us formulate this as a working principle , or perhaps even a working hypothesis that we have verified in all cases that we have looked into . in the case of large , or very large , numbers of occurrences of entities , the syntactically correct form will predominate . * from the 152,998 recipes we obtained 101,060 unique words . * in this list , ranked by decreasing frequency of occurrence , the 5000th ranked word had 162 occurrences ; the 15,000th ranked word had 18 occurrences ; and at the 56,229th ranked word ( 56% through the total list of 101,060 ranked words ) , the frequency of occurrence became thereafter ( for higher ranked words ) , 1 . * the top frequencies of occurrence are as follows : + .... and 890294 the 842542 to 531633 in 469003 with 313588 of 281320 recipe 280035 or 247912 ts 238309 for 226514 until 224481 add 207146 tb 200221 from 193659 minutes 190880 by 190815 on 183287 salt 162811 yield 162082 servings 161098 meal 155040 master 152980 via 149162 title 148802 tm 148766 categories 148612 into 145565 sugar 138670 heat 131846 water 130211 pepper 127901 oil 122997 chopped 119530 over 115212 butter 112329 at 106300 sauce 102881 is 101060 com 99841 flour 96793 stir 91580 mixture 91484 about 91279 it 89491 cook 87097 pan 86314 oz 83908 garlic 82776 cream 81633 place 80825 cheese 78986 bowl 76209 mix 76182 chicken 75807 onion 71516 cut 70808 lb 68376 posted 67769 fresh 66125 baking 64731 .... * the very final ranked words were as follows : + .... zwiebelfleisch 1 zwirek 1 zws 1 zygielbaum 1 zyla 1 zylman 1 zza 1 zzwc 1 zzzingers 1 zzzucchini 1 .... + the second last here arose from this entry in a recipe : `` title : wing zzzingers '' . the last word comes from this entry in a recipe : `` add zzzucchini , peppers and onion . '' . a power law ( see @xcite ) is a distribution ( e.g. of frequency of occurrence ) of the form @xmath2 where constant , for the data , @xmath3 ; and an exponential law is of the form @xmath4 . for a power law , @xmath5 , @xmath6 . a power law has heavier tails than an exponential distribution . in practice @xmath7 . for such values , @xmath8 has infinite ( i.e. arbitrarily large ) variance ; and if @xmath9 then the mean of @xmath8 is infinite . the density function of a power law is @xmath10 , and so @xmath11 , where @xmath12 is a constant offset . hence a log - log plot shows a power law as linear . power laws have been of great importance for modelling networks and other complex data sets ( see @xcite ) . figure [ fig1 ] shows a plot of rank ( most frequent , through to least frequent , the latter being a very large number of words that occur once only ) against the value of the frequency of occurrence . to understand the plot , we use log - log scaling . in a very similar way to the power law properties of large networks ( or file sizes , etc . ) we find an approximately linear regime , ending ( at the lower right ) in a large fan - out region . the slope of the linear region characterizes the power law . we find that the probability of having more than @xmath8 occurrences per word to be approximately @xmath13 for large @xmath8 ( since we find the slope of the fitted line in figure [ fig1 ] to be @xmath14 ) . just how well can correspondence analysis carry out the euclidean , factor determining mapping , if recipes are first aggregated together ? one reason for this question is computational time . for analysis of an @xmath15 data set , with @xmath16 words and @xmath0 recipes or recipes collections , then the dual spaces means that we analyze either the data set or its transpose . eigen - reduction is a cubic process . the computational time for analysis of the @xmath15 data set is @xmath17 , or for its transpose , @xmath18 . for a selected word set , if we can carry out the main analysis on a smaller @xmath0 , so much the better for us . aggregation is just concatenating the recipes . in the entire set of recipes , the number of words ( to repeat : of one or more characters in length ; punctuation deleted ; upper case set to lower case ; numerical and accented characters ignored ) was 101,060 . for initial assessment , we took the top - ranking 1000 words . so this illustrative case , using just 500 recipes , has the recipe set crossed by this 1000-word set , with the frequencies of occurrence tabulated . we aggregated the 500 recipes into 5 recipe - sets . these comprised recipes 1100 , 101200 , 201300 , 301400 , and 401 - 500 . each of these 5 recipe - sets had the frequencies of occurrences on the 1000-word set . here we analyze two cross - tabulation tables , of dimensions respectively @xmath19 , and @xmath20 . by construction of the latter , the column sums are identical . then due to the use of profiles , the average profile is the same in each case . if we represent the data matrix in frequency terms ( i.e. the frequencies of occurrences each divided by the grand total ) , the matrix can be denoted @xmath21 where @xmath22 is set of recipes , or of recipe - sets , @xmath23 is the word - set , @xmath24 is a recipe or a recipe - set , and @xmath25 is a word ) the column marginal distribution is @xmath26 , or in summation terms , @xmath27 . in correspondence analysis , the centre or origin of the euclidean factor space is given by @xmath28 in the recipe space , and by @xmath29 in the word space , such that we have the following view : starting with a probability distribution @xmath30 we want to explore how it differs from the product of marginal distributions , @xmath31 , and this in the @xmath32 metric of centre , the product @xmath31 . this is a decomposition of inertia of clouds in recipe and/or in word space . we have : @xmath33 where @xmath34 . figure [ cawithagg ] shows the principal factor display resulting from the correspondence analysis of ( 1 ) the @xmath35 recipes times words data , and ( 2 ) the @xmath20 recipe - sets times words data . note again that the analyses ( the eigen - reductions determining the factors ) were completely independent . the results of figure [ cawithagg ] can be explained in the following terms . we have a set of 500 items ( the recipes ) with quantitative description in a 1000-dimensional space . we have a derived , through concatenating , sets of 5 items in the same 1000-dimensional space . so the resultant euclidean mappings illustrate well the shared provenance in the two cases . the word analysis is less clear - cut . given the inherent dimensionality of the euclidean , factor space is min(@xmath36 ) , we have , on the one hand 1000 words projected into a 499-dimensional space , and on the other hand 1000 words projected into a 4-dimensional space . the locations of a given word in these two spaces are not directly related . we investigate supplementary elements to expedite the computation . carry out analysis in a small dimensional space , that nonetheless takes account of all relationships in an aggregated way . then complete the analysis on all the data by projecting into the euclidean , factor space . we took the 306 collected sets of , each , 500 recipes ( save for the case of 2 recipes short of this in one such collected set : hence we were dealing with 152,998 recipes ) . the number of words found in this data ( to repeat : of one or more characters in length ; punctuation deleted ; upper case set to lower case ; numerical and accented characters ignored ) was 101,060 . for initial assessment , we took the top - ranking 1000 words . the following principles and practices of correspondence analysis will now be availed of ( see @xcite ) : * in correspondence analysis , all frequency of occurrence values are divided by the grand total ( so that allows the viewpoint of empirical probabilities ) . if the row @xmath24 mass is then @xmath37 , the row @xmath24 profile is the row vector @xmath38 with each element divided by @xmath37 . now consider the 360-set , which is just all 500 recipes , @xmath38 , summed . the profile of this set is @xmath39 where @xmath40 . if one recipe is a multiplicative constant of another ( i.e. elementwise , for each word ) , then their profiles are the same . * the dual space relations allows supplementary rows or columns to be projected , after the fact , into the analysis . + ( in the latent semantic indexing context , `` folding in '' of newly presented rows or columns , that can be considered as a row or column set that is juxtaposed to the original @xmath41 matrix , is described in @xcite ) . * profiles are analyzed in correspondence analysis , meaning that row vector , and column vector , values have been divided by the associated row , or column , total . consider the row or column total as a mass , as is done in this context . by having all values first divided by the grand total of the rows / columns array , we have all values bounded by 0 and 1 . ( note that we require non - negative values for the mass to be workable for us here . ) in this way our 306-set of ( 500 ) recipes is the _ weighted mean _ of the associated recipes + to see this , consider the given frequency of occurrence data divided by the grand total ( over all recipes and words ) , on the set of recipes , that we will represent by matrix @xmath8 . consider the 500 recipes that are associated with one of the 306-sets . without loss of generality , call them @xmath42 . we will denote one such recipe by @xmath43 and just right now we are most interested in the set @xmath44 . @xmath23 is the set of words . then the _ profile _ of recipe @xmath45 is @xmath46 . + now consider the 360-set . it is the aggregation ( concatenation ) of the recipes . so , it is @xmath47 . its profile is @xmath48 . figure [ proj1fig ] presents an initial look at the use of the top ranked words ( in terms of frequency of occurrence in the entire data collection , hence : 306 collections of recipes , or 152,998 recipes ) . . lower left and right : these two sets of 500 recipes are projected into the analysis of the 306 recipe collections . what enables this to be accomplished is that the same 1000-word set is used in all cases here.,width=529 ] in figure [ proj2fig ] we look at the potential for handling massive data sets by projecting into a `` primary '' analysis . we have this analysis here on the @xmath49 recipe collections times 1000 highest ranked words . we project into the analysis the first @xmath35 recipe set , and the second @xmath19 recipe set . the scheme used is as follows . we take as a main matrix the one that crosses 306 rows with 1000 columns . then we have a supplementary row set , of 500 rows crossed with 1000 columns . finally we have a further supplementary row set , of 500 rows crossed with 1000 columns . the supplementary rows are projected into the correspondence analysis of the @xmath49 matrix , post factum . we note that it is best to carry out the eigen - reduction on the smaller of the row set and the column set . hence , relative to how we have presented our processing steps above , in fact we worked on the transposed matrices . ( for the @xmath49 matrix , diagonalization is carried out on a @xmath50 matrix . ) dataset , and black `` x '' are the projections , relative to principal factor plane projection of using the original @xmath35 data . upper right : a black dot , `` . '' , represents projections as supplementary elements of the 500 recipes , relative to the derived 5 recipe - sets ; and a red dot , `` . '' , represents the active correspondence analysis of these @xmath35 data . so the black dots are efficiently determined , but the red dots show the `` ground truth '' . lower panel : different set of 5 recipe - collections , as used previously ( e.g.upper left panel here ) , and then independently defined in order to enhance having `` like with like '' in the same recipe - collection . note that in all cases we are using a 2-dimensional ( albeit optimal ) display.,width=529 ] in figure [ bothways - diff5 ] , in the upper left we show with `` x '' ( i ) the projection as supplementary elements of a @xmath20 set of data derived from a @xmath35 set of data ; and shown with a red `` + '' ( ii ) the projections of the active correspondence analysis of the @xmath20 set of data . the @xmath35 set of data was the set of 500 recipes used before in fact the sequentially first of these batches of recipes ( and as indicated there were 306 of these batches ) . the word set is 1000 , being the top - ranked words in the entire word - set derived from all recipes . from the @xmath35 set of data , we took the first 100 recipes , then recipes 201300 , 301400 , and 401500 . we aggregated the recipes , which is equivalent to concatenating the recipes . this provided a frequency of occurrence cross - tabulation of @xmath20 . in figure [ bothways - diff5 ] , upper right , we show with a dot , `` . '' , ( i ) the 500 recipes with their projections found from using the @xmath35 data as supplementary elements relative to the @xmath20 data . we show also , using red dots , `` . '' , ( ii ) the active correspondence analysis of the @xmath35 data . in figure [ bothways - diff5 ] , lower panel , there is , with an `` x '' , ( i ) the projections of the active correspondence analysis of the @xmath20 set of data , as in the upper left . then , with a `` o '' , there is ( ii ) the same analysis on a matrix of the same dimensions , but this time with the 5 concatenated sets of recipes being based not on the given sequential order , but rather on the first factor projections of the full @xmath35 matrix . this was done in order to assess a different derived frequency of occurrence set of data . arising out of this discussion , we will draw a balance sheet on projections , through assessment of similarity of outcomes . we use the sums of squared euclidean distances , taken from the principal plane factor projections , which are euclidean . * quality of results based on sum of squared euclidean distances . * 1 . * concatenated recipes projected onto full recipe set , versus concatenated recipes : * @xmath20 projected onto @xmath35 , and active analysis of @xmath20 : * 0.03306077 * 2 . * full recipe set projected onto concatenated recipes , versus full recipe set : * @xmath35 projected onto @xmath20 , and active analysis of @xmath35 : * 0.283562 * next , @xmath20 is `` ordered '' i.e. based on a chosen sequencing and not an arbitrarily given one . the chosen sequencing is that of projections on the first factor . so , in the foregoing experiment , the 5 groups of recipes , each of 100 recipes , had their sets of 100 recipes in the given , arbitrary , sequence . now , the 5 groups of recipes , each of 100 recipes , have their sets of 100 recipes in a somewhat clustered sequence that is provided by these recipes projections on the first factor . 1 . * concatenated clustered recipes projected onto full recipe set , versus concatenated clustered recipes : * @xmath20 projected onto @xmath35 , and active analysis of @xmath20 : * 0.9963667 * 2 . * full recipe set projected onto concatenated clustered recipes , versus full recipe set : * @xmath35 projected onto @xmath20 , and active analysis of @xmath35 : * 1.097960 * what is noteworthy here is that the `` ordered '' , i.e. somewhat more clustered 100-strong recipe - sets , preform less well than arbitrarily - given 100-strong recipe sets . latter two results above vis - - vis the former two . ) we draw the conclusion that the lesser coherence associated with the latter is better when it comes to projecting other recipes into this space . we see a certain parallel between this outcome and the importance of incoherence in compressed sampling in signal processing . such incoherence is when @xcite ( p. 278 ) , there should be as much spread as possible between the sensing vector , on the one hand , and on the other hand , the `` sparsity atoms '' or basic components that we can use for the signal to be acquired . we also have an indication from the above that the bigger the better the analysis , in terms of observations , when it is a matter of projecting into the factor space . first result above , out of the four , being better than the second result . ) we conclude : the bigger the set of observations that we can work on , so much the better . also if we do need to work on a limited , representative set of aggregates , then lack of coherence ( i.e. lack of clustering ) is best . we use again the first 500-recipe set . with the word set , as before , derived from the entire set of recipes , we look at ( i ) the top ranked ( i.e. most frequently occurring ) 1000 words , ( ii ) the 42,052 word list such that word lengths are greater than 2 , and ( iii ) the full word set , with 101,060 words . in each case , we therefore used the 500 recipes cross - tabulated with the words in terms of frequency of occurrences . recall again that the word list used was derived from the entire set of recipes ( and not just these 500 ) . figures [ figevals1 ] and [ figevals2 ] portray the rapid falloff in inherent embedding dimensionality , based on the word set , and associated attribute dimensionality . in figure [ figevals1 ] , the ( effectively superimposed ) 42,052 and 101,060 word sets are on top , and the 1000 word set on the bottom . the reason for the two top curves being superimposed is the following . the full word set is just the words that occur twice or once in the recipes . hence they are very rare words . hence , too , while the 500 recipes here are characterized by , on the one hand , 42,052 words , and on the other hand , 101,060 words , nonetheless the extra data is extremely sparse consisting mostly of zero values . it is small wonder therefore that these two datasets , relating to words with more than two occurrences , and all words , give practically the same outcome . the top 8 actual eigenvalues are as follows , if we choose , on the basis of figure [ figevals1 ] to select , e.g. 8 factors as bearing the most important information : .... xx1c$evals[1:8 ] [ 1 ] 0.3441775 0.2973126 0.2854613 0.2463559 0.2445217 0.2348758 0.2255849 0.2212731 x1c$evals[1:8 ] [ 1 ] 0.2722012 0.1816320 0.1526613 0.1341513 0.1278123 0.1133759 0.1084320 0.1009735 .... our preference is to focus our interest on just the first three factors , given these eigenvalues . displaying the eigenvalues themselves . here the top curve relates to the set of 1000 words , and the bottom more or less fully superimposed curves relate to the 42,052 word set and using the entire word set of 101,060 words.,width=529 ] in section [ sectchar ] , we observed that : in the case of large , or very large , numbers of occurrences of entities , the syntactically correct form will predominate . in section [ sectproj ] , we concluded that : the bigger the set of observations that we can work on , so much the better . also if we do need to work on a limited , representative set of aggregates , then lack of coherence ( i.e. lack of clustering ) is best . finally , in section [ sectwordset ] , we concluded that : large , sparse word sets lead to a similar outcome ; and a well selected , smaller sized word set is important for best , i.e. smallest , reduced dimensionality mapping . the 152,998 recipes were found to have 101,060 unique words of 1 character or longer , having set all upper case to lower case , and having ignored punctuation , numeric and accented characters . the distribution of terms follows a power law ( zipf s law ) : the probability of having more than @xmath8 occurrences per word was found to be approximately @xmath13 . in total we have 152,998 recipes . a set of 247 recipe ingredients was assembled . they were chosen from a word list ( unique words ) drawn from the recipes and ordered by decreasing frequency of occurrence . there were just 122 recipes that did not contain at least one of these 247 ingredient terms . ( that could be due to use of less common ingredients , not figuring in our list of 247 ; or misspellings , of which there were a considerable number ; or some unusual text instead of a recipe . an example of the latter was a short overview of appropriate wines for types of food . ) figure [ f1f2rec ] shows the projections of recipes on the principal factor plane . figure [ f1f2ing ] shows the ingredients used , projected onto the principal factor plane . the ingredients with the strongest contributions to these factors are noted . looking further , we found that the strongest contribution by ingredients to factor 3 is `` chocolate '' ; and the strongest contribution to factor 4 is `` cheese '' . the solr ( lucence.apache.org/solr ) search server was used . 1 . the data to be indexed , and supported for search , is put into xml with the following structure : * ` < add > ` and ` < /add > ` at beginning and end . * each entry defined by ` < doc > ` and ` < /doc > ` . * a required field providing a unique identifier : + ` < field name=''id''>mm000001102.txt</field > ` . * other fields , such as the following for bounding box search : + ` < field name=''xcoord''>-0.7341409</field > ` + ` < field name=''ycoord''>-0.09961348</field > ` + ( note that these coordinates are the principal factor coordinates resulting from a correspondence analysis . i.e. , they are the factor 1 and factor 2 projections , respectively . ) * other fields such as : + ` < field name=''name''>''21 '' club rice pudding</field > ` * the main text field , tagged by : ` < field name=''recipe''> ... </field > ` . + the xml file was of size over 205.5 mb . it was put in directory ` example / exampledocs ` . 2 . two files in directory ` example / solr / collection1/conf ` need to be fully cognizant of the xml data to be used . these are ` schema.xml ` and ` solrconfig.xml ` . in the latter , changes made included the following , with regard to ` mlt ` or ` mlt ` , `` more like this '' option . + in the ` requesthandler ` there is : + .... < arr name="components " > < str > query</str > < str > facet</str > < str > mlt</str > < str > highlight</str > < str > stats</str > < str > debug</str > < /arr > < /requesthandler > ... < ! -- query settings -- > < str name="deftype" > edismax</str > < str name="qf " > recipe^1.0 < /str > < str name="df" > text</str > < str name="mm">100%</str > < str name="q.alt">*:*</str > < str name="rows">10</str > < str name="fl">*,score</str > < str name="mlt.qf " > recipe^1.0 < /str > < str name="mlt.fl" > recipe</str > < ! -- following : no . of nearest neigbours returned in mlt -- > < int name="mlt.count">1</int > < ! -- faceting defaults fm changed to off -- > < str name="facet" > off</str > .... 3 . in file ` schema.xml ` the fields used and their definitions need to be defined . this included : + ` < field name="recipe " type="text_general " indexed="true " stored="true " ` + ` termvectors="true"/ > ` + also : + .... < field name="xcoord " type="double " indexed="true " stored="true " / > < field name="ycoord " type="double " indexed="true " stored="true " / > .... 4 . for the browser - based querying , this is management by the ` velocity ` component , that is in subdirectory ` example / solr / collection1/conf / velocity ` . + the files changed there , to be appropriate for the data analyzed , were : ` browse.vm , footer.vm , header.vm , product-doc.vm , query.vm ` . display images ( see the principal factor plane in figures [ screen1 ] , [ screen2 ] ) used there are located in the named image directory , which is currently + ` example / solr - webapp / webapp / img ` . the solr server is started thus , in directory ` example ` , where in this example , 1 gb of memory is provided : ` java -xmx1024 m -jar start.jar ` to update , or commence , indexing , in the directory containing the xml data , ` example / exampledocs ` , the following command is issued . this supposes a running server . ` post.jar recipes-f1f2-1.xml ` the unique identifier field is the crucial aspect of what gets taken into the index , or updated . the following access address is used , based on the running server : ` http://localhost:8983/solr / browse ` through the upper right hand corner link to the admin screen , or directly using ` http://localhost:8983/solr/#/collection1 ` , there is availability of the log file ; querying can be carried out ; statistics of use and of the data can be accessed . note that as currently configured in this work , ` collection1 ` contains the indexed data . example of query : `` sugar beer pasta '' . a required term is specified in the query with a preceding plus sign , and a requirement not to have a term is specified with a minus sign preceding the term . the `` more like this '' option gives a number of nearest neighbours of the document , and its parameters are defined in the settings in ` schema.xml ` . figures [ screen1 ] and [ screen2 ] show examples of use . the correspondence analysis figure at the top is static . it and the table to its left are provided as navigation aids in search and discovery . the table to the left of the correspondence analysis principal factor plane is also static . an alternative could be a text cloud or cloudmap ( see @xcite ) . an example follows ( all to be placed on one line ) . for this , three `` more like this '' near neighbours were required ( i.e. , in file ` solrconfig.xml ` , there was this setting : ` < int name=''mlt.count''>3</int > ` ) . .... curl -o out.xml ' http://localhost:8983/solr / collection1/ browse?&q = id : mm078001428.txt&wt = xml&mlt = true ' .... in this case , results are saved to file ` out.xml ` . from a search for recipe ` mm078001428.txt ` , a `` more like this '' request is submitted . from the file ` out.xml ` here is a little utility to write out just the recipe identifiers ( all on one line ) : .... awk -v srch="\"id\ " > " ' begin{l = length(srch ) } end{print " \n " } { t = match($0,srch);if(!t){next}extr = substr($0,t+5,15);printf " % s " , extr } ' < out.xml .... in this case , this gives : .... mm078001428.txt mm110501451.txt mm158501305.txt mm161501159.txt .... using identifier ( and recipe ) `` mm078001428.txt '' , with name `` no fat , no salt , no sugar vanilla ice cream '' , we found its `` more like this '' best match to be `` mm110501451.txt '' , with name `` sugar - free cappuccino ice cream '' . using the latter , we found its best match to be the former , thereby showing consistency . more specifically , the two recipes in this case were mutual or reflexive nearest neighbours . see @xcite for the use of this principle ( of mutual nearest neighbours , and also nearest neighbour chains ) in agglomerative hierarchical clustering algorithms . we also sought best match recipes using the correspondence analysis factor space . this latter is of course euclidean . we used the full space dimensionality ( hence 247 less 1 due to linear dependence through centring the cloud of recipes , and the dual cloud of ingredients ) . for recipe `` mm078001428.txt '' , we got its closest neighbour as `` mm048501554.txt '' , `` fruit flavor milk shakes '' . for recipe `` mm110501451.txt '' , we got its closest neighbour as `` mm161501159.txt '' , `` vanilla ice cream - diabetic * ww no.2 '' is the nearest neighbour of point @xmath51 . point @xmath52 is the nearest neighbour of point @xmath53 . point @xmath54 is the nearest neighbour of point @xmath52 , and reciprocally @xmath52 is the nearest neighbour of point @xmath54.,width=377 ] in figure [ nns ] the geometric situation is depicted , irrespective of how nearest neighbour is defined ( it need not be a distance but some dissimilarity satisfying @xmath55 ; @xmath56 ; and the closer @xmath51 is to @xmath53 , i.e. the more alike they are , the smaller the value of @xmath57 ) . the squared euclidean distance in the full dimensional euclidean factor space gives for the pair of best match recipes furnished by solr , 18.94413 . then in the full dimensional factor space , the best match distance squared of `` mm078001428.txt '' , as noted above , was 4.934411 ; and the best match distance squared of `` mm110501451.txt '' was 6.103886 . to note that the correspondence analysis best match distances are supported by the data is to be unfair to what is , with solr , a different framework for determining best matches . this includes using all words , with stemming and lemmatization , compared to the 247 ingredients used in the correspondence analysis case . we note that the solr best match information lends itself to looking for clusters such that nearest neighbour chains are followed , and once a ( mutual or reciprocal ) nearest neighbour pair is found , they can be agglomerated with no impact on close - by recipes . this is due to bruynooghe s reducibility property . background on nearest neighbour chain , and reciprocal nearest neighbour based hierarchical agglomerative clustering can be found in @xcite . 99 s. deerwester , g.w . dumais , s.t . furnas , t.k . landauer and r. harshman , `` indexing by latent semantic analysis '' , _ journal of the american society for information science _ , 41 , 391407 , 1990 . n. eiron and k.s . mccurley , `` link structure of hierarchical information networks '' , proc . third workshop on algorithms and models for the web - graph ( waw 2004 ) , _ lecture notes in computer science _ , pp . 143155 , 2004 . i. fellows , word clouds , software in r written by ian fellows , version 2.2 , 2012 - 09 - 05 , http://log.fellstat.com/?cat=11 p. hall , j.s . marron and a. neeman , `` geometric representation of high dimension , low sample size data '' , _ journal of the royal statistical society b _ , 67 , 427444 , 2005 . b. le roux and h. rouanet , geometric data analysis , kluwer , 2004 . b. le roux and h. rouanet , multiple correspondence analysis , sage , 2010 . m. mitzenmacher , `` a brief history of generative models for power law and lognormal distributions '' , _ internet mathematics _ , 1 , 226251 , 2004 . a. morin , `` interactive text mining and applications '' , _ ntts 2011 new techniques and technologies for statistics 2011 _ , presentation ( 78 pp . ) at http://www.cros-portal.eu/book/export/html/17 f. murtagh , _ multidimensional clustering algorithms _ , physica - verlag , 1985 . f. murtagh and a. heck , _ multivariate data analysis _ , kluwer , 1987 . f. murtagh , `` on ultrametricity , data coding , and computation '' , _ journal of classification _ , * 21 * , 167184 , 2004 . f. murtagh , `` identifying the ultrametricity of time series '' , _ european physical journal b _ , * 43 * , 573579 , 2005 . f. murtagh , _ correspondence analysis and data coding with r and java _ , chapman & hall / crc , 2005 . f. murtagh , a. ganz , s. mckie , j. mothe and k. englmeier , `` tag clouds for displaying semantics : the case of filmscripts '' , _ information visualization journal _ , 9 , 253 - 262 , 2010 . e. neuwirth and l. reisinger , `` dissimilarity and distance coefficients in automation - supported thesauri '' , _ information systems _ , * 7 * , 4752 , 1982 . newman , `` the structure and function of complex networks '' , _ siam review _ , 45 , 167256 , 2003 . starck , f. murtagh and j. fadili , _ sparse image and signal processing : wavelets , curvelets , morphological diversity _ , cambridge university press , 2010 . ( 2nd edition forthcoming 2015 . ) recipe mm160001 collection , number 245 out of 500 in sequence from this collection , is as follows . .... title : the perfect roast chicken with roasted shallots and portob categories : lifetime tv , life5 yield : 4 servings 4 lb young roaster salt and pepper ; to taste 1 md onion ; halved and peeled 3 cloves garlic ; peeled and -smashed 1 bn fresh herbs ; rosemary thyme ; flat - leaf parsley 1/4 c olive oil 2 c chicken broth ; divided 8 lg shallots 8 portobello mushrooms ; stems -removed and ; cut in half 1/2 c dry white wine how to prepare the chicken : 1 . preheat the oven to 350 f. season the skin and the inside of the cavity of the chicken generously with salt and pepper . 2 . place the onion , garlic , and fresh herbs in the cavity and truss the chicken . in a roasting pan over medium - high heat , heat the oil until it begins to smoke . place the chicken , breast - side - down , in the oil and cook until all the sides off the chicken are completely browned . place the bird , breast - side - up into the oven and baste with 1/4 cup of the chicken stock . continue basting with 3/4 cup of the chicken stock every 10 minutes until the chicken is done , approximately 1 hour , or until the chicken reaches an internal temperature of 170 f. 5 . add the shallots 20 minutes after the chicken has been put into the oven and the mushrooms 40 minutes after the chicken has been put into the oven . remove the cooked chicken , shallots , and mushrooms from the oven and set on a platter , cover and let rest for 10 minutes . how to prepare the sauce : 1 . place the roasting pan back on the stove over medium - high heat and bring the juices in the pan to a boil . using a small ladle remove any of the fat that has risen to the top of the pan . 2 . add the white wine to the pan and , using a wooden spoon , scrape up the browned bits from the bottom of the pan . reduce the wine by half and add the remaining 1 cup of chicken stock . reduce the stock by half , season with salt and pepper and add the fresh thyme . carve the chicken and arrange on a platter with the shallots and mushrooms . spoon the pan juices over the meat . the perfect roast chicken with roasted shallots and portobello < a9 > 1997 lifetime entertainment services . all rights reserved . mc formatted using mc buster by barb at pk converted by mm_buster v2.0l . ----- .... we note the following in this recipe text : abbreviations ( `` lg '' , large ; `` lb '' , pounds ; `` c '' , cup ; etc . ) . source and other summarized or abbreviated data . a control character ( `` @xmath58a9@xmath59 '' ) . list of 247 ingredients used , with their frequencies of occurrence . .... salt sugar water pepper oil butter 167636 142624 134454 132012 126242 115387 sauce flour garlic cream cheese chicken 105909 99870 85080 83859 80835 78635 onion juice egg milk lemon eggs 73933 65773 62000 61125 55408 50086 bread onions rice chocolate olive vanilla 48789 44169 41645 40737 39539 38386 cake tomatoes parsley potatoes vinegar vegetable 35317 34894 32217 31423 31340 30735 wine tomato beans beef vegetables cloves 29192 28747 28180 28761 27203 27034 soup orange cinnamon margarine mushrooms salad 26558 25976 25656 24266 23693 23612 fish corn broth celery mustard pie 23267 21212 20832 21193 20192 20081 pasta fruit peppers soy chili basil 19743 18989 18695 18036 17148 16354 shrimp soda cookies syrup carrots honey 16332 16257 16216 15652 15197 14733 sodium cookie parmesan ice dressing cornstarch 14034 14173 13878 13928 13886 13848 thyme bacon pastry lime yeast potato 13636 13653 13309 13274 13347 12957 apple protein spinach casserole oregano cumin 12642 12129 11889 12106 11855 11996 nutmeg cholesterol raisins clove coconut pineapple 11756 11376 11019 11042 10892 10864 roast chips chop puree topping marinade 10758 10667 10637 10638 10542 10586 noodles loaf desserts cilantro yolks peanut 10441 10456 10371 10405 10362 10258 italian chile seasoning apples almonds peas 10285 10293 10350 10280 10216 10459 sesame turkey ham cabbage paprika leaf 10184 9975 9725 9736 9689 9703 mixer yogurt coriander carbohydrate sausage cayenne 9211 9067 8903 8765 8918 8892 lamb wheat bean walnuts cakes mint 8571 8639 8573 8470 8398 8256 lettuce mayonnaise pecans sherry cocoa pudding 8159 8000 7893 7950 7835 7843 cheddar grain salmon olives carrot shell 7877 7806 7724 7744 7767 7511 vegetarian broccoli zucchini salsa flakes grease 7405 7413 7290 7268 7104 7079 chinese shallots poultry mushroom steak rosemary 7079 6877 6969 6857 6927 6699 eggplant chiles rind curry coffee dill 6708 6804 6696 6798 6602 6683 spices breads buttermilk worcestershire starch seafood 6648 6670 6351 6273 6031 6135 pumpkin gelatin carbohydrates scallions almond chives 5820 5899 5749 5815 5662 5621 spice meats herbs tofu dessert pizza 5576 5591 5496 5409 5466 5398 strawberries juices muffins mexican tortillas chops 5373 5318 5290 5176 5139 4996 rum icing soups cornmeal emeril asparagus 5014 4917 5084 5032 4874 4740 sauces muffin fruits stuffing jelly salads 4792 4778 4737 4704 4756 4731 shells jalapeno mozzarella banana steaks greens 4645 4581 4484 4387 4354 4390 spaghetti broiler cucumber cherries toast cherry 4343 4346 4679 4351 4307 4358 chilies yolk gravy cider root tabasco 4324 4289 4258 4244 4304 4272 oats fiber veal tortilla sage dijon 4225 4135 4061 4085 4040 4040 bananas broil tuna molasses peaches peppercorns 3970 3970 3914 3949 3939 3928 candy duck tarragon fennel tart custard 3884 3805 3851 3789 3814 3774 maple scallops stew brandy berries pears 3749 3871 3802 3780 3777 3559 crab confectioners crackers biscuits bouillon peanuts 3611 3571 3570 3464 3526 3440 leeks beer turmeric stalks ricotta oranges 3392 3450 3385 3393 3274 3278 lentils raspberry cracker herb raspberries strawberry 3260 3216 3128 3113 3083 3093 jam 3151 .... figure [ textplot ] shows an improved presentation of figure [ f1f2ing ] , through movement of projections ( from the red dots , indicated by the stick lines ) , as implemented in the textplot program in the word cloud package in r @xcite . in the recipe set , there were 101,060 unique words . words were as defined by us : length @xmath60 ; all upper case set to lower case ; punctuation removed ; numeric and accented characters ignored . variations on spelling included the following example , with numbers of occurrence in the recipe set : .... zucchini 7155 zuchinni 86 zuccini 57 zucchinis 56 zuchini 40 zucchine 14 zucchinni 7 zucchin 6 zuckinni 5 zuchinnis 4 zuccinni 3 zuccchini 2 zuccihini 2 zuchhini 2 azucchini 1 brushingzucchini 1 lambzucchini 1 ofzucchini 1 pzucchini 1 zucchiniabout 1 zucchinii 1 zucchinnis 1 zucchni 1 zucchnini 1 zuccinis 1 zuchhinis 1 zuchine 1 zuchinis 1 zucinni 1 zzzucchini 1 .... we examine here , using the 152,998 @xmath1 247 ingredients set , the projections in the principal factor plane of singulars and plurals . recall that the principal factor plane , while being the best visualization of the data , with inherent dimensionality 246 , only accounts for just over 3.8% of the total inertia . the singulars used were as follows , with additionally their 23 corresponding plurals : `` strawberry '' , `` carrot '' , `` muffin '' , `` egg '' , `` apple '' , `` yolk '' , `` tortilla '' , `` bread '' , `` cookie '' , `` pepper '' , `` juice '' , `` vegetable '' , `` clove '' , `` carbohydrate '' , `` spice '' , `` soup '' , `` peanut '' , `` onion '' , `` olive '' , `` cracker '' , `` almond '' , `` bean '' , `` banana '' . figures [ sings ] , [ plurs ] , and also [ sandp ] with links drawn for all singular and plural pairs , show the close association in the principal factor plane . some singulars and plurals are admittedly somewhat separated , e.g. `` cracker '' , `` yolk '' , pointing to some difference in semantic clearly possible in practice and contextual use of singular and plural . overall the association between the terms is well exemplified in the figures .

bragg gratings ( bgs ) are light - controlling structures produced by a periodic variation of the refractive index along an optical fiber or waveguide . devices based on fiber gratings , such as dispersion compensators , sensors , elements of laser cavities , etc . , are widely used in optical systems ka:99 . gap solitons ( gss ) in fiber gratings , alias bg solitons stsi:94 , are supported through the balance between the bg - induced dispersion , which incorporates the bandgap in the system s linear spectrum , and the kerr nonlinearity of the fiber or waveguide . analytical solutions for bg solitons in the standard fiber - grating model are well known acwa:89,chjo:89 . slightly more than half of the analytical found family is stable , as first demonstrated , within the framework of the variational approximation , in ref . @xcite , and then in a consistent numerical form in ref . later , an accurate numerical technique for simulations of gss was elaborated in ref . @xcite , and a rigorous analysis of their stability was developed in ref . @xcite following the theoretical prediction , bg solitons were created in the experiment , using a relatively short bg written in the cladding of an optical fiber @xcite and a virtual optically - induced grating coen:04 . in particular , an essential physical achievement was the creation of slow bg solitons moving at the velocity much lower than the velocity of light @xmath0 in vacuum ( namely , @xmath1 @xcite ) . gss are one of fundamental species of solitons in optics , as well as in other nonlinear media . in particular , the gs concept was extended to bose - einstein condensates ( becs ) , where such solitons were predicted in a condensate loaded into a periodic potential ( optical lattice ) bako:02,oski:03,br:04 , and then created in the experiment @xcite . an issue of great significance to the fundamental studies and applications is the development of methods for control of bg solitons . one of them is _ apodization _ @xcite , i.e. , the use of a grating with the bragg reflectivity gradually varying along the fiber ( or waveguide ) . in particular , a possibility was predicted to slow down the soliton and eventually bring it to a halt in a properly apodized bg mama:04 . experimentally , it has been shown that the apodization helps to couple solitons into fiber gratings @xcite , and may also split them @xcite . the retardation of the bg solitons to the above - mentioned low velocity , @xmath1 , was also performed with the help of the apodization @xcite . technologies which make it possible to fabricate regular fiber gratings with periodic apodization , i.e. , an effective _ superlattice _ built on top of the bg , are well established @xcite . analysis of the light propagation in the so developed _ supergratings _ was performed in @xcite . it was shown that the supergrating gives rise to extra gaps in the system s spectrum ( rowland ghost gaps " ) . solitons in the gaps were looked for in an approximation based on supercoupled - mode equations " , which , essentially , assume that a soliton amounts to a slowly varying envelope of the supergrating s bloch function(s ) . as is known , such an approximation ( the averaging method ) generally applies to the description of gss near bandgap edges @xcite . a related problem was considered in @xcite , _ viz_. , gss in the bec loaded into a double - periodic optical lattice . the period doubling opens an additional narrow mini - gap " , where stable gss exist . the objective of our work is to investigate the existence of gss in harmonic superlattices created on top of the usual bg in fibers or waveguides with the intrinsic kerr nonlinearity , extending the analysis initiated in yame:05 . our model has periodic potentials in one spatial dimension . analytical approaches for gss in two or more spatial dimensions were also developed recently in @xcite . the paper is organized as follows . we first give our mathematical model of the supergrating and analyze the bandgap structure in the model , using its dispersion relation , in section [ s : prl ] . this constitutes a non - trivial extension of the classical spectral theory for the mathieu equation smjo:99 , and predicts parameter values at which gss may be expected to exist in the model . the sequence of new bandgaps not present in the unmodulated bg is produced . in sections [ s : mel ] and [ s : av ] we establish the existence of gss in the newly opened bandgaps , using a combination of analytical and numerical techniques . we find various families of gss in the supergrating , including those which do not exist in the ordinary gs model ( in particular , stable asymmetric solitons , and multi - soliton bound states ) . using the melnikov method , we first prove that gss in the original bg model have counterparts in the extended model . a similar treatment was used for a perturbed hamiltonian partial differential equations in @xcite . then , the method of averaging is used to demonstrate the emergence of gss in the new bandgaps . the most important result is that in each new bandgap , there is a single family of fundamental symmetric gss which _ entirely _ fills the gap . such gss near band edges in nonlinear schrdinger equations with periodic potentials were also studied by different approaches in hwakya:11,pesuki:04 . in section [ s : bif ] we discuss bifurcations of gss from a geometric point of view and illustrate this general approach in numerical computations for our problem . the majority of the newly established gss are _ stable _ solutions of the model equations . in particular , the family of fundamental gss in the new bandgaps is stable even for negative frequencies , when the ordinary gss are unstable @xcite . we do not study the stability of gss in this paper in detail , referring to @xcite for the verification of the stability by means of direct simulations . the standard model of the bragg grating ( bg ) in an optical fiber is based on the equations @xcite @xmath2 where @xmath3 and @xmath4 are amplitudes of the right- and left - propagating waves in the fiber . in these equations , @xmath5 is the coordinate along the fiber , and @xmath6 is time . a family of exact soliton solutions of ( [ e : standard ] ) have been found in @xcite . here we will consider a situation where the bg is subjected to periodic modulation in @xmath5 , with a period @xmath7 , which gives rise to a _ supergrating_. the general form of the extended model reads @xmath8 v+\mu \cos \left ( kx+\delta \right ) ~u+\left ( |v|^{2}+\textstyle\frac{1}{2}|u|^{2}\right ) u = 0 , \\ & iv_{t}-iv_{x}+\left [ 1-\varepsilon \cos \left ( kx\right ) \right ] u+\mu \cos \left ( kx+\delta \right ) ~v+\left ( |u|^{2}+\textstyle\frac{1}{2}|v|^{2}\right ) v = 0.\end{split } \label{e : uv}\ ] ] the real parameter @xmath9 accounts for the periodic modulation of the bg strength . in real fiber gratings , it may be typically @xmath10 cm ru:86,brst:97 , while the total length of the grating may be up to @xmath10 m. the modulation is implemented through the change imposed on the local variation of the refractive index in the fiber s cladding and may also affect the local index felt by each wave . this is accounted for by the perturbation parameter @xmath11 and a phase shift @xmath12 . in practice , the harmonic superlattice corresponding to eqs . ( [ e : uv ] ) may be implemented if the optical interference pattern , which is used to burn the grating into the fiber , is created with spatial beatings . we consider stationary gss , which are found via the ansatz @xmath13substituting this into eqs . ( [ e : uv ] ) , we obtain @xmath14 v+\mu \cos \left ( kx+\delta \right ) u+\left ( |v|^{2}+\textstyle\frac{1}{2}|u|^{2}\right ) u = 0 , \\ & \omega v - iv^{\prime } + \left [ 1-\varepsilon \cos ( kx)\right ] u+\mu \cos \left ( kx+\delta \right ) v+\left ( |u|^{2}+\textstyle\frac{1}{2}|v|^{2}\right ) v = 0,\end{split } \label{e : uvans}\]]where the prime stands for @xmath15 . we first analyze the bandgap structure of ( [ e : uv ] ) by looking for solutions of the linearized system in the form of @xmath16 with @xmath17 being the propagation constant . thus , we have to find periodic solutions of the linearization of ( [ e : uvans ] ) @xmath18 v+\mu \cos \left ( kx+\delta \right ) u = 0 , \\ & ( \omega + q)v - iv^{\prime } + \left [ 1-\varepsilon \cos ( kx)\right ] u+\mu \cos \left ( kx+\delta \right ) v = 0.\end{split } \label{e : uvlin}\]]note that in the unperturbed problem , with @xmath19 , the dispersion relation is given by @xmath20 , so that we find the well - known gap in the spectrum @xmath21 . let us consider the case of @xmath22 . for @xmath23 non - trivial solutions of ( [ e : uvlin ] ) emerge due to parametric resonance , similar to the situation in the classical mathieu equation @xcite . consequently , we expect new gaps to open up at points @xmath24 , @xmath25 . we will denote the corresponding gaps by @xmath26 , while the central gap will be denoted by @xmath27 . in the central gap @xmath27 , gss for @xmath28 are found as perturbations of the soliton for @xmath29 , established in acwa:89 . we will use melnikov s method @xcite to prove the existence of gap solitons in section [ s : mel ] . for the non - central gaps , however , we will apply the averaging method to investigate the emergence of ( small ) gss for @xmath28 in section [ s : av ] . straightforward perturbation theory for parametrically excited systems yields approximation to the gaps for small @xmath9 . for example , an approximation to the gap with @xmath30 close to @xmath31 , i.e. close to the unperturbed first gap at @xmath32 , can be obtained by looking for solutions of ( [ e : uvlin ] ) to lowest order approximation @xmath33 using this ansatz we derive at the solvability condition at order @xmath34 , @xmath35 \\ \qquad \qquad \qquad = \varepsilon ^{2}\left [ ( \omega ^{2}-q^{2})^{2}-k^{2}(\omega ^{2}+q^{2}+1)-1\right ] .\end{array } \label{e : solv}\]]observe that @xmath36 , since we are interested in the gap close to the unperturbed one with @xmath32 . moreover , note that at the boundaries of the gaps periodic solutions with wavenumber @xmath37 or @xmath38 exist @xcite . hence , being interested in solutions with @xmath39 we introduce @xmath40 and expand ( [ e : solv ] ) for small @xmath41 to obtain @xmath42 a new gap appears if this quadratic equation for @xmath43 has no real solutions . thus we obtain as the condition for the gap , that @xmath44 in terms of the original parameter @xmath30 we find that the gaps open up at @xmath45 and their widths are @xmath46 similar expressions can be obtained for the higher - order gaps @xmath47 for @xmath48 showing that the gap width of the band corresponding to @xmath49 scales with @xmath50 . we can treat the case of @xmath51 similarly . these perturbative results can be extended numerically by detecting parameter values , for which eqs . ( [ e : uvlin ] ) possesses periodic solutions with wavenumber @xmath37 or @xmath38 . it is well known that these parameters form the boundary of the gaps in the spectrum ( compare with above ) . following @xcite we have computed these parameter values as solutions of the corresponding eigenvalue problem in matlab . the results for several combinations of parameter values @xmath37 and @xmath52 are shown in fig . f : linear . in this figure the shaded and white areas stand for bands and gaps , respectively . it is found that varying @xmath37 merely changes the width of bands and gaps with respect to @xmath53 , whereas changing @xmath52 leads to more complex effects . in particular , for @xmath54 gaps opening at @xmath29 can vanish again when @xmath55 or @xmath52 are increased . in this and next sections we investigate the existence of gss in eq . ( e : uv ) , using both analytical and numerical techniques . these solitons are described by homoclinic solutions to the origin @xmath56 in eqs . ( e : uvans ) . before we present the computations let us take a closer look at properties of eqs . ( [ e : uvans ] ) . it will be important that these equations allow for a symmetry reduction by setting @xmath57 , where @xmath58 denotes complex conjugation . in the following , we will thus only consider the reduced equation @xmath59 let @xmath60 , where @xmath61 . ( [ e : u ] ) is rewritten as @xmath62b -\frac{3}{2}(a^{2}+b^{2})b , \\ b^{\prime}=&[\omega-1+\varepsilon\cos kx+\mu\cos(kx+\delta)]a + \frac{3}{2}(a^{2}+b^{2})a . \end{split } \label{e : ab}\ ] ] we will consider as a dynamical system in the three - dimensional phase space @xmath63 , with @xmath64 as the circle of length @xmath65 . note that if @xmath66 , then eqs . ( [ e : ab ] ) are _ reversible _ , that is , invariant under compositions of time reversal and the ( linear ) involutions , @xmath67 we refer to @xcite for a general review of reversible systems . of particular importance for us will be _ symmetric _ homoclinic orbits , which are mapped to itself under the action of @xmath68 or @xmath69 . it is well known that an orbit is symmetric if and only if it intersects the invariant plane of the involution . now we assume that @xmath70 and @xmath71 , and set @xmath72 for some @xmath73 . the cases of @xmath74 and @xmath75 will be treated in section [ s : av ] and a : bound , respectively . for @xmath29 eqs . ( [ e : ab ] ) become a planar hamiltonian system , @xmath76 with a hamiltonian @xmath77+\frac{3}{8}(a^2+b^2)^2,\ ] ] where @xmath78 and @xmath79 represent the canonical momentum and coordinates , respectively . the equilibrium at the origin is a hyperbolic saddle in ( e : ab0 ) and has a pair of homoclinic orbits @xmath80 which are symmetric under the involution @xmath81 . when @xmath55 and/or @xmath52 are nonzero but sufficiently small , the origin is still a hyperbolic saddle in and has two - dimensional stable and unstable manifolds which may intersect transversely . such intersection yields transverse homoclinic orbits to the saddle at the origin , which persist under variation of the parameters @xcite . along certain curves in the parameter space , however , the intersection of the manifolds may become tangential , resulting in bifurcations of the homoclinic orbits . see section [ s : bif ] for details . here we aim to prove the existence of such transverse homoclinic orbits in eqs . ( [ e : ab ] ) by means of the melnikov method @xcite . from formula ( 4.5.6 ) in @xcite , we derive the melnikov functions @xmath82 for @xmath83 as @xmath84\,a_{\pm } ( x)\,b_{\pm } ( x)\cos k(x+x_{0})dx \notag \\ & + 2\mu \int_{-\infty } ^{\infty } a_{\pm } ( x)\,b_{\pm } ( x)\cos [ k(x+x_{0})+\delta ] dx \notag \\ = & -\varepsilon \left ( \int_{-\infty } ^{\infty } [ 2\omega + 3(a_{\pm } ^{2}(x)+b_{\pm } ^{2}(x))]\,a_{\pm } ( x)\,b_{\pm } ( x)\sin kx\,dx\right ) \sin kx_{0 } \notag \\ & -2\mu \left ( \int_{-\infty } ^{\infty } a_{\pm } ( x)\,b_{\pm } ( x)\sin kx\,dx\right ) \sin ( kx_{0}+\delta ) , \label{e : mel}\end{aligned}\]]where we used the fact that @xmath85 and @xmath86 are even and odd functions of @xmath5 , respectively . substituting ( [ e : homo0 ] ) into ( e : mel ) and using the method of residues , we compute @xmath87^{2}}\sin kx\,dx \notag \\ & = \frac{\pi k}{3}\,\mathrm{cosech}\!\left ( \frac{k\pi } { 2\sin \theta } \right ) \sinh \!\left ( \frac{k\theta } { 2\sin \theta } \right ) \label{e : int1}\end{aligned}\]]and @xmath88a_{\pm } ( x)\,b_{\pm } ( x)\sin kx\,dx \notag \\ & = \frac{8}{9}\sin ^{5}\theta \int_{-\infty } ^{\infty } \frac{\sinh ( 2x\sin \theta ) } { [ \cosh ( 2x\sin \theta ) + \cos \theta ] ^{3}}\sin kx\,dx \notag \\ & = \frac{\pi k}{9}\,\mathrm{cosech}\!\left ( \frac{k\pi } { 2\sin \theta } \right ) \left [ -2\sinh \!\left ( \frac{k\theta } { 2\sin \theta } \right ) \cos \theta + k\cosh \!\left ( \frac{k\theta } { 2\sin \theta } \right ) \right ] . \label{e : int2}\end{aligned}\]]see [ a : cal ] for derivations of ( [ e : int1 ] ) and ( [ e : int2 ] ) . hence , the melnikov functions become @xmath89where we used the relation @xmath90 . let us assume that @xmath91 then we easily see that @xmath92 has simple zeros at @xmath93 where @xmath94 this implies that there exist transverse homoclinic orbits near @xmath95 and @xmath96 . in particular , when @xmath66 , we have @xmath97 so that the first and second orbit are symmetric under @xmath81 and @xmath98 , respectively . thus , we can prove that there exist gss with @xmath99 in ( [ e : uv ] ) . for @xmath100 , @xmath101 and @xmath102 : ( a ) @xmath103 ; ( b ) @xmath104 . , title="fig : " ] for @xmath100 , @xmath101 and @xmath102 : ( a ) @xmath103 ; ( b ) @xmath104 . , title="fig : " ] of ( [ e : uv ] ) in the central gap with @xmath102 , @xmath105 and @xmath101 . the solid curve shows the real part and the dashed curve the imaginary part of the solution . , title="fig : " ] ( a ) @xmath106 of ( [ e : uv ] ) in the central gap with @xmath102 , @xmath105 and @xmath101 . the solid curve shows the real part and the dashed curve the imaginary part of the solution . , title="fig : " ] ( b ) @xmath107 of ( [ e : uv ] ) in the central gap with @xmath105 , @xmath102 , @xmath108 and @xmath109 . , title="fig : " ] + ( a ) @xmath106 of ( [ e : uv ] ) in the central gap with @xmath105 , @xmath102 , @xmath108 and @xmath109 . , title="fig : " ] + ( b ) @xmath107 to illustrate the above analysis we now introduce the poincar section @xmath110 in phase space , and consider the return map induced by the flow of . we have computed the stable and unstable manifolds of the origin using the continuation tool ` auto ` @xcite with assistance of the ` hommap ` driver @xcite . in particular , to compute the stable and unstable manifolds , small segments ( @xmath111 ) in the stable and unstable subspaces of the origin were taken and solutions of ( [ e : ab ] ) with initial conditions on the segments were numerically integrated . the same approach was also used for computation of the one - dimensional unstable manifolds for a three - dimensional poincar map in @xcite . as detected in the theory , the stable and unstable manifolds are observed to intersect transversely and gss could be found . the stable and unstable manifolds on the poincar section @xmath112 for @xmath113 , @xmath114 , @xmath115 and @xmath102 are drawn in fig . [ f : im0 ] . intersections of these manifolds give rise to homoclinic orbits or gss , and symmetric gss in ( [ e : uv ] ) for the same parameter values are plotted in fig . [ f : big0 ] . these gss are not visibly different from the unperturbed ones of . in fig . [ f : big0 m ] asymmetric gss in are also plotted for @xmath116 , @xmath117 , @xmath114 , @xmath115 and @xmath102 . note that there are different asymmetric gss which have almost the same shapes as those in fig . [ f : big0 m ] but have peaks at a position shifting by @xmath118 . plots of gss throughout this paper show the real part of the solutions as a solid curve , whereas the imaginary part is plotted as a dashed curve . in order to investigate the existence of gss of ( [ e : uv ] ) with @xmath99 in the non - central gaps , we shall use the higher - order averaging method mu:88,mu:91 . let us describe a general setup for the averaging analysis . we seek homoclinic solutions to the origin of ( [ e : ab ] ) near @xmath119 in the limit of small @xmath120 . note that for @xmath121 the origin is an equilibrium in ( [ e : ab ] ) , at which the jacobian matrix has a pair of purely imaginary eigenvalues @xmath122 when @xmath123 . let @xmath124 and @xmath125 be non - negative integers which will be determined for each gap later . define a detuning parameter @xmath126 by @xmath127 let @xmath128 \displaystyle\frac{mk}{2(\omega _ { \pm m}+1)}\sin \frac{mk}{2}x & \displaystyle\cos \frac{mk}{2}x\end{pmatrix},\ ] ] which is the fundamental matrix of the linearization at the origin of the unperturbed system ( [ e : ab0 ] ) with @xmath129 . using the transformation @xmath130 \displaystyle\frac{m}{2(\omega _ { \pm m}+1)}\eta \end{pmatrix } \label{e : trans}\]]in ( [ e : ab ] ) , one obtains @xmath131b \\[1ex ] \lbrack \varepsilon ^{j}\omega + \cos kx+\bar{\mu}\cos ( kx+\delta ) + \frac{3}{2}\varepsilon ^{l}(a^{2}+b^{2})]a\end{pmatrix } , \label{e : xieta}\end{aligned}\]]where @xmath132 and @xmath78 and @xmath79 are represented by @xmath133 and @xmath134 via ( [ e : trans ] ) . homoclinic solutions in ( [ e : xieta ] ) , equivalently in ( [ e : ab ] ) , can be approximately obtained by the averaging method . these homoclinic solutions correspond to gps of ( [ e : uv ] ) with @xmath99 in the non - central gaps . while in the first gaps @xmath135 the standard ( i.e. , first - order ) averaging method is sufficient , we have to use the higher - order averaging method for the higher - order gaps . this difficulty occurs because the potential considered here consists of the pure harmonic component @xmath136 . but it is important to note that higher - order averaging is also needed for more general potentials with higher fourier components , if these higher components are small compared to the first one . this is the reason for us to present the higher - order method here . the necessary lengthy computations can be easily performed with computer algebra systems . specifically , we have used the program ` haverage.m ` for the computer software ` mathematica ` @xcite , developed in @xcite ( see also @xcite ) . we exemplarily present the results for the first three gaps @xmath135 , @xmath137 and @xmath138 . similar second - order averaging analyses can also calculate the boundaries of the central gap ( see [ a : bound ] ) . for the computations in the first gap at @xmath139 , we set @xmath140 and @xmath141 . the standard averaging procedure for ( [ e : xieta ] ) yields the first - order averaged system @xmath142 , \\ & \eta^{\prime}=\varepsilon\left [ \frac{\omega_{\pm 1}(1 + 2\omega)+\bar{\mu}\cos\delta}{k}\xi + \frac{\bar{\mu}\sin\delta}{k}\eta + \gamma_{\pm 1}(\xi^2+\eta^2)\xi\right ] , \end{split } \label{e : av1}\ ] ] where @xmath143 eqs . are easily obtained by averaging the right hand side of with respect to @xmath5 over @xmath144 $ ] and such computations can be done by the ` mathematica ` program ` haverage.m ` . in the averaged system ( [ e : av1 ] ) with @xmath145 , we perform a rotational transformation @xmath146 to obtain a system of a more convenient form @xmath147 , \\ & \tilde{\eta}^{\prime}=\varepsilon\left [ \frac{2\omega_{\pm 1}}{k}(\omega_{1\pm}+\omega)\tilde{\xi } + \gamma_{\pm 1}(\tilde{\xi}^2+\tilde{\eta}^2)\tilde{\xi}\right ] , \end{split } \label{e : av1a}\ ] ] where @xmath148 and @xmath149 with @xmath150.\ ] ] note that eq . ( [ e : av1a ] ) is still valid for @xmath151 if @xmath152 is taken when @xmath153 and if @xmath154 is taken when @xmath155 . the system ( [ e : av1a ] ) is hamiltonian with a hamilton function @xmath156 + \frac{1}{4}\gamma_{\pm 1}(\tilde{\xi}^2+\tilde{\eta}^2)^2\right\},\end{aligned}\ ] ] for which the origin @xmath157 is a saddle if @xmath158 , i.e. , @xmath159 we easily see that if @xmath160 then @xmath161 ( resp . @xmath162 ) and a closing of the gap region @xmath163 ( resp . @xmath164 ) occurs . see also fig . f : linear(d ) . if @xmath165 , then @xmath166 , and thus such a closing does not occur . let us assume that @xmath166 , i.e. , condition ( [ e : closing ] ) does not hold . a straightforward analysis of the level sets of @xmath167 shows the existence of two homoclinic orbits in ( [ e : av1a ] ) and hence in ( e : av1 ) . moreover , we can obtain analytical expressions of the homoclinic orbits as @xmath168 } { \cosh[(4\omega_{\pm 1}\omega_{1\pm}\sin\phi / k)\varepsilon x]+\cos\phi } , \right . \notag \\ & \qquad \left . \mp 2\sqrt{\frac{2\omega_{\pm 1}\omega_{1\pm}}{3k\gamma_{\pm 1}}}\sin\phi \cos\frac{1}{2}\phi\ , \frac{\cosh[(2\omega_{\pm 1}\omega_{1\pm } \sin\phi / k)\varepsilon x ] } { \cosh[(4\omega_{\pm 1}\omega_{1\pm } \sin\phi / k)\varepsilon x]+\cos\phi } \right ) , \label{e : homo}\end{aligned}\ ] ] where @xmath169 . it is easy to see that , since the coefficient in front of @xmath170 in @xmath167 is negative , the homoclinic orbits intersect the @xmath171-axis and are symmetric about it , forming a figure - of - eight configuration . in general , it is very difficult to make precise statements about the symmetry of homoclinic orbits in ( [ e : av1 ] ) . for some special cases , however , we have the following results : 1 . when @xmath172 and @xmath173 , the homoclinic orbits in the averaged system ( [ e : av1 ] ) are symmetric about the @xmath174-axis if @xmath175 and about the @xmath176-axis if @xmath177 in the negative gap , while they are always symmetric about the @xmath174-axis in the positive gap . 2 . when @xmath178 and @xmath179 , the homoclinic orbits in ( [ e : av1 ] ) are symmetric about the @xmath174-axis if @xmath180 and about the @xmath176-axis if @xmath181 in the positive gap , while they are always symmetric about the @xmath174-axis in the negative gap . 3 . when @xmath22 , the homoclinic orbits in ( [ e : av1 ] ) are symmetric about the @xmath174-axis in both the positive and negative gaps . from the existence of homoclinic orbits in the averaged system , we claim that there exist homoclinic orbits in and hence stationary gss in . for this consider ( [ e : xieta ] ) and ( e : av1 ) as dynamical systems defined in the phase space @xmath63 . recall that we set @xmath140 and @xmath141 in ( [ e : xieta ] ) . we introduce a cross section @xmath182 at @xmath183 and define the poincar maps @xmath184 for ( [ e : xieta ] ) and ( [ e : av1 ] ) , respectively . for both @xmath185 and @xmath186 the origin @xmath187 is a fixed point and has stable and unstable manifolds . moreover , the stable and unstable manifolds , @xmath188 and @xmath189 , coincide along the homoclinic orbits in ( [ e : av1 ] ) for @xmath186 . by the averaging theorem ( see theorem 4.1.1 of @xcite ) , the stable and unstable manifolds , @xmath190 and @xmath191 , for @xmath185 are @xmath192-close to @xmath188 and @xmath193 . since eq . ( e : xieta ) as well as eq . ( [ e : av1 ] ) are hamiltonian , the poincar maps @xmath194 are area - preserving . ( solid lines ) which are close to @xmath195 ( broken lines ) but do not intersect . the regions @xmath196 and @xmath197 with @xmath198 are hatched distinctly . ] now assume that the stable and unstable manifolds @xmath199 and @xmath200 do not intersect , and go inward and outward , respectively , as shown in fig . [ f : asu ] . we want to show that this yields a contradiction and define a region @xmath196 , which is encircled by two parts of @xmath199 and @xmath200 and a line connecting two points @xmath201 and @xmath202 on @xmath199 and @xmath200 . the region @xmath196 is mapped by the poincar map @xmath203 to another region @xmath204 which is encircled by two different parts of @xmath205 and @xmath200 and a curve connecting two points @xmath206 and @xmath207 on @xmath199 and @xmath200 . it is obvious that @xmath208 is larger than @xmath196 and contradicts the fact that @xmath203 is area - preserving . similarly , we can not assume that @xmath199 and @xmath200 go outward and inward , respectively . therefore , we can prove that @xmath199 and @xmath200 intersect and there exists a homoclinic orbit in the full system ( e : xieta ) and hence in ( [ e : u ] ) . on the other hand , eq . can be averaged up to @xmath209 for any integer @xmath210 but has very rapid oscillations compared with the unperturbed homoclinic orbits which vary very slowly with @xmath211 . applying a result by neishtadt @xcite ( see also @xcite ) , we see that the splitting distance between @xmath199 and @xmath200 is exponentially small with respect to @xmath212 at most . thus , there exist stationary gs solutions to the system of equations ( [ e : uv ] ) in the gap regions . moreover , assume that @xmath66 , i.e. , @xmath22 or @xmath213 , so that eqs . ( [ e : ab ] ) is reversible with respect to @xmath214 and @xmath215 , @xmath216 . then it follows from the above arguments that the homoclinic orbit in the averaged system ( [ e : av1 ] ) is symmetric about the @xmath176-axis ( resp . @xmath174-axis ) . by the persistence of symmetric orbits in reversible systems , there exist symmetric homoclinic orbits about the @xmath176-axis ( resp . @xmath174-axis ) in the full system ( [ e : xieta ] ) and with respect to @xmath81 and @xmath217 ( resp . @xmath98 and @xmath218 ) in ( [ e : ab ] ) and hence in ( [ e : u ] ) . in particular , the homoclinic orbits in ( [ e : u ] ) have an even ( resp . odd ) real part @xmath78 and odd ( resp . even ) imaginary part @xmath219 . we can use similar arguments to the central gap when condition ( [ e : con ] ) does not hold , and prove that there must be homoclinic orbits . however , the splitting distance between the stable and unstable manifolds is not exponentially small with respect to @xmath55 but @xmath220 at most since both of the perturbations in and unperturbed homoclinic orbits vary with @xmath221 . for @xmath179 . ] for the associated poincar map : ( a ) @xmath222 ; ( b ) @xmath223 ; ( c ) @xmath224.,title="fig : " ] for the associated poincar map : ( a ) @xmath222 ; ( b ) @xmath223 ; ( c ) @xmath224.,title="fig : " ] + for the associated poincar map : ( a ) @xmath222 ; ( b ) @xmath223 ; ( c ) @xmath224.,title="fig : " ] in addition , we can detect a non - transverse homoclinic orbit in ( [ e : ab ] ) when @xmath179 and @xmath225 as follows . fix @xmath179 . then the melnikov functions @xmath92 have graphs as shown in fig . [ f : mel ] , depending on @xmath226 , @xmath227 or @xmath228 . since @xmath229 represent signed measures of their distance ( see , e.g. , section 4.5 of @xcite ) , noting that @xmath230 and eqs . ( [ e : ab ] ) is reversible , we can draw the behavior of the stable and unstable manifolds @xmath231 for the poincar map of ( [ e : ab ] ) as shown in figs . [ f : su](a ) and ( b ) . using the fact that the poincar map is area - preserving and applying a discussion similar to the above one for ( e : xieta ) , we see that @xmath190 and @xmath191 coincide for some @xmath55 near @xmath232 , say @xmath233 , as shown in fig . f : su(c ) . thus , the point @xmath234 is very degenerate in the parameter space . this also suggests that new interesting behavior may occur at the point in the original pdes ( [ e : uv ] ) . when @xmath108 , @xmath104 , @xmath235 and @xmath102 : ( a ) @xmath236 ; ( b ) @xmath237 ; ( c ) @xmath238 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] when @xmath108 , @xmath104 , @xmath235 and @xmath102 : ( a ) @xmath236 ; ( b ) @xmath237 ; ( c ) @xmath238 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] + when @xmath108 , @xmath104 , @xmath235 and @xmath102 : ( a ) @xmath236 ; ( b ) @xmath237 ; ( c ) @xmath238 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] . the parameter values @xmath239 of fig . [ f : im0](b ) with @xmath102 is plotted as @xmath240 . ] in fig . [ f : im0 ] we show the results of numerical computations of the stable and unstable manifolds by ` auto97 ` with ` hommap ` near the homoclinic bifurcation point @xmath241 for @xmath242 , @xmath107 and @xmath102 . we see that as predicted by the theory , the invariant manifolds almost coincide at @xmath243 while they split when @xmath55 is different from @xmath232 . thus , the theoretical prediction for the homoclinic bifurcations is very precise . figure [ f : hbc ] also shows the approximate homoclinic bifurcation curves given by eq . , on which the stable and unstable manifolds coincide , in the @xmath244-space for @xmath245 . we set @xmath246 and @xmath247 , and perform a second - order averaging method to prove the existence of gap solitons of ( [ e : uvans ] ) with @xmath99 in the second gaps appearing at @xmath248 . to avoid the complexity of expressions , we give our results only for @xmath22 in the gaps of higher orders . the second - order averaged system has been obtained using the ` mathematica ` program ` haverage.m ` , and it reads @xmath249 \right . \\ & \qquad + \beta_{\pm 2}\left[k^{2}\xi^{2}+ ( \omega_{\pm 2}+1)^{2 } \eta^{2}\right]\big\}\ ; \eta , \\ \eta^{\prime}= & \displaystyle{\frac{\varepsilon^{2}}{(\omega_{\pm 2}+1)^{2 } } } \left\{\alpha_{\pm 2 } \left[(6(\omega_{\pm 2}+1)\omega+1)k^{2}- 5(\omega_{\pm 2}+1)^{2}\right ] \right . \\ & \qquad + \beta_{\pm 2}\left[k^{2}\xi^{2 } + ( \omega_{\pm 2}+1)^{2}\eta^{2}\right]\big\ } \ ; \xi , \end{split } \label{e : av2}\ ] ] with @xmath250 } { 16k^{2}(\omega_{\pm 2}+1)^{2}}.\end{aligned}\ ] ] note that @xmath251 and @xmath252 . ( e : av2 ) is a hamiltonian system with the hamilton function @xmath253k^{2}-5(\omega_{\pm 1/1}+1)^{2}\ } } { 2(\omega_{\pm 1/1}+1)^{2}}\xi^{2 } \notag \\ & \quad + \frac{\alpha_{\pm 1/1 } \{[6(\omega_{\pm 1/1}+1)\omega-5]k^{2}+(\omega_{\pm 1/1}+1)^{2}\ } } { 2k^{2}}\eta^{2 } \notag \\ & \quad + \frac{\beta_{\pm 1/1}}{4k^{2}(\omega_{\pm 2/1}+1)^{2 } } [ k^{2}\xi^{2}+(\omega_{\pm 1/1}+1)^{2}\eta^{2}]^{2}\biggr).\end{aligned}\ ] ] let @xmath254 the origin is a saddle of ( [ e : av2 ] ) and has a pair of homoclinic orbits if @xmath255 and @xmath256 when the signs ` @xmath257 ' and ` @xmath258 ' are taken in the subscript of @xmath53 in ( [ e : av2 ] ) , respectively , i.e. , @xmath259 or @xmath260 again , we find the averaged system to be hamiltonian with a saddle equilibrium at the origin , and a study of the corresponding level set of the hamiltonian shows the existence of a figure - of - eight of homoclinic orbits . these orbits intersect the @xmath174-axis . by the same argument as above , this implies the existence of gap solitons for ( [ e : u ] ) in the gaps @xmath137 having an even @xmath78-component and an odd @xmath79-component . we set @xmath261 , @xmath262 and @xmath263 , and perform the third - order averaging procedure for ( [ e : xieta ] ) using the ` mathematica ` program ` haverage.m ` to obtain the averaged system @xmath264k^{2 } -4(\omega_{\pm 3}+1)^{2}\right\}\eta \\ & + \frac{\varepsilon^{3}}{9k^{2 } } \left\{\alpha_{\pm 3}-\beta_{\pm 3}\left[9k^{2}\xi^{2 } + 4(\omega_{\pm 3}+1)^{2}\eta^{2}\right]\right\}\eta , \\ \eta^{\prime}= & \frac{\varepsilon^{2}}{4(\omega_{\pm 3}+1)^{2}}\rho_{\pm 3 } \left\{[32(\omega_{\pm 3}+1)\omega-9]k^{2 } -4(\omega_{\pm 3}+1)^{2}\right\}\xi \\ & + \frac{\varepsilon^{3}}{4(\omega_{\pm 3}+1)^{2 } } \left\{\alpha_{\pm 3}+\beta_{\pm 3}\left[9k^{2}\xi^{2 } + 4(\omega_{\pm 3}+1)^{2}\eta^{2}\right]\right\}\xi , \end{split } \label{e : av3o}\ ] ] where @xmath265 [ 81k^{4}-56(\omega_{\pm3}+1)^{2}k^{2}+16(\omega_{\pm3}+1)^{4 } ] } { 512k^{4}(\omega_{\pm 3}+1)^{2 } } , \\ \beta_{\pm 3 } = & \frac{81k^{4}+24(\omega_{\pm 3}+1)^{2}k^{2}+16(\omega_{\pm 3}+1)^{4 } } { 64k^{2}(\omega_{\pm 3}+1)^{2}}.\end{aligned}\ ] ] note that @xmath266 and @xmath267 since @xmath268^{2}+18(\omega_{\pm3}+1)^{2}>0.\end{aligned}\ ] ] let @xmath269 where @xmath270 note that @xmath271 and @xmath272 . ( [ e : av3o ] ) becomes @xmath273\}\eta , \\ \eta^{\prime}= & \frac{\varepsilon^{3}}{4(\omega_{\pm 3}+1)^{2}}\ { ( \gamma_{\pm 3}\nu+\alpha_{\pm 3 } ) + \beta_{\pm 3}[9k^{2}\xi^{2}+4(\omega_{\pm 3}+1)^{2}\eta^{2}]\}\xi , \end{split } \label{e : av3}\ ] ] where @xmath274>0.\ ] ] eq . ( [ e : av3 ] ) is a hamiltonian system with the hamilton function @xmath275^{2}\biggr ) . \label{e : h3}\end{aligned}\ ] ] we easily see that @xmath276 and set @xmath277 the origin is a saddle and has a pair of homoclinic orbits if @xmath278 , i.e. , @xmath279 especially , in the regions ( [ e : region3 ] ) the coefficients of @xmath280 and @xmath281 in the hamiltonian ( [ e : h3 ] ) are positive and negative , respectively , so that the pair of homoclinic orbits draws a vertical figure - eight in the @xmath282-phase plane . similar computations reveal the existence of gap solitons in the higher order gaps . these solutions are odd in @xmath78 ( and even in @xmath79 ) or vice versa , depending on whether the gap number is odd or even . a general proof of this is beyond the scope of this paper . in the following we illustrate and extend this perturbative results by numerical computations . when @xmath101 and @xmath102 : ( a ) @xmath283 , @xmath284 ; ( b ) @xmath285 , @xmath284 ; ( c ) @xmath286 , @xmath287 ; ( d ) @xmath288 , @xmath287 ; ( e ) @xmath289 , @xmath290 ; ( f ) @xmath291 , @xmath290 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] when @xmath101 and @xmath102 : ( a ) @xmath283 , @xmath284 ; ( b ) @xmath285 , @xmath284 ; ( c ) @xmath286 , @xmath287 ; ( d ) @xmath288 , @xmath287 ; ( e ) @xmath289 , @xmath290 ; ( f ) @xmath291 , @xmath290 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] + when @xmath101 and @xmath102 : ( a ) @xmath283 , @xmath284 ; ( b ) @xmath285 , @xmath284 ; ( c ) @xmath286 , @xmath287 ; ( d ) @xmath288 , @xmath287 ; ( e ) @xmath289 , @xmath290 ; ( f ) @xmath291 , @xmath290 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] when @xmath101 and @xmath102 : ( a ) @xmath283 , @xmath284 ; ( b ) @xmath285 , @xmath284 ; ( c ) @xmath286 , @xmath287 ; ( d ) @xmath288 , @xmath287 ; ( e ) @xmath289 , @xmath290 ; ( f ) @xmath291 , @xmath290 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] + when @xmath101 and @xmath102 : ( a ) @xmath283 , @xmath284 ; ( b ) @xmath285 , @xmath284 ; ( c ) @xmath286 , @xmath287 ; ( d ) @xmath288 , @xmath287 ; ( e ) @xmath289 , @xmath290 ; ( f ) @xmath291 , @xmath290 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] when @xmath101 and @xmath102 : ( a ) @xmath283 , @xmath284 ; ( b ) @xmath285 , @xmath284 ; ( c ) @xmath286 , @xmath287 ; ( d ) @xmath288 , @xmath287 ; ( e ) @xmath289 , @xmath290 ; ( f ) @xmath291 , @xmath290 . the red and blue curves represent the stable and unstable manifolds , respectively.,title="fig : " ] + ( a ) @xmath292 , @xmath293 , @xmath22 of ( [ e : uvans ] ) in the gaps @xmath135 for @xmath102 , symmetric with respect to @xmath217 or @xmath218 . gray curves in the panels correspond to solutions of the averaged system ( [ e : av1]).,title="fig : " ] + ( b ) @xmath294 , @xmath293 , @xmath22 of ( [ e : uvans ] ) in the gaps @xmath135 for @xmath102 . gray curves in the panels correspond to solutions of the averaged system ( [ e : av1 ] ) . , title="fig : " ] + ( a ) @xmath292 , @xmath293 , @xmath295 , @xmath296 of ( [ e : uvans ] ) in the gaps @xmath135 for @xmath102 . gray curves in the panels correspond to solutions of the averaged system ( [ e : av1 ] ) . , title="fig : " ] + ( b ) @xmath294 , @xmath293 , @xmath295 , @xmath296 of ( [ e : uvans ] ) in the gaps @xmath135 for @xmath297 , @xmath298 , @xmath299 and @xmath300 . , title="fig : " ] + ( a ) @xmath301 of ( [ e : uvans ] ) in the gaps @xmath135 for @xmath297 , @xmath298 , @xmath299 and @xmath300 . , title="fig : " ] + ( b ) @xmath302 we computed the stable and unstable manifolds of the origin and homoclinic orbits in ( [ e : ab ] ) , i.e. , gss with @xmath99 in ( [ e : uv ] ) , using the computer package , ` auto ` with ` hommap ` . in our computations we found gss due to the transverse intersection of the stable and unstable manifolds , thus confirming the analytical results . for several parameter values , the stable and unstable manifolds for the poincar section @xmath303 are shown in fig . [ f : im123 ] and gss in ( [ e : uv ] ) are plotted in figs . [ f : big123]-[f : big1 ] . in addition to the solutions of ( [ e : ab ] ) , the figures include homoclinic solutions of the averaged systems ( [ e : av1 ] ) , ( [ e : av2 ] ) and ( [ e : av3 ] ) , rescaled according to the transformation ( [ e : trans ] ) . as can be seen , they provide a good match to the envelope of the solutions to ( [ e : ab ] ) . in all figures presented , the parameter @xmath37 has been set to @xmath102 . in closing this section , we return to the parameter values of fig . f : linear(d ) . in this case a new bandgap opens up for small @xmath55 , but closes at @xmath304 . see also formula . our computations reveal that gss exist in the gaps both above and below this parameter value . in fig . [ f : big1a ] we show solution profiles at @xmath305 and @xmath306 . we further discuss the behavior of gss under variation of the system parameters . when the parameter values are varied , homoclinic tangencies , i.e. , homoclinic bifurcations , can occur , leading to the creation of additional gss and bifurcations of gss . we have found such bifurcations in all gap regions including the central one . ) in the gaps @xmath164 and @xmath27 for @xmath102 and @xmath109 : ( a ) and ( b ) @xmath283 , @xmath101 ; ( c ) @xmath283 , @xmath307 ; ( d ) @xmath104 , @xmath101 . in plates ( b ) and ( c ) , the point `` @xmath240 '' represents pitchfork and saddle - node bifurcations , respectively , title="fig:",scaledwidth=46.0% ] ) in the gaps @xmath164 and @xmath27 for @xmath102 and @xmath109 : ( a ) and ( b ) @xmath283 , @xmath101 ; ( c ) @xmath283 , @xmath307 ; ( d ) @xmath104 , @xmath101 . in plates ( b ) and ( c ) , the point `` @xmath240 '' represents pitchfork and saddle - node bifurcations , respectively , title="fig:",scaledwidth=46.0% ] + ) in the gaps @xmath164 and @xmath27 for @xmath102 and @xmath109 : ( a ) and ( b ) @xmath283 , @xmath101 ; ( c ) @xmath283 , @xmath307 ; ( d ) @xmath104 , @xmath101 . in plates ( b ) and ( c ) , the point `` @xmath240 '' represents pitchfork and saddle - node bifurcations , respectively , title="fig:",scaledwidth=46.0% ] ) in the gaps @xmath164 and @xmath27 for @xmath102 and @xmath109 : ( a ) and ( b ) @xmath283 , @xmath101 ; ( c ) @xmath283 , @xmath307 ; ( d ) @xmath104 , @xmath101 . in plates ( b ) and ( c ) , the point `` @xmath240 '' represents pitchfork and saddle - node bifurcations , respectively , title="fig:",scaledwidth=46.0% ] + as an example , we plot bifurcation diagrams of gss for eqs . ( [ e : uvans ] ) in the gaps @xmath164 and @xmath27 when @xmath292 or 0.5 and @xmath296 , in fig . [ f : bd ] . here @xmath308 represents the point at which each gs crosses the section @xmath309 . in the diagrams we observe saddle - node and pitchfork bifurcations of gss the occurrence of these bifurcations is explained from the behavior of the poincar map of ( [ e : ab ] ) for the section @xmath182 , as follows . + let us first assume that @xmath54 , such that the poincar map is not reversible . in this case , a tangency between the stable and unstable manifolds as shown in fig . [ f : hbmap](a ) leads to a saddle - node bifurcation , in which two new gss are created . see fig . [ f : bd](c ) for a numerical evidence . now suppose that @xmath22 and let us choose a cross - section , such that the poincar map is reversible with respect to @xmath81 and @xmath98 . we can distinguish two different bifurcation scenarios . if the stable and unstable manifolds have a cubic tangency on the invariant plane of @xmath81 or @xmath98 , but intersect this plane transversely as shown in fig . [ f : hbmap](b ) , then a pitchfork bifurcation of homoclinic orbits takes place . when the parameters are varied , a pair of homoclinic orbits appear , while the original homoclinic orbit still exists due to the persistence of symmetric orbits . see figs . [ f : bd](a ) and ( b ) . the original gs is symmetric , but the two new orbits are asymmetric . on the other hand , if the stable and unstable manifolds are tangent to the invariant plane at their quadratic tangency as shown in fig . [ f : hbmap](c ) , then a saddle - node bifurcation of symmetric gss occurs . see fig . f : bd(b ) . note that a saddle - node bifurcation of asymmetric gss can occur even though @xmath22 . when @xmath102 and @xmath109 : ( a ) @xmath283 , @xmath307 , @xmath310 ; ( b ) @xmath283 , @xmath311 , @xmath101 ; ( c ) @xmath104 , @xmath312 , @xmath101 . the red and blue curves represent the stable and unstable manifolds , respectively . in plate ( a ) the invariant manifolds have quadratic tangencies inside the circles.,title="fig : " ] when @xmath102 and @xmath109 : ( a ) @xmath283 , @xmath307 , @xmath310 ; ( b ) @xmath283 , @xmath311 , @xmath101 ; ( c ) @xmath104 , @xmath312 , @xmath101 . the red and blue curves represent the stable and unstable manifolds , respectively . in plate ( a ) the invariant manifolds have quadratic tangencies inside the circles.,title="fig : " ] + when @xmath102 and @xmath109 : ( a ) @xmath283 , @xmath307 , @xmath310 ; ( b ) @xmath283 , @xmath311 , @xmath101 ; ( c ) @xmath104 , @xmath312 , @xmath101 . the red and blue curves represent the stable and unstable manifolds , respectively . in plate ( a ) the invariant manifolds have quadratic tangencies inside the circles.,title="fig : " ] we illustrate these general ideas with computations for equation . figure [ f : im ] shows the stable and unstable manifolds of the poincar map of on the section @xmath182 for the parameter values when saddle - node or pitchfork bifurcations occur in figs . [ f : bd ] . figure [ f : im](a ) corresponds to fig . [ f : hbmap](a ) and depicts the situation when the map is not reversible ( @xmath178 ) and the stable and unstable manifolds have a quadratic tangency . similarly , in figs . [ f : im](b ) and ( c ) the poincar map is reversible , and the manifolds have cubic and quadratic tangencies on the invariant plane @xmath313 as in the sketches in figs . [ f : hbmap](b ) and ( c ) . of ( [ e : uvans ] ) born at the bifurcations of fig . [ f : bd](a)-(c ) for @xmath287 , @xmath283 , @xmath102 and @xmath101 . the gss in plates ( a ) and ( b ) are symmetric and born at the saddle - node bifurcation while the gss in plate ( c ) is asymmetric and born at the pitchfork bifurcation . ] of ( [ e : uvans ] ) born at the bifurcations of fig . [ f : bd](a)-(c ) for @xmath287 , @xmath283 , @xmath102 and @xmath101 . the gss in plates ( a ) and ( b ) are symmetric and born at the saddle - node bifurcation while the gss in plate ( c ) is asymmetric and born at the pitchfork bifurcation . ] of ( [ e : uvans ] ) born at the bifurcations of fig . [ f : bd](a)-(c ) for @xmath287 , @xmath283 , @xmath102 and @xmath101 . the gss in plates ( a ) and ( b ) are symmetric and born at the saddle - node bifurcation while the gss in plate ( c ) is asymmetric and born at the pitchfork bifurcation . ] of ( [ e : uvans ] ) born at the saddle - node bifurcation of fig . [ f : bd](d ) for @xmath105 , @xmath104 , @xmath101 and @xmath102 . ] of ( [ e : uvans ] ) born at the saddle - node bifurcation of fig . [ f : bd](d ) for @xmath105 , @xmath104 , @xmath101 and @xmath102 . ] figures [ f : newgs-1 ] and [ f : newgs0 ] show the profiles of gss born at the bifurcations in figs . [ f : bd](a)-(c ) and ( d ) , respectively . in particular , the gss in fig . [ f : newgs0 ] are bound states with two humps . standard results about transverse homoclinic orbits explain the existence of multi - pulse gss , see @xcite , and we also find such gss in the other gap regions . we note that bifurcations of symmetric homoclinic orbits in reversible systems that are _ autonomous _ have been discussed in buchto:96,kn:97 . our analysis of the poincar map also yields comparable bifurcation results in systems with periodic forcing like eqs . ( [ e : ab ] ) . in this paper we have studied the existence of gss ( gap solitons ) in a model of the bg ( bragg grating ) subjected to periodic modulation along the fiber , thus creating the supergrating . the corresponding mathematical model amounts to a parametrically forced extension of the standard bg model with the additional symmetry - breaking forcing term . as was known previously @xcite , the supergrating opens up a set of new bandgaps . using the melnikov method and averaging techniques , we have demonstrated that families of gss exist robustly in the new gaps . we have also analyzed bifurcations of gss and showed that , in addition to the fundamental gss , the model also supports families of bound states of the solitons . many of the gss found in the model are completely stable , as was verified by means of direct simulations for some of them @xcite . the present paper extends the results of @xcite and gives detailed mathematical proofs for them , although we did not analyze the stability problem here in detail . in particular , we have discussed the effects of the symmetry - breaking forcing term and demonstrated that it can lead to the closing and re - opening of the bandgaps . in the presence of the new term , gss exist robustly ( as generic solutions ) in all the bandgaps , but , as the symmetry of the equation is broken , they undergo many bifurcations . this research was initiated and many important parts were carried out when the authors worked with alan champneys or visited him at the university of bristol . we thank him for discussions and comments as well as his hospitality and support . k.y . acknowledges support from the japan society for the promotion of science , grant - in - aid for scientific research ( c ) nos . 18560056 , 21540124 and 22540180 . ] let @xmath314^j}\ ] ] and let @xmath315 for @xmath316 . to estimate ( [ e : ij ] ) , we consider the complex integral @xmath317 where @xmath318 is a rectangle contour , @xmath319 and @xmath320 , depicted in fig . [ f : a1 ] , with @xmath321 . the complex function @xmath322 is singular only at @xmath323 which are @xmath124th - order poles , inside of @xmath318 . by the residue theorem , we have @xmath324 where @xmath325 is the residue of @xmath326 at @xmath327 . we compute @xmath328 so that @xmath329 and @xmath330.\end{aligned}\ ] ] on the other hand , we can write @xmath331 since by @xmath332 @xmath333 and the second and fourth integrals in ( [ e : int3 ] ) tend to zero as @xmath334 , we have @xmath335 where we used ( [ e : fj ] ) . hence , we obtain @xmath336 and @xmath337,\ ] ] from which eqs . ( [ e : int1 ] ) and ( [ e : int2 ] ) follow . applying the third - order averaging method , we calculate the boundaries of the central gap . again , the ` mathematica ` program `` haverage.m '' is used to obtain the third - order averaged systems . let @xmath338 and @xmath339 . using the transformation @xmath340 , we rewrite ( [ e : ab ] ) as @xmath341\eta+o(\varepsilon^{5/2 } ) , \\ \eta^{\prime}= & \varepsilon^{1/2}[\cos kx+\bar{\mu}\cos(kx+\delta)]\xi + \varepsilon^{3/2}\left(-\nu\xi+\frac{3}{2}\xi^{3}\right)+o(\varepsilon^{5/2 } ) . \end{split } \label{e : abr}\ ] ] the third - order averaged system for ( [ e : abr ] ) is obtained as @xmath342 , \label{e : avr}\ ] ] where @xmath343 was taken as the small parameter . in the averaged system ( [ e : avr ] ) , the origin is a saddle if @xmath344 hence , we have the approximate right boundary of the central gap , @xmath345 let @xmath346 and @xmath339 . using the transformation @xmath347 , we rewrite ( [ e : ab ] ) as @xmath348\eta -\varepsilon^{3/2}\left(\nu\eta+\frac{3}{2}\eta^{3}\right)+o(\varepsilon^{5/2 } ) , \\ \eta^{\prime}= & -2\varepsilon^{1/2}\xi + \varepsilon^{3/2}[\cos kx+\bar{\mu}\cos(kx+\delta)]\xi+o(\varepsilon^{5/2 } ) , \end{split } \label{e : abl}\ ] ] the third - order averaged system for ( [ e : abl ] ) is obtained as @xmath349,\quad \eta^{\prime}=-2\varepsilon^{1/2}\xi,\ ] ] in which the origin is a saddle if @xmath350 thus , we have the approximate left boundary of the central gap , @xmath351 e. j. doedel , a. r. champneys , t. r. fairgrieve , yu . a. kuznetsov , b. sandstede , and x. j. wang , continuation and bifurcation software for ordinary differential equations , 1997 ( an upgraded version is available at ` http://cmvl.cs.concordia.ca/auto/ ` ) . t. dohnal , d. e. pelinovsky and g. schneider , coupled - mode equations and gap solitons in a two - dimensional nonlinear elliptic problem with a separable periodic potential , _ j. nonlinear sci . _ 19 ( 2009 ) 95131 . c. sim , averaging under fast quasiperiodic forcing , in : j. seimenis ( ed . ) , _ hamiltonian mechanics : integrability and chaotic behavior , proceedings of a nato advanced research workshop held in toru , june 28-july 2 , 1993 _ , plenum , new york , 1994 , pp .

gravitational waves ( gw ) , on very general grounds , appear to be a largely unavoidable consequence of a well established fact that no known interaction propagates instantly from source to observer : gravitation would be the first exception to this rule , should it be described by newton s theory . indeed , the newtonian gravitational potential @xmath0 satisfies poisson s equation @xmath1 where @xmath2 is the density of matter in the sources of gravitational fields , and @xmath3 is newton s constant . but , since ( [ eq1.1 ] ) contains no time derivatives , the time dependence of @xmath0 is purely _ parametric _ , i.e. , time variations in @xmath2 _ instantly _ carry over to @xmath0 , irrespective of the value of @xmath4 . so , for example , non - spherically symmetric fluctuations in the mass distribution of the sun ( such as e.g. those caused by solar storms ) would instantly _ and _ simultaneously be felt both in the nearby mercury and in the remote pluto quite independently of the _ quantitative _ relevance of such instant propagation effect in this particular example which is none in practice , its very existence is _ conceptually _ distressing . in addition , the asymmetry between the space and time variables in ( [ eq1.1 ] ) does not even comply with the basic requirements of special relativity . einstein s solution to the problem of gravity , general relativity ( gr ) , does indeed _ predict _ the existence of radiation of gravitational waves . as early as 1918 , einstein himself provided a full description of the polarisation and propagation properties of weak gws @xcite . according to gr , gws travel across otherwise flat empty space at the speed of light , and have _ two _ independent and _ transverse _ polarisation amplitudes , often denoted @xmath5 and @xmath6 , respectively @xcite . in a more general framework of so called _ metric theories _ of gravity , gws are allowed to have up to a maximum 6 amplitudes , some of them transverse and some longitudinal @xcite . the theoretically predicted existence of gws poses of course the experimental challenge to _ measure _ them . historically , it took a long while even to attempt the construction of a _ gravitational telescope _ : it was not until the decade of the 1960 s that j. weber first took up the initiative , and developed the first gravitational antennas . these were elastic cylinders of aluminum , most sensitive to _ short bursts _ ( a few milliseconds ) of gws . after analysing the data generated by _ two _ independent instruments , and looking for events in coincidence in both , he reported evidence that a considerably large number of gw flares had been sighted @xcite . even though weber never gave up his claims of _ real _ gw detection @xcite , his contentions eventually proved untenable . for example , the rate and intensity of the reported events would imply the happening of several supernova explosions per week in our galaxy @xcite , which is astrophysically very unlikely . it became clear that more sensitive detectors were necessary , whose design and development began shortly afterwards . in the mid 1970 s and early eighties , the new concept _ interferometric _ gw detector started to develop @xcite , which would later lead to the larger _ ligo _ and _ virgo _ projects , as well as others of more reduced dimensions ( _ geo-600 _ and _ tama _ ) , and to the future _ space _ antenna _ lisa_. in parallel , _ cryogenic _ resonant detectors were designed and constructed in several laboratories , and towards mid 1990 s the next generation of _ ultracryogenic _ antennas , _ nautilus _ ( rome ) , _ auriga _ ( padua ) , _ allegro _ ( baton rouge , louisiana ) and _ niobe _ ( australia ) , began _ taking data_. more recently , data exchange protocols have been signed up for multiple detector coincidence analysis @xcite . based on analogous physical principles , new generation _ spherical _ gw detectors are being programmed in brazil , holland and in spite of many years of endeavours and hard work , gws have proved elusive to all dedicated detectors constructed so far . however , the discovery of the binary pulsar _ psr 1913 + 16 _ by r. hulse and j. taylor in 1974 @xcite , and the subsequent long term detailed monitoring of its orbital motion , brought a breeze of fresh air into gw science : the measured decay of the orbital period of the binary system due to _ gravitational bremsstrahlung _ accurately conforms to the predictions of general relativity . hulse and taylor were awarded the 1993 nobel prize in physics for their remarkable work . as of 1994 @xcite , the accumulated binary pulsar data confirm gr to a high precision of a tenth of a percent . the binary pulsar certainly provides the most compelling evidence to date of the gw phenomenon as such , yet it does so thanks to the observation of a _ back action _ effect on the source . even though i do not consider accurate the statement , at times made by various people , that this is only _ indirect _ evidence of gws , it is definitely a matter of fact that there is more to gws than revealed by the binary pulsar for example , amplitude , phase and polarisation parameters of a gw can only be measured , according to current lore , with dedicated gw antennas . but how do gw telescopes interact with the radiation they are supposed to detect ? this is of course a fundamental question , and is also the subject of the present contribution , where i intend to review the _ theoretical foundations _ of this problem , and its solutions as presently understood . in section [ sec.2 ] , i summarise the essential properties of gws within a rather large class of possible theories of the gravitational interaction ; section [ sec.3 ] briefly bridges the way to sections [ sec.4 ] and [ sec.5 ] where interferometric and acoustic detector concepts are respectively analysed in some detail . section [ sec.5 ] also includes aboundant reference to new generation spherical detectors in its various variants ( solid , hollow , dual ) . for the sake of completeness , i have added a section ( section [ sec.6 ] ) with a very short summary of detector characterisation concepts , so that the interested reader gets a flavour of how sensitivities are defined , what do they express and how do gw signals compare with local noise disturbances in currently conceived detectors . section [ sec.7 ] closes the paper with a few general remarks . quite generally , a time varying mass - energy distribution creates in its surroundings a time varying gravitational field ( _ curvature _ ) . as already stressed in section [ sec.1 ] , we do not expect such time variations to travel instantly to distant places , but rather that they travel as _ gravity waves _ across the intervening space . now , how are these waves `` seen '' ? a _ observer may of course not feel any variations in the gravitational field where he / she is immersed , if he / she is in _ free fall _ in that field this is a consequence of the _ equivalence principle _ _ two _ nearby observers have instead the capacity to do so : for , both being in free fall , they can take each other as a reference to measure any relative _ accelerations _ caused by a non - uniform gravitational field , in particular those caused by a _ gravitational wave field_. we can rephrase this argument saying that gravitational waves show up as _ local tides _ , or gradients of the local gravitational field at the observatory . in the language of differential geometry , tides are identified as _ geodesic deviations _ , i.e. , variations in the four - vector connecting nearby geodesic lines . it is shown in textbooks , e.g. @xcite , that the geodesic deviation equation is @xmath7 where `` @xmath8 '' means _ covariant derivative _ , @xmath9 is proper time for _ either _ geodesic , @xmath10 is the riemann tensor , @xmath11 is a unit tangent vector ( again to either geodesic ) , and @xmath12 is the vector connecting corresponding points of the two geodesic lines . the gw fields considered in this paper will be restricted to a class of _ perturbations _ of the geometry of otherwise flat space - time , with the additional assumptions that they be [ cols= " < " , ] this is certainly not the most general definition of a gw yet it will suffice to our purposes here : any gws generated in astrophysical sources and reaching a man - made detector definitely satisfy the above requirements . the interested reader is referred to @xcite for a thorough treatment of more general gws . following the above assumptions , a gw can be described by perturbations @xmath13 of a lorentzian metric @xmath14 , i.e. , there exist _ quasi - lorentzian _ coordinates @xmath15 in which the space - time metric @xmath16 can be written @xmath17 with @xmath18 the actual effect of a gw on a pair of test particles is , according to ( [ eq2.1 ] ) , determined by the riemann tensor @xmath10 , and this in turn is determined by the functions @xmath19 . i now review briefly the different possibilities in terms of which is the theory underlying the physics of gravity waves , i.e. , which are the field equations which the @xmath19 satisfy . the vacuum field equations of general relativity are , as is well known @xcite , @xmath20 where @xmath21 is the ricci tensor of the metric @xmath22 . if quadratic and successively higher order terms in the perturbations @xmath19 are neglected then this tensor can be seen to be given by @xmath23 with @xmath24 new coordinates @xmath25 can be defined by means of transformation equations @xmath26 and these will still be quasi - lorentzian if the functions @xmath27 are _ sufficiently small_. more precisely , the gw components are , in the new coordinates , @xmath28 provided higher order terms in @xmath29 are neglected . thus `` sufficiently small '' means that the derivatives of @xmath30 be of the order of magnitude of the metric perturbations @xmath31 so that in the new coordinates @xmath32 the metric tensor _ also _ splits up as @xmath33 . it is now possible , see @xcite , to choose new coordinates in such a way that the _ gauge conditions _ @xmath34 hold . this being the case , equations ( [ eq2.3 ] ) read @xmath35 which are vacuum _ wave equations_. therefore gws travel across empty space at the speed of light , according to gr theory . _ plane wave solutions _ to ( [ eq2.9 ] ) satisfying ( [ eq2.8 ] ) can now be constructed @xcite which take the form @xmath36 for waves travelling down the @xmath37-axis . the label @xmath38 is an acronym for _ transverse - traceless _ , the usual denomination for this particular gauge . the physical meaning of the polarisation amplitudes in ( [ eq2.10 ] ) is clarified by looking at the effect of an incoming wave on test particles . consider e.g. two equal test masses whose center of mass is at the origin of @xmath38 coordinates ; let @xmath39 be their distance in the absence of gws , and @xmath40 the orientation ( relative to the @xmath38 axes ) of the vector joining both masses . making use of the geodesic deviation equation ( [ eq2.1 ] ) , with the riemann tensor associated to ( [ eq2.10 ] ) , it can be seen that the gw only affects the _ transverse _ projection of the distance relative to the wave propagation direction ( the @xmath37-axis ) ; in fact , if @xmath41 is such distance then some simple algebra leads to the result . but it can also be calculated at the position of _ either _ one of them this would only make up for a negligible _ second order _ difference . ] @xmath42\;. \label{eq2.12}\ ] ] '' and `` @xmath43 '' polarisation modes of a gw , according to gr theory . the dotted lines ( circles ) indicate the test particles position in the absence of gw signal . each step in the graph corresponds to a quarter of the period of the driving gw , as labeled atop . [ fig1],width=449 ] it is very important at this point to stress that the _ wavelength _ @xmath44 of the incoming gw must be much larger than the distance @xmath45 between the particles for ( [ eq2.12 ] ) to hold , i.e. , @xmath46 and this is a condition which must be _ added _ to the already made assumption that @xmath47 . a graphical representation of the result ( [ eq2.12 ] ) is displayed in figure [ fig1 ] : a number of test particles are evenly distributed around a circle perpendicular to the incoming gw , i.e. , in the @xmath48 plane . when a periodic signal comes in , the distances between those particles change following ( [ eq2.12 ] ) ; note that the changes are modulated by the angular factors , i.e. , according to the particles positions on the circle . the `` @xmath49 '' mode is characterised by a vanishing wave amplitude @xmath50 , while in the `` @xmath43 '' mode @xmath51 vanishes . although general relativity has never been questioned so far by experiment , there are in fact alternative theories , e.g. brans - dicke theory @xcite , which are interesting for a number of reasons , for example _ cosmological _ reasons @xcite . generally , these theories make their own specific predictions about gws , and they partly differ from those of gr just discussed . the term _ metric theory _ indicates that the gravitational interaction affects the _ geometry _ of space - time , i.e. , the metric tensor @xmath22 is a fundamental ingredient though other fields may also be necessary to complete the theoretical scheme . obviously , general relativity falls within this class of theories . the appropriate scheme to assess the physics of such more general class of gws was provided long ago @xcite . the idea is to consider only _ plane gravitational waves _ , which should be an extremely good approximation for astrophysics , given our great distance even to the nearest conceivable gw source , and to characterise them by their _ newman - penrose scalars _ @xcite . it appears that only _ six _ components of the riemann tensor out of the usual 20 are independent in a _ plane _ gw ; these are given by the four newman - penrose scalars @xmath52 of which @xmath53 and @xmath54 are real , while @xmath55 and @xmath56 are _ complex _ functions of the null variable @xmath57 see @xcite for all notation details . if a quasi - lorentzian coordinate system is chosen such that gws travel along the @xmath37-axis then @xmath58 , and one can calculate the scalars ( [ eq2.14 ] ) to obtain _ 2(t - z ) & = & -r_tztz , [ eq2.15a ] + _ 3(t - z ) & = & -(-r_txtz+r_tytz ) , [ eq2.15b ] + _ 4(t - z ) & = & -r_txtx+r_tyty+2r_txty , [ eq2.15c ] + _ 22(t - z ) & = & -r_txtx - r_tyty . [ eq2.15d ] general relativity is characterised by @xmath59 being the only non - vanishing scalar , while in brans - dicke theory @xmath60 also is different from zero see @xcite for full details . it is relevant to remark at this stage that the only non - trivial components of the riemann tensor of a plane gw are the so called `` _ _ electric _ _ '' components , @xmath61 , as we see in equations ( [ eq2.15a]d ) above . these are , incidentally , _ also _ the only ones which appear in the geodesic deviation equation ( [ eq2.1 ] ) , since one may naturally choose @xmath62 . but , because of the symmetries of the riemann tensor , only values of @xmath63 and @xmath64 different from zero , i.e. , @xmath65 , give non - zero contributions . ] has spherical symmetry , so it includes a combination of longitudinal and transverse excitations in like proportions . [ fig2],width=415 ] this fact helps us make a graphical representation of all _ six _ possible polarisation states of a general metric gw in the same manner as in figure [ fig1 ] . the result is represented in figure [ fig2 ] whose source is reference @xcite . the idea is to take a ring of test particles , let a gw pass by , and analyse the results of the displacements it causes in the distributions of those particles , just as done in section [ sub2.2 ] . note that the first three modes are _ transverse _ , while the other three are longitudinal see the caption to the figure . as already stressed , gr only gives rise to the two @xmath56 modes . we are now ready to discuss the objectives and procedures to detect gws : knowing their physical structure , one can design systems whose interaction with the gws be sufficiently well understood and under control ; suitable monitoring of the dynamics of such systems will be the source of information on whatever gw parameters may show up in a given observation experiment . there are _ two _ major detection concepts : _ interferometric _ and _ acoustic _ detection . historically , the latter came first through the pioneering work of j. weber , but interferometric gw antennas are at present attracting the larger stake of the investment in this research field , both in hardware and in human commitment . this is because much hope has been deposited in their capabilities to reach sufficient sensitivity to _ measure _ gws for the first time . interferometric detectors aim to measure phase shifts between light rays shone along two different ( straight ) lines , whose ends are defined by _ freely suspended _ test masses . this is done in a michelson layout , using mirrors , beam splitters and photodiodes . acoustic detectors are instead based on elastically linked test masses rather than freely suspended which _ resonantly _ respond to gw excitations . these _ qualitative _ ideas can be made quantitatively precise , but the process is a non - trivial one and has important subtleties which must be properly understood for a thorough assessment of the detector workings and readout . the next sections are devoted to explain with some detail which are the _ theoretical principles _ governing the behaviour and response of both kinds of gw antennas . a rather _ nave _ idea to measure the effect of an incoming gw is provided by the following argument see also figure [ fig3 ] for reference : let a gw having a `` @xmath49 '' polarisation ( assume gr for simplicity at this stage ) come in perpendicular to the local horizontal at a given observatory ; if three masses are laid down on the vertices of an ideally oriented isosceles right triangle then , as we saw ( figure [ fig1 ] ) , the catets shrink and stretch with opposite phases . if a beam of laser light is now shone into the system , and a beam - splitter attached to mass @xmath66 and mirrors attached to @xmath67 and @xmath68 , then one can think of measuring the distance changes between the masses by simple interferometry . '' polarised relative to the @xmath69 and @xmath70 axes , causes the end masses @xmath67 and @xmath68 to oscillate in phase opposition relative to the central mass @xmath66 . light is shone into the system , and suitable beam splitters and mirrors are attached to the masses ; length changes are then measured interferometrically , which directly lead to determine the gw amplitude . [ fig3],width=415 ] this may look like a very reasonable proposal for a detector yet the following criticism readily suggests itself : gravitation is concerned with _ geometry _ , i.e. , gravitational fields alter lengths and angles ; therefore gws will affect _ identically _ both distances between the masses of figure [ fig3 ] _ and _ the wavelength of the light travelling between them thus leading to a _ cancellation _ of the conjectured interferometric effect while the criticism is certainly correct , the conclusion is _ not_. the reason is that it overlooks the fact that gravitation is concerned with the geometry of _ space - time _ not just space . in the case of the above gw it so happens that , in @xmath38 coordinates , the time dimension of space - time is not _ warped _ in the gw geometry see the form of the metric tensor in equation ( [ eq2.10 ] ) while the transverse space dimensions _ are_. consequently an electromagnetic wave travelling in the @xmath48 plane experiences _ wavelength _ changes depending on the propagation direction , but it does _ not _ experience _ frequency _ changes . the net result of this is that the _ phase _ of the electromagnetic wave differs from direction to direction of the @xmath48 plane , and _ this _ makes a gw amenable to detection by interferometric principles . looked at in this way , the masses represented in figure [ fig3 ] only play a _ passive _ role in the detector , in the sense that they simply make the interference between the two light beams possible by providing physical support for the mirrors and beam splitters . in other words , the physical principles underlying the working of an interferometric gw detector must have to do with the _ interaction between gws and electromagnetic waves _ rather than with geodesic deviations of the _ masses_. admittedly , this is not the most common point of view @xcite . it can however be made precise by the following considerations , which are studied in depth in references @xcite and @xcite . according to the above , it appears that we must address the question of how gw - induced fluctuations in the geometry of a background space - time affect the properties ( amplitude and phase ) of a plane electromagnetic wave a _ light beam _ which travels through such space - time . it will be sufficient to consider that the light beam is a _ test _ beam , i.e. , its back action on the surrounding geometry is negligible . let then @xmath71 be the vector potential which describes an electromagnetic wave travelling in vacuum ; @xmath72 thus satisfies maxwell s equations @xmath73 where @xmath74 stands for the generalised dalembert operator : @xmath75 we need only retain _ first order _ terms in @xmath31 in the covariant derivatives , so that equation ( [ eq3.1 ] ) reads @xmath76 to this equation , _ gauge conditions _ must be added . we shall conventionally adopt the usual lorentz conditions , @xmath77 , which , to lowest order in the gravitational perturbations , read @xmath78 in addition to the weakness of the gw perturbation , it is also the case in actual practice that : * the gw typical frequencies , @xmath79 , are much smaller than the frequency of the light , @xmath80 : @xmath81 . * the wavelength of the gw is much larger than the cross sectional dimensions of the light beam . wave front distortions and beam curvature are effects which can also be safely neglected in first order calculations @xcite . finally , i shall make the simplifying assumption of perpendicular incidence , i.e. , the incoming gw propagates in a direction orthogonal to the interferometer s arms . we shall thus consider one of the interferometer arms aligned with the @xmath69-direction , and the other with the @xmath70-direction , while the incoming gw will be assumed to approach the detector down the @xmath82 . the interaction gw - light beam will thus occur in the @xmath83 plane , hence the gw perturbations can be suitably described by a function of time alone , i.e. , @xmath84 where @xmath79 is the frequency of the gw , which we can also safely assume to be plane - fronted , since its source will in all cases of interest be far removed from the observatory . in addition , for a beam running along the @xmath85 , the electromagnetic vector potential will only depend on the space variable @xmath69 and on time , i.e. , @xmath86 the following _ ansatz _ suggests itself for a solution to equations ( [ eq3.3 ] ) : a_t(t , x ) & = & _ 0(t ) , [ eq3.7a ] + a_x(t , x ) & = & _ 1(t ) , [ eq3.7b ] + a_y(t , x ) & = & a_2 e^i_2(t ) , [ eq3.7c ] + a_z(t , x ) & = & a_3 e^i_3(t ) , [ eq3.7d ] where @xmath87 , @xmath88 , @xmath89 and @xmath90 are small quantities of order @xmath91 , @xmath92 and @xmath93 are constants , and @xmath80 is the frequency of the light . clearly , these expressions reproduce the plane wave solutions to vacuum maxwell s equations in the limit of flat space - time , i.e. , when @xmath94 . let us now take an incoming gw of the form @xmath95 and substitute it into ( [ eq3.3 ] ) and ( [ eq3.4 ] ) , with the _ ansatz _ ( [ eq3.7a]-d ) , neglecting higher order terms in the ratio @xmath96 . then @xcite , _ both _ @xmath97 and @xmath98 are seen to satisfy the approximate differential equation @xmath99 the solution to this equation which is _ independent of the initial conditions _ is , to the stated level of accuracy , @xmath100 as shown in reference @xcite , we need not worry about either @xmath87 or @xmath88 at this stage because the _ longitudinal _ component of the electric field ( i.e. , @xmath101 ) is an order of approximation smaller than the transverse components , which are given by @xmath102 hence e_y(x , t ) & & a_2 , [ eq3.12a ] + e_z(x , t ) & & a_3 . [ eq3.12b ] these expressions beautifully show how the incoming gw causes a _ phase shift _ in an electromagnetic beam of light . note that this phase shift is a _ periodic _ function of time , with the frequency of the gw . if we consider a real interferometer , such as very schematically shown in figure [ fig3 ] , and call @xmath103 the _ round trip _ time for the light to go from @xmath66 to @xmath68 and back , then the accumulated phase shift is , according to these formulas , @xmath104 since there is an obvious symmetry between light rays travelling to the right and to the left for a gw arriving perpendicularly to them . the arguments leading to equations ( [ eq3.12a]-b ) can be very easily reproduced , _ mutatis mutandi _ , to obtain the phase shift experienced by a light ray travelling in the @xmath105 , rather than the @xmath69 direction everything in fact amounts to a simple interchange @xmath106 in the equations , which includes @xmath107 as this is equivalent to @xmath108 , see ( [ eq2.10 ] ) . the result is @xmath109 in the actual interferometer , provided it has equal arm lengths , the two laser rays recombine in the beam splitter with a net phase difference @xmath110 and this produces an _ interference signal _ , which is in principle measurable if the instrumentation is sufficiently sensitive . the reader may wonder how is it that the detector signal only depends on one of the gw amplitudes , @xmath51 , but not on the other , @xmath50 . the reason is that we have made a very special assumption regarding the orientation of the gw s polarisation axes relative to the light beam propagation directions . in a realistic case , even if perpendicular gw incidence happens , the detector s arms will not be aligned with the gw s natural axes , let alone the most likely case of _ oblique _ incidence . an important conclusion one should draw from this section is a _ conceptual _ one , that interferometric detectors are able to measure gw amplitudes and polarisations as a result of the interaction between the electromagnetic field of light rays and the background space - time geometry they travel across . beyond this , though , equation ( [ eq3.15 ] ) has very relevant _ quantitative _ consequences , too . for example , as stressed in reference @xcite , its range of validity is _ not _ limited to interferometer arms short compared to the gw wavelength . therefore , according to the formula , a _ null _ effect ( signal cancellation ) happens if the round trip time @xmath111 equals the period of the gw , @xmath112 . likewise , equation ( [ eq3.15 ] ) also tells us that _ maximum _ detector signal occurs when @xmath113 . all this happen to be true for arbitrary incidence and polarisation of the incoming gws as well . for gw frequencies in the 1khz range , the best detector should thus have arm lengths in the range of 150 kilometers and even longer for lower gw frequencies . no ground based gw antenna has ever been conceived of such dimensions yet there are intelligent ways to _ store _ the light in shorter arms for suitably tuned gw periods . i shall not go into details of these technical matters , see @xcite and @xcite for thorough information . at its ends to oscillate at the gw s frequency . resonant amplification is obtained when the latter equals the spring s characteristic frequency , @xmath80 . [ fig4],width=264 ] acoustic gw detectors work based on a completely different concept see figure [ fig4 ] : the idea is to set up test masses @xmath114 linked together by a spring of relaxed length @xmath39 , so that gw _ tides _ drive their oscillations around the equilibrium position , with significant mechanical amplification at the spring s characteristic frequency @xmath80 . the spring deformation @xmath115 thus obeys the following equation of motion : @xmath116 where @xmath117\,\sin^2\theta\ ; , \label{eq5.3}\ ] ] as follows from ( [ eq2.12 ] ) see also reference @xcite for a complete discussion of this case . this is the main idea , but in practice acoustic gw detectors are _ elastic solids _ rather than a single spring , i.e. , they do not have a single characteristic frequency but a whole spectrum . the response of an elastic solid to a gw excitation is assessed by means of the classical theory of elasticity , as described for example in @xcite . in such theory the deformations of the solid are given by the values of a vector field of displacements , @xmath118 , which satisfies the evolution equations @xmath119 where @xmath120 is the ( undeformed ) solid s density , and @xmath44 and @xmath121 are its lam coefficients , related to the poisson ratio and young modulus of the material the solid is made of @xcite . the function in the rhs of the equation is the density of _ external forces _ driving the motion of the system ; in the present case , these are the _ tides _ generated by the sweeping gw , i.e. , @xmath122 where @xmath123 are components of the riemann tensor evaluated at a fixed point of the solid , most expediently chosen at its center of mass . as already discussed in section [ sub2.2 ] , plane gws have at most six degrees of freedom , adequately associated with the six _ electric components _ of the riemann tensor , @xmath123 . the six components are _ one monopole _ amplitude and _ five quadrupole _ amplitudes , and this important structure is made clear by the following expression of the density of gw tidal forces : @xmath124 which is entirely equivalent to ( [ eq5.5 ] ) see reference @xcite , and uses the common @xmath125 notation convention to indicate multipole terms . it is very important to stress at this stage that @xmath126 are pure _ form factors _ , simply depending on the fact that tides are monopole - quadrupole quantities , while all the relevant _ dynamic _ information carried by the gw is encoded in the time dependent coefficients @xmath127 . according to these considerations , we see that the ultimate objective of an acoustic gw antenna is to produce values of @xmath127 or indeed to extract from the device s readout as much information as possible about those quantities . somewhat lengthy algebra permits to write down a _ formal solution _ to equations ( [ eq5.5 ] ) and ( [ eq5.6 ] ) in terms of an orthogonal series expansion @xcite : @xmath128\ ; , \label{eq5.7}\ ] ] where f_n^(l , m ) & & _ solid_n^ * ( ) ^(l , m)()^3x , [ eq5.8a ] + g_n^(l , m)(t ) & & _ 0^tg^(l , m)(t)_n(t - t)t , [ eq5.8b ] with @xmath114 the whole solid s mass ; @xmath129 is the ( possibly _ degenerate _ ) characteristic frequency of the elastic body , and @xmath130 the corresponding _ wavefunction _ , both determined by the solution to the _ eigenvalue _ problem @xmath131 with the boundary conditions that the surface of the solid be free of any tensions and/or tractions see @xcite for full details . historically , the first gw antennas were weber s elastic _ cylinders _ @xcite , but more recently , _ spherical _ detectors have been seriously considered for the next generation of gw antennas , as they show a number of important advantages over cylinders . i shall devote the next sections to a discussion of both types of systems , though clear priority will be given to spheres , due to their much richer capabilities and theoretical interest . first thing to study the response of an elastic solid to an incoming gw is , as we have just seen , to determine its characteristic oscillation _ modes _ , i.e. , its frequency spectrum @xmath132 and associated wavefunctions , @xmath133 . in the case of a cylinder this is a formidable task ; although its formal solution is known @xcite , cylindrical antennas happen in practice tobe narrow and long @xcite , and so approximate solutions can be used instead which are much simpler to handle , and sufficiently good see also @xcite . it appears that , in the _ long rod _ approximation , the most efficiently coupled modes are the _ longitudinal _ ones , and these have typical sinusoidal profiles , of the type @xmath134 for a rod of length @xmath135 whose end faces are at @xmath136 , and in which the speed of sound is @xmath137 . figure [ fig5 ] graphically shows the longitudinal deformations of the cylinder which correspond to ( [ eq5.10 ] ) , including _ transverse _ distortions which , though not reflected in the simplified equation ( [ eq5.10 ] ) , do happen in practice as a result of the poisson ratio being different from zero @xcite . an important detail to keep in mind is that _ odd n _ modes have maximum displacements at the end faces , while _ even n _ modes have nodes there . in fact , the latter do not couple to gws @xcite . it is also interesting to stress that the center of the cylinder is always a node this is relevant e.g. to suspension design issues @xcite . , width=302 ] a very useful concept to characterise the sensitivity of an acoustic antenna is its _ cross section _ for the absorption of gw energy . if an incoming gw flux density of @xmath138 watts per square metre and hertz sweeps the cylinder and sets it to oscillate with energy @xmath139 joules then the cross section is _ defined _ by the ratio @xmath140 which is thus measured in m@xmath141hz . simple calculations show that , for optimal antenna orientation ( perpendicular to the gw incidence direction ) this quantity is given by @xcite @xmath142 where @xmath114 is the cylinder s total mass . it is interesting to get a flavour of the order of magnitude this quantity has : consider a cylinder of al5056 ( an aluminum alloy , for which @xmath143 m/s ) , 3 metres long and 60 centimetres across , which has a mass of 2.3 tons ; the above formula tells us that , for the first mode ( @xmath144 ) , @xmath145 this is a _ very small _ number indeed , and gives an idea of how weak the coupling between gws and matter is . the weakness of the coupling gives an indication of how difficult it is to detect gws . by the same token , though , gws are very weakly damped as they travel through matter , which means they can produce information about otherwise invisible regions , such as the interior of a supernova , or even the big bang . equation ( [ eq5.12 ] ) is only valid for _ perpendicular _ gw incidence . if incidence is instead oblique then a significant damping factor of @xmath146 comes in , where @xmath147 is the angle between the gw direction and the cylinder s axis @xcite ) , if one notices that the energy of oscillation appearing in the numerator of ( [ eq5.12 ] ) is proportional to @xmath148 . ] . this is a severe penalty , and it also happens in interferometric detectors not optimally oriented @xcite . the first initiatives to construct and operate gw detectors are due to j. weber , who decided to use elastic cylinders . this philosophy and practice has survived him , and still today ( february 2002 ) all gw detectors in continuous data taking regimes are actually weber bars , though with significant sensitivity improvements @xcite derived , amongst other , from ultracryogenic and _ squid _ techniques . about ten years after weber began his research , r. forward published an article @xcite where he pondered in a semi - quantitative way the potential virtues of a _ spherical _ , rather than cylindrical gw detector . ashby and dreitlein @xcite estimated how the whole earth , as an _ auto - gravitating _ system , responds to gws bathing it , and later wagoner @xcite developed a theoretical model to study the response of an elastic sphere to gw excitations . interest in this new theoretical concept then waned to eventually re - emerge in the 1990s . the _ allegro _ detector group at louisiana constructed a room temperature prototype antenna @xcite , which produced sound experimental evidence that it is actually possible to have a working system capable of making _ multimode _ measurements i ll come to this in detail shortly , thus proving that a full - fledged spherical gw detector is within reach of current technological state of the art , as developed for weber bars . it was apparently the fears to find unsurmountable difficulties in this problem which deterred further research on spherical gw antennas for years @xcite . in this section i will give the main principles and results of the _ theory _ of the spherical gw detector , based on a formalism which has already been partly used in section [ sec.5 ] , and for whose complete detail the reader is referred to @xcite . as we have seen , first thing we need is the _ eigenmodes _ and _ frequency spectrum _ of the spherical solid . this is a classical problem , long known in the literature , the solution to which i will briefly review here , with some added emphasis on the issues of our present concern . the oscillation eigenmodes of a solid elastic body fall into two families : _ spheroidal _ and _ torsional _ modes @xcite . of these , only the former couple to gws , while torsional modes do not couple at all @xcite . spheroidal wavefunctions have the analytic form @xmath149 where @xmath150 are spherical harmonics @xcite , @xmath151 is the outward pointing normal , @xmath152 is the ` angular momentum operator ' , @xmath153 , @xmath154 and @xmath155 are somewhat complicated combinations of spherical bessel functions @xcite , and @xmath156 are ` quantum numbers ' which label the modes . the frequency spectrum appears to be composed of ascending series of _ multipole harmonics _ , @xmath157 , i.e. , for _ each _ multipole value @xmath158 there are an infinite number of frequency harmonics , ordered by increasing values of @xmath159 . for example , there are _ monopole _ frequency harmonics @xmath160 , @xmath161 , @xmath162 , etc . ; then _ dipole _ frequencies @xmath163 , @xmath164 , , then _ quadrupole _ harmonics @xmath165 , @xmath166 , and so on . each of these frequencies is @xmath167-fold degenerate , and this is a fundamental fact which makes of the sphere a theoretically ideal gw detector , as we shall shortly see . if the above expressions are substituted into ( [ eq5.8a]-b ) , then into ( [ eq5.7 ] ) , one easily obtains the sphere s response function as @xmath168 & + & \sum_{n=1}^\infty\,\frac{a_{n2}}{\omega_{n2}}\,\left[\sum_{m=-2}^2 \vec{u}_{n2m}(\vec{x})\,g_{n2}^{(2,m)}(t)\right]\ ; , \label{eq5.15}\end{aligned}\ ] ] where a_n0 & = & -_0^r a_n0(r)r^3r , [ eq5.16a ] + a_n2 & = & -_0^r r^3r , [ eq5.16b ] and @xmath169 ) are actually _ complex _ ( the spherical harmonics for @xmath170 are complex ) suitable combinations of them and their complex conjugates have been used to make plotting possible . these _ shapes _ are shared by _ all _ corresponding harmonics , i.e. , quadrupole profiles for example are those shown no matter the harmonic number @xmath171 of their frequency @xmath172 : simply , they oscillate faster for higher @xmath171 , but always keep the represented profile . [ fig6],width=7 ] the series expansion ( [ eq5.15 ] ) transparently shows that _ only _ monopole and quadrupole spherical modes can possibly be excited by an incoming gw . the monopole will of course not be excited at all if general relativity is the true theory of gravitation . a spherical solid is thus seen to be the _ best possible _ shape for a gw detector . this is because of the optimality of the _ overlap coefficients _ @xmath173 between the universal form factors @xmath126 and the sphere s eigenmodes @xmath174 , which comes about due to the clean _ multipole structure _ of the latter : only the @xmath158=0 and @xmath158=2 _ spheroidal _ modes couple to gws , hence _ all the gw energy _ is deposited into them , and only them . any other solid s shape , e.g. a cylinder , has eigenmodes most of which have some amount of monopole / quadrupole projections in the form of the coefficients ( [ eq5.8a ] ) , and this means the incoming gw energy is distributed amongst _ many modes _ , thus making detection less efficient . we shall assess quantitatively the efficiency of the spherical detector in terms of _ cross section _ values below . but , as just stated , quadrupole modes are _ more specifically , they are 5-fold degenerate , each degenerate wavefunction corresponding to one of the five integer values @xmath175 can take between @xmath176 and @xmath177 . monopole modes are instead non - degenerate . figure [ fig6 ] shows the shapes of all these modes @xcite see the caption to the figure for further details . _ degeneracy _ is a key concept for the _ multimode _ capabilities of the spherical detector . for , as explicitly shown by equation ( [ eq5.15 ] ) , monopole and quadrupole _ detector _ modes are driven by one and five _ gw amplitudes _ , respectively , i.e. , @xmath178 and @xmath179 . therefore , if one could _ measure _ the amplitudes of these modes , i.e. , the amplitudes of the deformations displayed in figure [ fig6 ] , then a _ complete deconvolution _ of the gw signal would be accomplished . this is a _ unique _ feature of the spherical antenna , which is not shared by any other gw detector : it enables the determination of _ all _ the gw amplitudes , not just a combination of them , no matter where the signal comes from . in section [ sub5.3 ] below i shall give a more detailed review of how the multimode capability can be implemented in practice . the general definition ( [ eq5.11 ] ) applies in this case , too . since cross section is a _ frequency _ dependent concept , and since quadrupole modes are degenerate , it is clear that energies deposited in _ each _ of the five degenerate modes of a given frequency harmonic must be added up to obtain @xmath180 for that mode . such energy must be calculated by means of volume integrals to add up the energies of all differentials of mass throughout the solid , the details of which i omit here . the final result turns out to be a remarkable one @xcite : cross sections factorise in the form @xmath181 where @xmath182 is a characteristic of the sphere s material[multiblock footnote omitted ] , and @xmath183 a dimensionless quantity associated with the @xmath184-th frequency harmonic ; finally , and this is the stronger theoretical point of this expression , @xmath185 is a coefficient which is characteristic of the underlying theory of gws , symbolically indicated with @xmath186 . for example , if the latter is general relativity ( gr ) then @xmath187 { \cal k}_2(\aleph ) = \mbox{\large $ \frac{16\,\pi^2}{15}$}\ ; , \end{array}\right . \label{eq5.19}\ ] ] while if it is e.g. brans - dicke @xcite then these expressions get slightly more complicated @xcite , etc . cc + cylinder & sphere + + @xmath188 = 910 hz & @xmath189 + @xmath190 & @xmath191 = 3.1 metres + @xmath192 = 2.3 tons & @xmath193 = 42 tons + @xmath194@xmath43@xmath195 cm@xmath141 hz & @xmath197 + ( * optimum orientation * ) & ( * omnidirectional * ) + sticking to gr , a few illustrative figures are in order . they are shown in table [ tab1 ] , where a material of aluminum al5056 alloy has been chosen . it appears that a sphere having the same fundamental frequency ( @xmath198 ) as a cylinder ( @xmath199 ) is about 20 times more massive , and this results in a significant improvement in cross section , since it is proportional to the detector mass . a spherical detector is therefore almost _ one order of magnitude _ more sensitive than a cylinder in the same frequency band obviously apart from the important fact that the sphere has _ isotropic _ sensitivity . . note the ( slightly ) better figure ( 17% ) for the lowest mode in the sphere , as well as the appreciable value in the second harmonic of the latter , in sharp contrast with the _ null _ coupling of this mode in the cylinder . third harmonics show a considerable reduction in sensitivity . [ fig7],width=9 ] but there is more to this . table [ tab1 ] also refers to the sphere s cross section in its _ second _ higher quadrupole harmonic frequency , @xmath200 almost twice the value of the first , @xmath198 . it is very interesting that cross section at this second frequency is _ only _ 2.61 times smaller than that at the first @xcite while , as stressed in section [ sub2.1 ] , it is _ zero _ for the second mode of the cylinder . figure [ fig7 ] shows a plot of the cross sections _ per unit mass _ of a cylinder and a sphere of like fundamental frequencies . it graphically displays the numbers given in the table , but also shows that , even per unit mass , the sphere is a better detector than a cylinder its cross section ` curve ' stays above the cylinder s . in particular , the _ first _ quadrupole resonance turns out to have a cross section which is 1.17 times that of the cylinder ( per unit mass , let me stress again ) @xcite , i.e. , @xmath20117% better . this constitutes the _ quantitative _ assessment of the discussion in the paragraph immediately following equation ( [ eq5.17 ] ) . in order to determine the actual gw induced motions of an elastic solid a _ motion sensing _ system must be set up . in the case of currently operating cylinders this is done by what is technically known as _ resonant transducer _ the idea of such device is to couple the _ large _ oscillating cylinder mass ( a few tons ) to a _ small _ resonator ( less than 1 kg ) whose characteristic frequency is accurately tuned to that of the cylinder . joint _ dynamics of the resulting system @xmath202cylinder + resonator@xmath203 is a two - mode _ beat _ of nearby frequencies given by @xmath204 where @xmath205 is the frequency of either oscillator when uncoupled to the other , and @xmath206 . the key concept of this device is the _ resonant energy transfer _ , which flows back and forth between cylinder and sensor with the period of the beat , i.e. , @xmath207 . this means that , because the sensor s mass is very small compared to that of the cylinder , the amplitude of its oscillations is enhanced by a factor of @xmath208 relative to those of the cylinder , whence a _ mechanical amplification _ factor is obtained _ before _ the sensor oscillations are converted to electrical signals , and further processed see a more detailed account of these principles in @xcite . the same principles can certainly be applied to make resonant motion sensors in a spherical antenna . in this case , however , a special bonus is there , associated to the _ degeneracy _ of the quadrupole frequencies : because all five quadrupole modes oscillate with the _ same _ frequency , it is possible to attach five ( or more ) identical resonators , tuned to a given quadrupole frequency , at suitable positions on the sphere surface , thus taking _ multiple samples _ of the sphere s motion . this makes possible to retrieve the oscillation amplitudes of the five degenerate modes figure [ fig6 ] , and thereby of the gw quadrupole amplitudes @xmath127 , since both are linearly related through equation ( [ eq5.15 ] ) . a _ single _ spherical antenna can thus deconvolve completely the quadrupole gw signal , and do so with _ isotropic sky coverage_. these characteristics are _ unique _ to the spherical detector , and they make it a _ theoretically _ superior system compared to either interferometers or weber bars . in addition , a sphere can naturally measure the amplitude of the non - degenerate monopole mode , as it is conceptually simple to sense the amplitude of an isotropically breathing pattern . , width=4 ] the conceptual idea of a resonant sensor is shown in figure [ fig8 ] , and the equations of motion for such a system are @xcite & = & ^2 + ( + ) ( ) + + & & _ b=1^j m_b_b^2 ^(3)(-_b)_b + _ gw(,t ) , [ eq5.20a ] + _ a(t ) & = & -_a^2 + _ a^gw(t ) , a=1, ,j , [ eq5.20b ] where @xmath209 and @xmath210 are the mass and characteristic frequency of the @xmath211-th resonator , @xmath212 is the gw tide on the sphere see ( [ eq5.6 ] ) , and @xmath213 is the gw induced tidal acceleration on the resonator itself , relative to the centre of the sphere ; @xmath214 is the three dimensional dirac density , i.e. , point - like connections between sphere and sensors are assumed . the mathematical detail of the analysis of these equations is somewhat sophisticated . the interested reader will find complete information in @xcite ; the rest of this section will be devoted to a brief discussion of the main conclusions of that analysis . first thing to stress is that equations ( [ eq5.20a]-b ) can not be solved analytically , they must instead be solved by a _ perturbative _ procedure . the small perturbation parameters are the ratios @xmath215 where @xmath114 is the sphere s total mass . actually , the analysis assumes that the resonators are all _ identical _ , any deviations from this being eventually assessed by suitable methods @xcite . the fundamental result links the spring deformations to the the gw amplitudes @xmath127 by the following formula , expressed in terms of _ laplace transforms _ noted with a caret ( @xmath216 ) : @xmath217 where it is assumed that the resonators frequency @xmath80 is tuned to either a monopole or a quadrupole harmonic of the sphere . the _ transfer functions _ @xmath218 naturally depend on whether a monopole or a quadrupole mode is selected for resonator tuning ; i will quote here only the quadrupole case , as it is the most interesting one see again full information in @xcite : @xmath219\right.\nonumber \\ & \times & \left.\frac{v_a^{(c)}v_b^{(c)*}}{\zeta_c}\right\}\ , y_{2m}(\vec{n}_b)\,\delta_{l2}\ ; , \label{eq5.23}\end{aligned}\ ] ] where @xmath220 is the @xmath221-th normalised eigenvector of the matrix @xmath222 , associated to its _ non - zero _ eigenvalue @xmath223 , @xmath224 is a legendre polynomial , and @xmath225 is the position of the @xmath226-th resonator on the sphere s surface . finally , @xmath227 equations ( [ eq5.22])([eq5.24 ] ) are the key to the gw signal deconvolution problem : they show that _ beats _ occur around the tuned frequency ( @xmath172 in this case ) , and that the resonators oscillate with amplitudes enhanced by a factor @xmath208 , as indeed expected . note that these beats have frequencies which depend on the resonators positions @xmath225 , as shown by the presence of the @xmath228 coefficient in ( [ eq5.24 ] ) . , width=9 ] the deconvolution problem consists in inferring the gw amplitudes @xmath229 from the telescope s readouts @xmath230 . thus _ at least _ 5 sensors must be attached to the sphere s surface if e.g. the 5 quadrupole amplitudes are looked for : once the corresponding five @xmath230 are _ measured _ the system ( [ eq5.22 ] ) is solved for @xmath231 , that s it . crucial at this point is to decide _ where _ to implant the resonators , as such decision bears fundamentally on the simplicity , or even the possibility of solving the posed problem . there are two major proposals in the literature for this , and they are displayed in figure [ fig9 ] . they permit the definition of so called _ mode channels _ , which are linear combinations of the system readouts @xmath230 which are directly proportional to the gw amplitudes @xmath231 . they happen to be of the form @xcite @xmath232 both for _ tiga _ and _ phc_. the actual result of these linear combinations is the following : @xmath233\,\hat g^{(2,m)}(s)\ ; , \label{eq5.26}\ ] ] i.e. , they are _ convolution products _ of the signal and the system beats . this formula appears to be very powerful , as it shows that suitable sensor systems enable a _ spherical detector to fully deconvolve _ all _ the gw amplitudes . one should however be careful about this conclusion , for the formula _ also _ indicates that the relevant information _ can only be obtained at the resonance frequencies _ @xmath234 in an ideal , non - dissipative system . in a real system , resonant linewidths are never infinitely sharp , they have instead a certain breadth ; this actually makes possible the observation across wider bandwidths , provided the amplifier noise can be kept sufficiently small i shall briefly come to this below in section [ sec.6 ] . the real merit of the just described spherical gw detector comes from its _ symmetry_. a _ hollow _ sphere does of course share the symmetry properties of a solid sphere , and one might therefore expect it to be an interesting alternative , too . a detailed study of the performance of a hollow spherical gw detector can be found in @xcite . the added bonus of a hollow sphere is that there is one more structural parameter one can adjust to enhance this or that property , and this is of course the _ thickness _ of the spherical shell . it appears that cross sections at e.g. different quadrupole harmonics have characteristic behaviours when plotted as functions of thickness ; also , one can decide to attach the motion sensors to the inner _ or _ to the outer side of the shell , as their gw induced oscillations have different amplitudes . , width=5 ] to actually build and suspend a hollow sphere in the laboratory may be a difficult task from a technological point of view . recently , though , a new concept spherical detector appeared in the literature @xcite : this is called _ dual sphere _ , and consists in a solid sphere nested inside a hollow one , concentric with it and with a narrow gap between them figure [ fig10 ] . an incoming gw drives _ both _ spheres into oscillation . clearly , the inner solid sphere has a higher first ( quadrupole ) mode frequency than the outer hollow one , therefore an incoming gw with a frequency _ between _ those two will drive the oscillations of the spheres with opposite phase , say ) to above it ; the transition region has a width of order @xmath235 centred at @xmath80 , where @xmath236 is the oscillator s mechanical quality factor , about 10@xmath237 or more in gw detectors . ] , thus enhancing the signal by a rough factor of 2 . the motion sensing in either hollow or dual spheres is conceptually analogous to that in the solid sphere , with the added flexibility that in the hollow piece one can choose to sense the displacements of either side of the shell . actually , though , it appears that non - resonant detectors appear constitute a better choice in dual spheres , for this enables a significant bandwidth enlargement @xcite . let me briefly discuss now , for completeness , a few essentials of gw detector sensitivity . so far i have only discussed the _ theoretical _ basis of the workings of gw detectors , whether interferometric or acoustic , yet have made no mention of the _ practical _ difficulties of getting them actually working the extreme weakness of any expected gw signals arriving in the earth @xcite is in fact a source of such truly difficult problems that it has prevented gw detectors from sighting a real signal in the last 40 years , since the times of j. weber . detector noise _ has to date overwhelmed any signals possibly hitting the antennas , and therefore the technological challenge has been for years , and still is today , to reduce that noise to the level where it can be filtered out with a meaningful signal to noise ratio @xcite . during the last decade or so , a number of people in different countries worldwide have managed to get important gw detection research projects funded which constitute a major step forward in detector technology . their goal is to improve the sensitivity to the point where a significant _ event rate _ becomes available to the gw astronomer . we thus find such laboratories as _ virgo _ ( a french - italian collaboration ) , _ ligo _ ( usa alone ) , or _ lisa _ ( a space mission , jointly funded by _ nasa _ and _ esa _ , the european space agency ) . in addition to those , somewhat smaller experiments are _ geo-600 _ ( a german - british venture ) and _ tama _ ( the japanese project , currently making strong progress in both and stability ) . so much for _ interferometric _ antennas . but endeavours have not declined in the _ acoustic detector _ arena , either . in fact , the five acoustic detectors of the weber type ( _ nautilus , explorer , auriga , allegro and niobe _ ) constitute the _ only working systems _ in the world today . unfortunately , though , they are only sensitive to catastrophic events happening in our galaxy , with a far too low occurrence rate . these detector systems are periodically upgraded , and their sensitivity has gone up a few orders of magnitude since their origins . as already stressed in section [ sec.1 ] , new generation _ spherical _ gw detectors are being programmed in brazil , holland and italy , and these should suddenly improve over bars by at least one more order of magnitude . , width=10 ] figure [ fig.11 ] contains a recent plot of gw sensitivity of various earth based detectors , therefore in the frequency range near 1khz . on the other hand , _ lisa _ will be sensitive at frequencies far away from the range plotted in figure [ fig.11 ] see _ lisa _ s web site at http://sci.esa.int/home/lisa and figure [ fig.12 ] below . what we see in ordinates in plots like the one in figure [ fig.11 ] , which are standard in gw science , is a _ spectral density _ or , rather , its _ square root_. it is to be understood as follows . _ noise _ of whatever origin causes any detector to generate random outputs _ stochastic _ time series , as this is technically known . this noise competes with any gw signals which may be present in the antenna readout , obscuring their detection . noise of course does not come from the sky , it rather gets added to the gw signal at almost all the different stages of the detection process . for example , the gw induced oscillations of a given mass compete with thermal oscillations of that mass and its suspension systems ; then there is noise in the conversion from mechanical or optical gw signals to electrical signals ; then there is noise introduced by the electronics in the amplifiers of the latter ; and so forth in the end , the detector output is a certain physical magnitude ( e.g. a voltage ) which consists of a number of various superimposed sources of noise plus , possibly , a gw signal _ converted _ to volts by the detector hardware . one infers the value of the _ actual _ dimensionless gw amplitude @xmath238 , say , from the output voltage if one knows the precise physics of the transduction process and this is obviously the case in any useful instrument . conversely , it is also possible to _ back - convert _ all the sources of noise voltage picked up across the various detector stages to an equivalent dimensionless `` gw noise '' , which gets directly added to real gw signals , and travels through an ideally noiseless detector . this artifact is expedient because , for a fixed antenna , it enables a quick assessment of the detectability of a given gw signal , normally calculated by methods of gravitation theory , by direct comparison with suitably constructed detector characteristic curves such as those in figure [ fig.11 ] . let us then call @xmath239 any one of the detector s readout channels , _ back - converted _ to a gw amplitude by the above described procedure . we split this up into a gw signal proper @xmath238 plus a noise term @xmath240 : @xmath241 for stationary gaussian noise , the statistical properties of @xmath240 are encoded in its _ spectral density _ function , @xmath242 , which is the fourier transform of the _ autocorrelation function _ @xmath243 where @xmath244 stands for _ ensemble average _ it can be shown @xcite that the _ optimum filter _ to extract the signal @xmath238 from the system readout @xmath239 is the so called _ matched filter _ , whose transfer function is the signal s fourier transform @xmath245 divided by the spectral density @xmath242 , and the _ detection threshold _ can be set from the integrated signal to noise ratio : @xmath246 the gw signal @xmath238 is a dimensionless quantity , as it measures a perturbation of the minkowski metric , see equation ( [ eq2.2 ] ) . therefore the spectral density @xmath242 has dimensions of time , or inverse frequency , hz@xmath247 , according to the definitions ( [ eq5.28 ] ) . now , signal to noise ratio as defined by the integral ( [ eq5.29 ] ) is made up of the contributions of the ratio between the signal power @xmath248 to the noise power @xmath242 at all frequencies ; the idea of the graphical representation in figure [ fig.11 ] is thus to show which is the level of noise at different frequencies by means of an rms quantity , such as is the square root of the noise spectral density . the appropriate units for this representation are accordingly hz@xmath249 , as indicated . , width=12 ] different interesting sources of gws are being considered by other authors in this volume , so i will not go into such matters here . it is nevertheless instructive to present an example graph of a few signals on top of various detectors sensitivity curves in order to get a picture of the actual possibilities of each instrument , and also to grasp what are the spectral orders of magnitude of different gw signals . one such plot is presented in figure [ fig.12 ] . this is , let me insist , the standard way to assess the detectability of the different gw sources . this article is a brief review on the nature of the interactions of gws with _ test _ particles and _ test _ electromagnetic fields , as they specifically happen in currently conceived detection devices . while the fundamental principles are not new in themselves , their application in actual systems is still in many cases subject of research , as we have seen . a thorough understanding of these matters is absolutely essential for an adequate interpretation of the antenna readouts , the more so if one considers the extreme _ weakness _ of any signals reaching us from even the most intense sources . i have omitted any detailed discussion of the practical problems faced by real detector building . this is a major research field in itself , of an intrinsically sophisticated and multidisciplinary nature involving such fundamental issues as quantum measurement limits and techniques , or quantum optics @xcite . but it is of course not directly related to astrophysics or relativity i have however considered appropriate to summarily brief the interested reader on how detector sensitivities are defined , and on which are the detection thresholds in current state of the art gw detectors : this is a key issue for an astrophysicist / relativist wishing to assess the detectability of a given gw signal , whether by existing instruments or future planned . gw detection endeavours have been the subject of intensive research during the last 4 decades , though the ultimate goal of sighting gw events has still not been accomplished . while this may look like a major failure , one may not forget that detector sensitivities have gone up by a remarkable _ six orders of magnitude _ ( in energy , three in gw amplitude ) since the very first telescopes constructed by j. weber . also , a look at the trend in this development indicates that we are getting closer to the obejctive . j. weber : ` supernova 1987a rome and maryland gw radiation antenna observations ' . in : _ gravitational wave experiments _ , proceedings of the 1st edoardo amaldi conference , ed . by e. coccia , g. pizzella , f. ronga ( world scientific , singapore 1995 ) , p. 416 aguiar _ et al . _ : ` the first phase of the brazilian graviton project ' . in : _ gravitational waves _ , proceedings of the 3rd edoardo amaldi conference , ed . by s. meshkov ( world scientific , singapore 2000 ) , p. 413 paik : ` electromechanical transducers and bandwidth of resonant mass gw detectors ' . in : _ gravitational wave experiments _ , proceedings of the 1st edoardo amaldi conference , ed . by e. coccia , g. pizzella , f. ronga ( world scientific , singapore 1995 ) , p. 201 lobo : ` spherical gravitational wave detectors and geometry ' . in : _ gravitational waves _ , proceedings of the 2nd edoardo amaldi conference , ed . by e. coccia , g. pizzella , g. veneziano ( world scientific , singapore 1998 ) , p. 168 thorne : ` gravitational radiation ' . in : _ 300 years of gravitation _ , ed . by s.w . hawking and w. israel ( cambridge university press , cambridge 1988 ) , see also more recent reviews by k.s . thorne , e.g. gr - qc/9704042 , or visit the website ` http://fermi.phys.ualberta.ca/~ccgrra/thorne/index.html `

the classical theory of dispersion is a subject with a long and noble history @xcite . although the main features of the phenomenon can be described by treating matter as a continuum characterized by macroscopic quantities such as the electric and magnetic polarizations , it is clear that a truly fundamental theory has to be based on a microscopic model of matter . we shall now try to summarize some crucial aspects of the problem in an historical perspective , before illustrating the new features of our present approach . by treating an elementary electric dipole as an oscillator subject to a linear restoring force , it is possible to obtain a simple expression for the molecular polarizability , i.e. the complex frequency - dependent linear coefficient which relates the microscopic dipole moment to the amplitude of the incident electromagnetic radiation . in order to correctly apply this simple model to the description of the behavior of a large system of mutually interacting dipoles , one has however to consider that the field acting on each microscopic oscillator can not be simply identified with the macroscopic electromagnetic field in the medium . in fact , while the latter simply represents the average of the microscopic field over a region much larger than the intermolecular spacing , the former has to be carefully calculated by evaluating and summing , on the site occupied by the considered dipole , the retarded fields generated by all the other dipoles of the medium . this `` exciting '' field ( as we shall refer to in the following , although the names `` effective '' or `` local '' field have also been employed in the literature ) was theoretically estimated by lorentz already at the end of nineteenth century @xcite by dividing the medium into two regions separated by a virtual sphere surrounding the considered dipole . he restricted his attention to the typical situation in which the wavelength of the macroscopic electromagnetic field is of a larger order of magnitude than the average intermolecular spacing , so that one can take for the virtual sphere a radius intermediate between the two . he then argued that the influence of the portion of the medium lying outside the sphere can be fairly approximated as that of a continuous distribution of electric dipole moment , whereas the sum of the forces exerted by all the dipoles situated inside the sphere can be assumed to vanish in most cases on the basis of symmetry considerations . a rather similar analysis , leading to equivalent conclusions , was also carried out by planck @xcite . with these arguments one can derive the well - known lorentz lorenz formula @xcite , relating the macroscopic dielectric constant of the medium to the molecular polarizability , and it is thus possible to deduce an approximate expression for the dispersion relation of an array of oscillators in the long - wavelength regime ( which generally includes the optical frequencies ) . a detailed microscopic theory of dispersion in a crystalline solid , although with neglect of radiation reaction , was formulated by ewald @xcite . he considered a rectangular parallelepiped as the unit cell of the bravais lattice , and his results were subsequently generalized by born to more general crystal structures @xcite . the mathematical methods used by these authors ( one has to keep in mind that the theory of distributions was not yet existing at that time ) led however to rather clumsy expressions for the exciting field , which could be numerically evaluated only in the limit of an infinitely large ratio between wavelength and lattice constant , i.e. still essentially in the continuum approximation . in this way the previous results by lorentz and planck were recovered for structures with tetrahedral symmetry . furthermore , in the case of parallelepipeds of unequal edges , ewald was able to perform in the same limit a numerical calculation relating the ratio between the edges to the phenomenon of double refraction . finally , ewald extended his analysis of the model also to the study of x - ray diffraction @xcite , but he made use to this purpose of other important simplifications which are possible only in the opposite limit of a radiation frequency much higher that the characteristic frequencies of the crystal . many investigations were later devoted to the application of quantum mechanics to the theory of light dispersion , and the results of ewald and born were apparently considered to be the final word about the problem of the mutual interaction of a large array of classical resonators . we are going to prove that , on the contrary , a deeper analysis reveals important properties of this fundamental dynamical system which have been for many decades completely overlooked . in the present paper we shall study a system of infinitely many charged particles , subject to linear restoring forces towards their equilibrium positions at the sites of a regular lattice , and interacting with each other through the retarded electromagnetic fields . our outset will therefore be similar to ewalds s , but with inclusion in the equations of motion of the usual `` triple - dot '' radiation reaction term , which corresponds to the nonrelativistic form of the lorentz dirac equation @xcite . the only approximation that we shall use is that of small oscillations : this will allow us still to deal with a system of linear equations . we shall provide a general and rigorous procedure for the calculation of the exciting field , avoiding to introduce at any stage of the procedure the continuum approximation . this will be accomplished by a method involving the careful subtraction of two divergent quantities ( representing respectively the total field and the field generated by the dipole under consideration ) , which appears to be more powerful than that used by ewald and born , and presents some formal analogy with the renormalization techniques of quantum field theory . using our procedure we shall show that for an infinite regular lattice there exists a continuous set of normal modes which describe the propagation of plane electromagnetic waves . a remarkable result will be that the inclusion of the radiation damping term in the equations of motion of the oscillating particles , instead of giving rise to an extinction coefficient for the wave , as is commonly believed according to the standard approximated treatments of the model , is on the contrary essential for justifying the presence of undamped collective waves . such a result in fact constitutes an extension to the three dimensional case of an analogous one already obtained by two of the present authors for the case of a rectilinear chain of one - dimensional oscillators @xcite . it relies upon a remarkable identity which was originally formulated in a different context by wheeler and feynman @xcite . these authors deduced it from the hypothesis of the `` complete absorber '' , which they introduced in order make their time - reversible action - at - a - distance electrodynamics compatible with the observed phenomenon of radiation reaction . for the physical system here considered we are going to prove in a simple and direct way that , although no absorption mechanism is present in the model , this `` wheeler feynman identity '' actually holds as a purely mathematical property of the entire class of solutions on which we are interested . by numerically studying the dispersion relations for the crystal , as resulting from the exact solutions of the model , it will also be shown that completely new features appear for frequencies in a region about the proper frequency of the oscillators . in such a region the wavelength can in fact become as short as the lattice spacing , so that the approximations adopted in the previous literature become unavailable , and only an exact solution can give predictions about the behavior of the system . it is found that inside the interval of frequencies where undamped wave propagation was believed to be impossible , plane waves can actually propagate , with very low group velocities , along certain lattice directions and for appropriate wave polarization . let us consider an infinite three - dimensional simple bravais lattice , that is an array of points @xmath0 where @xmath1 denotes the triple of relative integers @xmath2 , and @xmath3 , @xmath4 , @xmath5 are a set of primitive translation vectors for the lattice @xcite . we choose the orientation of the @xmath6 in such a way that @xmath7 . @xmath8 clearly represents the volume of the primitive cell . each point @xmath9 is the equilibrium position of a point particle ( electron ) of mass @xmath10 and electric charge @xmath11 , which is subject to an elastic force of the form @xmath12 where the vector @xmath13 represents the instantaneous coordinates of the point particle . although it is not strictly necessary for the mathematical self - consistency of the model , in order to reproduce in a more realistic way the situation of solid state physics we can assume that @xmath9 is also the seat of a static positive ion of charge @xmath14 . as is traditionally the case for the classical models of dispersion , we shall neglect the coulomb interaction between the ion and its associated electron , since classical mechanics fails at such short distance scales , and we shall instead identify the proper oscillator frequency @xmath15 with a characteristic excitation frequency of the optical electron in the atomic ground state . according to quantum mechanics , it could even be possible to obtain a more realistic model by associating to each atom a set of fictitious oscillators @xcite , one for each allowed quantum transition from the ground state of energy @xmath16 to an excited state of energy @xmath17 , with proper frequencies @xmath18 . in order to assign to the contribution of each oscillator an appropriate weight , one has then to make the substitution @xmath19 , where the `` oscillator strengths '' @xmath20 are coefficients subject to the sum rule @xmath21 ( we are considering here atoms with a single optically active electron ) . an explicit calculation from first order perturbation theory provides @xmath22 @xmath23 being the position operator and @xmath24 the unit vector representing the direction of oscillation . for the sake of simplicity we shall consider in the following calculations a single oscillator per atom , in accordance with the original literature on classical dispersion , although the extension to the case of multiple oscillators presents no conceptual difficulty . the charge and current densities associated with the electron - ion pair are given respectively by @xmath25 \label{rho } \\ { \bf j}_{{\bf n}}({\bf x},t)&= & e{\bf \dot{z}}_{{\bf n}}(t ) \delta ^{3}({\bf x}-{\bf z}_{{\bf n}}(t))\ , , \label{jey}\end{aligned}\ ] ] where @xmath26 denotes dirac s delta function . the retarded potentials generated by the charge - current density @xmath27 ( with @xmath28 ) are defined as @xmath29 where @xmath30 is the retarded green function @xcite and @xmath31 is the fourier transform of @xmath32 . we are here using the four - dimensional notation so that , for instance , @xmath33 denotes the four - vector @xmath34 , and @xmath35 , @xmath36 , @xmath37 . summation over repeated indices is always implicitly understood . the retarded potentials satisfy the equation @xmath38 and the lorentz gauge condition @xmath39 introducing then the retarded fields according to the usual relations @xmath40 and putting @xmath41 , we can write the ( nonrelativistic ) equation of motion of the electrons as @xmath42 where @xmath43 represent the exciting fields . the notation used in the two last equations means that the summation index @xmath44 runs over all the values in @xmath45 except @xmath46 . the last term in eq.([eqmoto ] ) describes the radiation reaction force , according to the lorentz dirac prescription . see ref . @xcite for a comparison between the expression of the atomic polarizability resulting from this classical equation in the dipole approximation , and the corresponding result obtained for a two - level atom in electric - dipole interaction with the quantized electromagnetic field . we suppose the amplitude of the oscillations to be small enough , so that at every stage we can neglect all terms of order higher than one in the @xmath47 and their time derivatives of any order . it follows that eq . ( [ eqmoto ] ) simplifies to @xmath48 where the retarded fields , included into @xmath49 according to eq . ( [ erregende ] ) , are to be calculated in the dipole approximation , whereby each @xmath50 depends linearly on @xmath51 and its time derivatives . of course , eq . ( [ eqmotod ] ) actually represents an infinite system of coupled linear equations , since we have one such equation for each @xmath52 . we shall look for a global solution of the form ( in customary complex notation ) @xmath53\ , , \label{normmode}\ ] ] representing a plane wave with amplitude @xmath54 , frequency @xmath55 and wavevector @xmath56 . the parameters @xmath55 and @xmath56 can always be chosen so that @xmath57 and @xmath56 belongs to the first brillouin zone of the crystal . we shall proceed as follows : for a generic motion of the form ( [ normmode ] ) we shall calculate the resulting expression for the exciting field @xmath49 ; then by substituting this expression into eq . ( [ eqmotod ] ) we shall find out that the equations of motion of all the particles can be simultaneously satisfied , provided that @xmath56 and @xmath55 satisfy a well defined dispersion relation . in the dipole approximation , that is to first order in @xmath54 , the charge and current densities become @xmath58 \\ { \bf j}_{{\bf n}}({\bf x},t ) & = & -ie\omega { \bf c } \delta ^{3}({\bf x}-{\bf r}_{{\bf n}})\exp \left [ i(\bm{\kappa } \cdot { \bf r}_{{\bf n}}-\omega t)\right]\end{aligned}\ ] ] whence , according to the definition ( [ jtilde ] ) @xmath59 \label{rhojtilde } \\ { \bf \tilde{j}}_{{\bf n}}(k ) & = & -2\pi ie \omega { \bf c}\delta ( k^{0}-\omega /c)\exp \left [ i(\bm{\kappa}-{\bf k}){\bf \cdot r}_{{\bf n } } \right]\ , . \label{jeytilde}\end{aligned}\ ] ] substituting these expressions into eq . ( [ potential ] ) and then applying eq . ( [ field ] ) we get @xmath60 \nonumber \\ & & \times \int \frac{d^{3}{\bf k}}{(2\pi ) ^{3}}\,\frac { ( \omega /c)^{2}{\bf c}-({\bf k}\cdot { \bf c}){\bf k } } { { \bf k}^{2}-(\omega / c)^{2}-i\varepsilon } \nonumber \\ & & \times\exp \left [ i{\bf k\cdot ( x}-{\bf r}_{{\bf m}})\right]\ , . \label{eenne}\end{aligned}\ ] ] it follows @xmath61 \ , , \label{erreg1}\ ] ] where the second rank tensor @xmath62 has components @xmath63\end{aligned}\ ] ] and we have used the notation @xmath64 to denote the vector with components @xmath65 . by substituting the expressions ( [ normmode ] ) and ( [ erreg1 ] ) into eq . ( [ eqmotod ] ) , the original system of infinitely many coupled equations is transformed into the single vectorial equation @xmath66 which admits solutions with nonvanishing @xmath54 when @xmath67 = 0\ , . \label{determ}\ ] ] the above equation determines an implicit relation between @xmath56 and @xmath55 , which constitutes the sought for dispersion relation of the crystal . once a solution of the form ( [ normmode ] ) has been obtained , it is easy to write down the expression for the macroscopic quantities which can be associated to the normal mode . the macroscopic polarization density is in fact given by the continuous function of space which interpolates the microscopic displacement vectors of the individual dipoles : @xmath68\ , . \label{pmacr}\ ] ] the macroscopic fields @xmath69 , @xmath70 and @xmath71 are then given by the solutions of the usual macroscopic maxwell equations inside the ( nonmagnetic ) medium : @xmath72 from eq . ( [ maxwell1 ] ) it follows @xmath73 . then by taking the curl of eq . ( [ maxwell2 ] ) and eliminating @xmath74 with the aid of eq . ( [ maxwell4 ] ) one easily obtains @xmath75 from this , eq . ( [ pmacr ] ) and eq . ( [ maxwell2 ] ) , we conclude @xmath76 \label{emacro } \\ { \bf b}_{\rm mac}({\bf x},t)&=&\frac{e}{v}\frac{(\omega /c ) \bm{\kappa } \times { \bf c}}{\bm{\kappa}^2-(\omega /c)^2 } \nonumber \\ & & \times\exp \left [ i(\bm{\kappa}\cdot { \bf x}-\omega t)\right]\,.\end{aligned}\ ] ] since ( [ determ ] ) is a complex equation , it is _ a priori _ to be expected that , in order that a real solution for @xmath55 may exist , the vector @xmath56 must necessarily be assigned an imaginary component , which represents an extinction coefficient for the wave . a fundamental observation can however be made at this point , showing that this is not actually the case and that eq . ( [ determ ] ) determines @xmath55 as a real function of the real independent variable @xmath56 . to this purpose , let us introduce the advanced potentials @xmath77 defined by a formula analogous to ( [ potential ] ) with , in place of @xmath78 , the advanced green function @xmath79 a completely general result about the lorentz dirac equation asserts that the self - force , given by the expression involving the triple time - derivative of the particle position , is equal to the electromagnetic force exerted on the particle by one half the difference between the retarded and advanced fields generated by the particle itself @xcite . using this result in the dipole approximation , we have that the last term of eq.([eqmotod ] ) can be expressed as @xmath80 where @xmath81 and @xmath82 . note that the field @xmath83is a solution of the homogeneous ( i.e. source - free ) field equation and therefore , at variance with @xmath84 , it is regular at the particle position @xmath9 . using eq . ( [ dirac ] ) and the identity @xmath85 , we can rewrite eq . ( [ eqmotod ] ) as @xmath86 where @xmath87 the summation being extended to all the values @xmath88 in the last equation . we have @xmath89 with @xmath90 \nonumber \\ & = & \frac{\delta ( x^{2 } ) \varepsilon ( x^{0})}{4\pi } = \frac{i}{2(2\pi ) ^{3}}\int d^{4}k\,e^{ik\cdot x } \delta ( k^{2})\varepsilon ( k^{0})\end{aligned}\ ] ] and @xmath91 for a normal mode solution the above threefold series can be evaluated by using eqs . ( [ rhojtilde]-[jeytilde ] ) and the relation @xmath92 where the @xmath93 , for @xmath88 , are the points of the reciprocal lattice @xcite , defined as @xmath94 here @xmath95 indicates the completely antisymmetric tensor with @xmath96 . we obtain @xmath97 we see that the integrand on the r.h.s . of eq . ( [ ameno ] ) is a singular function with support on the cone @xmath98 . on the other hand , for a given real @xmath56 , the functions @xmath99 and @xmath100 are different from zero on this cone only when there exists @xmath88 such that @xmath101 this proves that for all values of @xmath55 , except those belonging to the discrete set of singular values @xmath102 , the `` wheeler feynman identity '' @xmath103 holds at any point @xmath33 of spacetime . on the other hand , the results that will be obtained in the next section show immediately that , for a given @xmath56 , the exciting field diverges for @xmath104 , so that none of these frequency values can possibly correspond to a normal mode solution . the wheeler feynman identity is therefore established in complete generality for all physical solutions expressible as linear combinations of normal modes . as an immediate consequence of this identity one has that the last term on the r.h.s . of eq.([eqmotod2 ] ) vanishes . recalling eq . ( [ dirac ] ) , this result can also be put in the form @xmath105 a relation physically equivalent to this one was obtained by oseen ( although without a rigorous mathematical proof ) already in 1916 @xcite . we can then write @xmath106\,,\ ] ] with @xmath107 \,.\end{aligned}\ ] ] the symbol @xmath108 indicates the principal value of the integral . note that for real @xmath56 and @xmath55 the function @xmath109 is real , since taking the complex conjugate amounts to making the substitution @xmath110 @xmath111 in the summation index . from these considerations it follows that the equation of motion ( [ eqmotod1 ] ) can be rewritten as @xmath112 and the original complex equation ( [ determ ] ) is converted into the real equation @xmath113 = 0\,,\ ] ] which determines a dispersion relation between the real variables @xmath114 and @xmath55 . note that this remarkable result , which allows for the propagation of undamped plane waves in the crystal , holds just as a consequence of the inclusion of the lorentz dirac radiation reaction term in the equations of motion . this does not appear surprising , when one recalls that the expression of this term was determined just in order to insure global energy conservation for the complete system of particles and field @xcite . the expression for the retarded field produced by an oscillating dipole is well - known , and it can in fact be obtained by explicit calculation of the integral on the r.h.s . of eq . ( [ eenne ] ) . it might seem therefore that the most direct way of calculating the exciting field @xmath49 would be to substitute such an expression into eq . ( [ erregende ] ) , as was done in ref.@xcite for the one - dimensional case . it turns out however that in the three - dimensional case the series on the r.h.s . of eq.([erregende ] ) does not converge in a proper sense . it is possible indeed to assign to the sum an unambiguous meaning via the prescription @xmath115 but the above formula is not convenient for numerical computations purposes ( the sum converges slowly for nearly vanishing @xmath116 ) , and furthermore gives us no insight into the physical content of the results . we shall therefore follow a different path , which consists in converting the sum over the points @xmath9 into a sum over the points @xmath93 of the reciprocal lattice . after some manipulations we shall obtain an absolutely convergent series , which in typical cases can be numerically computed with little effort . furthermore , our final expression will provide an explicit expansion in powers of @xmath117 ( @xmath118 being the mean lattice parameter and @xmath119 the wavelength ) , in such a way that the well - known result provided by the old theories in the long - wavelength limit will appear to be just the zero - order approximation of the general result . we start from the relation @xmath120 \ , , \label{erreg2}\ ] ] where @xmath121 is the total retarded field . using eqs . ( [ eenne ] ) and ( [ recipro ] ) we obtain @xmath122 ( { \bf g}_{{\bf m}}+\bm{\kappa})}{({\bf g}_{{\bf m}}+ \bm{\kappa})^{2}-(\omega /c)^{2}-i\varepsilon } \nonumber \\ & & \times\exp \left [ i({\bf g}_{{\bf m}}+\bm { \kappa } ) \cdot { \bf x } -i\omega t\right ] \ , . \label{etot}\end{aligned}\ ] ] note that , according to eq . ( [ emacro ] ) , the term for @xmath123 of the above series is equal to the macroscopic field @xmath124 . this of course corresponds to the fact that , for @xmath125 , the macroscopic field just represents the average over a unit lattice cell of the total microscopic retarded field . we observe also that @xmath126 , so that @xmath127 \\ & = & \exp [ i\bm { \kappa } \cdot { \bf r}_{{\bf n}}+i({\bf g}_{{\bf m } } + \bm{\kappa})\cdot ( { \bf x}-{\bf r}_{{\bf n}})]\,.\end{aligned}\ ] ] therefore , substituting eqs . ( [ etot ] ) and ( [ eenne ] ) into eq . ( [ erreg2 ] ) , and then comparing the resulting expression with eq . ( [ erreg1 ] ) , we get @xmath128 \equiv m^{{\rm ret}}_{ij}({\bf q},f)\,,\end{aligned}\ ] ] where we have introduced the dimensionless variables @xmath129 , @xmath130 , and @xmath131 . let us now shortly rewrite the r.h.s.of the above equation as @xmath132 , with @xmath133 . the regularity of the function @xmath134 at @xmath135 descends obviously from the fact that @xmath136 is regular at @xmath137 , as can be seen from the very definition ( [ erregende ] ) of @xmath49 . recalling that @xmath138 we can write @xmath139 reintroducing the explicit expression for @xmath140 , we thus obtain @xmath141 e^{-\eta { \bf k}^{2 } } \label{mret0}\end{aligned}\ ] ] with @xmath142 by isolating the term for @xmath143 of the series on the r.h.s.of eq . ( [ mret0 ] ) we get the expression @xmath144 which is related to the macroscopic field inside the crystal according to eq . ( [ emacro ] ) : @xmath145\ , . \label{emacro2}\ ] ] we can then write @xmath146 \,,\end{aligned}\ ] ] with @xmath147 the series @xmath148 and the integral @xmath149 are both divergent in the limit @xmath150 . we are going to study them separately and to split each of them into a divergent and a convergent part . the two divergent parts must of course cancel each other , as we shall check directly . we shall then be able to express @xmath151 as the sum of a finite term and of an absolutely convergent series . by expanding the function @xmath152 in powers of @xmath153 ( from now on @xmath154 stands for @xmath155 ) we can write @xmath156 where , for @xmath157 , @xmath158 is a homogeneous function of @xmath159 of degree @xmath160 , and @xmath161 for @xmath162 . the term of degree @xmath163 has been multiplied by @xmath164 [ where @xmath165 for @xmath166 , @xmath167 for @xmath168 ) ] in order that all terms appearing in the above equation be integrable in a neighborhood of @xmath169 . we have @xmath170 \bigg\ } \\ c_{ij}^{(3 ) } & = & \frac{1}{k^{4}}\bigg\ { -2{\bf q}\cdot { \bf k } ( q_{i}q_{j}-f^{2}\delta _ { ij } ) \\ & & + ( q_{i}k_{j}+q_{j}k_{i})\left [ f^{2}-{\bf q}^{2 } + \frac{4({\bf q}\cdot { \bf k})^{2}}{k^{2}}\right ] \\ & & -4\frac{k_{i}k_{j}}{k^{2 } } { \bf q}\cdot { \bf k}\left [ f^{2}-{\bf q}^{2 } + \frac{2({\bf q}\cdot { \bf k})^{2}}{k^{2}}\right ] \bigg\ } .\end{aligned}\ ] ] this decomposition , besides separating the terms which give rise to divergent contributions to @xmath148 and @xmath149 , leads naturally to an asymptotic expansion of these quantities for long wavelengths . in fact , if we suppose that the refraction index @xmath171 is of order unity ( as it is reasonable to expect for frequencies not too close to resonance ) , we have @xmath172 , and it is immediate to check that @xmath173 , @xmath174 for @xmath175 . introducing @xmath176 , we can rewrite eq . ( [ recipro ] ) in terms of dimensionless quantities as @xmath177 let us then denote with @xmath178 the fourier transform operator , which transforms the generic function @xmath179 into the function @xmath180 ( { \bf x})=\int d^{3}{\bf k}\,e^{i{\bf k } \cdot { \bf x}}f({\bf k})\,.\ ] ] if the function @xmath181 is continuous at all points @xmath182 , we have from eq . ( [ recipro2 ] ) @xmath183 ( 2\pi { \bf y}_{{\bf n}})\ , . \label{fourier}\end{aligned}\ ] ] applying this formula and eq . ( [ gauss ] ) we obtain @xmath184 for @xmath185 , where @xmath186 indicates a term such that @xmath187 for any arbitrarily large @xmath188 . let us then decompose @xmath189 as @xmath190 where @xmath191 is a quadratic symmetric tensor satisfying @xmath192 . since @xmath193 is discontinuous for @xmath194 , we can not directly apply eq . ( [ fourier ] ) to evaluate its contribution to the series @xmath148 . however , since @xmath195 we can use eq . ( [ fourier ] ) to obtain @xmath196 it follows that @xmath197 for @xmath185 , where @xmath198 is a finite symmetric dimensionless tensor , satisfying the condition @xmath199 . the numerical values of @xmath200 in general depend on the particular crystal structure considered . in appendix a we derive the following equivalent expression in terms of a sum over the points of the direct lattice : @xmath201 from eqs . ( [ fourier0 ] ) and ( [ beta0 ] ) we conclude @xmath202 + \beta_{ij}^{(0)}+o(\eta ^{\infty } ) \ , . \label{c0s}\end{aligned}\ ] ] on the other hand , it is almost immediate to see that @xmath203 in fact , for obvious symmetry reasons , the l.h.s . must be of the form @xmath204 , but then the condition @xmath205 implies @xmath206 . it follows that @xmath207 and so @xmath208 we thus conclude that at order zero in @xmath209 the tensor @xmath151 is given by @xmath210 for an isotropic crystal the optical behavior at long wavelengths must be invariant under spatial rotations . this means that the symmetric tensor @xmath211 must be a multiple of the identity matrix , but since its trace @xmath212 vanishes , we see that isotropy implies @xmath213 . in such a case , recalling eqs . ( [ pmacr ] ) and ( [ emacro2 ] ) , we derive from eq . ( [ m0 ] ) that at this order of approximation @xmath214 the above expression for the exciting field is the same that was obtained by lorentz and planck with the argument of the virtual sphere mentioned in the first section , and was later confirmed through more rigorous analysis by ewald and born . the same matrix @xmath215 of eq . ( [ m0 ] ) of course also provides the zeroth order approximation for the tensor @xmath216 . therefore recalling eq.([eqmotoel ] ) one obtains @xmath217 where @xmath218 is the so - called `` plasma frequency '' of the material . the above equation can be put into the form @xmath219 , where the tensor @xmath220 represents the electric susceptibility . the dielectric function can then by obtained as @xmath221 . it appears from eq . ( [ chi ] ) that at the present order of approximation the lattice of resonators behaves in general as a biaxial crystal . let us denote with @xmath222 ( @xmath223 ) the three ( in general distinct ) eigenvalues of the symmetric tensor @xmath224 , with @xmath225 the eigenvalues of @xmath226 , and with @xmath227 the indexes of refraction for transversal waves with electric fields polarized along the mutually orthogonal directions of the corresponding eigenvectors . we then have that , for @xmath228 , the ( approximately ) frequency independent quantities @xmath229 that were introduced by havelock @xcite as a measure of the phenomenon of structural double refraction , can be calculated according to our model as @xmath230 according to eqs . ( [ beta0 ] ) or ( [ beta0x ] ) these parameters depend solely on the lattice structure . as we mentioned in section i , ewald was able to devise a method to calculate @xmath231 when the unit cell of the lattice is a rectangular parallelepiped @xcite . although his formulas look considerably more complicated , we have checked that the numerical results , that he obtained for particular values of the ratio between the edges of the cell , are in very good agreement ( considering the tools available at that time for numerical computation ) with the general result provided by eq . ( [ dij ] ) . in the following subsection we are going to show that eq.([eqmacr ] ) has to be significantly corrected when the size of the lattice parameter is not negligible with respect to the wavelength . it will be found that it is no longer possible in that case to give for the dielectric tensor a general expression as a function of the frequency only , since , as was already recognized by ewald @xcite , the form of the dispersion relation involves both the direction of polarization and the direction of the wavevector @xmath56 . our analysis will allow us to study quantitatively in detail the behavior of such a dispersion relation . since @xmath232 , as well as @xmath233 , are odd functions of @xmath159 , they clearly give no contributions to either @xmath148 or @xmath149 . let us now consider @xmath234 . if we introduce the fourth order completely symmetric tensor @xmath235 satisfying the condition @xmath236 , we obtain after some simple algebraic manipulation @xmath237 \\ & & + \frac{4}{7k^{4}}\left [ 4\delta _ { ij}p_{hl}^{(2)}({\bf k } ) q_{h}q_{l}-6q_{h}\left ( q_{i}p_{jh}^{(2)}({\bf k})+q_{j}p_{ih}^{(2 ) } ( { \bf k})\right ) + p_{ij}^{(2)}({\bf k})(7f^{2}-3{\bf q}^{2})\right ] \\ & & + \frac{4}{k^{6}}p_{ijhl}^{(4)}({\bf k})q_{h}q_{l}\,.\end{aligned}\ ] ] by integrating eq . ( [ fourier0 ] ) with respect to @xmath116 we obtain @xmath238 where @xmath239 is a finite dimensionless constant dependent on the crystal structure . similarly , if we put @xmath240 integration of eq . ( [ beta0 ] ) provides @xmath241 where @xmath242 is another finite symmetric tensor such that @xmath243 . let us then define @xmath244 we have @xmath245 whence , using eq . ( [ fourier ] ) , @xmath246 it follows that @xmath247 where the integration constants @xmath248 are completely symmetric tensors satisfying the condition @xmath249 for @xmath250 . we show in appendix a that @xmath251 a preliminary determination of @xmath252 and @xmath253 with the aid of the two above formulas can considerably improve the accuracy in the numerical calculation of @xmath254 according to eq . ( [ gamma2 ] ) . from eqs . ( [ alpha ] ) , ( [ beta2 ] ) and ( [ gamma2 ] ) it follows that @xmath255 \\ & & -m_{ij}^{(2)}({\bf q},f)+o(\eta ) \,,\end{aligned}\ ] ] with @xmath256 \nonumber \\ & & -\frac{4}{7}\big [ 4\delta _ { ij}\beta _ { hl}^{(2)}q_{h}q_{l}-6q_{h } \left(q_{i } \beta_{jh}^{(2)}+q_{j } \beta _ { ih}^{(2)}\right ) \nonumber \\ & & + \beta _ { ij}^{(2)}(7f^{2}-3{\bf q}^{2})\big ] -4\gamma _ { ijhl}^{(2)}q_{h}q_{l}\ , . \label{m2}\end{aligned}\ ] ] by the same argument used to deduce eq . ( [ b0 ] ) we have @xmath257 in a similar way one finds that @xmath258 in fact , since the l.h.s . is an invariant tensor , it must be of the form @xmath259 , but then again the condition @xmath260 implies @xmath206 . it follows that @xmath261 \int d^{3}{\bf k}\,\frac{\exp ( -\eta k^{2})}{15k^{2 } } \\ & = & \frac{2\pi^{3/2}}{15\sqrt{\eta } } \left [ 3q_{i}q_{j}-\delta _ { ij } ( { \bf q}^{2}+10f^{2})\right ] \,,\end{aligned}\ ] ] so that @xmath262 finally , we have @xmath263 since both the sum and the integral on the r.h.s . of the two above equations are absolutely convergent . an explicit calculation carried out in appendix b shows that @xmath264 in accordance with eq . ( [ oseen ] ) . we can therefore conclude that @xmath265 with @xmath266 as we had anticipated in the previous section , the matrix @xmath267 is symmetric and real for real @xmath268 and @xmath181 , whereas @xmath269 just cancels the radiation reaction term in the equation of motion . since @xmath270 , we can rewrite eq . ( [ eqmotoel ] ) as @xmath271 or , in dimensionless variables , @xmath272 where @xmath273 , and @xmath274 is the classical electron radius . in order to apply the general theory developed until now to a concrete situation , let us consider in more detail the particular case of a simple cubic lattice , for which the primitive translation vectors of eq . ( [ lattice ] ) are given by @xmath275 for @xmath223 , @xmath276 being the unit vectors of the three coordinate axes and @xmath277 the lattice parameter . for this lattice one has @xmath278 , and the first brillouin zone corresponds to the cubic region @xmath279 . obvious symmetry considerations imply that @xmath280 , @xmath281 . then from the general condition @xmath282 one deduces that @xmath283 , and so @xmath284 the symmetry also implies that @xmath285 , where @xmath286 and the non - tensor @xmath287 is defined so that @xmath288 for @xmath289 , @xmath290 otherwise . since @xmath291 , @xmath292 , the condition @xmath293 implies @xmath294 . equation ( [ m2 ] ) thus becomes @xmath295 \\ & & + 4\gamma ^{(2 ) } ( 2q_{i}q_{j}+{\bf q}^{2}\delta_{ij}-5q_{i}^{2}\delta _ { ij})\,.\end{aligned}\ ] ] a numerical calculation based on eq . ( [ alpha ] ) shows that @xmath296 while from eqs . ( [ gamma2]-[gm2ijhl ] ) one obtains @xmath297 for application in numerical computations , the above results can also be put in the useful form @xmath298 -2c(l)(2q_{i}q_{j}+{\bf q}^{2}\delta _ { ij}-5q_{i}^{2 } \delta_{ij } ) \\ & & + \frac{4\pi } { 105l}\left [ -q_{i}q_{j}(3{\bf q}^{2}-7f^{2})+ ( |{\bf q}|^{4}+21{\bf q}^{2}f^{2}+70f^{4})\delta_{ij}\right ] + o(l^{-2})\,,\end{aligned}\ ] ] where we have introduced the functions @xmath299 with @xmath300 . as an example of application of these formulas , we have numerically calculated a few dispersion curves for an ideal simple cubic lattice with parameter @xmath301 and proper frequency of the oscillators @xmath302 , where @xmath303 and @xmath304 are the bohr radius and the bohr frequency respectively . in fig . 1 we show the dependence of @xmath305 as a function of @xmath306 , as determined according to eq . ( [ maineq ] ) for three different directions of propagation of the plane waves . the results can be compared with those obtained from the standard formula valid in the long wavelength limit , which , as can be seen from eqs . ( [ maineq ] ) and ( [ m0 ] ) , is given by @xmath307 one can see that , according to the above equation , @xmath54 has necessarily to be either orthogonal or parallel to @xmath268 . in the first case ( transversal polarization ) one obtains the dispersion relation @xmath308 while the second case ( longitudinal polarization ) can occur only when @xmath309 , independently of the value of @xmath310 . these relations are represented in the figure by the thick solid lines . the curve corresponding to transversal polarization presents a vertical asymptote for @xmath311 , whereas that for longitudinal polarization is obviously just a vertical straight line at @xmath312 . according to the standard theory no propagation is possible in the frequency interval between @xmath313 and @xmath314 . a further dispersion curve for transversal polarization is present for @xmath315 , according to both the approximated and the exact theory , but its onset in the figure is practically indistinguishable from the frequency axis . note that , for the particular numerical values that we have considered , one has @xmath316 and @xmath317 , so that @xmath318 , @xmath319 and @xmath320 . as one could expect , one can notice that the predictions of our exact theory depart in a significant way from those of the approximated one as soon as @xmath321 becomes of order of magnitude comparable to unity . this situation of course corresponds to the fact that the lattice parameter is no longer negligible with respect to the wavelength . a new feature which is revealed by the exact theory is the appearance , even for a perfect cubic lattice , of an anisotropic behavior , consisting in a remarkable dependence of the dispersion relation on the direction of propagation and polarization of the plane waves . the two thin solid curves at the opposite sides of the figure , which both refer to the direction of propagation @xmath322 $ ] , correspond to transversal ( the lower frequency one ) and longitudinal polarization respectively . if one considers the dependence of @xmath323 as a function of @xmath310 which is described by the former curve , one can see that the frequency reaches a maximum which corresponds to a zero of the group velocity @xmath324 and then slowly decreases over a wide region of the @xmath310 axis extending until the edge of the brillouin zone . this means that there exists a relatively small interval of the frequency axis with the property that , for each @xmath55 belonging to this interval , there are two transversal modes with direction of propagation along a crystal axis and different values of the wavelength . a similar phenomenon occurs also for the lowest - frequency branch of the dashed curve , which corresponds to a wave again polarized along a crystal axis , but propagating along the @xmath325 $ ] direction . a completely different dispersion relation is shown instead by a wave propagating in the same direction , but transversely polarized along the @xmath326 $ ] direction . this is illustrated by the intermediate branch of the dashed line . for this curve the frequency monotonically increases with @xmath310 over the entire brillouin zone , but with a slope and therefore a group velocity of the associated waves which drops to a relatively small value when @xmath323 approaches and then exceeds @xmath313 . at the edge of the brillouin zone this branch joins continuously , at a point with vertical tangent , the third branch of the same curve , which corresponds to longitudinal polarization . this branch extends up to @xmath314 , where a new transversal branch begins . it follows that , at distinction with the modes propagating along the @xmath322 $ ] direction , there is no forbidden frequency zone for waves propagating along the @xmath325 $ ] direction . of course , no hint about the existence of such a phenomenon could have been derived from the standard theory . a very similar behavior is shown by the dash - dot curve , corresponding to waves propagating along the @xmath327 $ ] direction . here again the branch with transversal and that with longitudinal polarization join continuously at the edge of the brillouin zone . note that , whereas a wave propagating along the @xmath325 $ ] direction can be transversely polarized along two directions ( the @xmath328 $ ] and @xmath326 $ ] ) with distinct symmetry properties and so also with different dispersion relations , the two possible transversal polarizations are degenerate for waves propagating along either the @xmath322 $ ] or @xmath327 $ ] directions . for all the curves displayed in fig . 1 , the polarization vector @xmath54 is either parallel or orthogonal to the wavevector @xmath268 . we would like to point out however that this fact is not a general property of the the solutions of eq . ( [ maineq ] ) , but is rather a consequence of the special symmetry of the particular directions of propagation that we have considered . note also that the continuous links between the transversal and the longitudinal branches of the curves can be seen as an obvious consequence of the particular degeneracy of the matrix @xmath329 when @xmath330 and @xmath331 . in this paper we have presented a new refined version of the classical dispersion theory of electromagnetic radiation in a crystalline solid , which has been completely described at a microscopic level as an infinite regular array of charged oscillators , without ever making use of the continuum approximation . it has been shown that , when the wavelength is comparable with the lattice parameter , the predictions provided by our calculations differ in a remarkable way from those derived from the old standard approximated formulas . in particular , in the region near the proper frequency of the oscillators , which was believed to represent a forbidden gap for light propagation , the phenomenon of `` slow light '' ( i.e. light propagation with small group velocity ) is found instead to take place along appropriate crystal directions . the model we have studied is however interesting in itself also from a purely theoretical point of view , independently of its phenomenological implications . it is in fact an exceptional case of a completely solvable model ( although in the dipole approximation ) in which the radiation reaction force acting on classical charged point particles is fully taken into account . actually , it turns out that the inclusion of this force has fundamental implications on the qualitative properties of the solutions , allowing for the existence of undamped collective oscillations . another interesting feature is that , although the model was formulated in terms of the usual maxwell lorentz electrodynamics with lorentz dirac selfinteraction , it is also compatible with the action - at - a - distance electrodynamics which was proposed by wheeler and feynman in 1945 @xcite . in fact , when the wheeler feynman identity is verified , the two theories lead to the same equations of motion for the charged particles . of course the investigations that we have presented here can be developed and extended in several directions . one of these , which appears to be particularly significant also from a phenomenological point of view , is the study of the behavior of a semi - infinite lattice occupying only one half of the full three - dimensional space . this investigation should clarify the relationship between the normal modes of the crystal which we have described in the present work and the observable electromagnetic radiation propagating in the free half - space , i.e. the phenomena of refraction and reflection at the surface of the crystal . in this way it is to be expected that classical fundamental results , such as the ewald oseen extinction theorem @xcite , will be justified on the basis of the detailed microscopic dynamics of the complete system . let us introduce the operator @xmath332 , which acts on the generic function @xmath179 by operating the convolution with a gaussian of width @xmath333 : @xmath334 ( { \bf p})=\int d^{3}{\bf k}\frac{\exp ( -|{\bf p}-{\bf k}|^{2}/4\tau ) } { 8(\pi \tau ) ^{3/2}}f({\bf k})\,.\ ] ] according to eq . ( [ gauss ] ) , if @xmath181 is continuous in @xmath335 , then @xmath336 ( { \bf p})=f({\bf p})$ ] . furthermore we have for any @xmath181 @xmath337 ( { \bf x})= e^{-\tau { \bf x}^{2 } } [ \hat{f } f ] ( { \bf x})\,.\ ] ] the function @xmath338 , defined by eq . ( [ b0def ] ) , is continuous for @xmath339 and we have @xmath340 ( { \bf 0},\eta ) = \int d^{3 } { \bf k}\frac{\exp ( -{\bf k}^{2}/4\tau ) } { 8(\pi \tau ) ^{3/2}}b_{ij}^{(0 ) } ( { \bf k},\eta ) = 0\,.\ ] ] therefore , applying eq . ( [ fourier ] ) to the function @xmath341 , we can write @xmath342 ( { \bf h}_{{\bf m}},\eta ) \nonumber \\ & = & \lim_{\tau \rightarrow 0^{+}}\sum_{{\bf n } } [ \hat{f } \hat{c}_{\tau } b_{ij}^{(0 ) } ] ( 2\pi { \bf y}_{{\bf n}},\eta ) \nonumber \\ & = & \lim_{\tau \rightarrow 0^{+}}\sum_{{\bf n}}e^{-\tau { \bf y}_{{\bf n}}^{2 } } [ \hat{f } b_{ij}^{(0 ) } ] ( 2\pi { \bf y}_{{\bf n}},\eta ) \ , . \label{app_a1}\end{aligned}\ ] ] using the formula @xmath343 with standard arguments it is easy to show that @xmath344 ( { \bf x},\eta ) \rightarrow -\frac{3}{4\pi x^{5}}p_{ij}^{(2)}({\bf x})+o(\eta ^{\infty } ) \ ] ] for @xmath185 , @xmath345 , while according to eq . ( [ b0 ] ) @xmath346 ( { \bf 0},\eta ) = \int d^{3}{\bf k}\,b_{ij}^{(0)}({\bf k},\eta ) = 0\,.\ ] ] therefore from eq . ( [ app_a1 ] ) one deduces immediately eqs . ( [ beta0 ] ) and ( [ beta0x ] ) . in a very similar way , by the same argument we used to deduce eq . ( [ bijhl ] ) we have @xmath347 ( { \bf 0},\eta ) & = & 0 \\ \small[\hat{f } b_{ijhl}^{(0)}\small ] ( { \bf 0},\eta ) & = & 0\,,\end{aligned}\ ] ] whereas one can show that @xmath348 ( { \bf x},\eta ) & \rightarrow & \frac{15}{8\pi x^{7}}p_{ijhl}^{(4)}({\bf x})\left ( 1 - 14 \frac{\eta } { x^{2}}\right ) \\ & & + o(\eta ^{\infty } ) \end{aligned}\ ] ] for @xmath185 , @xmath345 . applying again eq . ( [ fourier ] ) we then have @xmath349 ( { \bf h}_{{\bf m}},\eta ) \nonumber \\ & = & \lim_{\tau \rightarrow 0^{+}}\sum_{{\bf n}}[\hat{f } \hat{c}_{\tau } b_{ijhl}^{(0 ) } ] ( 2\pi { \bf y}_{{\bf n}},\eta ) \nonumber \\ & = & \lim_{\tau \rightarrow 0^{+}}\sum_{{\bf n}}e^{-\tau { \bf y}_{{\bf n}}^{2 } } [ \hat{f } b_{ijhl}^{(0 ) } ] ( 2\pi { \bf y}_{{\bf n}},\eta ) \nonumber \\ & = & \gamma _ { ijhl}^{(0)}-\gamma _ { ijhl}^{(-2)}\eta + o(\eta ^{\infty } ) \,,\end{aligned}\ ] ] with @xmath252 and @xmath350 given by eqs . ( [ g0ijhl ] ) and ( [ gm2ijhl ] ) respectively . hence , by integrating with respect to @xmath116 , eq . ( [ gamma2 ] ) is finally obtained . we have @xmath352 \bigg\ } . \label{cbar}\end{aligned}\ ] ] let us put @xmath353 . since @xmath152 is a second - rank tensor , from invariance considerations it is readily found that we must have @xmath354 with @xmath355 after performing a rotation on the integration space @xmath159 , in such a way that the unit vector @xmath356 of the third coordinate axis is directed along the vector @xmath268 , we obtain @xmath357 where @xmath358 for @xmath359 , where we have used the relation @xmath360 we have also for @xmath361 , @xmath362@xmath363 whence @xmath364 in a similar way we obtain @xmath365 so that @xmath366 from eqs . ( [ aqf]-[i2 ] ) we get @xmath367 then by substituting these expressions into eqs . ( [ celle ] ) and ( [ cbar ] ) we finally obtain eq . ( [ ibar ] ) .

over the last fifteen years there has been consistent theoretical interest in possible small deviations from the exact local lorentz invariance ( li ) of general relativity as well as a flourishing of observational tests . the theoretical interest is driven primarily by ideas in the quantum gravity ( qg ) community that lorentz invariance may not be an exact local symmetry of the vacuum . the possibility of outright lorentz symmetry violation ( lv ) or a different realization of the symmetry than in special relativity has arisen in string theory @xcite , loop qg @xcite , non - commutative geometry @xcite , space - time foam @xcite , some brane - world backgrounds @xcite , and condensed matter analogues of `` emergent gravity '' @xcite . as well , allowing the fundamental theory of gravity to be non - relativistic in the ultraviolet can make gravity renormalizable while avoiding some of the other pathologies that plague renormalizable gravitational actions with higher derivative terms @xcite . constructing useful tests for the various active models and ideas is therefore vital . since there are so many theoretical models around , a good approach is to work within a calculable framework where all possible lv terms are parameterized . each theory then picks out a certain combination of terms which can be constrained . in this vein , a standard method is to simply analyze a lagrangian containing the standard model fields and all lv operators of interest that can be constructed by coupling the standard model fields to new lv tensor fields that have non - zero vacuum expectation values . all renormalizable lv operators that can be added to the standard model in this way are known as the standard model extension ( sme ) @xcite . these operators all have mass dimension three or four and can be further classified by their behavior under cpt . higher mass dimension operators can be systematically explored as well , which is useful in case the naive eft hierarchy breaks down due to other new physics ( e.g. susy ) or quantum gravity introducing a custodial mechanism for the renormalizable operators in the infrared . the cpt odd dimension five kinetic terms for qed coupled to a non - zero vector field were written down in @xcite while the full set of dimension five operators with a vector were analyzed in @xcite . the dimension five and six cpt even kinetic terms for qed for particles coupled to a non - zero background vector , which we are primarily interested in here , were partially analyzed in @xcite . the full set of dimension five and six operators for qed has recently been introduced in @xcite . it is notable that susy forbids renormalizable operators for matter coupled to non - zero vectors @xcite but permits certain nonrenormalizable operators at mass dimension five and six . many of the parameterized lv operators have been very tightly constrained via direct observations ( see @xcite for reviews ) . in particular , the dimension five and six cpt even operators for lv with a vector field have been recently directly energies . susy , however , is an example of new relevant physics that can change this . ] constrained in the hadronic sector by exploiting ultra - high energy cosmic - ray observations @xcite performed by the pierre auger observatory ( pao ) . indeed , the construction and successful operation of this instrument has brought uhecrs to the interest of a wide community of scientists and it is expected to allow , in the near future , the assessment of several problems of uhecr physics and also to test fundamental physics ( in particular lorentz invariance in the qed sector ) with unprecedented precision @xcite . the uhecr constraints @xcite rely on the behavior of particle reaction thresholds with lv , which are one of the best methods in the eft approach to constrain nonrenormalizable lv operators . many lv operators give modified dispersion relations for free particles , where the energy as a function of momentum deviates slightly from the special relativistic form . for threshold reactions what matters is not the size of the lv correction to the energy compared to the absolute energy of the particle , but instead the size of the lv correction to the mass of the particles in the reaction . hence the lv terms usually become important when their size becomes comparable to the mass of the heaviest particle . if the lv term scales with energy as @xmath7 , then this critical energy is @xmath8 @xcite . according to this reasoning , the larger the particle mass the higher is the energy at which threshold lv effects come into play . this is why @xmath9 electrons and positrons , but not protons , can be used to constrain @xmath10 lv @xcite , and why uhe protons are needed to obtain constraints on hadronic lv with @xmath11 scaling ( which corresponds to cpt even mass dimension five and six operators ) . from this point of view , neutrinos , with their tiny mass of order @xmath12 @xcite , are in principle the most suited particles to provide strong constraints on lv , at least for reactions involving _ only _ neutrinos . one such reaction is that of neutrino oscillation . indeed , for a decade neutrino oscillations have proven to be excellent tests of the sme and other lv models @xcite as when the lv corrections are near the neutrino mass , the oscillation pattern can change as a function of energy , direction and mass . icecube may even be able to probe dimension six operators with time of flight techniques with tev neutrinos from distant gamma ray bursts @xcite . uhecr experiments have also the capability of placing constraints on sme parameters by exploiting neutrino oscillations @xcite . one can construct complementary and in some measures even more sensitive neutrino tests of higher dimension operators by leveraging observables which deviate more strongly from their special relativistic values as the neutrino travel distance increases . one such observable is found to be related to uhe neutrino spectrum observations . despite the threshold being low for lv effects to kick in , neutrinos with ultra - high energy are necessary to achieve a signal , as they interact so weakly that the phase space for a lv reaction must be huge to generate an appreciable rate . this requirement implies that , since the lv terms and hence the phase space grow with energy , very large energies are needed . indeed , for renormalizable operators , where the phase space does not grow quickly enough , reactions of the type we consider here never achieve the necessary rate given current bounds on their coefficients , even for uhe neutrinos over cosmological distances @xcite . however , for the cpt even non - renormalizable operators the situation is different , and we find that for energies of order @xmath4 there can be a significant modification of the spectrum . next generation neutrino detectors such as anita @xcite and supereuso @xcite are sensitive to neutrinos of energies @xmath13 . further experiments , like the planned arianna @xcite and iceray @xcite , will cover the range @xmath14 . in the present work we study how limits on the absolute scale of non - renormalizable cpt even lv neutrino parameters can be obtained from uhe neutrino observations . in particular we will determine the energy scale where the neutrino spectrum might begin to deviate as a function of the size of the lv coefficients . with three neutrino species as well as assorted light leptons possibly involved , the spectrum as a function of various lv parameters can only be computed by a detailed parameterized numerical search . at this stage , where lv in this sector is only speculative , we feel that such a search is unwarranted . however , we will show the `` best case '' scenario for a lv neutrino signal with dimension six operators as well as a scenario where a signal in the uhe spectrum is much more subtle even though lv is still relatively strong . this paper is structured as follows . in section [ sec : theo ] we describe the theoretical lv framework in which we will derive the lv effects on the uhe neutrino spectrum . in sections [ sec : nulv ] and [ sec : nusplitting ] we will give general information on the standard understanding of uhe neutrino generation by uhecrs , and we will present and detail the main lv reactions possibly affecting their spectrum . section [ sec : bestcase ] is devoted to present results in our test cases . in section [ sec : otherprocesses ] we discuss the possible role of other processes than neutrino splitting . in section [ sec : conclusions ] we will report our final remarks and conclusions . in order to study the phenomenological consequences of lv induced by qg , the existence of a dynamical framework in which to compute reaction thresholds and rates is essential . we assume that the low energy effects of lv can be parameterized in terms of a local eft . furthermore , for simplicity we assume that only boost invariance is broken , while rotations are preserved ( see @xcite for further comments on rotation breaking in this context ) . therefore we introduce lv by coupling standard model fields to a non - zero vector . more complicated forms could of course be chosen , however this would introduce significant direction dependence . with low statistics already for neutrino observatories , disentangling direction dependence would be a difficult task . we focus on the cpt even mass dimension five and six operators involving a vector field @xmath15 ( which we assume to describe the preferred reference frame in which the cmb is seen as isotropic ) coupled to a dirac neutrino @xmath16 in a mass eigenstate with mass @xmath17 that are quadratic in matter fields and hence modify the free field equations . neglecting the left - handed helicity of neutrinos for a moment , the lagrangian for a generic fermion is the usual dirac term plus @xcite @xmath18 } \psi~,\end{aligned}\ ] ] where @xmath19 is a timelike unit vector describing the preferred frame , @xmath20 and @xmath21 are the usual right and left projection operators , @xmath22 , and @xmath23 is the gauge covariant derivative . the @xmath24 coefficients are dimensionless . for fermions , at @xmath25 the helicity eigenstates are almost chiral , with mixing due to the particle mass and the dimension five operators . since we will be interested in high energy states , we re - label the @xmath24 coefficients by helicity , i.e. @xmath26 . the resulting high energy dispersion relation for positive and negative helicity particles can easily be seen from ( [ eq : actionfermion ] ) to involve only the appropriate @xmath27 or @xmath28 terms . for compactness , we denote the helicity based dispersion by @xmath29 . therefore at high energies we have the dispersion relation @xmath30 where @xmath31 and @xmath32 . we have dropped the @xmath33 terms as the @xmath34 operator present in these terms makes the correction to the equations of motion proportional to @xmath35 and hence tiny . the dimension five fermion operators induce two corrections , one proportional to @xmath36 and one corresponding to a change in the limiting speed of the fermion away from @xmath37 . the second effect , generated by @xmath38 is naturally of order @xmath39 due to the mass suppression for neutrinos and hence can also be disregarded . furthermore , we can drop the positive helicity coefficients , as the dominant signal will be from left - handed neutrinos produced by standard model couplings . finally , since cpt is conserved , and neutrinos and antineutrinos exist only in opposite states of helicity , @xmath40 . all these lovely properties of neutrinos mean we have just one lv coefficient , while in other cases @xcite further assumptions , such as parity preservation , are needed to reduce the number of free parameters . the final dispersion relation we assume in this work for each mass eigenstate neutrinos and antineutrinos is @xmath41 where @xmath42 denotes the mass eigenstate of the neutrino . there is no reason why each mass eigenstate needs have the same coefficient , and indeed whether or not they do makes a dramatic difference in the observed spectrum . in particular , we will study two cases : in the `` flavor blind '' case all the @xmath43 have roughly the same magnitude , which translates in all the lv effects being essentially independent of flavor ; in the `` flavor dependent '' case the @xmath43 are instead different for different mass states , which makes the lv effects depend upon flavor states . among the `` flavor blind '' cases we can find a `` best case '' scenario , in which the effects of lv are maximal , as well as a `` worst case '' scenario , in which lv is present but is ineffective in the context of uhe neutrinos we are discussing now . a fairly accurate general estimate of the minimum energy in which lv corrections in equation ( [ eq : finaldisp ] ) is relevant is obtained , as stated previously , by comparing the largest mass of the particles entering in the lv reaction with the magnitude of the lv correction in these equations @xcite . in our case , assuming @xmath44 , the typical energy at which lv contributions start to be relevant is of order @xmath45 , assuming @xmath46 . this energy of 10 tev allows observatories such as icecube to possibly constrain these operators in the future using neutrino oscillations . we instead focus on changes to the uhe neutrino spectrum induced by these operators . if we neglect exotic sources of uhe neutrinos ( as suggested in many top - down models for the production of uhecrs , now disfavored by the current experimental photon limits ) , the `` cosmogenic '' neutrino flux is created @xcite via the decay of charged pions produced by the interaction of primary nucleons with cmb photons above @xmath47 , the greisen - zatsepin - kuzmin ( gzk ) effect @xcite . hires @xcite and auger @xcite spectral observations seem to confirm the presence of a gzk suppression in the uhecr spectrum . although the suppression of the uhecr spectrum could be also due to the maximal accelerating power of uhecr sources , the fact that it occurs at just the right energy for being gzk taking place during propagation , and the results @xcite on the correlation of the uhecr arrival directions with the large scale distribution of matter within @xmath48 , seem to favor the gzk explanation . motivated by these considerations , we make here the hypothesis that the gzk reaction is at work during uhecr propagation . we have checked numerically that the process of pion decay is not strongly affected by the above lv if only neutrinos are lv . one might ask in general if lv effects in the hadronic sector , which we have so far neglected , can matter in the production of uhe neutrinos . we can neglect the hadronic sector because of the tight constraints already placed on such lv operators @xcite , which are stronger than what will be considered here . violation of lorentz invariance however introduces new phenomena in the propagation of uhe neutrinos . a partial list of these effects includes : modified @xmath49 oscillations : : since the lv is parameterized in the mass eigenstate , the lv terms act as contributions to an effective mass and contribute to neutrino oscillations . cosmogenic neutrinos are not the right phenomena with which to study modified oscillations as they have oscillated many times during their flight from an unknown source , making it extremely hard to derive oscillation constraints . icecube and other neutrino observatories may be sensitive to these operators for atmospheric neutrinos . @xmath49-erenkov : : erenkov radiation in vacuum via charge - radius coupling or gravity : @xmath50 . this possibility has been already investigated for renormalizable operators @xcite . the rate is too small for both renormalizable operators ( at current limits ) or non - renormalizable operators , even for cosmogenic neutrinos . it also involves additional lv coefficients for photon emission and perhaps new modes for graviton emission @xcite , thereby complicating the analysis and , at best , yielding less definitive limits . @xmath49-splitting : : : @xmath51 . this effect is what we focus on in the following , as it exclusively involves the neutrino sector and has a high enough rate to be seen at uh energies . @xmath49-pair emission : : : @xmath52 . here , @xmath53 represents some other fermion species besides neutrinos . electrons are the only fermion species light enough to have an effect compared to @xmath49-splitting at uh energies . we initially will ignore this reaction , but we will consider the possible effects of electron pair and hadronic emission in section [ sec : otherprocesses ] . to understand how a neutrino can split into three , it suffices to calculate the threshold energy for this reaction to occur , which in a lorentz invariant scenario would be infinite . let us compute the threshold for the decay @xmath54 , where @xmath55 , @xmath56 are neutrinos of mass @xmath57,@xmath58 and @xmath59 are the momenta . in the threshold configuration the momenta of the outgoing particles are all aligned and parallel to the direction of the incoming neutrino momentum @xcite , hence we choose all momenta to be in the @xmath60-direction . let @xmath61,@xmath62 and @xmath63 be the fraction of initial momentum carried respectively by the outgoing @xmath55 , by @xmath56 , and by @xmath64 , @xmath65 and @xmath66 . for a general dispersion relation with lv term @xmath67 the threshold equation ( the energy conservation equation imposing momentum conservation ) can then be written as @xmath68 = m_{\nu_a}^{2}\frac{1-x}{x } + m_{\nu_b}^{2}\left(\frac{1}{y } + \frac{1}{t}\right)~. \label{eq : genthres}\ ] ] when li is exact , the left hand side vanishes and there is no solution . with lv , in the simple case where @xmath11 and @xmath69 , we obtain the equation @xmath70 the minimum of the right - hand side in the equation above is 9 and is attained at @xmath71 , hence the threshold energy is given by @xmath72 note that @xmath73 must be positive in order for there to be splitting , i.e. we must deal with superluminal neutrinos . if @xmath74 then the threshold energy must be solved for numerically as the lv parameter and mass are generically different . however , the existence of a finite threshold can still be shown for appropriate values of @xmath73 and @xmath75 . in the latter case , the coefficients do not need to be all positive for there to be a threshold , however it is necessary that @xmath76 , where @xmath77 is a positive number that depends on the incoming momentum . thresholds for negative coefficients have been investigated previously @xcite and there are strong limits on the lv coefficients . ) neutrino splitting is forbidden for superluminal neutrinos in the case @xmath78 studied e.g. in @xcite . although the model presented in @xcite does not have an eft description , a simple threshold computation under the assumption of energy - momentum conservation and equal values of masses and lv parameters for neutrinos and antineutrinos shows that if neutrino splitting is possible also in that framework , values @xmath79 can be probed by @xmath80 neutrinos . ] however , for uhe neutrinos it is more difficult to extract constraints , as we shall discuss in section [ sec : notsogood ] . neutrinos created in a flavor eigenstate are , of course , a combination of mass eigenstates . in order to maximize the neutrino splitting signal one would want _ all _ mass eigenstates to split ( `` flavor blind '' scenario ) . the best case scenario is therefore when @xmath81 for each @xmath42 and each @xmath43 is of the same order of magnitude ( so all states begin to decay at the same energy ) . as our best case scenario then we take @xmath43 equal and positive for all @xmath42 . let us now consider , within the above flavor blind scenario " , energies well above threshold so that neutrinos do effectively split , and the best case combination of the @xmath43 . the splitting rate can be calculated from the neutrino width @xmath82 where the sum runs over the open splitting channels and @xmath83 represents the partial width for the channel @xmath84 . @xmath83 is simply @xmath85 where @xmath86 represents the phase space of the final states and @xmath87 the matrix element for the process . the dominant channel for neutrino splitting is via the tree level neutral current interaction . the matrix element for this interaction is @xmath88 where @xmath89 is the charged current electro - weak coupling constant , @xmath90 is the usual spin projector , @xmath91 is the weinberg angle and @xmath92 represent the 4-momentum components of the @xmath93 boson . @xmath94 and @xmath95 are the spinors associated with @xmath55 and @xmath56 , their functional form can be found in @xcite . @xmath96 is at most the order of @xmath97 and hence for any incoming momenta @xmath98 @xmath99 . after some brute force algebra , we end with @xmath100 where @xmath101 is the angle between incoming and outgoing @xmath55 and @xmath102 is a complicated function of @xmath101 . the phase space , @xmath103 is given by @xmath104 we integrate over @xmath105 and we are left with @xmath106 substituting @xmath107 and @xmath103 back into ( [ eq : partialwidth ] ) we have @xmath108 we now turn to estimating the size of the remaining integral well above threshold , as uhe neutrinos are far in excess of the neutrino splitting threshold around 20 tev . temporarily , we will set @xmath109 and re - insert it at the end . we first assume that the opening angle for all three neutrinos is roughly equal , as is true over most of phase space . in this case , @xmath102 can be approximated as @xmath110 . the opening angle @xmath101 vanishes as @xmath111 , and is small even for uhe neutrinos . @xmath102 then reduces to @xmath112 . indeed , @xmath101 is given by the typical transverse momenta @xmath113 divided by the longitudinal momenta @xmath114 . the characteristic size of @xmath113 can be estimated by energy conservation , recognizing that when energy is conserved any `` excess '' energy from lv goes into the energy needed to create transverse momenta . formally this implies that @xmath115 i.e. @xmath116 . @xmath114 itself for any particle can range from almost 0 to almost the initial energy @xmath117 well above threshold . the available phase space volume for the remaining outgoing particles can therefore be approximated ( although somewhat overestimated ) as a cylinder with length @xmath117 and radius @xmath118 , which has a total volume of @xmath119 . similarly , @xmath101 can be estimated as @xmath120 . the @xmath121-function in energy simply removes one factor of energy from our equation . putting these expressions back into ( [ eq : partialwidth2 ] ) we find @xmath122 or , simplified @xmath123 where we used @xmath124 . if we now want to restore the @xmath125 factor , we notice that we must put one @xmath125 for each @xmath126 . in the end , the total rate is computed by adding the partial rates for each neutrino flavor , which is equivalent to multiply by 3 . hence , we have @xmath127 this width can be turned into a decay length as @xmath128 this makes clear we need to push the required energies above @xmath129 ( with @xmath130 ) for the rate to be appreciable . as a final remark , we notice that the decay length in eq . ( [ eq : nudeclength ] ) strongly depends on both the energy and @xmath1 . therefore , the error about its actual magnitude we might have made in our estimate will reflect in very small errors in the determination of the energy at which lv effects start to be relevant as well as of the constraint on @xmath1 . at such high energies the z could be real - i.e. there is a resonance in the matrix element . even in this regime , however , the neutrino decay time can be computed easily , as the only hypothesis one has to relax is that the z 4-momentum @xmath131 satisfies @xmath99 . the magnitude of @xmath96 can be easily computed exploiting the kinematic equations . we obtain @xmath132 . the final decay length is then @xmath133~. \label{eq : clength}\ ] ] a comparison between the two different decay lengths , eq . ( [ eq : nudeclength ] ) and eq . ( [ eq : clength ] ) , can be found in fig . [ fig : declength ] . the z resonance is hit at @xmath134 . ) /red dashed line ) , and with ( eq . ( [ eq : clength])/black solid line ) the z boson resonance . ] we notice that even though the two computations lead to very different results above the resonance , they will not lead to any appreciable effect in the neutrino spectra , as at such energies the decay lengths are anyway much smaller than the propagation distance of cosmological neutrinos . in the flavor blind case above every mass eigenstate undergoes splitting . it is easy to construct a scenario where only one mass eigenstate decays - one eigenstate has a positive @xmath135 lv coefficient while the other two have no lv . while this is unnatural in some sense , it serves as a nice example to illuminate a method to easily hide lv . the neutrino spectrum produced by the gzk process has a distinct distribution in flavor states of 1/3 @xmath136 and 2/3 @xmath137 . therefore , in order to calculate a spectrum with lv we need to convert this flavor spectrum into a mass spectrum , i.e. we need to choose a neutrino mixing model . for our purposes we choose tribimaximal mixing @xcite , which satisfies current experimental constraints and provides a simple mixing matrix . with tribimaximal mixing the gzk neutrino spectrum has equal distribution over all mass eigenstates . now that we have spelled out out test models we shall proceed discussing the results for both scenarios . the net effect of neutrino splitting is very simple . it kills one neutrino of energy @xmath138 and creates 3 neutrinos of average energy @xmath139 , provided the parent neutrino is above threshold and has a reasonable life time . as we showed in sec . [ sec : computation ] , for a life time shorter than the age of the universe , the neutrino energy has to be above @xmath129 , i.e. we need to probe uhe neutrinos . the effects of neutrino splitting on the uhe neutrino spectrum are twofold and can be understood qualitatively as follows . flux suppression at uh energies : : the splitting is effectively an energy loss process for uhe neutrinos . if the rate is sufficiently high , the energy loss length can be below 1 mpc . let us call @xmath140 the energy at which this happens . then , being gzk neutrinos produced mainly at distances larger than 1 mpc , we do not expect any neutrino to be detected at earth with @xmath141 . + the mere observation of neutrinos up to a certain energy @xmath142 would imply a constraint , according to eq . [ eq : nudeclength ] @xmath143 flux enhancement at sub - uh energies : : neutrinos lose energy by producing lower energy neutrinos . eventually these neutrinos will become stable , either because their energy is below threshold , or because their lifetime is larger than their propagation time . accordingly , we expect an enhancement of the neutrino flux at energies below @xmath144 . in spite of being qualitatively straightforward , however , this analysis is not powerful enough to provide us with constraints on lv . in order to obtain meaningful constraints , we have to resort to full montecarlo simulations of the uhecr propagation from sources to the earth . we simulated then the propagation of uhecr protons in the inter galactic medium using the monte carlo package crpropa @xcite , suitably modified to take into account lv in the neutrino sector . the simulation parameters are the following : we simulated unidimensional uhecr proton propagation , with source energy spectrum @xmath145 , from a spatially uniform distribution of sources located at redshift @xmath146 according to the waxman & bahcall ( wb ) distribution used in @xcite . the injection proton spectrum was tuned to fit auger data @xcite . figure [ fig : spectra_eta ] shows the outcome of the simulations for different values of the lv parameter @xmath125 in the best case scenario , together with experimental sensitivities from some existing and planned observatories , as well as the waxman & bahcall bound @xcite for reference .- rays , in the hypothesis that the sources are optically thin to the escape of uhe particles and that both @xmath147-rays and neutrinos are originated from uhecr interactions with radiation backgrounds . it is worth mentioning that this bound might be strongly affected by qg effects , as shown in @xcite . ] results are in agreement with qualitative expectations previously discussed : above a certain energy the neutrino spectrum displays a sharp cut off , while at lower energy a peak appears . the peak position and strength , as well as the position of the cut off , depend on the value of the lv parameter @xmath125 . if @xmath130 the peak is found at @xmath148 and overwhelms the li spectrum by a factor of roughly 10 , while the cut off energy is at @xmath149 . also the evolution of the neutrino spectra with @xmath125 can be seen in the same fig . [ fig : spectra_eta ] . although at present it is not possible to draw firm conclusions for the constraints on lv in the neutrino sector , we notice that future experiments , if not already auger , will be able to probe the fluxes we predict , hence they will be able to cast constraints on @xmath125 . in the future , the arianna experiment @xcite will be able to probe fluxes down to @xmath150 in the energy range @xmath151 in the full configuration @xcite . if such a sensitivity will be achieved , a constraint of order @xmath5 will be cast , according to fig . [ fig : spectra_eta ] . moreover , the iceray experiment @xcite planned at the south pole is expected to observe roughly 4 neutrino events per year of data taking for the wb model preferentially studied here , in the range @xmath152 @xcite . hence , constraints of order @xmath153 are expected after few years of data taking by this experiment . we finally notice here that by exploiting this strategy nothing can be said about the case @xmath154 . if only one mass eigenstate decays , the energy momentum conservation equations , rates , and lifetimes all remain the same . we therefore can apply our montecarlo spectrum for this scenario to just any one of the mass eigenstates ( given our assumption of tribimaximal mixing no neutrino mass eigenstate is preferred ) . the only difference is that only 1/3 of the total flux undergoes neutrino splitting . the resulting spectrum is shown in fig . [ fig : onemasssplits ] . note that even for @xmath155 the departure from the li spectrum is significantly reduced , as one would expect , with the magnitude of the deviation at the peak being lowered to a @xmath156 excess over the li flux . hence , with present sensitivity , absence of a signal in the neutrino spectrum does not mean that lv is ruled out for neutrinos , even for @xmath2 positive coefficients . we notice that the cutoff feature present in the flavor blind case has now effectively disappeared . then , observations of a uhe neutrino flux up to some maximal energy do not allow to draw conclusions about lv in this case scenario . we now return to the question of negative coefficients and neutrino splitting . so far , in both our flavor blind and flavor dependent scenarios all the coefficients have been positive . as argued in section [ sec : nusplitting ] , neutrinos can split even with negative coefficients when the relation @xmath76 holds . however , one can immediately see that therefore at least one of the mass eigenstates is always stable if the coefficients are negative . this implies that no equivalent of the best case scenario " ( all the mass eigenstates decay ) is possible for negative @xmath43 and that , consequently , the deviation from the li spectrum will be suppressed by at least 1/3 . therefore cosmogenic neutrinos are not effective probes of the @xmath157 region of parameter space . predictions of cosmogenic neutrino spectra are known to be plagued by several uncertainties , as they depend strongly on the evolution of uhecr sources with redshift , which can not be directly probed with uhecrs due to the gzk attenuation effect . as a consequence , also our predictions for the uhe neutrino flux in the presence of lv can be affected . we check this by simulating different uhecr sources evolution models for the best case scenario " . in fig . [ fig : max ] we show the results for the model outlined in @xcite , which is expected to give the largest flux . as a result , the presence and the strength of the bump feature are demonstrated to be weakly dependent on the underlying uhecr source distribution , with the choice of the source distribution affecting by a factor of 2 the height of the bump . this was somewhat expected from our qualitative considerations : as long as uhecr sources are located at distances much larger than 1 mpc from earth , the magnitude of the bump depends only on how many neutrinos were originally produced at energy larger than the lv cutoff energy . on the other hand , the cutoff feature is definitely model independent , as long as the unknown maximal energy at which uhecrs are accelerated is enough to produce neutrinos at energy larger than @xmath158 . as we mentioned in the previous sections , there are other processes , besides splitting , leading to neutrinos losing energy in the inter - galactic medium . in particular , the hadronic decay modes of the z@xmath159 could lead to the production of uhe protons , thereby mimicking a `` z - burst '' effect @xcite . although this model was conceived for other purposes than constraining lv , and has now lost most of its attraction due to experimental results on proton spectra above the gzk threshold , it is notable that the same mechanism could in principle help constrain lv in the neutrino sector . however , we can argue with a very simple argument that this is not the case . let us consider the li neutrino flux , which corresponds in fig . [ fig : spectra_eta ] to the @xmath160 case . the minimal process leading to proton production in this context is @xmath161 , whose threshold energy , if @xmath130 , is at @xmath162 . processes involving more particles are naturally suppressed , however they represent the majority of the allowed channels and hence they might be relevant to our case . even if we do not know the energy spectrum of the produced protons , we know that their mean energy is roughly 1/80 times the energy of the parent neutrino @xcite . we can then compute the maximal expected flux of uhecrs produced by this z - burst - like mechanism by converting all the neutrinos with energy @xmath163 above threshold into 2 protons of energy @xmath164 . the result is shown in fig . [ fig : zburst ] . this mechanism would indeed contribute to the uhecr spectrum at @xmath165 , producing an excess no larger than 5% , which however is below the systematic uncertainties of uhecr experiments . therefore , no constraints on lv can be implied using this technique . electron / positron pair production would also contribute to the cutoff in the neutrino spectrum . the threshold for pair production of electrons is at roughly @xmath166 ev . at energies near the existing neutrino splitting cutoff , @xmath167 ev we are well above threshold . the mass of the electrons is therefore irrelevant in calculating the decay rate ( just as the neutrino mass was ) and hence the electron pair production process has nearly the same rate as the neutrino splitting process . if neutrinos produce electron / positron pairs ( which after all depends on lv in the electron sector ) it can only increase the steepness of the cutoff . it will not contribute to the excess neutrino flux at @xmath168 ev and hence one needs to probe energies one order of magnitude larger than @xmath169 to derive constraints exploiting the presence or the absence of a cutoff feature . in this work we have investigated possible signals of higher dimension lv cpt even operators in the uhe neutrino spectrum , in particular the effect of the neutrino splitting on the uhe neutrino spectrum . this process provides a clean test as it does not involve other lv operators apart from the neutrinos ones . in addition , since the dominant neutrino field is left - handed , there are no complications due to different lv for different chirality of fermion , as there are in the uhecr case , and one ends up with one lv coefficient for each neutrino mass eigenstate . in the flavor blind scenario , where every mass eigenstate undergoes splitting approximately at the same energy , there is both a precocious fall off of the neutrino flux at uhe as well as a significant excess in the uhe neutrino flux at energies as low as @xmath168 ev . noticeably , this kind of energies are well within reach of current and future uhecr experiments and about order of magnitude below those so far used for lv tests with hadronic and electromagnetic uhecrs . according to our study , existing or planned uhe neutrino experiments have the potential to probe lv in the neutrino sector for coefficients @xmath170 . however , we have discovered a serious difficulty with deriving constraints in the absence of a positive lv signal event . the distribution in mass eigenstates is roughly equal for uhe neutrinos in realistic mixing scenarios . although it might seem somewhat unnatural that lv has different effects on different mass states of the same particle field , if this is the case then it is possible that lv can exist / be strong for one mass eigenstate yet be almost invisible for uhe neutrino detectors . in particular , the observation of a neutrino flux up to some maximal energy does not imply a firm conclusion on lv , at least with present accuracy . however , let us note that if a neutrino is ever detected at energy @xmath163 , we can rule out a flavor blind @xmath171 according to eq . ( [ eq : constraint_naive ] ) . on the other hand , the bump feature can lead to constraints on @xmath172 , as this bump should be observable in any such scenario of lv in the neutrino sector . in this case , to obtain a @xmath2 constraint on lv requires at least a 50% accuracy in the determination of the neutrino spectrum in the energy range @xmath173 , which can be achieved by future experiments . we thank j. kelley and b. mcelrath for useful discussions . this work was supported by the deutsche forschungsgemeinschaft through the collaborative research centre sfb 676 `` particles , strings and the early universe : the structure of matter and space - time '' . lm acknowledges support from the state of hamburg , through the collaborative research program `` connecting particles with the cosmos '' within the framework of the landesexzellenzinitiative ( lexi ) . 999 v. a. kostelecky and s. samuel , `` spontaneous breaking of lorentz symmetry in string theory , '' phys . d * 39 * , 683 ( 1989 ) . j. r. ellis , n. e. mavromatos and d. v. nanopoulos , phys . b * 665 * , 412 ( 2008 ) [ arxiv:0804.3566 [ hep - th ] ] . r. gambini and j. pullin , `` nonstandard optics from quantum spacetime , '' phys . d * 59 * , 124021 ( 1999 ) . c. rovelli and s. speziale , phys . d * 67 * , 064019 ( 2003 ) [ arxiv : gr - qc/0205108 ] . j. alfaro and g. palma , phys . d * 67 * ( 2003 ) 083003 [ arxiv : hep - th/0208193 ] . s. m. carroll , j. a. harvey , v. a. kostelecky , c. d. lane and t. okamoto , `` noncommutative field theory and lorentz violation , '' phys . lett . * 87 * , 141601 ( 2001 ) . [ arxiv : hep - th/0105082 ] . j. lukierski , h. ruegg and w. j. zakrzewski , annals phys . * 243 * ( 1995 ) 90 [ arxiv : hep - th/9312153 ] . g. amelino - camelia and s. majid , int . j. mod . phys . a * 15 * ( 2000 ) 4301 [ arxiv : hep - th/9907110 ] . m. chaichian , p. p. kulish , k. nishijima and a. tureanu , phys . b * 604 * , 98 ( 2004 ) [ arxiv : hep - th/0408069 ] . g. amelino - camelia , j. r. ellis , n. e. mavromatos , d. v. nanopoulos and s. sarkar , `` potential sensitivity of gamma - ray burster observations to wave dispersion in vacuo , '' nature * 393 * , 763 ( 1998 ) . [ arxiv : astro - ph/9712103 ] . c. p. burgess , j. cline , e. filotas , j. matias and g. d. moore , `` loop - generated bounds on changes to the graviton dispersion relation , '' jhep * 0203 * , 043 ( 2002 ) . [ arxiv : hep - ph/0201082 ] . c. barcelo , s. liberati and m. visser , _ `` analogue gravity , '' _ living rev . * 8 * , 12 ( 2005 ) . [ arxiv : gr - qc/0505065 ] . horava , phys . d * 79 * , 084008 ( 2009 ) [ arxiv:0901.3775 [ hep - th ] ] . g. amelino - camelia , arxiv:0806.0339 [ gr - qc ] . g. amelino - camelia , int . j. mod . d * 11 * ( 2002 ) 35 [ arxiv : gr - qc/0012051 ] . d. colladay and v. a. kostelecky , phys . d * 58 * , 116002 ( 1998 ) [ arxiv : hep - ph/9809521 ] . r. c. myers and m. pospelov , phys . rev . * 90 * ( 2003 ) 211601 [ arxiv : hep - ph/0301124 ] . p. a. bolokhov and m. pospelov , phys . d * 77 * , 025022 ( 2008 ) [ arxiv : hep - ph/0703291 ] . d. mattingly , pos * qg - ph * , 026 ( 2007 ) . v. a. kostelecky and m. mewes , phys . rev . d * 80 * ( 2009 ) 015020 [ arxiv:0905.0031 [ hep - ph ] ] . s. groot nibbelink and m. pospelov , phys . lett . * 94 * , 081601 ( 2005 ) [ arxiv : hep - ph/0404271 ] . d. mattingly , living rev . * 8 * , 5 ( 2005 ) [ arxiv : gr - qc/0502097 ] . t. jacobson , s. liberati and d. mattingly , annals phys . * 321 * ( 2006 ) 150 [ arxiv : astro - ph/0505267 ] . s. liberati and l. maccione , arxiv:0906.0681 [ astro-ph.he ] . to appear in annual review of nuclear and particle science . j. collins , a. perez , d. sudarsky , l. urrutia and h. vucetich , phys . lett . * 93 * , 191301 ( 2004 ) [ arxiv : gr - qc/0403053 ] . l. maccione , a. m. taylor , d. m. mattingly and s. liberati , jcap * 0904 * ( 2009 ) 022 [ arxiv:0902.1756 [ astro-ph.he ] ] . m. galaverni and g. sigl , phys . , * 100 * , 021102 ( 2008 ) . arxiv:0708.1737 [ astro - ph ] . l. maccione and s. liberati , jcap * 0808 * , 027 ( 2008 ) [ arxiv:0805.2548 [ astro - ph ] ] . m. galaverni and g. sigl , phys . d * 78 * ( 2008 ) 063003 [ arxiv:0807.1210 [ astro - ph ] ] . t. kifune , astrophys . j. * 518 * ( 1999 ) l21 [ arxiv : astro - ph/9904164 ] . r. aloisio , p. blasi , p. l. ghia and a. f. grillo , phys . d * 62 * ( 2000 ) 053010 [ arxiv : astro - ph/0001258 ] . g. amelino - camelia and t. piran , phys . d * 64 * ( 2001 ) 036005 [ arxiv : astro - ph/0008107 ] . f. w. stecker and s. t. scully , astropart . * 23 * ( 2005 ) 203 [ arxiv : astro - ph/0412495 ] . l. gonzalez - mestres , nucl . suppl . * 190 * , 191 ( 2009 ) [ arxiv:0902.0994 [ astro-ph.he ] ] . s. t. scully and f. w. stecker , astropart . phys . * 31 * ( 2009 ) 220 [ arxiv:0811.2230 [ astro - ph ] ] . f. w. stecker and s. t. scully , new j. phys . * 11 * ( 2009 ) 085003 [ arxiv:0906.1735 [ astro-ph.he ] ] . t. jacobson , s. liberati and d. mattingly , phys . d * 67 * ( 2003 ) 124011 [ arxiv : hep - ph/0209264 ] . l. maccione , s. liberati , a. celotti and j. g. kirk , jcap * 0710 * ( 2007 ) 013 [ arxiv:0707.2673 [ astro - ph ] ] . amsler w. m. _ et al . _ [ particle data group ] phys . b * 667 * , 1 ( 2008 ) . s. r. coleman and s. l. glashow , phys . d * 59 * ( 1999 ) 116008 [ arxiv : hep - ph/9812418 ] . m. c. gonzalez - garcia , f. halzen and m. maltoni , phys . d * 71 * , 093010 ( 2005 ) [ arxiv : hep - ph/0502223 ] . j. s. diaz , v. a. kostelecky and m. mewes , phys . d * 80 * , 076007 ( 2009 ) [ arxiv:0908.1401 [ hep - ph ] ] . s. yang and b. q. ma , arxiv:0910.0897 [ hep - ph ] . m. c. gonzalez - garcia and f. halzen , jcap * 0702 * ( 2007 ) 008 [ arxiv : hep - ph/0611359 ] . a. bhattacharya , s. choubey , r. gandhi and a. watanabe , arxiv:0910.4396 [ hep - ph ] . s. r. coleman and s. l. glashow , phys . b * 405 * ( 1997 ) 249 [ arxiv : hep - ph/9703240 ] . p. gorham _ et al . _ [ anita collaboration ] , arxiv:0812.2715 [ astro - ph ] . a. petrolini , arxiv:0909.5220 [ astro-ph.im ] . a. santangelo and a. petrolini , arxiv:0909.5370 [ astro-ph.he ] . s. w. barwick , j. phys . * 60 * ( 2007 ) 276 [ arxiv : astro - ph/0610631 ] . s. w. barwick , nucl . instrum . a * 602 * ( 2009 ) 279 . p. allison _ et al . arxiv:0904.1309 [ astro-ph.he ] . v. s. beresinsky and g. t. zatsepin , phys . b * 28 * ( 1969 ) 423 . f. w. stecker , astrophys . space sci . * 20 * ( 1973 ) 47 . r. engel , d. seckel and t. stanev , phys . d * 64 * ( 2001 ) 093010 [ arxiv : astro - ph/0101216 ] . d. v. semikoz and g. sigl , jcap * 0404 * ( 2004 ) 003 [ arxiv : hep - ph/0309328 ] . k. greisen , phys . * 16 * ( 1966 ) 748 ; zatsepin , kuzmin , sov.phys.jetp 4 , 78 . r. abbasi _ et al . _ [ hires collaboration ] , arxiv : astro - ph/0703099 . m. roth [ pierre auger collaboration ] , arxiv:0706.2096 [ astro - ph ] . j. abraham _ et al . _ [ pierre auger collaboration ] , science * 318 * ( 2007 ) 938 [ arxiv:0711.2256 [ astro - ph ] ] . t. jacobson and d. mattingly , phys . d * 70 * ( 2004 ) 024003 [ arxiv : gr - qc/0402005 ] . d. mattingly , t. jacobson and s. liberati , phys . d * 67 * ( 2003 ) 124012 [ arxiv : hep - ph/0211466 ] . t. j. konopka and s. a. major , new j. phys . * 4 * , 57 ( 2002 ) [ arxiv : hep - ph/0201184 ] . j. m. carmona and j. l. cortes , phys . b * 494 * ( 2000 ) 75 [ arxiv : hep - ph/0007057 ] . e. armengaud , g. sigl , t. beau and f. miniati , astropart . * 28 * ( 2007 ) 463 [ arxiv : astro - ph/0603675 ] . j. n. bahcall and e. waxman , phys . b * 556 * ( 2003 ) 1 [ arxiv : hep - ph/0206217 ] . t. p. a. collaboration , phys . d * 79 * ( 2009 ) 102001 [ arxiv:0903.3385 [ astro-ph.he ] ] . e. waxman and j. n. bahcall , phys . d * 59 * ( 1999 ) 023002 [ arxiv : hep - ph/9807282 ] . j. n. bahcall and e. waxman , phys . d * 64 * ( 2001 ) 023002 [ arxiv : hep - ph/9902383 ] . g. amelino - camelia , m. arzano , y. j. ng , t. piran and h. van dam , jcap * 0402 * ( 2004 ) 009 [ arxiv : hep - ph/0307027 ] . d. fargion , b. mele and a. salis , astrophys . j. * 517 * ( 1999 ) 725 [ arxiv : astro - ph/9710029 ] . r. aloisio , v. berezinsky , p. blasi , a. gazizov , s. grigorieva and b. hnatyk , astropart . * 27 * ( 2007 ) 76 [ arxiv : astro - ph/0608219 ] . p. f. harrison , d. h. perkins and w. g. scott , phys . b * 530 * ( 2002 ) 167 [ arxiv : hep - ph/0202074 ] .

the possible modification of nucleon properties in the nuclear medium is currently a much discussed topic in low energy hadron physics . one way to consider such problems is to describe the nucleon as a topological soliton and then study the influence of baryonic matter on the properties of such solitons , making use of the independent particle picture which has been so successful in describing many properties of nuclei . there exist already some works where nucleons described as skyrme type solitons embedded in infinite nuclear matter have been considered @xcite . the results of these studies were in qualitative agreement with experimental indications and with results of other authors using different approaches . on the quantitative level , however , there is a too large renormalization of the nucleons effective mass in nuclear matter . the quantitative value of this renormalization is about @xmath2 for the normal nuclear matter density , @xmath3 @xcite . it is therefore difficult to relate this modification of the nucleon self energy in the medium to the nucleon mass in free space because such calculations consider only the special case when a nucleon is placed in the center of the heavy nucleus . the infinite nuclear matter approach allows one only to consider properties of nucleons placed near the center of heavy nuclei where the density is constant . one can expect that taking into account non - spherical effects , i.e. deformation of the skyrmion in the finite nucleus , would improve the results also on the quantitative level density changes play an important role when the nucleon is placed at sufficiently large distances from the center of nucleus . the role of deformation effects ( for an early work see @xcite ) has already been investigated in connection with the nucleon nucleon interaction in ref . the results presented in this work demonstrated the necessity of non - spherical deformed solutions of the field equations to obtain a reliable description of the @xmath4 potential . a similar approach to the deformation for two interacting skyrmions has been used in the work @xcite , where changes in the nucleon shape are investigated by letting the nucleon deform under the strong interactions with another nucleon . on the other hand , rotational and vibrational excitations of deformed skyrmions have also been considered for baryon number equal to one ( @xmath5 ) system @xcite . in a series of papers @xcite axially symmetric few skyrmion systems with baryon number @xmath6 larger than one were investigated , because spherically symmetric configurations in the skyrme model do not give bound skyrmionic systems @xcite even for the solutions of the model within the finite nuclei as considered in ref . @xcite ( for similar or related work , see also ref . @xcite ) . in the works @xcite the deformation is introduced by a modification of the unit spherical vector and additionally through the modification of the profile function @xcite in the expression of the ansatz for the chiral field ( for more precise definitions , see below ) . the results of all investigations for @xmath7 systems lead to the conclusion that the system has to be deformed to obtain the ground state . some other papers that deal with deformation effects and non - spherical configurations are collected in refs . @xcite . in the present work we consider properties of the deformed nucleon embedded in light , medium heavy and heavy nuclei . deformation effects are introduced by the distortion of the profile function of chiral field under the action of the external field ( which parameterizes the baryonic density within a given nucleus ) . in contrast to most previous investigations , we do not consider a constant nuclear density as appropriate for nuclear matter but rather the distance - dependent density as given for finite nuclei . we calculate the modifications of the nucleon properties in finite nuclei , and their dependence on the distance between the topological center of the skyrmion inside the nucleus and the geometrical center of the nucleus under consideration will be considered . the paper is organized in the following way . after the formulation of the problem in section [ sec : form ] , we discuss the classical soliton mass and distribution of baryon charge for a skyrmion inside finite nuclei in section [ sec : mass ] . based upon the results for the mass and baryon charge distribution of the deformed classical soliton , we present in section [ sec : quant ] the quantization of axially symmetric systems with baryon number @xmath5 . section [ sec : prop ] is devoted to the quantum properties of the solitons , in particular the effective mass of nucleons in finite nuclei and the consequences for @xmath0 mass splittings . we also consider the density dependence of the nucleons electromagnetic radii and calculate the intrinsic quadrupole moment of the deformed nucleon . finally , in section [ sec : concl ] the conclusions and an outlook for further investigations are given . our starting point is the medium modified skyrme lagrangian @xcite @xmath8 ^ 2 + { { \displaystyle}\frac}{f_{\pi}^2m_{\pi}^2}{16}\alpha_s(\vec r)\mbox{tr}[u+u^{\dagger}-2]\,\ , , \label{lag } \end{array}\ ] ] where @xmath9 , @xmath10mev is the pion decay constant , @xmath11 is a dimensionless parameter and @xmath12mev is the pion mass . the medium functionals @xmath13 and @xmath14 are expressed via the functionals @xmath15 and @xmath16 of @xmath17-wave and @xmath18-wave pion nucleon scattering lengths and the nuclear density @xmath19 : @xmath20 here , @xmath16 has the meaning of the pion dipole susceptibility of the medium . the influence of the medium enters through these functionals ( for explicit expressions , see section [ sec : sus ] ) . note that we employ here the same approximations as in @xcite . since the fourth order term can be obtained from the infinite mass limit of @xmath21 meson exchange , a possible density dependence of that term could enter via in medium modified @xmath21 meson properties as it was done e.g. in @xcite . similarly , one could add a sixth order term @xmath22 arising from @xmath23 meson exchange or an symmetric fourth order term due to scalar mesons . we do not consider such terms here . also , since the empirical information on possible changes of vector meson properties in the medium has not yet lead to an established picture , we treat the coupling @xmath11 as a density independent number . this could be released in more sophisticated model but goes beyond the scope of the present investigation . for a more detailed discussion about how the specific form of the lagrangian in eq . ( [ lag ] ) arises , we refer to ref . @xcite . the ansatz for the chiral field @xmath24 has the hedgehog form @xmath25 \label{anzsph}\ ] ] for the skyrmion placed at the center of the nucleus . when the skyrmion is placed at a distance @xmath26 from the center of the spherical nucleus ( see fig . [ fig1 ] ) , the chiral field @xmath24 can be written in the form @xmath27\ , , \label{ansatz}\ ] ] where @xmath28 is the unit isotopic vector correlated with the unit space vector @xmath29 . in the more general case , @xmath30 is a functional of some angular functions @xmath31 , @xmath32 @xcite . the dependence @xmath33 $ ] and the dependence of the profile function @xmath34 on the direction of the radial co - ordinate space vector @xmath35 characterizes the deformation of the skyrmion . as it was mentioned above , in the works @xcite such deformation was taken into account when considering properties of multi - skyrmionic systems . these modifications are dictated by the topological properties of the model , in particular the @xmath36 dependence is dictated by the constraint that the baryon number is larger than one , @xmath7 . here we consider the @xmath5 case and simply set @xmath37 . consequently , we can represent the vector @xmath28 in the following form @xmath38 however , this is not the only deformation effect . we also modify the profile function @xmath34 by allowing for a @xmath39 dependence , @xmath40 note that any azimuthal @xmath41 dependence of @xmath34 is removed due to the axial symmetry of the system ( see the next subsection ) . obviously these deformation effects also exist in the case when a skyrmion is placed at the center of nucleus , @xmath42 , but in this case the deformation has only a radial dependence . in the previous work @xcite this dependence was determined by solving the equations for the profile function of the ansatz ( [ anzsph ] ) in nuclear matter , @xmath43 . it is clear that in the present case the deformation has in general both a radial and an angular dependence . the angular dependence will come into play when the skyrmion is located at some distance from the center of the nucleus . the radial part of the deformation represents the breathing mode of the skyrmion and the angular part represents its deviations from the spherical form . finally , taking into account our choice of @xmath28 and @xmath34 ( eqs . ( [ greattheta ] ) and ( [ profform ] ) ) , the following ansatz for the chiral field @xmath24 arises @xmath44~. \label{ansatzlast}\ ] ] this form will be used in the present work . as a result some modifications beyond the one induced by the breathing mode will appear . for example , one expects that an intrinsic quadrupole moment of the skyrmion emerges . this will be discussed in more detail below . before discussing the in - medium skyrmion properties , we point out that throughout this paper we neglect the effects due to fermi motion of the solitons ( nucleons ) . in a strict large-@xmath45 counting , such effects are suppressed . however , since we do not expect deformation effects to be large , one can doubt the validity of this formal suppression . at present , it is simply not known how to include such effects systematically because one not only would we have to boost the solitons ( where the boost velocity could be obtained in the local density approximation ) but also the coupling to the rotational modes would considerably complicate the quantization procedure to obtain physical states . while the neglect of fermi motion might to some extent alter the quantitative results of our study , we believe , however , that the qualitative ones would remain unaltered . the mass functional in the static case can be obtained from the lagrangian ( [ lag ] ) by using the ansatz ( [ ansatzlast ] ) @xmath46d^3\vec r\ , . \end{array}\ ] ] to simplify the calculations one introduces the variable @xmath47 putting the origin of the coordinate system @xmath48 at the topological center of the skyrmion @xmath49 ( see fig . [ fig1 ] ) . then the expression for the mass functional can be written as @xmath50d^3\vec r\,'\ , . \label{massr } \end{array}\ ] ] we note that the distance @xmath51 between the skyrmion and the nucleus is a matter of convention . it depends on how one determines the center of the soliton . if the skyrmion is deformed its mass and topological centers may not coincide . in this paper we determine @xmath51 as the distance between the center or the nucleus under consideration and topological center of the skyrmion under consideration . from figure [ fig1 ] it is easy to see that the system has an axial symmetry . one can choose the direction of the @xmath52 axis along the direction of the vector @xmath26 so that the integrand in eq . ( [ massr ] ) has no azimuthal @xmath53 dependence . consequently the profile function @xmath34 only depends on the variables @xmath54 and @xmath55 . therefore , the lagrangian ( [ lag ] ) in the static case can be written in the form @xmath56\alpha_p(r\,',\theta',r)\\ \quad \\\qquad\qquad -{{\displaystyle}\frac}{1}{4e^2 } [ ( \partial_i\vec n\cdot\partial_i \vec n)^2- ( \partial_i\vec n\cdot\partial_j \vec n)^2 ] \sin^4f\\ \quad \\\qquad\qquad -{{\displaystyle}\frac}{1}{2e^2 } [ ( \partial_i\vec n\cdot\partial_i \vec n)\partial_j f\partial_jf- ( \partial_i\vec n\cdot\partial_j \vec n ) \partial_i f\partial_jf ] \sin^2f\\ \quad \\\qquad\qquad -{{\displaystyle}\frac}{m_\pi^2f_\pi^2}{4}(1-\cos f)\alpha_s(r\,',\theta',r)~ , \label{lstatic } \end{array}\ ] ] which ( after some algebraic manipulations ) leads to following mass functional for convenience . ] @xmath57 f_\theta^2+{{\displaystyle}\frac}{m_\pi^2}{e^2f_\pi^2}\left(\sin{{\displaystyle}\frac}{f}{2}\right)^2 \tilde r^2\alpha_s(\tilde r,\theta,\tilde r)\\ \quad \\\qquad \qquad \left . + \left[\sin^2f\left({{\displaystyle}\frac}{\sin^2\theta}{\sin^2\theta}+\theta_\theta^2\right ) + { { \displaystyle}\frac}{\tilde r^2\alpha_p(\tilde r,\theta,\tilde r)}{4}\right ] f_{\tilde r}^2\right\ } \sin \theta d\theta d\tilde r\ , . \label{masstat } \end{array}\ ] ] here @xmath58 , @xmath59 , @xmath60 and we have introduced the dimensionless variables @xmath61 , @xmath62 . the corresponding baryon number of the skyrmion is given by @xmath63 the static soliton solution is found by minimizing the functional ( [ masstat ] ) . for doing that , one has to solve a system of coupled differential equations @xmath64 for the functions @xmath65 and @xmath31 subject to the boundary conditions determined from the baryon number condition @xmath5 ( [ baryonnumber ] ) . to simplify the numerical computations we perform various approximations . first , consider the profile function @xmath66 for an undeformed ( spherical ) skyrmion . it is well known that @xmath66 can be parameterized to a very good approximation as @xmath67 for the free skyrmions with baryon number @xmath5 . the value of the parameter @xmath68 is obtained from the constraint of having the minimum mass functional for the free skyrmion . this parameterization in principle can be used for a bound skyrmion and we apply this ansatz in constructing our profile function ( despite some short - coming , see section [ sec : mass ] ) . also , from previous calculations we know that the dependence of the profile function on the density is weak , see e.g. @xcite . renormalization of the mass functional ( [ masstat ] ) mainly occurs via the medium functionals @xmath69 and @xmath70 . furthermore , from fig . ( [ fig1 ] ) we see that changes of @xmath55 mainly affect the value of medium functionals and @xmath70 are functionals of the medium density @xmath71 . ] and following the results from ref . @xcite , we assume changes of profile function to be small . using these facts we can represent our profile function as @xmath72\right\}~ , \label{proflast}\ ] ] where the @xmath73 functions are chosen to maintain periodicity in @xmath39 . for the function @xmath74 we choose the parameterization @xmath75 in order to avoid singularities from factors like @xmath76 in the expression of the mass functional ( [ masstat])@xcite . in eqs . ( [ proflast ] ) and ( [ thetalast ] ) , @xmath68 , @xmath77 , @xmath78 , @xmath79 , @xmath80 and @xmath81 , @xmath82 , @xmath83 , @xmath80 are variational ( or in other words deformation ) parameters . our aim is the minimization of mass the functional ( [ masstat ] ) using the expressions ( [ proflast ] ) and ( [ thetalast ] ) for the profile function . before doing that , we have to specify the medium dependence that enters through the @xmath17- and @xmath18-wave pion - nucleon functionals @xmath84 and @xmath85 , respectively . the medium functionals @xmath84 and @xmath85 introduced in eqs . ( [ lag ] ) , ( [ medfunc ] ) have the form @xcite . for asymmetric nuclei , additional isovector effects will be present but are not considered here . ] @xmath86 where @xmath87 is a kinematical factor , @xmath88mev the mass of the nucleon and @xmath89 is the migdal parameter which takes into account the short range pair correlations of the dipole centers . the empirical parameters @xmath90 , @xmath91 @xcite can be taken from analyses of low energy pion - nucleus scattering data . we choose the following parameterization for the density of the finite spherical nuclei considered here @xcite : @xmath92 \exp\left\{-{{\displaystyle}\frac}{r^2}{r_{0}^{2}}\right\}\,,\quad & \mbox{for $ 4 < a < 20$}\,,\\ \quad \\ \rho(r ) = \left({{\displaystyle}\frac}{a-1}{a}\right){{\displaystyle}\frac}{\rho_0 } { 1+\exp\left\{{{\displaystyle}\frac}{r - r'}{a}\right\}}\ , , & \mbox{for $ a\ge 20$}\ , . \label{density } \end{array}\ ] ] in the actual calculations ( minimization procedure ) we have considered the nuclei @xmath93c , @xmath94o , @xmath95ca , @xmath96fe , @xmath97au and @xmath98pb . the first two represent _ light _ , the next two _ medium - heavy _ and the last two _ heavy _ nuclei . the parameter @xmath99 depends on the type of nucleus considered . its value is @xmath100 fm for @xmath93c and @xmath101 fm for @xmath94o . @xmath102 is the charge of the nucleus and @xmath3 is the normal nuclear matter density . furthermore , @xmath103 fm and @xmath104 fm . @xmath105 is the baryon number of the nucleus and the factor @xmath106 allows one to consider properties of a given nucleon which is one of the nucleons of the nucleus . for the accuracy of our calculations , such a simple rescaling of the nuclear density should be sufficient . the density is pameterized as @xmath107 as dictated by the axial symmetry of the system . for the input parameters of the skyrme model we use @xmath108 mev and @xmath109 in order to reproduce the free space masses of the nucleon and the @xmath0 as in the seminal work @xcite . one could also work with the empirical value of @xmath110 and adjust @xmath11 to the @xmath111 mass splitting . in that case , however , the nucleon mass would be too large , but relative changes would be affected considerably less @xcite . for definiteness , we prefer here to work with the physical nucleon mass . our calculations have been performed using different sets of the variational parameters @xmath68 , @xmath77 , @xmath78 , @xmath112 , @xmath82 , @xmath113 . we represent the skyrmion size parameter @xmath68 as @xmath114 where @xmath115 is the value for the skyrmion in free space ( @xmath116 fm ) . the minimization procedure of the mass functional ( [ masstat ] ) showed that considering the set of parameters @xmath117 is sufficiently accurate . the strength of the parameter @xmath118 decreases with increasing index @xmath119 . for example , @xmath79 is one order of magnitude smaller than @xmath77 at the distances where deformation effects are maximal . the parameter @xmath120 decrease very fast with increasing index @xmath121 . for example , @xmath82 is one order of magnitude smaller than @xmath81 and @xmath83 is almost zero . for controlling our results and keeping a high accuracy , we have performed calculations with the following set of parameters : @xmath122 for the reasons just mentioned , we only show results of @xmath77 , @xmath78 and @xmath81 . the dependence of the skyrmion mass from the distance between the centers of the nucleus considered and the soliton ( which represents one of the nucleons within the nucleus ) is shown in figs . [ co ] , [ cafe ] , [ aupb ] , for the light , medium - heavy and heavy nuclei , in order . the bands labeled `` 1 '' in these figures represent the soliton mass normalized to its free space value and for convenience we display by the bands denoted `` 2 '' the nuclear density normalized to its value at the center of the nucleus . these bands are shown in the top panels of these figures . the behavior of the soliton mass is similar for all cases when the soliton is placed within various nuclei , i.e. the mass increases with falling medium density and approaches its free space value at the border of the nucleus . it has its minimum value at the center of the nuclei . this is even the case for the light nuclei , which do not have their maximum density exactly in their center . the deformation parameters which represent the deviations of the nucleon shape from the spherical one are presented in the bottom panels of figs . [ co ] , [ cafe ] , [ aupb ] . the bands denoted `` 3 '' represent the dependence of the relative value of the radial deformation parameter @xmath123 from soliton location inside nucleus and the bands `` 4 '' and `` 5 '' give the dependence for @xmath77 and @xmath78 , respectively . the band `` 6 '' represents the parameter @xmath81 multiplied by factor 10 . the other deformation parameters are very small and we do not show them here . the leading parameter @xmath77 has its maximum value at the distances where the density of the nuclei falls off by @xmath12420% for @xmath93c and @xmath94o , @xmath124 40% for @xmath95ca , @xmath96fe , @xmath97au and @xmath98pb , whereas the radial deformation parameter @xmath125 has its maximal value at the center of the nuclei . the parameters @xmath78 and @xmath81 have maximum values where the density of the nucleus falls off @xmath12480% . in these regions , the density gradient is maximal leading to the largest angular deformation . from this one concludes that the skyrmions inside nuclei is in a deformed state . the topological baryon charge distribution for a skyrmion embedded in the nuclei @xmath93c and @xmath96fe is shown in figs . [ c12bc ] , [ fe56bc ] , respectively , where figures * a * ) represent the baryon charge distribution in the @xmath126 plane and figures * b * ) give its projection onto this plane . each area at the bottom figure represents the projection of surfaces of equal baryon charge density . the point @xmath127 indicates the topological center of the skyrmion and the center of nucleus is situated at the negative side of the @xmath128 axis . one can see that opposite directions in the @xmath128 direction are not equivalent in this plane . stated differently , going for the same distance in opposite directions in @xmath128 gives different values of the baryon charge density . we remark that the holes at the origin of the system ( @xmath127 ) are caused by the parameterization of the profile function ( see eq . ( [ proflast ] ) ) . in this approximation , @xmath129 as @xmath130 . this derivative takes a finite value if one directly solves the equation coming from minimization of the mass functional ( [ masstat ] ) not using the ansatz eq . ( [ profpar ] ) , thus no such holes would appear . however , the effects discussed so far and in the following are genuine and independent of the ( dis)appearance of these computational artifacts . the light area at positive values of the axis @xmath128 is the area of maximal density . the mass distribution , i.e. integrand in the expression ( [ masstat ] ) shows a behavior similar to the one of baryon charge distribution . an exotic behavior of the mass and baryon charge distributions is observed when the skyrmion is placed at sufficiently large distances from the center of the nucleus , where density changes are significant ( see figs . [ c12bc ] , [ fe56bc ] ) . one can see that the maximum of the baryon density is shifted away from the center of the nuclei . when the skyrmion is located at the center of nucleus , the baryon density distribution is symmetric and thus the shift becomes more effective with increasing of @xmath131 . the shift of the density maximum from the topological center of the skyrmion in opposite direction to the direction of center of the nucleus can be explained in the following way . the mass of skyrmion decreases in the nuclear medium and if the density is decreasing , the skyrmion mass increases . while we are going from the center of the nucleus to its surface the density fall - off and mass concentration of the skyrmion occurs at the side of the smaller nuclear density as seen from its topological center . a similar behavior of the baryon charge density for other nuclei is observed , so we do not present it here . finally , at the end of this section we remark that rotations of the figures * b ) * around the @xmath128axis give three dimensional surfaces with an equal baryon charge distribution . from this we conclude that the soliton has two equal components of the moment of inertia which are not equal to the third one , @xmath132 . we will use this fact in the next section . as it is well known , baryons emerge in the skyrme model as quantized solitons @xcite . this quantization is performed in terms of an adiabatic rotation of the soliton and gives rise to the appropriate quantum numbers for the emerging physical states . in such an approach , the grand spin @xmath133 , i.e. the sum of angular momentum @xmath134 and isospin @xmath135 , is conserved . this leads to a tower of states with @xmath136 . it is well known that in this case only the nucleon ( @xmath137 ) and the @xmath0 ( @xmath138 have physical meaning . all other particles ( higher rotational states ) should be considered artifacts of the model and will be discarded . to quantize the solitons of our model we use the standard canonical quantization procedure performing time dependent rotations in isospace and in co - ordinate space @xmath139 where the matrix @xmath140 specifies the rotation in isospace and @xmath141 the one in co - ordinate space @xcite . the parameters of the rotation matrices @xmath140 and @xmath142 serve as collective variables describing the rotational degrees of freedom with respect to a body - fixed system of reference . the latter has its origin at the topological center @xmath49 of the skyrmion and points in the direction of the distance vector @xmath26 from the center of the nucleus to the center of the skyrmion ( see fig . [ fig1 ] ) . the quantization procedure used here is similar to the standard one @xcite but now the axial symmetry of the system should be taken into account @xcite . the angular velocities of isorotations @xmath143 and space rotations @xmath144 can be determined from the respective equations @xmath145 where dot means differentiation with respect to the time component , and @xmath146 is the totally antisymmetric tensor in three dimensions . using now the anstze ( [ ansatzlast ] ) , ( [ rotansatz ] ) in the lagrange density ( [ lag ] ) and integrating over the whole space one gets following expression @xmath147 @xmath148 \sin\theta d\theta d\tilde r\,,\\ \label{angmom2 } i_{\omega\omega}^{(12)}(r ) & = & { { \displaystyle}\frac}{\pi}{2e^3f_\pi}{\displaystyle}\int\limits_0^\infty\int\limits_{0}^\pi \left[{{\displaystyle}\frac}{\tilde r^2}{4}{{\displaystyle}\frac}{\sin^2\theta \cos^2\theta}{\sin^2\theta}\sin^2f+ \left({{\displaystyle}\frac}{\tilde r^2}{4}+{{\displaystyle}\frac}{(3+\cos2\theta)}{2 } { { \displaystyle}\frac}{\sin^2\theta}{\sin^2\theta}\sin^2f\right)f_\theta^2 \right.{\nonumber}\\ & & \qquad\qquad\left . + { { \displaystyle}\frac}{\tilde r}{2}\left ( \tilde r^2 + 4{{\displaystyle}\frac}{\sin^2\theta}{\sin^2\theta}\sin^2f\right ) f_\theta f_{\tilde r}\sin(\theta-\theta ) \right.{\nonumber}\\ & & \qquad\qquad\left . + { { \displaystyle}\frac}{\tilde r^2}{4}\left(2\,[\,2-\cos2(\theta\!-\!\theta)+\cos2\theta\ , ] { { \displaystyle}\frac}{\sin^2\theta}{\sin^2\theta}\sin^2f+\tilde r^2\sin^2(\theta\!-\!\theta ) \right ) f_{\tilde r}^2 \right.{\nonumber}\\ & & \qquad\qquad\left . + { { \displaystyle}\frac}{1}{4}\sin^2f\left(\tilde r^2 + 2(3+\cos2\theta){{\displaystyle}\frac}{\sin^2\theta}{\sin^2\theta } \sin^2f \right)\theta_\theta^2 \right . { \nonumber}\\ & & \qquad\qquad\left . -{{\displaystyle}\frac}{\tilde r^2}{2}\sin^2f[-3+\cos2(\theta-\theta)\ , ] f_{\tilde r}^2\theta_\theta^2\right ] \sin\theta d\theta d\tilde r\,,\\ \label{angmom3 } i_{\omega\omega}^{(12)}(r ) & = & { { \displaystyle}\frac}{\pi}{2e^3f_\pi}{\displaystyle}\int\limits_0^\infty\int\limits_{0}^\pi \sin^2f\left [ \left({{\displaystyle}\frac}{f_\theta^2}{2}+{{\displaystyle}\frac}{\tilde r^2}{8 } ( 1 + 4f_{\tilde r}^2)\right){{\displaystyle}\frac}{\cos\theta}{\sin\theta}\sin2\theta \right . { \nonumber}\\ & & \qquad\qquad\left . + \left({{\displaystyle}\frac}{\tilde r^2}{4}+ { { \displaystyle}\frac}{\sin^2\theta}{\sin^2\theta}\sin^2f- \tilde rf_{\tilde r } [ f_\theta\sin(\theta-\theta)-\tilde rf_{\tilde r}]\right)\theta_\theta \right . { \nonumber}\\ & & \qquad\qquad\left . + { { \displaystyle}\frac}{\cos\theta}{2\sin\theta}\theta_\theta^2\sin^2f\sin2\theta\right ] \sin\theta d\theta d\tilde r\,,\\ i_{\omega\omega}^{(33)}(r ) & = & { { \displaystyle}\frac}{\pi}{2e^3f_\pi}{\displaystyle}\int\limits_0^\infty\int\limits_{0}^\pi \sin^2f\left ( { { \displaystyle}\frac}{\tilde r^2}{2}+ 2f_\theta^2 + 2\tilde r^2f_{\tilde r}^2 + 2\theta_\theta^2\sin^2f\right)\sin^2\theta \sin\theta d\theta d\tilde r\ , , \label{angmom4}\end{aligned}\ ] ] where @xmath149 is static mass functional ( eq . ( [ masstat ] ) ) and @xmath150 , @xmath151 , @xmath152 , @xmath153 are four moments of inertia of the skyrmion . in deriving equations ( [ angmom])-([angmom4 ] ) we have made use of the axial symmetry of the system . defining now canonical conjugate variables in the body - fixed reference system @xmath154 one obtains the hamiltonian of the system @xmath155}+{{\displaystyle}\frac}{\hat t_3 ^ 2}{4i_{\omega\omega}^{(33 ) } } \nonumber\\ & = & m+{{\displaystyle}\frac}{\hat { \bf t}^2i_{\omega\omega}^{(12 ) } + \hat { \bf j}^2i_{\omega\omega}^{(12 ) } + 2\hat { \bf t } \cdot \hat{\bf j}i_{\omega\omega}^{(12 ) } -\left(i_{\omega\omega}^{(12 ) } + i_{\omega\omega}^{(12 ) } -2i_{\omega\omega}^{(12)}\right)\hat t_3 ^ 2 } { 4\left[i_{\omega\omega}^{(12)}i_{\omega\omega}^{(12)}- \left(i_{\omega\omega}^{(12)}\right)^2\right]}+{{\displaystyle}\frac}{\hat t_3 ^ 2}{4i_{\omega\omega}^{(33)}}\ , . \label{hamiltonian}\end{aligned}\ ] ] here @xmath156 is the isospin operator , @xmath157 the spin operator . if in addition one considers states with @xmath158 , where @xmath159 is the grand spin , and sandwiches the hamiltonian between wave functions @xmath160 in the body - fixed system of reference , the energy of the system takes the form @xmath161 \left(t(t+1)-t_3 ^ 2\right ) } { 4\left[i_{\omega\omega}^{(12)}i_{\omega\omega}^{(12)}- \left(i_{\omega\omega}^{(12)}\right)^2\right ] } + { { \displaystyle}\frac}{t_3 ^ 2}{4i_{\omega\omega}^{(33)}}\ , , \end{array } \label{energy}\ ] ] where we used the constraint @xmath162 which is due to the axial symmetry of ansatz ( [ ansatzlast ] ) . the degeneracy on the quantum number @xmath163 is partially lifted , i.e. the energy of the skyrmion only coincides for states with @xmath164 . for example , the following pairs of baryons have the same masses : @xmath165 but the baryons in the states @xmath166 and @xmath167 do _ not _ have the same masses . the above specified quantum numbers refer to the body - fixed system of reference . in contrast to a deformed skyrmion in an isotropic background ( e.g. the vacuum or isotropic and isosymmetric nuclear matter ) @xcite , our present case corresponds to a deformed nucleon or delta in an unisotropic background given by the nuclear density profile . therefore , the projection onto good quantum numbers in the laboratory system of reference does not follow here from a simple rotation of the intrinsically deformed object . in fact , only the third components of angular momentum and isospin in the body - fixed system , @xmath163 and @xmath168 , are good quantum numbers . as a check , let us consider whether we can obtain the canonical spherically symmetric solutions from our deformed ones . first , it is easy to see that @xmath169 where @xmath170 is the moment of inertia of the spherically symmetric hedgehog ansatz , @xmath171 consequently @xmath172 and one obtains the well known energy expression for a baryon in the spherically symmetric case @xmath173 more generally , the following relation for all observables can be shown to hold : @xmath174 this shows that our construction has the proper limit when all deformation parameters are set to zero . predictions for the masses of the quantized solitons are presented in fig . [ nmass ] , where we show the nucleon mass normalized to its free space value . note that the distance @xmath131 between the centers of the nucleon and the nucleus is normalized to the corresponding radius @xmath175 of the nucleus for given baryon number @xmath105 . we show the behavior of the nucleon mass only for one representative nucleus of the three classes , i.e. for @xmath94o , @xmath96fe and @xmath97au . the other nuclei from the same class exhibit a similar behavior . for the radii of the nuclei we use formula @xmath176 the behavior of the nucleon mass in the nucleus is similar for all light , medium heavy and heavy nuclei . when the density decreases , the mass of the nucleon increases approaching its free space value at the edge of the nucleus . we conclude that the mean effective mass of the nucleon in real nuclei will not decrease as much as @xmath2 in contrast to the case of homogeneous nuclear matter with normal density @xmath177 . the finite nuclei approach for the medium essentially improves the results , especially for light nuclei , i.e. taking into account realistic mass distributions one obtains more reasonable values for the mass renormalization . for example , this renormalization is about @xmath178 for a nucleon placed in the center of @xmath93c or @xmath94o . in principle , one could now calculate an average medium modification in a given nucleus by simply folding the calculated effective mass with the nuclear density . we do not show the results of such calculations here because the calculation could be further refined . if one uses e.g. the shell structure of the nuclei , then all nucleons will be situated at some distances ( within the shells ) from their mass center and the mean effective value of the nucleon mass in such systems will be further reduced for nucleons in heavy nuclei . further improvements of the results can be made by including a dilaton field @xcite . we also note that a similar renormalization of the nucleon mass was found in ref.@xcite . as it was noted in the previous section , the mass of the @xmath0 states depends on the absolute value of the third component of isospin . we found small deviation from degeneracy for delta states with @xmath179 and @xmath180 . the difference of the masses @xmath181 of @xmath0 in the states @xmath166 and @xmath167 ( see eq . ( [ states ] ) ) has a negative value near the center of the nucleus and after reaching its minimum begins to increase towards positive values reaching at some distance its maximal value and then smoothly drops to zero . for example , it has its minimal value @xmath182mev at @xmath183 fm and maximal value @xmath184mev at @xmath185 fm for deltas in @xmath93c . for the case of @xmath96fe , the corresponding values are @xmath186(@xmath187 @xmath188mev and @xmath186(@xmath189 @xmath190mev . we remind the reader that all these numbers refer to states defined in the body - fixed system of reference as discussed before . for the relative mass difference @xmath191 we obtained a behavior similar to the one of nucleon mass ( cf . fig.([nmass ] ) ) and do not graphically present it here . the difference @xmath191 decreases when the density increases and has its minimal value at the center of the nucleus . more specifically , @xmath192 has the value 0.867 for @xmath93c , 0.615 for @xmath96fe and 0.609 for @xmath97au , in the respective center of the nucleus . note that in free space @xmath193mev in the current approach . as to the ratio of @xmath0 and @xmath194 masses , @xmath195 , it is almost constant inside the nucleus and approximately equal to that ratio in free space . the calculation of the electromagnetic currents reduces to the calculation of the isoscalar and the isovector currents . in topological models and in particular in the skyrme model , the isoscalar current is equal to the model independent baryon number current ( for a discussion on the role of vector mesons , see ref.@xcite ) @xmath196 with @xmath197 the totally antisymmetric tensor in four dimensions . similarly , the isovector current , which is the noether current associated to the global transformations @xmath198 ( @xmath199 are @xmath200 matrixes ) , has the form in the equation ( [ veccur ] ) while there is one over the index @xmath201 . ] @xmath202 + [ r_{\nu},[r_{\mu},r_{\nu}]]\}\ , , \label{veccur}\ ] ] where @xmath203 then , the operator of the electric charge of the nucleon is determined by the zeroth components of the isoscalar and the isovector currents , @xmath204 where @xmath205 is the third component of the isovector current and @xmath206 is the third component of the isotopic spin operator . sandwiching this current between corresponding nucleon states , we obtain the proton and neutron charge distributions . the normalized isoscalar root - mean - square radius along each axis , @xmath207 , ( normalized with the respect to the isoscalar charge ) can be determined from the zeroth component of the baryon current and has the form @xmath208 where @xmath209 ( @xmath210 ) indicates the projection of the radius vector on the corresponding axis . similarly , the isovector root - mean - square radii ( normalized to the isovector charge ) , @xmath211 , are determined by the zeroth component of the vector current @xmath205 @xmath212 where @xmath213\,.\ ] ] one can see that the influence of the medium enters indirectly via the modified profile functions . results for the in medium radii are collected in table [ table ] . we give the isoscalar @xmath214 and isovector @xmath215 root - mean - square radii of the nucleons along the @xmath216 and @xmath128 axes inside the nuclei @xmath94o , @xmath96fe and @xmath97au , evaluated in the center of each nucleus and at those distances where the intrinsic quadrupole moment of the nucleons has its first and second extremum , correspondingly ( see the discussion below ) . we note that due to the axial symmetry @xmath217 . clearly , these radii swell , as first pointed out in ref.@xcite in the analysis of data on quasi - elastic electron - nucleus scattering . however , this swelling is not isotropic and depends very sensitively on the distance from the center of the nucleus . the concept of a common swelling by one scale factor is certainly much too simple . our results are therefore not in contradiction with the y - scaling analysis of ref.@xcite ( see also the discussion in @xcite on this issue ) . as noted before due to the deformation the nucleon can have a nonvanishing _ intrinsic _ quadrupole moment @xcite @xmath218 that is characterized by the third component @xmath163 of the angular momentum in the body - fixed frame of reference . the corresponding isoscalar and isovector intrinsic quadrupole moments follow from ( [ isosr ] ) and ( [ isovr ] ) by appropriate projection @xmath219 consequently , the proton and neutron intrinsic quadrupole moments follow from @xmath220 the dependence of the isoscalar and isovector intrinsic quadrupole moments as a function of @xmath221 is shown in fig . again we present results only for the nuclei @xmath94o , @xmath96fe and @xmath97ag . for the other nuclei the dependences are similar . one can see that intrinsic quadrupole moments are small . this smallness of the intrinsic quadrupole moments means that nucleons in the nucleus are weakly deformed . the intrinsic quadrupole moments has two extrema . near the center of nucleus they have negative values and a corresponding minimum at that distances where density changes are @xmath12420% for @xmath94o and @xmath1247% for @xmath96fe and @xmath97au , respectively . then with increasing density the intrinsic quadrupole moments become positive and have a maximum at the distance where the density of the nucleus has fallen by @xmath222% . note also that the isoscalar intrinsic quadrupole moment is about two times smaller than the isovector one . consequently near the center of the nucleus protons are flattened along the axis @xmath128 ( oblate ) and and near the border of the nucleus they are extended ( prolate ) . in contrast neutrons near the center of nucleus slightly prolate and near the border of nucleus oblate . consequently , at some intermediate distance the nucleons must have spherical form . finally , let us recall that the observable static quadrupole moment in the laboratory system of an intrinsically deformed baryon in an isotropic background is linked to its intrinsic quadrupole moment by the wigner - eckhardt theorem as follows @xcite @xmath223 where the relations are valid for maximally projected states with @xmath224 . thus for a deformed baryon of total angular momentum @xmath225 the quadrupole moment in the _ laboratory _ system is zero for its ground state , i.e. for the nucleon . it would be interesting to investigate the transition quadrupole moments to excited states , extending the work of @xcite . as mentioned before , our case corresponds to a deformed baryon in an unisotropic background . a simple projection to the laboratory system is therefore not possible . in summary , we have considered properties of a nucleon embedded in finite nuclei making use of the skyrme model . deformation effects are taken into account by the distortion of chiral profile function under the action of the nuclear medium , parameterized in terms of the functionals eq.([funcrho ] ) . nucleons and deltas emerge as quantized solitons after performing an adiabatic rotation in co - ordinate and isospace which takes into account the axial symmetry of the system ( leading to two different moments of inertia , @xmath226 . the main results of this study can be summarized as follows : * the density dependence of the nucleon mass shows a more realistic behavior than in the case of a uniform density as e.g. for homogeneous nuclear matter . the effective nucleon mass has its minimum at the center of the nucleus and approaches its free space value at the surface of the nucleus . * the nucleons in finite nuclei are weakly deformed . they acquire a small intrinsic quadrupole moment which , however , is strongly dependent on the distance from the center of the nucleus , e.g. the proton ( neutron ) deformation changes from a oblate ( prolate ) to a prolate ( oblate ) form as it is moved toward the surface of the nucleus . * similarly , there is a direction - dependent swelling of the isoscalar and isovector root - mean - square radii . we have stressed that the concept of a uniform swelling factor is too simple a concept to apply to real nuclei . * as consequence of the axial symmetry of the system , the @xmath227 @xmath0 states ( @xmath228 ) and the @xmath229 ones ( @xmath230 ) have slightly different masses in finite nuclei . let us briefly give an outlook for further studies based on these results . as it was mentioned above this approach can be used for model investigations of the shell structure of nuclei and to study the properties of a remote nucleon in such a system ( halo nuclei ) . in the latter case the nucleon in the finite nucleus should not be considered as one within the nucleus but weakly coupled to the nucleus at large distances . also , the model naturally allows to study nucleon properties in the lightest nuclei , especially the electromagnetic structure of nucleons in @xmath231he , which is of current experimental interest . work along these lines is under way . the work of u. yakhshiev has been supported by intas ys fellowship @xmath23200 - 51 . we thank andy jackson for a useful correspondence . 99 a.m. rakhimov , m.m . musakhanov , f.c . khanna and u.t . yakhshiev , phys . rev . * c58 * , 1738 ( 1998 ) . a.m. rakhimov , f.c . khanna , u.t . yakhshiev , m.m . musakhanov , nucl . phys . * a643 * , 383 ( 1998 ) ; + m. musakhanov , a. rakhimov , u. yakhshiev , z. kanokov , phys . atom . j. nucl . phys . ) * 62 * , 1988 ( 1999 ) . meiner , phys . lett . * 62 * , 1013 ( 1989 ) ; phys . lett . * b220 * , ( 1989 ) ; nucl . * a503 * , 801 ( 1989 ) . i. klebanov , nucl . phys . * b262 * , 133 ( 1985 ) ; + n.k . glendenning , phys . rev . * c34 * , 1072 ( 1986 ) ; + e. wst , g.e . brown and a.d . jackson , nucl . phys . * a 468 * , 450 ( 1987 ) ; + a.s . goldhaber and n.s . manton , phys . b198 * , 231 ( 1987 ) ; + t.s . walhout , nucl . phys . * a484 * , 397 ( 1988 ) ; nucl . phys . * a 519 * , 816 ( 1990 ) ; + a.d . jackson , a. wirzba and l.s . castillejo , phys . lett . * b198 * , 315 ( 1987 ) ; nucl . phys . * a486 * , 634 ( 1988 ) ; + m. kugler and s. shtrikman , phys . lett . * b208 * , 491 ( 1988 ) ; phys . rev . * d40 * , 3421 ( 1989 ) ; + l.s . et al . _ , a501 * , 801 ( 1989 ) ; + h. forkel _ et al . _ , . phys . * a504 * , 818 ( 1989 ) ; + a.d . jackson , c. weiss and a. wirzba , nucl . a529 * , 741 ( 1991 ) ; + a. wirzba , in `` baryons as skyrme solitons '' , g. holzwarth ( ed . ) , world scientific , singapore , 1993 ; + w.k . baskerville , nucl . phys . * a596 * , 611 ( 1996 ) ; + p. amore , nuovo cim . * a111 * , 493 ( 1998 ) . v. vento , g. baym and a.d . jackson , phys . b102 * , 97 ( 1981 ) . u.b . kaulfu and ulf - g . meiner , phys . rev . * d31 * , 3024 ( 1985 ) . a. rakhimov , t. okazaki , m.m . musakhanov , f.c . khanna , phys . b378 * , 12 ( 1996 ) . c. hajduk , b. schwesinger , phys . lett . * b140 * , 172 ( 1984 ) ; + c. hajduk , b. schwesinger , nucl . phys . * a453 * , 620 ( 1986 ) . kopeliovich , b.e . stern , sov . j. jetp lett . * 45 * , 165 ( 1987 ) ; + v.b . kopeliovich , sov . j. nucl.phys . * 47 * , 1495 ( 1988 ) ; + b.e . stern , sov . j. nucl.phys . * 49 * , 530 ( 1989 ) ; + v.b . kopeliovich , russ . j. nucl.phys . * 56 * , 160 ( 1993 ) . nikolaev , sov . * 20 * , 401 ( 1989 ) ; + v.a . nikolaev , o.g . tkachev , sov . * 21 * , 1499 ( 1990 ) ; + r.m . nikolaeva , v.a . nikolaev , o.g . tkachev , sov . * 23 * , 542 ( 1992 ) . t. kurihara , h. kanada , t. otofuji , s. saito , prog . . phys . * 81 * , 858 ( 1989 ) . bogomolniy , v.a . fateev , sov . . phys . * 37 * , 228 ( 1983 ) . yakhshiev , n.a.taylanov , uzb . j. phys . * 2 * , 114 ( 2000 ) . e. braaten and l. carson , phys . * 56 * , 1897 ( 1986 ) ; + e. braaten and l. carson , phys . rev . * d38 * , 3525 ( 1988 ) ; + e. braaten , s. townsend and l. carson , phys . lett . * b235 * , 147 ( 1990 ) ; + n.s . manton , phys . b192 * , 177 ( 1987 ) ; + m.f . atiyah and n.s . manton , commun . phys . * 153 * , 391 ( 1993 ) ; + r.a . leese and n.s . manton , nucl . a572 * , 575 ( 1994 ) ; + c.j . houghton , n.s . manton and p.m. sutcliffe , nucl . * b510 * , 507 ( 1998 ) . i. zahed _ et al . _ , phys . rev . * d31 * , 1114 ( 1985 ) ; + c .- k . lin , nucl . * a449 * , 673 ( 1986 ) ; + j. wambach , h.w . wyld and h.m . sommermann , phys . lett . * b186 * , 272 ( 1987 ) ; + t. otofuji _ . phys . * 78 * , 527 ( 1987 ) ; + i. mishustin , sov . jetp * 71 * , 21 ( 1990 ) ; + f. leblond and l. marleau , phys . rev . * d58 * , 054002 ( 1998 ) . a. wirzba and v. thorsson , nucl . * a589 * , 633 ( 1985 ) . t. ericson and w. weise , _ pions and nuclei _ ( claredon , oxford , 1988 ) . a. akhiezer , a.g . sitenko , v.k . tartakovski , _ nuclear electrodynamics _ ( springer verlag , 1994 ) . adkins , c.r . nappi and e. witten , nucl . phys . * b228 * , 552 ( 1983 ) ; + g.s . adkins and c.r . nappi , _ ibid . _ * b233 * , 109 ( 1984 ) . j. schechter , phys rev . * d21 * , 3393 ( 1980 ) ; h. gomm _ et al . _ , phys . rev . * d33 * , 3476 ( 1986 ) . meissner , a. rakhimov and u.t . yakhshiev , phys . lett . * b473 * , 200 ( 2000 ) . meiner , phys . rep . * 161 * , 213 ( 1988 ) . noble , phys . * 46 * , 412 ( 1981 ) . i. sick , phys b157 * , 13 ( 1985 ) . a. bohr and b.r . mottelson , _ nuclear structure , vol . ii , nuclear deformations _ ( w.a . benjamin , reading , massachusetts , 1975 ) . . [ table ] the isoscalar @xmath214 and isovector @xmath215 nucleon root - mean - square radii along the @xmath216 and @xmath128 axes inside various nuclei . the values of these quantities are given for the center of the nuclei and at those distances where the nucleon intrinsic quadrupole moments have their first and second extremum values , correspondingly . @xmath131 is the distance between centers of the nucleus and skyrmion topological center and the last column is the ratio of the density of nucleus @xmath233 to its value at the center @xmath234 . we note that @xmath235 . for comparison we also give the free space values of the root - mean - square radii . [ cols="^,^,^,^,^,^,^",options="header " , ]

having in mind the geometric interpretation of the group structure on a cubic ( see figure [ fig : kubika ] ) , [ fig : kubika ] the question of finding an analogous construction of the group structure in higher genera arises . in this paper , we show that such a geometric construction exists in the case of hyperelliptic jacobians . our ideas are continuation of those of reid , donagi and knrrer , see @xcite , @xcite , @xcite and also @xcite , @xcite . further development , realization , simplification and visualization of their constructions is obtained by using the ideas of billiard dynamics on pencils of quadrics developed in @xcite . the key projective - geometric feature of that billiard dynamics is the double reflection theorem , see theorem [ th : drt ] below . there are four lines belonging to a certain linear space and forming _ the double reflection configuration _ : these four lines reflect to each other according to * the billiard law at some confocal quadrics*. this billiard configuration appears to be the exact genus two generalization of three points belonging to a line in the elliptic case . in higher genera , we construct the corresponding more general billiard configuration , again by using double reflection theorem . this configuration , which we call _ @xmath3-brush _ , is in one of the equivalent formulations , a certain billiard trajectory of length @xmath6 and the sum of @xmath3 elements in the brush is , roughly speaking , the final segment of that billiard trajectory . the milestones of this paper are @xcite and @xcite and the key observation giving a link between them is _ that the correspondence @xmath7 in lemma 4.1 and corollary 4.2 from @xcite is _ * the billiard map at the quadric * @xmath8 . thus , after observing and understanding the billiard nature behind the constructions of @xcite , @xcite , @xcite , we become able to use the billiard tools to construct and study hyperelliptic jacobians , and particularly their real part . any real hyperelliptic jacobian may be realized as a set @xmath0 of lines simultaneously tangent to given @xmath1 quadrics @xmath9 , ... , @xmath10 of some confocal family in the @xmath2-dimensional euclidean space . it is well known that such a set @xmath0 is invariant under the billiard dynamics determined by quadrics from the confocal family . by using double reflection theorem and some other billiard constructions we construct a group structure on @xmath0 , _ a billiard algebra_. the usage of billiard dynamics in algebro - geometric considerations appears to be , as usually in such a situation , of a two - way benefit . we derive a fundamental property of @xmath0 : _ any two lines in @xmath0 can be obtained from each other by at most @xmath1 billiard reflections at some quadrics from the confocal family_. the last fact opens a possibility to introduce new hierarchies of notions : of _ @xmath3-skew lines in @xmath0 , @xmath11 _ and of _ @xmath3-weak poncelet trajectories of length @xmath12_. the last are natural quasi - periodic generalizations of poncelet polygons . by using billiard algebra , we obtain complete analytical description for them . these results are further generalizations of our recent description of cayley s type of poncelet polygons in arbitrary dimension , see @xcite . let us emphasize that the method used here , based on billiard algebra differs from the methods exposed in @xcite . the interrelations between billiard dynamics , subspaces of intersections of quadrics and hyperelliptic jacobians developed in this paper , enable us to obtain higher dimensional generalizations of several classical results . to demonstrate the power of our method , we present here generalizations of weyr s poncelet theorem ( see @xcite ) and also griffiths - harris space poncelet theorem ( see @xcite ) in arbitrary dimension are derived and presented here . we also give an arbitrary - dimensional generalization of the darboux theorem @xcite . let us mention that one of the plane versions of darboux theorem has been recently rediscovered , with some improvements , in @xcite . in that paper , the obtained configuration is called _ poncelet grid _ , while we name it here _ poncelet - darboux grid _ , which is , by our opinion , more historically justified . the paper is organized as follows . section [ sec : preliminaries ] consists of preliminaries . it starts with a review of confocal families of quadrics and their most important properties , in the @xmath2-dimensional euclidean and projective space . the poncelet theorem in the three - dimensional projective space over an arbitrary field is proved , giving accent on projective definitions and methods . we review the one reflection theorem and the double reflection theorem ( drt ) . the latter appears to be one of the main projective tools used in the paper . the notions of virtual reflections and virtual reflection configuration are recalled and some of their basic properties from @xcite are reviewed . we recall the definition of the generalized cayley s curve from @xcite , and discuss its form in the real case . section [ sekcija : bilijarskaalgebra ] is the kernel of the paper . in section [ sekcija : morfizam ] , relying on the simple lemma [ lema : veza ] , we begin to build a bridge between knrrer s constructions from @xcite and the billiard notions from our previous article @xcite and this paper . then , we construct a birational morphism between the generalized cayley s curve and the reid - donagi - knrrer curve ( see @xcite,@xcite,@xcite ) . in section [ sec : genus2 ] , we construct a group structure on the set of lines in @xmath13 that are simultaneously tangent to two given confocal quadrics . this group structure is naturally connected with the billiard law . using its properties , some beautiful properties of confocal families are derived , e.g. theorem [ th : zvezda ] . in section [ sec : general ] , set @xmath0 of lines simultaneously tangent to given @xmath1 quadrics @xmath9 , ... , @xmath14 of some confocal family in @xmath15 is in several steps endowed with a group structure called _ billiard algebra_. in this construction , we first define the addition operation on the set of all finite billiard trajectories with the same initial line . then , observing equivalence relations @xmath16 and @xmath17 in the set of finite billiard trajectories , we prove that the addition operation is compatible with them , and that the operation on the equivalence classes has additional algebraic properties . at the end , we are ready to use these constructions in order to obtain the billiard algebra on the set of lines . section [ sec : algebra ] is aimed to illustrate the billiard algebra introduced in section [ sekcija : bilijarskaalgebra ] . section [ sec : weak ] starts with introduction of the notions of _ @xmath3-skew lines _ and _ @xmath3-weak poncelet trajectories_. we prove the poncelet type theorem for such trajectories and deduce the corresponding analytic condition of cayley s type . in section [ sekcija : wgh ] , higher - dimensional generalizations of some classical results connected to poncelet s porism are stated and proved . we introduce _ generalized weyr chains _ and show their correspondence with poncelet polygons ( theorem [ th : gen.weyr ] and proposition [ propozicija : weyr.poncelet ] ) . this correspondence is then used to demonstrate the generalization to higher dimensions of the griffiths - harris space version of poncelet theorem ( theorem [ teorema : uopstena.grifits.haris ] ) . section [ sec : grid ] starts with recalling and a few comments on darboux s theorem on grids of conics associated to a poncelet polygon on an arbitrary liouville surface . then , for the case when the surface is the eucledean plane , we prove in theorems [ th : gen.grida ] and [ th : grida ] a generalization of the darboux s theorem . the essential generalization to higher dimensions is demonstrated in theorem [ th : gen.grida.d ] . section [ sec : conclusion ] contains concluding remarks . in this section , we are going to define families of confocal quadrics in the @xmath2-dimensional euclidean space @xmath15 and summarize their basic properties connected with the billiard reflection law . [ def : e.confocal ] _ a family of confocal quadrics _ in the @xmath2-dimensional euclidean space @xmath15 is a family of the form : @xmath18 where @xmath19 , , @xmath20 are real constants . let us notice that a family of confocal quadrics in the euclidean space is determined by only one quadric . from now on , we are going to consider the non - degenerate case when @xmath19, ,@xmath20 are all distinct . [ th : jacobi ] any point of the @xmath2-dimensional euclidean space is the intersection of exactly @xmath2 quadrics of the confocal family ( [ eq : confocal.family ] ) . the quadrics are perpendicular to each other at the intersecting points . ( see figure [ fig:3konf ] ) . , width=226,height=226 ] from here , it follows that to every point @xmath21 we may associate a @xmath2-tuple of distinct parameters @xmath22 , such that @xmath23 belongs to quadrics @xmath24 , , @xmath25 . additionally , if we order @xmath26 and @xmath27 , then : @xmath28 _ jacobi elliptic coordinates _ of point @xmath29 are values @xmath22 that satisfy : @xmath30 [ th : chasles ] any line in @xmath15 is tangent to exactly @xmath1 quadrics from a given confocal family . tangent hyper - planes to these quadrics , constructed at the points of tangency with the line , are orthogonal to each other . we say that two lines @xmath31 , @xmath32 intersecting in point @xmath33 _ satisfy the billiard reflection law _ on quadric @xmath34 if they are coplanar with the line normal to @xmath34 at @xmath35 , determine the congruent angles with this line . _ a billiard trajectory within @xmath34 _ is a polygonal line with vertices on @xmath34 , such that each pair of its consecutive segments satisfies the billiard reflection law on @xmath34 and lies on the same side of the tangent hyper - plane to @xmath34 at their common endpoint . the billiard reflection on a quadric is tightly connected with the corresponding confocal family . two lines that satisfy the reflection law on a quadric @xmath34 in @xmath15 are tangent to the same @xmath1 quadrics confocal with @xmath34 . from the previous theorem , it follows that all segment of a billiard trajectory within @xmath34 are tangent to the same @xmath1 quadrics confocal with @xmath34 . we call these @xmath1 quadrics _ caustics _ of the given trajectory . [ generalized poncelet theorem ] consider a closed billiard trajectory within quadric @xmath34 in @xmath15 . then all other billiard trajectories within @xmath36 , that share the same @xmath1 caustics , are also closed . moreover , all these closed trajectories have the same number of vertices . this theorem , concerning the plane case ( @xmath37 ) , is due to jean - victor poncelet and dates from the beginning of the xixth century @xcite . its generalization to the three - dimensional case is proved by darboux @xcite at the end of the same century . the generalization to an arbitrary dimension is obtained in @xcite at the end of xxth century . having the generalized poncelet theorem in mind , it is natural to ask about an explicit condition on a quadric and a set of confocal caustics , such that the corresponding billiard trajectories are closed . this condition , for @xmath37 , was given by cayley @xcite . the condition for an arbitrary @xmath2 is obtained by the authors of this paper in @xcite . [ generalized cayley condition ] the condition on a billiard trajectory inside ellipsoid @xmath38 in @xmath15 , with nondegenerate caustics @xmath39 , , @xmath40 from the family ( [ eq : confocal.family ] ) , to be perodic with period @xmath41 is : @xmath42 where @xmath43 and all @xmath44 are distinct and positive . in this section , we are going to present the generalization of notions of confocal families of quadrics and related billiard trajectories to the projective space over an arbitrary field of characteristic not equal to @xmath45 . also , we give a proof of the full poncelet theorem in the three - dimensional projective space , based on @xcite . we begin with the definition of confocal family of quadrics in the projective space , as it was done in @xcite . [ def : p.confocal ] a family of quadrics in the projective space is _ confocal _ if for any hyper - plane the set of poles with respect to these quadrics is a line . it can be shown that the dual to a confocal family is a pencil of quadrics . in contrast to the euclidean case , notice that it follows from there that a confocal family in the projective space is determined by two quadrics . on the other hand , if we introduce the metrics in a convenient manner in the real projective space @xmath46 , the affine part of a given confocal family in @xmath46 will satisfy definition [ def : e.confocal ] . in this sense , definitions [ def : e.confocal ] and [ def : p.confocal ] are equivalent . as well as in the euclidean case , any line in @xmath47 is also tangent to @xmath1 quadrics from a given confocal family . we will refer this statement by _ chasles theorem _ , similarly as in the euclidean case , although the second part of theorem [ th : chasles ] has no sense without metrics . now , we are able to define the billiard reflection in a pure projective manner , without metrics . let @xmath9 and @xmath48 be two quadrics that meet transversely . denote by @xmath49 the tangent plane to @xmath9 at point @xmath23 and by @xmath50 the pole of @xmath49 with respect to @xmath48 . suppose lines @xmath31 and @xmath32 intersect at @xmath23 , and the plane containing these two lines meet @xmath49 along @xmath51 . if lines @xmath52 are coplanar and harmonically conjugated , we say that rays @xmath31 and @xmath32 _ obey the reflection law _ at the point @xmath23 of the quadric @xmath9 with respect to the confocal family which contains @xmath9 and @xmath48 . if we introduce a coordinate system in which quadrics @xmath9 and @xmath48 are confocal in the euclidean sense , reflection defined in this way is same as the standard one . suppose rays @xmath31 and @xmath32 obey the reflection law at @xmath23 of @xmath9 with respect to the confocal system determined by quadrics @xmath9 and @xmath48 . let @xmath31 intersects @xmath48 at @xmath53 and @xmath54 , @xmath49 is a tangent plane to @xmath9 at @xmath23 , and @xmath50 its pole with respect to @xmath55 . then lines @xmath56 and @xmath57 respectively contain intersecting points @xmath58 and @xmath59 of ray @xmath32 with @xmath55 . converse is also true . let rays @xmath31 and @xmath32 obey the reflection law of @xmath9 with respect to the confocal system determined by quadrics @xmath9 and @xmath48 . then @xmath31 is tangent to @xmath48 if and only if is tangent @xmath32 to @xmath48 ; @xmath31 intersects @xmath55 at two points if and only if @xmath32 intersects @xmath55 at two points . next assertion is crucial for proof of the poncelet theorem . [ double reflection theorem ] [ th : drt ] suppose that @xmath9 , @xmath48 are given quadrics and @xmath31 line intersecting @xmath9 at the point @xmath60 and @xmath48 at @xmath54 . let @xmath61 , @xmath62 be tangent planes to @xmath9 , @xmath48 at points @xmath60 , @xmath54 respectively , and @xmath63 , @xmath64 their poles with respect to @xmath48 and @xmath9 . denote by @xmath65 second intersecting point of the line @xmath66 with @xmath9 , by @xmath59 intersection of @xmath67 with @xmath48 and by @xmath32 , @xmath68 , @xmath69 lines @xmath70 , @xmath71 , @xmath72 . then pairs @xmath73 ; @xmath74 ; @xmath75 ; @xmath76 obey the reflection law at points @xmath60 ( of @xmath9 ) , @xmath54 ( of @xmath48 ) , @xmath59 ( of @xmath48 ) , @xmath65 ( of @xmath77 ) respectively . if the line @xmath31 is tangent to a quadric @xmath34 confocal with @xmath9 and @xmath48 , then rays @xmath32 , @xmath68 , @xmath69 also touch @xmath34 . using double reflection theorem , we can prove the full poncelet theorem in the three - dimensional space . let @xmath78 be confocal quadrics and @xmath79 , ( @xmath80 ) lines such that pairs @xmath81 obey the reflection law at point @xmath82 of @xmath83 . let @xmath84 be a plane tangent to @xmath85 at @xmath82 , @xmath86 its pole with respect to @xmath34 and @xmath87 . if line @xmath31 intersects @xmath34 at @xmath88 , then , by one reflection theorem , the second intersection point @xmath59 of line @xmath67 with quadrics @xmath34 belongs to @xmath32 . similarly , having point @xmath89 , we construct the point @xmath90 . it follows that @xmath91 or @xmath92 . suppose @xmath91 . it can be proved that , if , for a given polygon @xmath93 and quadric @xmath34 , @xmath94 holds , then @xmath94 for any surface @xmath34 from the confocal family . suppose @xmath95 are rays @xmath96 reflected of @xmath34 at points @xmath89 . by double reflection theorem , @xmath95 and @xmath97 meet at point @xmath98 and obey the reflection law . in this way , we obtained a new polygon @xmath99 . if @xmath31 touches quadrics @xmath100 confocal with @xmath101 , then all sides of both polygons touch them . in this way , a two - parameter family of poncelet polygons is obtained . apart from the real motion of the billiard particle in @xmath102 , it is of interest to consider _ virtual reflections_. these reflections were discussed by darboux in @xcite ( see chapter xiv of book iv in volume 2 ) . in this section , we review some recent results from @xcite . formally , in the euclidean space , we can define the _ virtual reflection _ at the quadric @xmath34 as a map of a ray @xmath51 with the endpoint @xmath103 ( @xmath104 ) to the ray complementary to the one obtained from @xmath51 by the real reflection from @xmath34 at the point @xmath103 . on figure [ fig : virtual.real.reflection ] , ray @xmath105 is obtained from @xmath51 by the real reflection on the quadric surface @xmath34 at point @xmath103 . ray @xmath106 is obtained from @xmath51 by the virtual reflection . line @xmath12 is the normal to @xmath34 at @xmath103 . let us remark that , in the case of real reflections , exactly one elliptic coordinate , the one corresponding to the quadric @xmath36 , has a local extreme value at the point of reflection . on the other hand , on a virtual reflected ray , this coordinate is the only one not having a local extreme value at the point of reflection . in the @xmath45-dimensional case , the virtual reflection can easily be described as the real reflection from the other confocal conic passing through the point @xmath103 . in higher dimensional cases , the virtual reflection can be regarded as the real reflection from the line normal to @xmath34 at @xmath103 ( see figure [ fig : virtual.real.reflection ] ) . the notions of real and virtual reflection can not be straightforwardly extended to the projective space , since there we essentially use that the field of real numbers is naturally ordered . nevertheless , it turns out that it is possible to define a certain configuration connected with real and virtual reflection , such that its properties remain in the projective case , too . let points @xmath107 ; @xmath108 belong to quadrics @xmath77 , @xmath48 in @xmath47 . [ def : vrc ] we will say that the quadruple of points @xmath109 constitutes a _ virtual reflection configuration ( vrc ) _ if pairs of lines @xmath110 , @xmath111 ; @xmath112 , @xmath113 ; @xmath110 , @xmath114 ; @xmath111 , @xmath113 satisfy the reflection law at points @xmath115 , @xmath116 of @xmath9 and @xmath117 , @xmath118 of @xmath48 respectively , with respect to the confocal system determined by @xmath9 and @xmath48 . if , additionally , the tangent planes to @xmath119 at @xmath107 ; @xmath108 belong to a pencil , we say that these points constitute a _ double reflection configuration ( drc ) _ ( see figure [ fig : virtual_reflection ] ) . [ fig : virtual_reflection ] now , the darboux s statement can be generalized and proved as follows : [ th : virt.refl ] let @xmath9 , @xmath48 be two quadrics in the projective space @xmath47 , @xmath115 , @xmath116 points on @xmath9 and @xmath117 , @xmath118 on @xmath48 . if the tangent hyperplanes at these points to the quadrics belong to a pencil , then @xmath109 constitute a virtual reflection configuration . furthermore , suppose that the projective space is defined over the field of reals . introduce a coordinate system , such that @xmath77 , @xmath48 become confocal ellipsoids in the euclidean space . if @xmath9 is placed inside @xmath48 , then the sides of the quadrilateral @xmath120 obey the real reflection from @xmath48 and the virtual reflection from @xmath48 . we are going to conclude this section with the statement converse to the previous theorem . let pairs of points @xmath115 , @xmath116 and @xmath117 , @xmath118 belong to confocal ellipsoids @xmath9 and @xmath48 in euclidean space @xmath15 , and let @xmath121 , @xmath122 , @xmath123 , @xmath124 be the corresponding tangent planes . if a quadruple @xmath125 is a virtual reflection configuration , then planes @xmath121 , @xmath122 , @xmath123 , @xmath124 belong to a pencil . [ def : gen.cayley ] let @xmath51 be a line not contained in any quadric of the given confocal family in the projective space @xmath47 . the _ generalized cayley curve _ @xmath126 is the variety of hyperplanes tangent to quadrics of the confocal family at the points of @xmath51 . this curve is naturally embedded in the dual space @xmath127 . on figure [ fig : gen.cayley.curve ] , we see the planes which correspond to one point of the line @xmath51 in the @xmath128-dimensional space . the generalized cayley curve is a hyperelliptic curve of genus @xmath129 , for @xmath130 . its natural realization in @xmath131 is of degree @xmath132 . the natural involution @xmath133 on the generalized cayley s curve @xmath126 maps to each other the tangent planes at the points of intersection of @xmath51 with any quadric of the confocal family . it is easy to see that the fixed points of this involution are hyperplanes corresponding to the @xmath1 quadrics that are touching @xmath51 and to @xmath134 degenerate quadrics of the confocal family . now , let us make a few remarks on the real case . [ lema : polinom ] suppose a line @xmath51 is tangent to quadrics @xmath135 from the family ( [ eq : confocal.family ] ) . then jacobian coordinates @xmath136 of any point on @xmath51 satisfy the inequalities @xmath137 , @xmath138 , where @xmath139 let @xmath23 be a point of @xmath51 , @xmath22 its jacobian coordinates , and @xmath140 a vector parallel to @xmath51 . the equation @xmath141 is quadratic with respect to @xmath142 . its discriminant is : @xmath143 where @xmath144 by @xcite , @xmath145 for each of the coordinates @xmath146 , ( @xmath147 ) , the quadratic equation has a solution @xmath148 ; thus , the corresponding discriminants are non - negative . this is obviously equivalent to @xmath149 . the equation of the generalized cayley curve corresponding to a confocal family of the form ( [ eq : confocal.family ] ) can be written as : @xmath150 it is important to note that the constants @xmath121 , , @xmath151 , corresponding to the quadrics that are touching @xmath51 can not take arbitrary values . more precisely , following @xcite , we can state : there exists a line in @xmath15 that is tangent to @xmath1 distinct non - degenerate quadrics @xmath152 , , @xmath40 from the family ( [ eq : confocal.family ] ) if and only if the set @xmath153 can be ordered as @xmath154 , such that @xmath155 @xmath156 . it was observed in @xcite that the generalized cayley curve is isomorphic to the veselov - moser isospectral curve . this curve is also naturally isomorphic to the curves studied by knrrer , donagi and reid ( rdk curves ) . we begin next section with the construction of this isomorphism , in order to establish the relationship between billiard law and algebraic structure on the jacobian @xmath157 . the aim of this section is to construct in @xmath159 an algebraic structure that is naturally connected with the billiard motion . in section [ sekcija : morfizam ] , we show that the generalized cayley s curve is isomorphic to the rdk curve . then , in order to get better understanding and intuition , we are going first , in section [ sec : genus2 ] to describe in detail the billiard algebra , and prove some of its nice geometrical properties , for the case of dimension @xmath128 , i.e. when the corresponding curve is of genus @xmath45 . the general construction is given in the section [ sec : general ] . now we are going to establish the connection between generalized cayley s curve defined above and the curves studied by knrrer , donagi , reid and to trace out the relationship between billiard constructions and the algebraic structure of the corresponding abelian varieties . suppose a line @xmath51 in @xmath15 is tangent to quadrics @xmath135 from the confocal family ( [ eq : confocal.family ] ) . denote by @xmath159 the family of all lines which are tangent to the same @xmath1 quadrics . note that according to the corollary of the one reflection theorem , the set @xmath159 is invariant to the billiard reflection on any of the confocal quadrics . we begin with the next simple observation . [ lema : veza ] let the lines @xmath51 and @xmath160 obey the reflection law at the point @xmath50 of a quadric @xmath34 and suppose they are tangent to a confocal quadric @xmath9 at the points @xmath63 and @xmath161 . then the intersection of the tangent spaces @xmath162 is contained in the tangent space @xmath163 . it follows from the one reflection theorem : since the poles @xmath63 , @xmath161 and @xmath164 of the planes @xmath165 , @xmath166 and @xmath167 with respect to the quadric @xmath9 are colinear , the planes belong to a pencil . following @xcite , together with @xmath1 affine confocal quadrics @xmath135 , one can consider their projective closures @xmath168 and the intersection @xmath169 of two quadrics in @xmath170 : @xmath171 @xmath172 denote by @xmath173 the set of all @xmath174-dimensional linear subspaces of @xmath169 . for a given @xmath175 , denote by @xmath176 the closure in @xmath177 of the set @xmath178 . it was shown in @xcite that @xmath176 is a nonsingular hyperelliptic curve of genus @xmath1 . note that for @xmath179 , i.e. when the curve @xmath126 is of the genus @xmath45 , an isomorphism between @xmath180 and the jacobian of the hyperelliptic curve was established in @xcite . the projection @xmath181 maps @xmath182 to a subspace @xmath183 of the codimension @xmath45 . @xmath184 is tangent to the quadrics @xmath185 that are dual to @xmath186 . thus , the space dual to @xmath184 , denoted by @xmath187 , is a line tangent to the quadrics @xmath188 . we can reinterpret the generalized cayley s curve @xmath189 , which is a family of tangent hyperplanes , as a set of lines from @xmath159 which intersect @xmath51 . namely , for almost every tangent hyperplane there is a unique line @xmath160 , obtained from @xmath51 by the billiard reflection . having this identification in mind , it is easy to prove the following there is a birational morphism between the generalized cayley s curve @xmath126 and reid - donagi - knrrer s curve @xmath176 , with @xmath190 , defined by @xmath191 where @xmath160 is a line obtained from @xmath51 by the billiard reflection on a confocal quadric . it follows from the previous lemmata and lemma 4.1 and corollary 4.2 from @xcite . thus , lemma [ lema : veza ] gives a link between the dynamics of ellipsoidal billiards and algebraic structure of certain abelian varieties . this link provides a two way interaction : to apply algebraic methods in the study of the billiard motion , but also vice versa , to use billiard constructions in order to get more effective , more constructive and more observable understanding of the algebraic structure . in the following section , we are going to use this link in constructing _ the billiard algebra _ , which is a group structure in @xmath159 . before we proceed in general case , we want to emphasize the billiard constructions involved in the first nontrivial case , of genus two . we formulate _ the leading principle _ : _ the sum of the lines in any virtual reflection configuration is equal to zero if the four tangent planes at the points of reflection belong to a pencil . _ recall that , by definition [ def : vrc ] , such a configuration of four lines in a vrc , with tangent planes in a pencil , we are going to call _ a double reflection configuration ( drc)_. * neutral element * let us fix a line @xmath192 . * opposite element * first , we define @xmath193 . for a given line @xmath194 , define @xmath195 where @xmath196 is the hyperelliptic involution of the curve @xmath197 . [ prop : saglasnost- ] for any @xmath194 , both lines @xmath23 and @xmath198 are obtained from @xmath199 by the reflection on the same quadric @xmath200 from the confocal family . moreover , let @xmath201 be the unique plane orthogonal to @xmath199 from the pencil determined by the tangent planes to @xmath200 at the intersection points with @xmath199 , and @xmath202 the unique quadric from the confocal family such that @xmath201 is tangent to it . then the intersection point of @xmath201 with @xmath199 belongs to @xmath203 . follows from the degenerate case of double reflection theorem applied on quadrics @xmath200 and @xmath203 with @xmath204 . see figure [ fig : saglasnost- ] . , width=226,height=151 ] plane @xmath201 and quadric @xmath203 do not depend on @xmath23 . line @xmath199 reflects to itself on @xmath205 . suppose now that @xmath206 , but @xmath23 does nt belong to @xmath197 . thus @xmath23 does not intersect @xmath199 and the two lines generate a linear projective space . this space intersects @xmath159 along the divisor @xmath207 ( see , for example , @xcite , @xcite . ) according to the double reflection theorem , one can see that the lines @xmath199 , @xmath23 , @xmath208 , @xmath209 form a double reflection configuration . we define @xmath198 such that it forms a double reflection configuration with @xmath199 , @xmath210 , @xmath211 . from the definition , it is immediately seen that @xmath212 for every @xmath213 . the following property is a consequence of proposition [ prop : saglasnost- ] and the double reflection theorem . [ prop : x .- x ] lines @xmath23 and @xmath198 intersect each other and they satisfy the reflection law on @xmath203 . the following example is an illustration for the construction we have just made . take the line @xmath199 to be orthogonal to one of the coordinate hyperplanes . then @xmath214 coincide with this hyperplane and , additionally , among caustics @xmath9 , , @xmath14 there can not be two quadrics of the same type . according to @xcite , exactly the case of all quadrics of different type corresponds to the case where @xmath159 consists of only one real connected component , isomorphic to the connected component of zero of @xmath215 . in this case , for any line @xmath213 , the opposite element @xmath198 may be defined as the line symmetric to @xmath23 with respect to plane @xmath203 . * addition * we are going to define operation @xmath216 define @xmath217 , for all @xmath218 . for @xmath219 , define @xmath220 as the line that forms a double reflection configuration with @xmath221 . obviously , @xmath222 . notice that @xmath223 are unique lines from @xmath159 that intersect both @xmath220 and @xmath199 ( see @xcite ) . thus , we have : [ lema : rastavljanje ] each line @xmath224 can be in the unique way represented as the sum of two lines that intersect @xmath199 . now , suppose @xmath225 , @xmath226 ( see figure [ fig : partial1 ] ) . as already explained , take lines @xmath227 such that @xmath228 . construct @xmath229 as above , since both pairs @xmath230 and @xmath231 intersect @xmath199 . now both @xmath232 and @xmath233 intersect @xmath3 . thus , the three lines belong to a drc with the fourth line @xmath50 . we put by definition @xmath234 the following lemma gives a very important property of the operation . [ lema : saglasnost ] let @xmath235 be lines in @xmath197 , @xmath159 respectively and @xmath236 the quadric from the confocal family such that @xmath3 and @xmath199 reflect to each other on it . then the lines @xmath237 and @xmath198 intersect each other , their intersection point belongs to quadric @xmath236 , and these two lines satisfy the billiard reflection law on @xmath236 . , width=226,height=170 ] follows from @xcite , @xcite and lemma [ lema : veza ] . lemma [ lema : saglasnost ] is going to play an important role in proving basic properties of the operation : [ lema : asoc ] for all @xmath238 , the associative law holds : @xmath239 . denoting , like on figure [ fig : saglasnost ] , by @xmath198 , @xmath232 , @xmath233 lines forming drcs with triplets @xmath240 , @xmath241 , @xmath242 , and applying lemma [ lema : saglasnost ] , we see that @xmath243 is the unique line forming a drc with triplets @xmath244 , @xmath245 , @xmath246 . the same holds for line @xmath247 , thus the two lines coincide . [ lema:-(p+q+s)=(-p)+(-q)+(-s ) ] let @xmath238 . then @xmath248 . it is straightforward to prove that @xmath249 if and only if two of the lines @xmath250 are inverse to each other . in this case , the equality we need to prove immediately follows . thus , suppose @xmath251 does not intersect @xmath199 and that @xmath252 are lines forming a drc with @xmath253 . by lemma [ lema : saglasnost ] , we have that @xmath251 is the unique line intersecting @xmath254 , @xmath255 and @xmath256 . reflecting @xmath257 on @xmath203 and applying the double reflection theorem , we get that @xmath258 intersects lines @xmath259 , @xmath260 , @xmath261 . thus , it is equal to @xmath262 , by lemma [ lema : saglasnost ] . suppose now two lines @xmath263 are given , none of which is intersecting @xmath199 . we define their sum as follows . first , we represent @xmath23 as a sum of two lines intersecting @xmath264 : @xmath265 , @xmath219 . then , we define : @xmath266 we need to show that this definition is correct . [ lema : dobrodef ] let @xmath219 , and @xmath267 . then @xmath268 if @xmath269 , it is enough to apply lemma [ lema : asoc ] and the commutativity property for the addition in @xmath197 . if @xmath270 , then the lines @xmath271 and @xmath140 intersect with billiard reflection , as well as lines @xmath272 and @xmath140 do . thus , there is a unique , fourth line in the intersection of the space generated by @xmath273 $ ] and @xmath159 . at one hand , this line is equal to @xmath274 , at the other it is equal to @xmath275 . now , from lemmata [ lema : asoc][lema : dobrodef ] , it follows that we have constructed on @xmath159 a commutative group structure that is naturally connected with the billiard law . we started this section with the leading principle , and used it as a natural motivation for the construction of the algebra . now , we are going to show that this is justified , i.e. that the statement formulated as the leading principle really holds in the group structure . [ th : leading ] the sum of the lines in any double reflection configuration is equal to zero . let @xmath276 be the lines of a drc . take that pairs @xmath252 and @xmath277 satisfy the reflection law on @xmath9 , and pairs @xmath278 , @xmath279 on @xmath48 . then , by lemma [ lema : saglasnost ] , we have : @xmath280 where @xmath281 , @xmath282 are obtained from @xmath199 by the billiard reflection on @xmath85 , @xmath283 . obviously , @xmath284 . from ( [ eq : drc ] ) , we get : @xmath285 thus , we only need to check the case when @xmath286 and @xmath287 . then we have : @xmath288 and , from the definition of the addition operation , it follows that lines @xmath289 constitute a closed billiard trajectory with consecutive reflections on @xmath290 . on the other hand , @xmath291 is the unique line in @xmath158 that , besides @xmath199 , intersects both @xmath292 and @xmath293 . thus , @xmath294 this means that the double reflection configuration @xmath276 is degenerated , i.e. two of the lines in the configuration coincide . say @xmath295 , then both @xmath296 and @xmath2 are obtained from @xmath297 by the billiard reflection on @xmath9 . so , @xmath298 , @xmath299 and the theorem follows . let us consider a billiard trajectory @xmath300 with the initial line @xmath301 and the reflections on quadrics @xmath302 from the confocal family . applying double reflection theorem to @xmath303 , we obtain different trajectories with sharing initial segment @xmath199 and final segment @xmath304 . the reflections on each such a trajectory are also on the same set of quadrics @xmath305 , but their order may be changed . let us denote by @xmath306 , , @xmath307 all possible lines obtained after the first reflection on all these trajectories . then , in our algebra , the final segment @xmath304 can be calculated from these ones : [ prop : elln ] @xmath308 follows from theorem [ th : leading ] . the following interesting property also may be proved from lemmata [ lema : asoc][lema : dobrodef ] . let line @xmath309 be obtained from @xmath199 by consecutive reflections on three quadrics @xmath9 , @xmath55 , @xmath310 . then the @xmath311 lines obtained from @xmath199 by reflections of the these quadrics may be divided into @xmath45 groups such that : for each @xmath312 , the lines obtained from @xmath199 by the reflections on @xmath85 are in different groups ; all trajectories with three reflections on @xmath9 , @xmath48 , @xmath310 starting with @xmath199 and ending with @xmath23 contain lines only from one of the groups . we will finish this subsection with a very beautiful and non - trivial theorem on confocal families of quadrics . [ th : zvezda ] let @xmath313 be a family of confocal quadrics in @xmath314 . there exist configurations consisting of @xmath315 planes in @xmath316 with the following properties : * the planes may be organized in @xmath317 triplets , such that each plane in a triplet is tangent to a different quadric from @xmath313 and the three touching points are collinear . every plane in the configuration is a member of two triplets . * the planes may be organized in @xmath311 quadruplets , such that the planes in each quadruplet belong to a pencil and they are tangent to two different quadrics from @xmath313 . every plane in the configuration is a member of two quadruplets . moreover , such a configuration is determined by three planes tangent to three different quadrics from @xmath313 , with collinear touching points . denote by @xmath199 the line containing the three touching points and @xmath250 the lines obtained from @xmath199 by the billiard reflection on the given quadrics from @xmath313 . we construct lines @xmath318 as explained before lemma [ lema : saglasnost ] ( see figure [ fig : saglasnost ] ) . the planes of the configuration are tangent to corresponding quadrics at points of the intersection of the lines . the configuration of the planes in the dual space @xmath319 is shown on figure [ fig : zvezda ] . here , each plane is denoted by two lines that reflect to each other on . it would be interesting to describe the variety of such configurations as a moduli - space . the genus two case we have just considered in detail gives us the necessary experience of using the billiard constructions to build a group structure . moreover , it provides us with the case @xmath320 in the general construction we are going to present now . proposition [ prop : elln ] is going to serve as the motivation for defining the operation in the set @xmath159 for higher - dimensional . thus , let us now describe in detail the construction that preceded this proposition in section [ sec : genus2 ] . suppose that quadrics @xmath9 , , @xmath321 from the confocal family in @xmath15 are given . let @xmath322 , @xmath31 , , @xmath304 be lines in @xmath159 such that each pair of successive lines @xmath96 , @xmath323 satisfies the billiard reflection law at the quadric @xmath324 ( @xmath325 ) ; thus the lines form billiard trajectory @xmath326 . let us , for @xmath327 , define lines @xmath307 , @xmath328 , , @xmath329 by the procedure , as follows : \1 . set @xmath330 . @xmath331 , for @xmath332 is the unique line that constitutes a drc with @xmath333 , @xmath334 and @xmath335 . let us note that , in this way , each line in the sequence @xmath336 is obtained from the previous one by the reflection from @xmath337 respectively . considering only the initial subsequences @xmath338 , , @xmath339 @xmath340 , we may in the same way define lines @xmath341 , , @xmath342 . notice that , for each @xmath343 , @xmath341 intersects @xmath338 , and these two lines obey the reflection law on @xmath344 . thus , we have constructed a mapping @xmath345 from the set @xmath346 of billiard trajectories with the fixed initial line @xmath301 to the ordered sets of lines from @xmath159 which intersect @xmath199 : @xmath347 where line @xmath341 intersects @xmath199 according to the billiard law on the quadric @xmath344 . doing the opposite procedure , we define morphism @xmath348 , the inverse of @xmath345 , which assigns to an @xmath12-tuple of lines intersecting @xmath199 the unique billiard trajectory of length @xmath12 with @xmath301 as the initial line : @xmath349 hence , the mapping @xmath348 gives _ the billiard representation _ of an ordered set of lines intersecting @xmath264 . in order to consider just divisors of the curve @xmath350 , instead of ordered @xmath12-tuples of lines , we need to introduce the following relation @xmath16 between billiard trajectories : we say that two billiard trajectories are _ @xmath16-equivalent _ if one can be obtained from the other by a finite set of _ double reflection moves_. the double reflection move transforms a trajectory @xmath351 into trajectory @xmath352 if the lines @xmath353 form a double reflection configuration . it follows directly from the double reflection theorem that two trajectories @xmath354 with the same initial segment @xmath322 are @xmath16-equivalent if and only the @xmath12-tuples @xmath355 and @xmath356 may be obtained from each other by a permutation . thus , the mapping @xmath345 may be considered as a mapping from @xmath357 to the set of positive divisors on @xmath197 and it represents _ the divisor representation _ of billiard trajectories . the following lemma , which follows from @xcite , is a sort of riemann - roch theorem in this approach . [ lema : dijagonala ] a minimal billiard trajectory of length @xmath3 from @xmath23 to @xmath140 is unique , up the relation @xmath16 . if there are two non @xmath16-equivalent trajectories of the same length @xmath358 from @xmath23 to @xmath140 , then there is a trajectory from @xmath23 to @xmath140 of length @xmath359 . now , we are able to introduce an operation , _ summation _ , in the set @xmath360 . given two billiard trajectories @xmath361 , we define their sum by the equation : @xmath362 this operation is associative and commutative , according to the double reflection theorem . the set @xmath363 with the summation operation is a commutative semigroup . this semigroup is isomorphic to the semigroup of all effective divisors on the curve @xmath197 that do not contain the points corresponding to the caustics . note that in @xmath360 , the trajectory consisting only of the single line @xmath199 is a neutral element . define now the following equivalence relation in the set of all finite billiard trajectories . two billiard trajectories are _ @xmath17-equivalent _ if they have common initial and final segments , and they are of the same length . the set of all classes of @xmath17-equivalent billiard trajectories of length @xmath12 , with the fixed initial segment @xmath199 , we denote by @xmath364 , and @xmath365 [ prop : beta+ ] the relation @xmath17 is compatible with the addition of billiard trajectories . to prove this , we will need the following important lemma : [ lema : par.nepar ] let @xmath366 be a closed billiard trajectory , and @xmath367 a line that intersects @xmath31 . construct iteratively lines @xmath368 such that quadruples @xmath369 @xmath370 form double reflection configurations . then @xmath371 . we are going to proceed with the induction . for @xmath372 , the lines lines @xmath373 form a double reflection configuration , and the statement follows by the double reflection theorem . now , suppose @xmath374 . @xmath375 and @xmath376 are two billiard trajectories of the same length from @xmath31 to @xmath339 . if they are @xmath16-equivalent , then the claim follows by the double reflection theorem . if they are not @xmath16-equivalent , then , by lemma [ lema : dijagonala ] , there is a billiard trajectory @xmath377 . the statement now follows from the inductive hypothesis applied to trajectories @xmath378 and @xmath379 _ proof of proposition [ prop : beta+]_. we need to prove the following : @xmath380 clearly , it is enough to prove this relation for the case when @xmath381 and the length of @xmath382 is equal to @xmath45 . suppose that @xmath383 , where @xmath208 is obtained from @xmath199 by the reflection on quadric @xmath384 . then trajectories @xmath385 , @xmath386 are obtained by adding one segment to @xmath387 , @xmath388 respectively . these segments satisfy the reflection law on @xmath389 with the final segments of @xmath387 , @xmath388 . since @xmath390 , their final segments coincide and the statement follows from lemma [ lema : par.nepar ] applied to the trajectory @xmath391 where @xmath392 , @xmath393 . @xmath394 we wish to use the constructed algebra on the set of billiard trajectories , in order to obtain an algebraic structure on @xmath158 , such that the operation is naturally connected with the billiard reflection law . from @xcite we get [ teorema : dostiznost ] for any two given lines @xmath23 and @xmath140 from @xmath159 , there is a system of at most @xmath1 quadrics from the confocal family , such that the line @xmath140 is obtained from @xmath23 by consecutive reflections on these quadrics . the divisor representation of the corresponding billiard trajectory of length @xmath395 will be called _ the @xmath3-brush of @xmath140 related to @xmath23_. now , we can define a group structure in @xmath159 associated with a fixed line in this set . _ neutral element _ let us fix a line @xmath396 _ inverse element _ let @xmath23 be an arbitrary line in @xmath159 , and @xmath397 the divisor representation of the minimal billiard trajectory connecting @xmath199 with @xmath23 . define @xmath198 as the final segment of the billiard trajectory @xmath398 , where @xmath399 is the hyperelliptic involution of @xmath197 . _ addition _ for two lines @xmath23 and @xmath140 from @xmath159 , denote their brushes related to @xmath199 as @xmath400 and @xmath401 . define @xmath402 where @xmath403 is the final segment of the billiard trajectory @xmath404 . from all above , similarly as in the genus @xmath45 case , we get the set @xmath159 with the operation defined above is an abelian group . let us observe that another kind of divisor representation of billiard trajectories may be introduced , that associates a divisor of degree @xmath405 to a given trajectory . such a representation was used in @xcite and explicitly described in @xcite . the positive part of this representation coincide with the divisor obtained by @xmath348 , and the negative part is invariant for the hyperelliptic involution . moreover , as it follows from @xcite , two trajectories with the initial segment @xmath199 will have the same final segments if and only if their representations are equivalent divisors . thus , it follows that the group structure on @xmath159 is isomorphic to the quotient of the jacobian of the curve @xmath197 by a finite subgroup which is generated by the points corresponding to the caustic quadrics . now , as a consequence of theorem [ teorema : dostiznost ] we are able to introduce the following hierarchies of notions . [ definicija : s - mimoilaznost ] for two given lines @xmath23 and @xmath140 from @xmath159 we say that they are _ @xmath3-skew _ if @xmath3 is the smallest number such that there exist a system of @xmath406 quadrics @xmath344 , @xmath407 from the confocal family , such that the line @xmath140 is obtained from @xmath23 by consecutive reflections on @xmath344 . if the lines @xmath23 and @xmath140 intersect , they are _ @xmath405-skew_. they are _ @xmath408-skew _ if they coincide . [ definicija : s - poncelet ] suppose that a system @xmath409 of @xmath12 quadrics @xmath9 , , @xmath321 from the confocal family is given . for a system of lines @xmath410 , @xmath411 , , @xmath412 in @xmath159 such that each pair of successive lines @xmath413 , @xmath414 satisfies the billiard reflection law at @xmath324 @xmath415 , we say that it forms an _ @xmath3-weak poncelet trajectory of length @xmath12 associated to the system @xmath409 _ if the lines @xmath410 and @xmath412 are @xmath3-skew . for @xmath3-weak poncelet trajectories we will also sometimes say _ @xmath416-resonant billiard trajectories_. periodic trajectories or generalized classical poncelet polygons are @xmath408-weak poncelet trajectories or , in other words , @xmath417-resonant billiard trajectories , and they are described analytically in @xcite , @xcite . our next goal is to get complete analytical description of @xmath3-weak poncelet trajectories of length @xmath418 , generalizing in such a way the results from @xcite . here , we are going to use fully the tools and the power of billiard algebra . to fix the idea , let us consider first the system @xmath409 consisting on @xmath418 equal quadrics @xmath419 from the confocal family . suppose a system of lines @xmath410 , @xmath420 , , @xmath421 in @xmath159 forms an @xmath3-weak poncelet trajectory of length @xmath418 associated to the system @xmath409 . then @xmath422 with some line @xmath423 which intersects @xmath410 . again , from the condition that @xmath421 and @xmath410 are @xmath3-skew we get @xmath424 with some lines @xmath425 which intersect @xmath410 . from the last two equations we come to the conclusion [ propozicija : kejli1 ] the existence of an @xmath3-weak poncelet trajectory of length @xmath418 is equivalent to existence of a meromorphic function @xmath426 on the hyperelliptic curve @xmath126 such that @xmath426 has a zero of order @xmath418 at @xmath427 , a unique pole at infinity " @xmath428 , and the order of the pole is equal to @xmath429 . now , we are going to derive explicit analytical condition of cayley s type . as in @xcite we consider the space @xmath430 of all meromorphic functions on @xmath126 with the unique pole at the infinity point @xmath428 of the order not exceeding @xmath12 . let @xmath431 be one of the bases of this space , @xmath432 . consider the vectors @xmath433 where @xmath434 and vectors @xmath435 with @xmath436 . from the condition ( see @xcite ) @xmath437<n - g+1\ ] ] we get the condition @xmath438<r+s - g+2=r+s - d+3.\ ] ] now , we can rewrite it in the form usual for the cayley s type conditions . [ teorema : kejlijevi.uslovi ] the existence of an @xmath3-weak poncelet trajectory of length @xmath418 is equivalent to : @xmath439 when @xmath440 , and @xmath441 when @xmath442 . with @xmath443 , we denoted the coefficients in the taylor expansion of function @xmath444 in a neighbourhood of @xmath35 , where @xmath445 is the equation of the generalized cayley curve , with the polynomial @xmath446 given by ( [ eq : polinom ] ) . denote by @xmath447 the linear space of all meromorphic functions on @xmath126 , having a unique pole at the infinity point @xmath428 , with the order not exceeding @xmath448 . by the riemann - roch theorem : * @xmath449 + 1 $ ] if @xmath450 , * @xmath451 if @xmath448 is even and greater than @xmath452 , * @xmath453 if @xmath448 is even and greater than @xmath452 . we may choose the following bases in each of the three cases : * @xmath454 , where @xmath455\le d$ ] ; * @xmath456 , for @xmath457 ; * @xmath458 , for @xmath459 . now , the statement follows from the considerations preceding the theorem . for @xmath460 , the inequalities in theorem [ teorema : kejlijevi.uslovi ] become the conditions for periodic billiard trajectories ( see @xcite ) . the condition for existence of a @xmath461-weak poncelet trajectory of length @xmath418 is equivalent to : @xmath462 @xmath463 if @xmath464 , then , as it follows from the proof of theorem [ teorema : kejlijevi.uslovi ] , an @xmath3-weak poncelet trajectory of length @xmath418 may exist only if the hyper - elliptic curve @xmath189 is singular . in this section we obtain higher dimensional generalizations of the results of @xcite , @xcite , @xcite . nice exposition of those classical results one can find in @xcite . the dual version of @xcite is given in @xcite . each quadric @xmath34 in @xmath170 contains at most two unirational families of @xmath417-dimensional linear subspaces . such unirational families are usually called _ rulings of the quadric_. [ th : gen.weyr ] let @xmath9 , @xmath48 be two general quadrics in @xmath170 with the smooth intersection @xmath169 and @xmath465 , @xmath466 their rulings . if there exists a closed chain @xmath467 of distinct @xmath417-dimensional linear subspaces , such that @xmath468 , @xmath469 @xmath470 and @xmath471 @xmath472 , then there are such closed chains of subspaces of length @xmath473 through any point of @xmath180 . each of the unirational families @xmath474 determines an involution @xmath475 on abelian variety @xmath180 . such an involution interchanges two @xmath174-intersections of an element of @xmath476 with @xmath169 . denote by @xmath477 their composition and by @xmath478 . since @xmath479 is a translation on @xmath180 satisfying @xmath480 we see that @xmath479 is of order @xmath12 and the theorem follows . we will call the chains considered in theorem [ th : gen.weyr ] _ generalized weyr s chains_. the theorem can be adjusted for nonsmooth intersections , but we are not going into details . instead , we consider the case of two quadrics ( [ eq : kvadrika1 ] ) and ( [ eq : kvadrika2 ] ) as in section [ sekcija : morfizam ] . by using the projection @xmath481 we get : [ propozicija : weyr.poncelet ] a generalized weyr chain of length @xmath473 projects into a poncelet polygon of length @xmath473 circumscribing the quadrics @xmath168 and alternately inscribed into two fixed confocal quadrics ( projections of @xmath482 ) . conversely , any such a poncelet polygon of the length @xmath473 circumscribing the quadrics @xmath168 and alternately inscribed into two fixed confocal quadrics can be lifted to a generalized weyr chain of length @xmath473 . it follows from lemma 4.1 and corollary 4.2 of @xcite and lemma [ lema : veza ] . thus , we obtained in a correspondence between generalized weyr chains and poncelet polygons subscribed in @xmath1 given quadrics and alternately inscribed in two quadrics from some confocal family . such poncelet polygons have been completely analytically described , among others , in @xcite ( see example 4 from there ) . let us note that correspondence between classical weyr s theorem in @xmath314 and classical poncelet theorem about two conics in a plane was observed by hurwitz in @xcite . nevertheless the projection we used here is not a straightforward generalization of the one used by hurwitz . polygonal lines , circumscribed about one conic and alternately inscribed in two conics , appear as an example in @xcite . our result , applied to the lowest dimension , gives the condition for the closeness of such a polygonal line , see corollary 1 from @xcite . let us also mention that if rulings @xmath483 and @xmath484 are connected by a generalized weyr s chain of length @xmath473 , then the same is true for @xmath466 , @xmath483 and also for the pair @xmath485 , @xmath486 of the complementary rulings of @xmath9 and @xmath48 . now we are able to present a new higher - dimensional generalization of the griffiths - harris space poncelet theorem from @xcite . [ teorema : uopstena.grifits.haris ] let @xmath487 and @xmath488 be the duals of two general quadrics in @xmath170 with the smooth intersection @xmath169 . denote by @xmath489 pairs of unirational families of @xmath417-dimensional subspaces of @xmath490 . suppose there are generalized weyr s chains between @xmath483 and @xmath466 and between @xmath483 and @xmath486 . then there is a finite polyhedron inscribed and subscribed in both quadrics @xmath9 and @xmath48 . there are infinitely many such polyhedra . the polyhedra from the previous theorem can be described in more details as arrangements of @xmath2-dimensional faces " which are bitangents of the quadrics @xmath9 and @xmath48 . their intersections are @xmath417-dimensional edges " which alternately belong to the rulings of @xmath487 and @xmath488 . the intersections of edges " are @xmath174-dimensional vertices " which belong to @xmath180 . the proof of the last theorem is based on the following lemma from @xcite . [ lema : odredjenost ] let @xmath491 be a quadric of the rank not less than @xmath132 and @xmath492 a linear subspace of the dimension @xmath493 . if @xmath34 is singular , suppose additionally that @xmath23 does not contain the vertex . then , for each ruling @xmath494 of @xmath34 , there is a unique @xmath417-dimensional linear subspace @xmath495 such that @xmath496 and @xmath497 . in @xcite ( volume 3 , book vi , chapter i ) , darboux proved that liouville surfaces are exactly these having an orthogonal system of curves that can be regarded in two or , equivalently , infinitely many different ways , as geodesic conics . these coordinate curves are analogues of systems of confocal conics in the euclidian plane . knowing this , essential properties of conics may be generalized to all liouville surfaces ( see @xcite , and also @xcite for a review and clarifications ) . here is the citation of one of these properties : _ `` considerons un polygone variable do nt tous les cts sont des lignes godsiques tangentes a une mme courbe coordonne . le dernier sommet de ce polygone dcrira aussi une des courbes coordonnes et il en sera de mme de points dintersection de deux cts quelconques de ce polygone . '' _ this statement is not only the generalization of poncelet theorem to liouville surfaces . additionally , for a poncelet polygon circumscribed about a fixed coordinate curve , with each vertex moving along one of the coordinate curves , darboux stated here that the intersection point of two arbitrary sides of the polygon is also going to describe a coordinate curve . let us note that a weaker version of this claim , although with some improvements , has been recently rediscovered in @xcite , and a more elementary proof of the result from @xcite has been published in @xcite . the main theorem of @xcite corresponds only to a fixed poncelet polygon associated to a pair of ellipses in the euclidean plane . in a sense , this gives a new argument in favour of our observation in @xcite and @xcite , that the work of darboux connected with billiards and poncelet theorem seems to be unknown to nowadays mathematicians . this section is devoted to a multi - dimensional generalization of the darboux theorem , related to billiard trajectories within an ellipsoid in the @xmath2-dimensional euclidean space . before starting with higher - dimensional generalizations , let us give some improvements together with a simpler proof of the statement corresponding to the plane case . [ th : gen.grida ] let @xmath498 be an ellipse in @xmath499 and @xmath500 , @xmath501 be two sequences of the segments of billiard trajectories @xmath498 , sharing the same caustic . then all the points @xmath502 @xmath503 belong to one conic @xmath504 , confocal with @xmath498 . moreover , under the additional assumption that the caustic is an ellipse , we have : if both trajectories are winding in the same direction about the caustic , then @xmath504 is also an ellipse ; if the trajectories are winding in opposite directions , then @xmath504 is a hyperbola . ( see figure [ fig : grida_elipsa ] . ) for a hyperbola as a caustic , it holds : if segments @xmath505 , @xmath506 intersect the long axis of @xmath498 in the same direction , then @xmath504 is a hyperbola , otherwise it is an ellipse . ( see figure [ fig : grida_hip ] . ) the statement follows by application of the double reflection theorem . namely , the lines @xmath507 and @xmath508 intersect at a point that belongs to one ellipse and one hyperbola from the confocal family . they satisfy the reflection law on exactly one of these two curves , depending on the orientation of the billiard motion along the lines . now , by the double reflection theorem , @xmath19 and @xmath509 satisfy the reflection law on the same conic . the same is true , for any pair @xmath505 , @xmath506 , by the induction . for the second part of the theorem , it is sufficient to observe that the winding direction about an ellipse is changed by the reflections on the hyperbolae , and preserved by the reflections on the ellipses from the confocal family . if an oriented line is placed between the foci , then the direction it is intersecting the axis containing the foci is changed by the reflections of the ellipses and preserved by the reflections on the hyperbolae . let @xmath500 , @xmath510 be two sequences of the segments of billiard trajectories within the ellipse @xmath498 , sharing the same caustic . if the caustic is an ellipse and the trajectories are winding in the opposite directions about it , then all the points @xmath502 @xmath511 are placed on two centrally symmetric half - branches of a hyperbola confocal with @xmath498 . denote by @xmath512 the ellipse which is the common caustic of the given billiard trajectories . it is possible to introduce a metric @xmath513 on @xmath512 , such that @xmath514 if and only if the tangent lines at @xmath515 and at @xmath516 intersect on the same ellipse of the confocal family . in particular , this means that the points where two consecutive billiard segments touch the caustic @xmath517 are always at the fixed distance from each other . take that @xmath507 and @xmath508 intersect on the upper left half - branch of the hyperbola @xmath504 ( see figure [ fig : grida_elipsa ] ) . then it may be proved from the double reflection theorem , that each touching points of these two segments with caustic @xmath517 are at equal @xmath513-distances from the upper left intersection point of @xmath504 and @xmath512 . since @xmath513 is symmetric with respect to the coordinate centre , they are also equally distanced from the lower right intersection point , but not from the other two intersection points . the same holds for any pair @xmath518 . thus , their intersections lies on two centrally symmetric half - branches of the hyperbola . now , the generalized claim about poncelet - darboux grids is immediately following from theorem [ th : gen.grida ] . [ th : grida ] let @xmath519 be the sequence of segments of a billiard trajectory within the ellipse @xmath498 . then each of the sets @xmath520 , @xmath521 , @xmath522 belongs to a single conic confocal to @xmath498 . if the caustic of the trajectory @xmath523 is an ellipse , then the sets @xmath524 are placed on ellipses and @xmath525 on hyperbolae . if the caustic is a hyperbola , then the sets @xmath526 , @xmath525 are placed on ellipses for @xmath343 even and on hyperbolae for @xmath343 odd . to show the statement for @xmath524 , take @xmath527 , @xmath528 and apply theorem [ th : gen.grida ] . for @xmath529 , take @xmath527 , @xmath530 . notice that theorem [ th : grida ] is more general then the one given in @xcite , since we do not suppose that the billiard trajectory is closed . also , the statement can be formulated for an arbitrary conic , not only for an ellipse . the main statement proved in @xcite and @xcite is a special case consequence of theorem [ th : grida ] : let @xmath523 be a closed billiard trajectory within an ellipse , with the elliptical caustic . each set @xmath524 lie on an ellipse confocal to @xmath498 , and @xmath525 on a confocal hyperbola . _ ( see figure [ fig : grida_elipsa ] ) . _ let us show one interesting property of poncelet - darboux grids . let @xmath523 be a billiard trajectory within ellipse @xmath498 , with the elliptical caustic @xmath512 . then the ellipse containing the set @xmath524 depends only on @xmath343 , @xmath498 and @xmath512 . in other words , this ellipse will remain the same for any choice of billiard trajectory within @xmath498 with the caustic @xmath512 . this claim may be proved by the use of the double reflection theorem , similarly as in theorem [ th : gen.grida ] . nevertheless , we are going to show it in another way , that gives a possibility of explicit calculation of the ellipse containing @xmath524 . to the billiard within @xmath498 with the fixed caustic @xmath531 , we may associate an elliptic curve ( see @xcite ) . each point of the curve corresponds to a conic from the confocal family . on the other hand , the billiard motion may be viewed as the linear motion on the jacobian of the curve ( i.e. on the curve itself , since this is the elliptic case ) , with translation jumps corresponding to reflections . more precisely , the translation is exactly by this value on the elliptic curve , which is associated to the ellipse that a segment is reflected on . thus , if the point @xmath532 on the elliptic curve is associated to the ellipse @xmath498 , then the set @xmath524 is placed on the ellipse corresponding to @xmath533 . although theorem [ th : gen.grida ] gives certain progress in understanding of poncelet - darboux grids , the essential breakthrough in this matter represents , in our opinion , the study of the higher - dimensional situation . this analysis is based on introduction of higher - dimensional analogues of grids and our notion of @xmath3-skew lines . [ th : gen.grida.d ] let @xmath500 , @xmath501 be two sequences of the segments of billiard trajectories within the ellipsoid @xmath498 in @xmath15 , sharing the same @xmath1 caustics . suppose the pair @xmath534 is @xmath3-skew , and that by the sequence of reflections on quadrics @xmath535 the minimal billiard trajectory connecting @xmath507 to @xmath508 is realized . then , each pair @xmath536 is @xmath3-skew , and the minimal billiard trajectory connecting these two lines is determined by the sequence of reflections on the same quadrics @xmath535 . this may be proved by the use of the double reflection theorem , similarly as in theorem [ th : gen.grida ] . this theorem also can be stated for an arbitrary quadric , not only for ellipsoids . let @xmath498 be an ellipsoid in @xmath15 , @xmath537 quadrics confocal to @xmath498 , and @xmath343 an integer . suppose that there exists a trajectory @xmath523 of the billiard within @xmath498 , having the caustics @xmath537 , such that the pair @xmath538 is @xmath3-skew and that the minimal billiard trajectory connecting @xmath338 to @xmath339 is realized by the sequence of reflections on quadrics @xmath539 confocal to @xmath498 . then , for any billiard trajectory @xmath540 within @xmath498 with the caustics @xmath9 , , @xmath14 , the pairs of lines @xmath541 are @xmath3-skew and the minimal billiard trajectory between @xmath542 and @xmath543 is realized by the sequence of reflections on @xmath539 . _ `` one of the most important and also most beautiful theorems in classical geometry is that of poncelet ( ... ) his proof was synthetic and somewhat elaborate in what was to become the predominant style in projective geometry of last century . slightly thereafter , jacobi gave another argument based on the additional theorem for elliptic functions . in fact , as will be seen below , the poncelet theorem and additional theorem are essentially equivalent , so that at least in principle poncelet gave a synthetic derivation of the group law on an elliptic curve . because of the appeal of the poncelet theorem it seems reasonable to look for higher - dimenisonal analogues ... although this has not yet turned out to be the case in the poncelet - type problems ... '' _ usually , the poncelet theorem is associated with the points of finite order on an elliptic curve . however , our analysis from the section [ sekcija : bilijarskaalgebra ] shows that a poncelet polygon should be understood as a pair of equivalent divisors and through a billiard analogue of the abel theorem it can further be associated to a meromorphic function of a hyperelliptic curve . on the other hand , the billiard analogue of the riemann - roch theorem gives uniqueness of the billiard trajectory with minimal number of lines connecting two given lines in @xmath159 . thus , the poncelet story in higher dimensions is at least as rich as a story of meromorphic functions on hyperelliptic curves . such a deep relationship between hyperelliptic jacobians and pencils of quadrics through integrable billiard systems opens a new view to the well - known , but still amazing role of elliptic coordinates in the theory of integrable systems and in the separation of variables in hamilton - jacobi method . the question of synthetic approach to the addition law on nonhyperelliptic jacobians remains open and could lead to new methods of separation of variables . * the family of all lines in @xmath15 which are tangent to given @xmath1 confocal quadrics ( see section [ sekcija : morfizam ] ) * the generalized cayley curve ( see definition [ def : gen.cayley ] in section [ gen.lebeg ] ) * the @xmath2-dimensional euclidean space * the set of all @xmath174-dimensional linear subspaces of the intersection @xmath169 of two quadrics in @xmath170 ( see lemma [ lema : veza ] in section [ sekcija : morfizam ] ) * the @xmath2-dimensional projective space * the dual @xmath2-dimensional projective space * a quadric * the quadric dual to @xmath34 * a quadric from the family ( [ eq : confocal.family ] ) of confocal quadrics in @xmath15 ( see definition [ def : e.confocal ] from section [ sec : confocal.e ] ) the research was partially supported by the serbian ministry of science and technology , project _ geometry and topology of manifolds and integrable dynamical systems_. one of the authors ( m.r . ) acknowledges her gratitude to prof . v. rom - kedar and the weizmann institute of science for the kind hospitality and support during the work on this paper . the authors would like to thank the referee for the valuable remarks and suggestions , which improved the manuscript . v. m. buchhstaber , v. z. enolskii , d. v. leykin _ hyperelliptic kleinian functions and applications _ , 10 , part 2 , pp . 1 - 125 , ed . i. m. krichever , s. p. novikov , gordon and breach , 1997 v. dragovi , m. radnovi , _ cayley - type conditions for billiards within @xmath343 quadrics in @xmath544 _ , j. of phys . a : math . gen . * 37 * ( 2004 ) , 1269 - 1276 . v. dragovi , m. radnovi , _ corrigendum : cayley - type conditions for billiards within @xmath343 quadrics in @xmath544 _ , j. of phys . a : math . gen . * 38 * ( 2005 ) , 7927 . arxiv : math - ph/0503053

the advent of precision cosmology with measurements of the cosmic microwave background ( cmb ) radiation has brought a new focus on the physics of the @xmath0 universe . the connection between two fossil relics of the early universe cmb fluctuations and galaxies offers the possibility for improved understanding of both as well as tighter constraints on cosmological parameters . these lectures present an overview of the physics of structure formation starting after the inflationary epoch , going through recombination , and extending to the present day . cosmological perturbation theory is often regarded as a highly technical subject beyond the scope of graduate courses in cosmology . however , these lectures may suggest how , with minor simplification , it can be integrated into a treatment of large scale structure . the potential benefits are many a detailed understanding of cmb anisotropy , a quantitative explanation for the key features of the cold dark matter transfer function , and an extension of the intuition developed from the newtonian theory of large scale structure to the physics of the post - inflationary universe . the aim of the current notes is more modest , but it is hoped that they provide some direction for those wishing to delve more deeply into cosmological perturbation theory and structure formation . the goal of cosmological perturbation theory is to relate the physics of the early universe ( e.g. inflation ) to cmb anisotropy and large - scale structure and to provide the initial conditions for numerical simulations of structure formation . the physics during the period from the end of inflation to the beginning of nonlinear gravitational collapse is complicated by relativistic effects but greatly simplified by the small amplitude of perturbations . thus , an essentially complete and accurate treatment of relativistic perturbation evolution is possible , at least in the context of simple fluctuation models like inflation . numerous articles have been written on this topic . see [ @xcite ] for references and for more detailed treatments similar in style to the present lecture notes . the starting point for cosmological perturbation theory is the metric of a perturbed robertson - walker spacetime , @xmath1dx^\mu dx^\nu \nonumber\\ & = & a^2(\tau)\left[-d\tau^2+\gamma_{ij}(\vec x\,)dx^idx^j+ h_{\mu\nu}(\vec x,\tau)dx^\mu dx^\nu\right]\ . \label{rwmet}\end{aligned}\ ] ] spatial coordinates take the range @xmath2 ; @xmath3 ( or @xmath4 for all three ) is a comoving spatial coordinate ; @xmath5 is conformal time ; @xmath6 is the cosmic expansion scale factor ; units are chosen so that the speed of light is unity ; and @xmath7 is the 3-metric of a maximally symmetric constant curvature space . conformal time is related to the proper time measured by a comoving observer ( i.e. one at fixed @xmath4 ) by @xmath8 . for cosmological perturbation theory it is more convenient than proper time . the metric perturbations are given by @xmath9 . the robertson - walker model is characterized by one function of time , @xmath6 , and one constant , the curvature constant @xmath10 . the expansion factor obeys the friedmann equation , @xmath11 where a dot denotes @xmath12 and @xmath13 is the total mean energy density . note that the usual hubble parameter is @xmath14 . the spatial part of the robertson - walker metric takes a simple form in spherical coordinates @xmath15 : @xmath16 the angular radius @xmath17 takes one of three forms depending on the sign of @xmath10 : @xmath18 these cases are referred to as closed , flat , and open , respectively . many authors choose units of length such that @xmath19 or 0 . however , it is convenient to let @xmath10 have units ( inverse length squared ) . from the friedmann equation , @xmath20 where the subscript 0 denotes the present value and @xmath21 is the total density parameter ( including dark energy ) . note that the range of @xmath22 is @xmath23 in the flat and open models but is @xmath24 $ ] in the closed model . the flat robertson - walker model favored by inflation is conformal to minkowski spacetime , i.e. the metrics are identical up to an overall conformal factor @xmath25 . in linear perturbation theory , the metric perturbations @xmath26 are regarded as a tensor field residing on the background robertson - walker spacetime . as a symmetric @xmath27 matrix , @xmath26 has 10 degrees of freedom . because of the ability to make continuous deformations of the coordinates , 4 of these degrees of freedom are gauge- ( coordinate- ) dependent , leaving 6 physical degrees of freedom . a proper treatment of cosmological perturbation theory requires clear separation between physical and gauge degrees of freedom . the metric degrees of freedom in linear pertubation theory were classified by lifshitz in 1946 [ @xcite ] . lifshitz presented the scalar - vector - tensor decomposition of the metric . it is based on a @xmath28 split of the components , which we rewrite as follows : @xmath29 here , @xmath30 is the matrix inverse of @xmath31 . the trace part of @xmath32 has been absorbed into @xmath33 so that @xmath34 has only 5 independent components . with the space - time ( @xmath28 ) split , we use @xmath31 ( or @xmath30 ) to lower ( or raise ) indices of spatial 3-vectors and tensors . for convenience , spatial derivatives will be written using the 3-dimensional covariant derivative @xmath35 defined with respect to the 3-metric @xmath31 ; it is a three - dimensional version of the full 4-dimensional covariant derivative presented in general relativity textbooks . for example , if @xmath36 , we can choose cartesian coordinates such that @xmath37 and @xmath38 . the scalar - tensor - vector split is based on the decomposition of a vector into longitudinal and transverse parts . for any three - vector field @xmath39 , we may write @xmath40 the curl and divergence are defined using the spatial covariant derivative , e.g. @xmath41 . the longitudinal / transverse decomposition is not unique ( e.g. one may always add a constant to @xmath42 ) but it always exists . the terminology arises because in the fourier domain @xmath42 is parallel to the wavevector while @xmath43 is transverse ( perpendicular to the wavevector ) . note that @xmath44 for some scalar field @xmath45 . thus , the longitudinal / transverse decomposition allows us to write a vector field in terms of a scalar ( the longitudinal or irrotational part ) and a part that can not be obtained from a scalar ( the transverse or rotational part ) . a similar decomposition holds for a two - index tensor , but now each index can be either longitudinal or transverse . for a symmetric tensor , there are three possibilities : both indices are longitudinal , one is transverse , or two are transverse . these are written as follows : @xmath46 where @xmath47 the first term in equation ( [ divtens ] ) is a longitudinal vector while the second term is a transverse vector . the divergence of the doubly - transverse part , @xmath48 , is zero . for a traceless symmetric tensor , the doubly and singly longitudinal parts can be obtained from the gradients of a scalar and a transverse vector , respectively : @xmath49 although we will not prove it here ( see [ @xcite ] for the details ) , under an infinitesimal coordinate transformation @xmath50 and @xmath51 can change while @xmath48 is invariant . similarly , the longitudinal and transverse parts of @xmath52 are both gauge - dependent . now we have the mathematical background needed to perform the scalar - tensor - vector split of the physical degrees of freedom of the metric . the `` tensor mode '' represents the part of @xmath32 that can not be obtained from the gradients of a scalar or vector , namely @xmath48 . the tensor mode is gauge - invariant and has two degrees of freedom ( five for a symmetric traceless @xmath53 matrix , less three from the condition @xmath54 ) . physically it represents gravitational radiation ; the two degrees of freedom correspond to the two polarizations of gravitational radiation . gravitational radiation is transverse : a wave propagating in the @xmath55-direction can have nonzero components @xmath56 and @xmath57 but no others . the tensor mode @xmath48 behaves like a spin-2 field under spatial rotations . note that my definition of the tensor mode strain @xmath34 here differs by a factor of 2 from the conventional @xmath58 in [ @xcite ] . the `` vector mode '' behaves like a spin-1 field under spatial rotations . it corresponds to the transverse vector parts of the metric , which are found in @xmath43 and @xmath51 . each part has two degrees of freedom . although we will not prove it here ( see [ @xcite ] for the details ) , it is possible to eliminate two of these degrees of freedom by imposing gauge conditions ( coordinate conditions ) . the synchronous gauge of lifshitz [ @xcite ] is one popular choice , with @xmath59 . another choice is the `` poisson '' or transverse gauge [ @xcite ] which sets @xmath60 . in either case , there are two physical degrees of freedom and they correspond physically to gravitomagnetism . although this effect is less well known than gravitational radiation or newtonian gravity , it produces magnetic - like effects on moving and spinning masses , such as the precession of a gyroscope in the gravitational field of a spinning mass ( lense - thirring precession ) . this phenomenon has not yet been discovered experimentally but should be measured by the gravity probe b satellite to be launched in 2002 [ @xcite ] . the `` scalar mode '' is spin-0 under spatial rotations and corresponds physically to newtonian gravitation with relativistic modifications . the scalar parts of the metric are given by @xmath33 , @xmath61 , @xmath42 , and @xmath50 . any two of these may be set to zero by means of a gauge transformation . one popular choice is @xmath62 , also known as the conformal newtonian gauge . it corresponds to the scalar mode in the * transverse gauge * , defined by the gauge conditions @xmath63 this gauge in linearized general relativity is the gravitational analogue of coulomb gauge in electromagnetism , where the magnetic vector potential is transverse , @xmath64 . in the gravitational transverse gauge , @xmath65 is transverse and @xmath34 is doubly transverse . this gauge is convenient for developing intuition although not necessarily the best for computation . the variables @xmath66 correspond to gauge - invariant variables introduced by bardeen [ @xcite ] as linear combinations of metric variables in other gauges . in linear perturbation theory , the scalar , vector , and tensor modes evolve independently . the vector and tensor modes produce no density perturbations and therefore are unimportant for structure formation , although they do perturb the microwave background . in the remainder of these lectures , only the scalar mode will be considered . the einstein equations give the equations of motion for the metric perturbations in terms of the energy - momentum tensor , the source of relativistic gravity . here we consider the case of a perfect fluid ( or several perfect fluid components combined ) , for which @xmath67 where @xmath68 and @xmath69 are the proper energy density and pressure in the fluid rest frame and @xmath70 is the fluid 4-velocity . we split @xmath71 into time and space components as we did for the metric . it is convenient to use mixed components because the conformal factor @xmath72 then cancels , yielding @xmath73\ , \nonumber \\ t^0_{\ \,i}&=&\left[\bar\rho(\tau)+\bar p(\tau)\right]v_i(\vec x,\tau ) = -\left[\bar\rho(\tau)+\bar p(\tau)\right]\nabla_i w\ , \\ t^i_{\ \,j}&=&\left[\bar p(\tau)+\delta p(\vec x,\tau)\right ] \delta^i_{\ \,j}\ , \nonumber\end{aligned}\ ] ] where @xmath74 and @xmath75 are respectively the energy ( or mass ) density and pressure of the robertson - walker background spacetime , @xmath76 is the fluid 3-velocity ( assumed nonrelativistic ) , and @xmath77 is a velocity potential . we are assuming that the velocity field is irrotational or , if it is not , that the matter fluctuations are linear so that the longitudinal and transverse velocity components evolve independently . this is a good approximation prior to the epoch of galaxy formation . we are also making another approximation , namely that the shear stress ( the non - diagonal part of @xmath78 ) is negligible compared with the pressure . this is a bad approximation for relativistic neutrinos , whose free - streaming leads to a non - isotropic momentum distribution after neutrino decoupling at a temperature of about 1 mev . ( a similar but smaller effect occurs for photons after they decouple at a temperature of about 0.3 ev . ) however , the effect of this error is modest during the radiation - dominated era ( photons contribute more pressure than neutrinos , and are isotropic ) and is unimportant during the matter - dominated era , when the nonrelativistic matter density is the major source of gravity . we introduce this approximation here in order to simplify the treatment ; full calculations include the effects of neutrino shear stress [ @xcite ] . when the stress tensor is isotropic , the einstein equations in transverse gauge imply that the two scalar potentials @xmath33 and @xmath61 are equal [ @xcite ] and the metric takes the simple form @xmath79\ . \label{cnmet}\ ] ] this is a simple cosmological generalization of the standard weak - field limit of general relativity , which is recovered by setting @xmath80 and @xmath81 . given the stress - energy tensor , we can now obtain field equations for @xmath82 using the einstein equations . the only independent equation for the robertson - walker background is the friedmann equation ( [ friedmann ] ) . for the perturbations , we obtain @xmath83\ , \ ] ] @xmath84 and @xmath85 there are more equations than in newtonian gravity because the einstein equations have local energy - momentum conservation built into them . equation ( [ poisson ] ) is remarkably like the poisson equation of newtonian gravity , especially when one realizes that the factor of @xmath72 is needed to convert the laplacian from comoving to proper spatial coordinates . otherwise there are three differences . first , the spatial curvature adds a scale to the laplacian in a curved space . second , the density source is @xmath86 rather than @xmath68 because @xmath33 represents a metric perturbation on top of the background robertson - walker spacetime . third , momentum density ( through @xmath77 ) is a source for gravity in general relativity . however , this term is comparable with the density term only on length scales comparable to or larger than the hubble length @xmath87 . two of our field equations are unfamiliar from newtonian dynamics . equation ( [ momcon ] ) is reminiscent of the zeldovich approximation [ @xcite ] in that the velocity potential is proportional to the gravitational potential in the linear regime when @xmath88 . however , it is a more general result whose newtonian limit may be derived by combining the continuity equation with the time derivative of the poisson equation . equation ( [ phidd ] ) is more surprising , as it gives the gravitational potential through a second - order time evolution equation as opposed to the action - at - a - distance poisson equation ( [ poisson ] ) . the reader may well wonder about the compatibility of the two equations as well as the physical validity of the poisson equation in general relativity , but equations ( [ poisson])([phidd ] ) are all correct . see [ @xcite ] for a full discussion . in brief , causality is restored by the tensor mode . the time evolution equation for the potential can be regarded as arising from local energy - momentum conservation ( which is built into the einstein equations ) combined with the poisson equation . indeed , an equation similar to equation ( [ phidd ] ) can be derived in newtonian gravity from the fluid and poisson equations . its utility arises from the fact that when pressure perturbations are gravitationally unimportant ( i.e. when matter and/or dark energy dominate over radiation ) , the time evolution of perturbations can be obtained from a single differential equation depending only on the background curvature , pressure , and density . to conclude this section , we present a simple model for understanding cmb anisotropy and large scale structure . we reduce the matter and energy content of the universe to a vacuum energy ( with @xmath89 contributing to the mean expansion rate but not to the fluctuations ) and two fluctuating components , cdm and radiation ( photons and neutrinos , the former coupled to baryons by electron scattering until recombination ) . the cdm and radiation are treated as perfect fluids , ignoring the free - streaming of neutrinos and the diffusion and streaming of photons during and after recombination . the fluid equations follow from local energy - momentum conservation , @xmath90 . for a perfect fluid , with the scalar mode only ( no vorticity and no gravitational radiation ) , the fluid equations in the metric of equation ( [ cnmet ] ) are @xmath91=0\ ] ] and @xmath92 + 4{\dot a\over a}(\rho+p)v^i + \nabla_ip+(\rho+p)\nabla_i\phi=0\ .\ ] ] these are familiar from newtonian fluid dynamics , with some extra terms . the damping ( @xmath93 ) terms arise from hubble expansion and the use of comoving coordinates . the @xmath94 term in the continuity equation ( [ conteq ] ) arises from the deformation of the spatial coordinates , i.e. the @xmath95 contribution to the metric . ( the flux term in eq . [ conteq ] describes transport relative to the comoving coordinate grid , which is deforming when @xmath96 . ) the energy flux ( or momentum density , they are equal ) includes pressure in relativity because of the @xmath97 work done by compression . equations ( [ conteq ] ) and ( [ eulereq ] ) apply separately to perfect fluid component . this section discusses the evolution of perturbations from the end of inflation ( taken to be at @xmath98 for all practical purposes ) through recombination using a two - fluid approximation . we approximate the matter and energy content in the universe as being two perfect fluids : cold dark matter ( a zero - temperature gas ) and radiation ( photons coupled to electrons and baryons before recombination , plus neutrinos which are approximated as behaving like photons ) . the net energy density perturbation is thus @xmath99 where @xmath100 and , if we neglect the contribution of baryons , @xmath101 . we could add a third component for the baryons , but this complicates the presentation without adding essential new behavior . in the interest of economy , we leave baryons out . we define the relative density contrasts @xmath102 and @xmath103 . the total pressure is simply @xmath104 where @xmath105 is the spatially constant negative pressure of vacuum energy ( cosmological constant ) , should any be present . as we noted in the previous section , density perturbations couple gravitationally only to the scalar mode of metric perturbations , and the scalar mode can not generate transverse vector fields . as a result , the peculiar velocity fields of cdm and radiation are longitudinal and are fully characterized by their potentials @xmath106 and @xmath107 for cdm and radiation , respectively . as a final simplifying assumption , we will assume that we are studying effects on distance scales less than the curvature length @xmath108 , which is at least 5 gpc and possibly infinitely large . with these assumptions , we can linearize the fluid equations ( [ conteq ] ) and ( [ eulereq ] ) for @xmath109 and @xmath110 to obtain @xmath111 these equations are easy to understand . the @xmath94 terms have the same origin as in equation ( [ conteq ] ) , namely the deformation of the comoving coordinate system . the velocity potential terms are related to the velocity divergence , @xmath112 . the radiation density fluctuations grow at @xmath113 the rate of matter fluctuations to compensate for the fact that the photon number density is proportional to @xmath114 . as expected , the velocity equations show that matter velocities redshift ( for @xmath115 ) with @xmath116 while radiation velocities ( e.g. the speed of light ) do not redshift . ( momenta redshift @xmath117 but the speed of light is unaffected by expansion . ) finally , radiation pressure gives rise to an acoustic restoring force @xmath118 which shows up as the @xmath119 term in the radiation longitudinal velocity equation . equations ( [ linfluid ] ) are coupled by gravity , which can be determined from the poisson equation ( [ poisson ] ) , which becomes @xmath120\ .\ ] ] we can solve equations ( [ linfluid ] ) and ( [ poisson2 ] ) most easily by expanding the spatial dependence in eigenfunctions of the laplacian @xmath121 . in flat space these are plane waves @xmath122 , and this is a good approximation even if the background is curved , provided @xmath123 . this approximation is valid for almost all applications except large angular scale cmb anisotropy . note that @xmath124 is a comoving wavevector ; the physical wavelength is @xmath125 . by expanding the spatial dependence in plane waves , we now have a fourth - order linear system of ordinary differential equations in time . we are free to define linear combinations of the variables @xmath126 as our fundamental variables . in terms of the mechanisms for generating primeval perturbations , the most natural variables are the metric perturbation @xmath33 and the specific entropy @xmath127 where @xmath128^{-1}\ ] ] is the effective one - fluid sound speed of the matter and radiation . ( acoustic signals actually do not propagate at this speed ; they propagate with speed @xmath129 through the radiation and they do not propagate at all through cdm . ) for a two - component radiation plus cdm universe , the solution of the friedmann equation ( [ friedmann ] ) gives @xmath130 the radiation - dominated era ends and matter - dominated era begins at redshift @xmath131 . a cosmological constant has no significant effect on @xmath132 provided that , during the times of interest , the vacuum energy density is much less than the radiation or cold dark matter density . the fluid and poisson equations can now be combined to give a pair of second - order ordinary differential equations , @xmath133 it is interesting to note that @xmath33 and @xmath134 evolve independently aside from the source term each provides to the other . the coupling implies that entropy perturbations are a source for the growth of gravitational perturbations and vice versa . for a given wavenumber @xmath135 , there are two key times in the evolution of @xmath33 and @xmath134 : the sound - crossing time @xmath136 and the time of matter - radiation equality at @xmath137 . [ to be precise , @xmath138 at @xmath139 . ] for @xmath140 , @xmath33 and @xmath134 decouple and they both have solutions that decay rapidly with @xmath5 as well as solutions that are finite as @xmath141 . the latter are conventionally called `` growing modes '' while the former are called decaying . the growing mode solution in the radiation era @xmath142 is @xmath143i(\vec k\,)\ , \\ \eta(\vec k,\tau)&=&i(\vec k\,)+9\left[\ln\omega\tau + { \cal c}-{\rm ci}(\omega\tau)+{1\over2}(\cos\omega\tau-1 ) \right]a(\vec k\,)\ , \nonumber\end{aligned}\ ] ] where @xmath144 is the phase of acoustic waves in the radiation fluid , @xmath145 is the euler constant , and @xmath146 is the cosine integral defined by @xmath147 in equation ( [ phietagrow ] ) we have neglected all terms @xmath148 aside from the lowest - order contribution of @xmath149 to @xmath33 . the general growing - mode solution contains two @xmath135-dependent integration constants , @xmath150 and @xmath149 , which represent the initial ( @xmath141 ) amplitude of @xmath33 and @xmath134 , respectively . conventionally they are called the `` adiabatic '' ( so - called because the entropy perturbation @xmath151 at @xmath98 , although `` isentropic '' would be more appropriate ) and `` isocurvature '' ( so - called because @xmath115 at @xmath98 ) mode amplitudes . in addition , there are decaying modes whose amplitude at @xmath152 is negligible unless the initial conditions were fine - tuned or there are external sources such as topological defects to continually regenerate the decaying mode . the solution of equations ( [ phietagrow ] ) is valid only for inflation or other mechanisms that generate fluctuations exclusively in the early universe , but not for topological defects . as a result of reheating , inflation produces fluctuations with spatially constant ratios of the number densities of all particle species . for cdm and radiation , for example , @xmath153 hence @xmath154 and thus @xmath151 until radiation pressure forces begin to separate the radiation from the cdm , which takes a sound - crossing time or @xmath155 . thus , although inflation produces the adiabatic mode of initial fluctuations , the differing equations of state @xmath156 for the two - fluid system lead to entropy perturbations proportional to the initial adiabatic mode amplitude @xmath150 ( see the second of eqs . [ phietagrow ] ) . physically , the number density of particles of all species are proportional to each other in the adiabatic mode as long as there has been too little time for particles to travel a significant difference via their thermal speeds . however , the differing thermal speeds of different particle species ( e.g. photons and cold dark matter ) eventually cause the particle densities to evolve differently , so that the number density ratios are no longer constant and the entropy perturbation is no longer zero . first - order phase transitions , on the other hand , change the equation of state without moving matter or energy on macroscopic scales , hence @xmath157 initially implying @xmath115 initially but @xmath158 . this isocurvature mode is characterized by the initial entropy perturbation amplitude @xmath149 . as the first of equations ( [ phietagrow ] ) shows , the isocurvature mode generates curvature fluctuations when the universe becomes matter - dominated . the isocurvature cdm inflationary model has a fine balance such that @xmath159 initially . this may involve nonlinear fluctuations of @xmath160 at early times , but @xmath161 is negligible early in the radiation - dominated era @xmath162 . for this reason , the isocurvature mode was originally called isothermal [ @xcite ] . on small scales , @xmath163 , equations ( [ phietaeqs ] ) may be solved using a wkb approximation . for our purposes it suffices to take the @xmath164 limit of equations ( [ phietagrow ] ) : @xmath165 the radiation density fluctuation scales as @xmath166 which oscillates in proportion to @xmath167 for the adiabatic mode and grows in proportion to @xmath168 for the isocurvature mode . note from equation ( [ phisubhor ] ) that the isocurvature mode induces oscillations in the potential with a @xmath169 phase shift relative to the adiabatic mode . this shift translates into a shift in the positions of the acoustic peaks in the cmb angular power spectrum . it is instructive to consider the general adiabatic perturbation in the radiation - dominated era including the decaying mode . this problem is easy to analyze and shows the role of acoustic waves in the radiation - dominated era . the two linearly independent solutions of the first of equations ( [ phietaeqs ] ) for @xmath162 and adiabatic initial conditions ( @xmath151 ) are @xmath170 where the subscripts label the growing and decaying modes , respectively . they are the real and imaginary parts of @xmath171 for a given spatial harmonic , @xmath172 represents a wave traveling in the @xmath173-direction . thus , the growing mode @xmath174 represents a superposition of waves traveling along @xmath124 and @xmath175 , i.e. a standing wave @xmath176\ .\ ] ] the waves travel at the sound speed @xmath177 in the radiation fluid . the entropy perturbation induced by a traveling wave potential perturbation is easily calculated in the radiation - dominated era using the green s function method applied to the second of equations ( [ phietaeqs ] ) with @xmath178 for the source term . the result is @xmath179 the real part of this equation gives the @xmath150 contribution to @xmath134 in equations ( [ phietagrow ] ) . this section presents a simplified treatment of cmb anisotropy , with the aim of highlighting the essential physics without getting lost in all of the details . more complete treatments are found in [ @xcite ] . the microwave background radiation is fully described by a set of photon phase space distribution functions . ignoring polarization ( a few percent effect ) , all the information is included in the intensity or in @xmath180 where @xmath181 is the number density of photons of conjugate momentum @xmath182 at position and time ( @xmath183 . the conjugate momentum is related to the proper momentum measured by a comoving observer , @xmath184 , so that @xmath185 along a photon trajectory in the absence of metric perturbations . remarkably , despite metric perturbations and scattering with free electrons , the cmb photon phase space distribution remains blackbody ( planckian ) to exquisite precision : @xmath186\ ] ] where @xmath187 k is the current cmb temperature and @xmath188 is the temperature variation at position @xmath189 for photons traveling in direction @xmath190 . the phase space density is blackbody but the temperature depends on photon direction as a result of scattering and gravitational processes occurring along the line of sight . the phase space density may be calculated from initial conditions in the early universe through the boltzmann equation @xmath191 where the right - hand side is a collision integral coming from nonrelativistic electron - photon elastic scattering . in the absence of scattering , the phase space density is conserved along the trajectories @xmath192\phi\ .\ ] ] note that the photon energy changes via gravitational redshift ( the first term in @xmath193 ) or a time - changing potential , but in either case the energy of all photons is changed by the same factor . the photon direction changes by transverse deflections due to gravity ; @xmath194 is twice the newtonian value , a result well - known in gravitational lens theory [ @xcite ] . because @xmath195 for the unperturbed background ( the anisotropy arises due to the metric perturbations ) , gravitational lensing affects the cmb anisotropy only through nonlinear effects on small scales [ @xcite ] . the procedure for computing cmb anisotropy is to linearize the boltzmann equation assuming @xmath196 [ @xcite ] . until recently , the temperature anisotropy @xmath197 was expanded in spherical harmonics and the boltzmann equation was solved as a hierarchy of coupled equations for the various angular moments [ @xcite ] . in 1996 , seljak and zaldarriaga [ @xcite ] developed a much faster integration method call cmbfast based on integrating the linearized boltzmann equation along the observer s line of sight before the angular expansion is made : @xmath198 _ { \rm ret}\ ] ] where @xmath199 is the present conformal time , @xmath22 is the radial comoving coordinate of equation ( [ rwdl ] ) , subscript `` ret '' denotes evaluation using retarded time @xmath200 , @xmath201 is the relative density fluctuation in the photon gas ( @xmath202 in our two - fluid approximation ) , @xmath203 is the mean electron velocity ( i.e. the baryon velocity , which equals @xmath204 in our two - fluid approximation ) , @xmath205 is the thomson cross section , and the thomson optical depth is @xmath206 we have left out small terms ( `` pol . term '' ) due to polarization and the anisotropy of thomson scattering in equation ( [ cmblos ] ) . the terms in equation ( [ cmblos ] ) are easy to understand . the @xmath207 factor accounts for the opacity of electron scattering , which prevents us from seeing much beyond a redshift of 1100 , where @xmath208 . the cmb anisotropy appears to come from a thin layer called the photosphere , just like the radiation from the surface of a star . the two gravity terms give the effective emissivity due to the photon energy change caused by a varying gravitational potential . for a blackbody distribution , if all photon energies are increased by a factor @xmath209 , the distribution remains blackbody with a temperature increased by the same factor . thus , the line - of - sight integration of the fractional energy change of equation ( [ geodes ] ) translates directly into a temperature variation . the terms proportional to the thomson opacity @xmath210 are the effective emissivity due to thomson scattering . bearing in mind that the photon - baryon plasma is in nearly perfect thermal equilibrium ( the temperature variations are only about 1 part in @xmath211 ) , the photons we see scattered into the line of sight from a given fluid element have the blackbody distribution corresponding to the temperature of that element . recalling that the energy density of blackbody radiation is @xmath212 , we see that @xmath213 is simply the fractional temperature variation of the fluid element . in other words , if we carried a thermometer to the photosphere and measured its reading in direction @xmath214 , it would read a fraction @xmath215 different from the mean . now the energy density is defined in the fluid rest frame , while the fluid is moving with 3-velocity @xmath203 , so the temperature measured by a stationary thermometer ( with fixed @xmath3 ) is changed by a doppler correction @xmath216 . ( if the photosphere is approaching us , @xmath217 , the radiation is blue - shifted . ) comparing the cmb with a star , when we look at the surface of a star we see the temperature of the emitting gas ( to the extent that local thermodynamic equilibrium applies ) , corresponding to the @xmath215 term . for ordinary stars the doppler boosting and gravitational redshift effects are negligible , although they are appreciable for supernovae and white dwarfs , respectively . for the cmb , on the other hand , all four emission terms in equation ( [ cmblos ] ) are comparable in importance . it is instructive to approximate the photosphere as an infinitely thin layer by adopting the instantaneous recombination approximation , according to which the free electron fraction and hence opacity drops suddenly at @xmath218 , the conformal time of recombination at @xmath219 : @xmath220 substituting into equation ( [ cmblos ] ) and integrating by parts the gravitational redshift term gives the result first obtained in another form by sachs and wolfe in 1967 [ @xcite ] : @xmath221 where we have dropped the argument @xmath199 from @xmath197 , @xmath222 , and @xmath223 is the velocity potential for the photons ( @xmath224 in the two - fluid approximation discussed above ) . we could have used the last of equations ( [ linfluid ] ) to write the sum of the intrinsic ( @xmath215 ) and gravitational ( @xmath33 ) contributions as @xmath225 . the combination of derivatives @xmath226 is the time derivative along the past light cone ; the results are evaluated at recombination . while it is interesting that the primary anisotropy depends on the velocity potential derivative along the line of sight , there appears to be no fundamental significance to the simple dependence in equation ( [ cmbsw ] ) , aside from the fact that cmb anisotropy is produced by departures from hydrostatic equilibrium . ( in hydrostatic equilibrium , @xmath227 . ) if the radiation gas were in hydrostatic equilibrium in a static gravitational potential , there would be no primary cmb anisotropy . indeed , it can be shown from purely thermodynamic arguments that @xmath228 $ ] ( hence @xmath229 ) for a photon gas in equilibrium . sachs and wolfe [ @xcite ] showed that for adiabatic perturbations in the matter - dominated era , on scales much larger than the acoustic horizon ( @xmath230 in eq . [ phietagrow ] ) , the sum of the photospheric terms in equation ( [ cmbsw ] ) ( the terms evaluated at recombination ) is @xmath231 . thus , on angular scales much larger than @xmath232 ( roughly the size of the acoustic horizon ) , the cmb anisotropy is a direct measure of the gravitational potential on the photosphere at recombination . the angular power spectrum gives the mean squared amplitude of the cmb anisotropy per spherical harmonic component . thus , we expand the anisotropy in spherical harmonic functions of the photon direction : @xmath233 observations of our universe give definite numbers for the @xmath234 ( with experimental error bars , of course ) . theoretically , however , we can only predict the probability distribution of the @xmath234 . for statistically isotropic fluctuations ( i.e. having no preferred direction a priori ) , the @xmath234 are random variables with covariance @xmath235 where the kronecker delta s make the @xmath234 s uncorrelated ( @xmath236 if @xmath237 and 0 otherwise ) . the variance of each harmonic is given by the angular power spectrum @xmath238 ; rotational symmetry ensures that , theoretically , it is independent of @xmath239 . to calculate the angular power spectrum , we expand @xmath82 ( and the other variables ) in plane waves ( assuming that the background is flat , @xmath36 ) , @xmath240 where @xmath241 with a minus sign because @xmath214 is the radial unit vector at the origin . note that many cosmologists insert a factor @xmath242 in the fourier integral going from @xmath135-space to @xmath243-space ; take heed of the conventions when they matter . in equation ( [ fourierexp ] ) we have used the spherical wave expansion of a plane wave in terms of spherical bessel functions @xmath244 and legendre polynomials @xmath245 . this gives @xmath246 where , in the instantaneous recombination approximation , @xmath247 j_l(k\chi_{\rm rec})+kw_r(\vec k,\tau_{\rm rec } ) j'_l(k\chi_{\rm rec})\nonumber \\ & & + \int_0^{\chi_{\rm rec}}d\chi\ , 2\partial_\tau\phi(\vec k,\tau_0-\chi)j_l(k\chi)\ .\end{aligned}\ ] ] the temperature expansion coefficient takes a simple form in terms of @xmath248 : @xmath249 where @xmath250 is a unit vector in the direction of @xmath124 . to get the angular power spectrum now we must relate @xmath251 to the initial random field of potential or entropy fluctuations that induced the cmb anisotropy . let us assume we have adiabatic fluctuations , for which @xmath252 . we define the cmb transfer function @xmath253 which depends only on the magnitude of @xmath124 because the equations of motion have no preferred direction . to compute ensemble averages of products , we will need the two - point function for a statistically homogeneous and isotropic random field , whose variance defines the power spectrum : @xmath254 note that many cosmologists use an alternative definition of the power spectrum with a factor @xmath255 inserted on the right - hand side . the power spectrum so defined is greater by a factor @xmath255 than the one given here . [ the extra factors of @xmath255 in eq . [ fourierexp ] are then important . ] the definition presented here is conventional in field theory and it ensures that @xmath256 is the contribution per unit volume in @xmath135-space to the variance of @xmath257 . because of this interpretation , the power spectrum is also known as the spectral density . see [ @xcite ] for further discussion . for scale - invariant fluctuations ( equal power on all scales , as predicted by the simplest inflationary models ) , @xmath258 . combining equations ( [ covalm ] ) , ( [ almk ] ) , and ( [ dldef ] ) gives the formal expression for the angular power spectrum as an integral over the three - dimensional power spectrum : @xmath259 it is difficult to get a simple approximation to @xmath260 that makes this analytically tractable , except in the limit of large angular scales where the intrinsic and gravitational contributions to @xmath261 dominate , with @xmath262 . then @xmath263 and the integral may be evaluated analytically for @xmath264 with fixed @xmath265 . when @xmath266 ( the scale - invariant spectrum ) , the result gives @xmath267 or equal power on all angular scales . the phenomenon of acoustic peaks in @xmath268 is due to the acoustic oscillations in @xmath33 and @xmath223 which modify the potentials at recombination from their scale - invariant primeval forms . this presentation has been based on the traditional fourier space representation of the potential and other fields . the physical interpretation of the cmb two - point function is somewhat clearer in real ( position ) space , although a more detailed analysis based on green s functions is needed before one can reap the rewards . for an introduction to the position - space perspective see [ @xcite ] this section summarizes briefly the steps needed to go from fluctuation evolution in the early universe to predictions of galaxy formation and large scale structure . the presentation will focus on techniques rather than model results . however , in order to simplify the discussion , we will focus on the favored cold dark matter family of models ( with curvature and vacuum energy as options ) with adiabatic density fluctuations from inflation . during the radiation - dominated era , cdm density fluctuations grow only logarithmically . once the universe becomes matter - dominated at @xmath270 ( eq . [ ytau ] ) , equation ( [ phidd ] ) shows that the potential becomes frozen in as long as vacuum energy ( or other sources of pressure ) and curvature are dynamically unimportant . thus , from equation ( [ poisson ] ) , the cdm density fluctuation amplitude grows linearly with @xmath271 since @xmath272 : @xmath273 we have used subscript `` m '' to include all cold matter including baryons , which cluster with the cdm after recombination . we have also included the velocity potential term @xmath274 in equation ( [ poisson ] ) as part of @xmath275 as it is in bardeen s gauge - invariant density perturbation variable @xmath276 [ @xcite ] . ( the velocity potential term is anyway negligible on scales much smaller than the hubble distance @xmath87 . ) to fully specify the cdm density fluctuation spectrum we now need to know the gravitational potential , or equivalently the density fluctuations , at the end of the radiation era . there is a crucial physical scale , the jeans length for the radiation gas at @xmath152 , @xmath277 ( including the effects of baryons decreases this by up to 15% from what is given here . ) perturbations of wavelength longer than @xmath278 have always been jeans unstable , so that @xmath279 for @xmath280 . note that only the acoustic horizon not the causal horizon or hubble scale enters as a length scale in the cdm transfer function . to get the amplitude for shorter - wavelength perturbations which have suffered acoustic damping ( i.e. they were jeans stable for a period of time ) , it is not enough to use the radiation - era solution for the gravitational potential in equation ( [ deltacphi ] ) , because the cdm density fluctuations dominate over those of radiation before the universe becomes matter - dominated . instead , we turn to equations ( [ poisson ] ) and ( [ phietagrow ] ) for @xmath164 in the radiation era . for the adiabatic mode this gives @xmath281 , \quad\tau<\tau_e\ \hbox{and}\ \omega\tau\gg1.\end{aligned}\ ] ] the gravitational impulse created by the acoustic oscillations of the radiation fluid causes the cdm to oscillate slightly but as @xmath282 grows the cosine integral quickly dies and the cdm grows logarithmically . this growth is important for structure formation as it allows the cdm perturbations to dominate over the radiation perturbations enough for galaxies to form in less than a factor 1000 expansion since recombination when the large - scale radiation fluctuations were only a few parts in @xmath211 . combining results , we obtain an approximate solution for the cdm transfer function @xmath283 giving the amplitude of cdm density fluctuations relative to the primeval potential @xmath150 : @xmath284\ , \quad\ k\gg k_e\ , } \ ] ] where @xmath285 bardeen et al . [ @xcite ] give a fitting formula for @xmath286 , which , when normalized to 1 as @xmath287 , is conventionally called the cdm transfer function . however , given the measurements of the gravitational potential made possible from the cmb , it is preferable to retain the normalization of equation ( [ tmlim ] ) with the potential @xmath150 factored out in equation ( [ tmdef ] ) . the cdm linear power spectrum follows by analogy with equation ( [ pspect ] ) for the potential . the result , for a matter - dominated universe , is @xmath288 ^ 2t_m^2(k ) p_a(k)\ .\ ] ] the constant - curvature scale - invariant spectrum of potential fluctuations @xmath289 was first proposed by harrison [ @xcite ] for its naturalness properties long before inflationary cosmology gave it a concrete basis . with this spectrum , the cdm density spectrum has the well - known limiting behavior @xmath290 ^ 2\ , \quad k\gg k_e\ .}\ ] ] the power spectrum is useful for calculating the variance of the smoothed density field . if we convolve @xmath291 with a spherical window function whose fourier transform is @xmath292 normalized so that @xmath293 , then @xmath294 from this , we see that the cdm density fluctuations have logarithmically divergent power at small scales . thus , structure formation in a cdm model proceeds by hierarchical or bottom - up clustering with small scales collapsing by gravitational instability before being incorporated into larger objects . however , the slowness of the logarithm means that the structure formation proceeds rapidly from small to large scales . figure [ fig : pk ] compares the linear power spectra extrapolated to @xmath295 ( i.e. redshift 0 ) for several flat cdm models . the curves are all normalized so that the density smoothed with a 8 @xmath296 mpc spherical `` tophat '' has standard deviation @xmath297 in agreement with observed galaxy clustering for a `` linear bias factor '' @xmath298 . the @xmath269cdm model with @xmath299 has the most power on large scales , possibly too much ( although an acceptable fit can be found with slightly higher @xmath300 ) . the scdm model with @xmath301 has too little power on large scales . when combined with the cmb anisotropy , this provides a strong argument for dark energy ( e.g. a cosmological constant ) [ @xcite ] . the curves labeled `` bbks '' show the linear power spectra for the @xmath269cdm and scdm models in the limit @xmath302 ( with @xmath303 and 1.0 for the two models , respectively ) . they are obtained from the fitting formula given by equation ( g3 ) of [ @xcite ] and agree well with full calculations using the cosmics code [ @xcite ] , which was used for the other curves in figure [ fig : pk ] . baryons reduce the amount of small - scale power relative to scales @xmath304 because they can not experience gravitational instability on scales smaller than the photon - baryon jeans length until after recombination . the oscillations in the @xmath269cdm power spectrum are also due to baryons and are similar to the cosine integral contributions to @xmath305 in equation ( [ delsubhor ] ) . they arise from the acoustic oscillations of the baryons between @xmath306 and @xmath218 . when the power spectra are normalized at small scales ( as they effectively are with @xmath307 ) , increasing the baryon fraction increases the amount of large - scale power . aside from the baryon effect , the main parameter affecting the cdm spectrum ( and the only one in the bbks formula ) is @xmath308 . as may be checked using equation ( [ kradj ] ) , the peak of the cdm power spectrum is closely equal to @xmath308 . measurements of galaxy clustering actually constrain @xmath309 mpc@xmath310 where @xmath311 is a commonly used parameter . however , as figure [ fig : pk ] shows , a second parameter , the baryon fraction @xmath312 , also significantly affects the cdm spectrum . the results given above concerned the shape of the cdm power spectrum , which is frozen in after the universe becomes matter - dominated however , the amplitude grows as @xmath6 as assumed in equation ( [ tmdef ] ) only if @xmath313 . while this is a good approximation soon after recombination , it breaks down at small redshift when @xmath314 today . in this case we must include dark energy , curvature , and any other important contributor to the expansion rate of the universe . the evolution of cdm density fluctuations for @xmath315 is given by equations ( [ poisson ] ) and ( [ linfluid ] ) ignoring the radiation terms . on subhorizon scales for @xmath316 we have @xmath317 because the coefficients are spatially homogeneous , the time and space dependences of the solution factor : @xmath318 . for @xmath319 , @xmath320 grows with time and @xmath321 decays . for @xmath313 the growing solution is @xmath322 . for @xmath314 the solutions depend on the curvature @xmath10 ( sometimes expressed as @xmath323 ) and on the equation of state of the dark energy . see peebles [ @xcite ] and peacock [ @xcite ] for examples . peculiar velocities offer another probe of large - scale structure that is sensitive directly to all matter , dark or luminous . the divergence of peculiar velocity @xmath324 depends on the growth rate of density fluctuations : @xmath325 measurements of galaxy velocities actually give @xmath326 , so that the ratio of velocity divergence to density fluctuation gives the logarithmic growth rate of density fluctuations . an empirical fit to a range of models gives @xmath327 . in practice , the galaxy number density may be a biased tracer of the total matter density , conventionally parameterized by the bias factor @xmath328 : @xmath329 . comparing galaxy clustering and velocities then allows determination of @xmath330 . linear theory breaks down once dark matter clumps become significantly denser than the cosmic mean . once this happens , full numerical simulations are needed to accurately follow the nonlinear dynamics of structure formation . moreover , radiative processes , star formation , and supernova feedback have a strong effect on the baryonic component once galaxies begin forming . although these processes can only be included in simulations in a phenomenological way , numerical simulations are still the best way to advance our theoretical understanding of structure formation deep into the nonlinear regime . here we present only a brief synopsis of numerical simulation methods . for a comprehensive review of both the methods and their applications see [ @xcite ] . particle methods are used for representing the dark matter . the phase space of dark matter is represented by a sample of particles evolving under their mutual gravitational field : @xmath331\ ] ] where @xmath332 is the dirac delta function . in practice , the particles are spread over a softening distance to avoid unphysical two - body relaxation due to large - angle scattering . ( a simulation may use particles of mass @xmath333 , leading to unphysical tight binaries if the forces are not softened . ) the art of dark matter simulation is mainly in calculating forces much faster than @xmath334 for a softening distance several orders of magnitude smaller than the simulation volume . a variety of algorithms and codes are available ; see [ @xcite ] for details . the initial conditions for dark matter positions and velocities are given from linear theory by the zeldovich approximation [ @xcite ] , which relates the comoving position to its initial value @xmath335 , a lagrangian variable ( i.e. one that labels each particle ) : @xmath336 the initial density fluctuation field @xmath337 is obtained as a sample of a gaussian random field with appropriate power spectrum . baryon physics is added by numerical integration of the fluid equations ( [ conteq ] ) and ( [ eulereq ] ) with the addition of heating and cooling , ionization , etc . there are two classes of methods in widespread use in cosmology : smoothed - particle hydrodynamics ( sph ) and grid - based methods . sph is an extension of the n - body dynamics of equations ( [ nbody ] ) to include pressure forces and thermodynamics . it has the advantage of concentrating the spatial resolution and computational effort where the baryons are densest , with the drawback of relatively large numerical viscosity and diffusivity owing to the particle discreteness . grid - based methods use finite - difference techniques to approximate the fluid equations on a ( usually ) regular lattice . high accuracy methods developed for aerodynamic and other applications provide excellent resolution of shock waves , but with a resolution limited by the grid . adaptive mesh refinement techniques are now being used to refine the resolution where needed , resulting in the ability to follow structure formation to much higher densities than previously possible [ @xcite ] . we conclude this introduction to numerical simulation methods with a mosaic of gravitational potential and dark matter density from a high - resolution simulation of the mixed cold+hot dark matter model . figure [ fig : pd ] shows the linear theory and nonlinear simulation results . on the scales shown , @xmath269cdm would be almost indistinguishable . the gravitational potential shown in figure [ fig : pd ] is a statistical fractal at the end of inflation , viz . a gaussian random field with power spectrum @xmath338 . during the radiation era , acoustic damping smooths the potential on scales smaller than @xmath339 . because the density fluctuations are related to the potential by two spatial derivatives , the corresponding linear density field is dominated by high spatial frequency components . indeed , if not for the smoothing imposed by finite spatial resolution , the variance of density fluctuations would be logarithmically divergent . it is instructive to compare the post - recombination density with the inflationary potential . the density is almost proportional to the potential but with opposite sign . this is because the density / potential transfer function ( [ tmlim ] ) varies only logarithmically on small scales . the nonlinear evolution of the density field causes the mass to cluster strongly into wispy filaments and dense clumps ; most of the volume has low density . the intermittency of structure is a sign of strong deviations from gaussian statistics . numerical simulations show that most of the mass falls into dense self - gravitating clumps . as a crude approximation , let us suppose that the mass profile of every clump is spherically symmetric . in this case every gravitationally bound shell will eventually fall into the center of the clump , at a time that is dependent on the enclosed mass and energy of the shell . for linear growing mode initial conditions , collapse occurs when linear theory would predict that the mean density contrast interior to the shell is @xmath340 [ @xcite ] . thus , in the spherical model , the distribution of clump masses is simply dependent on the statistics of the smoothed linear density field . press and schechter [ @xcite ] devised an elegant method for computing the mass distribution of gravitationally bound clumps using the simple spherical model . their procedure is first to smooth the linear density field by convolution with a spherical `` tophat '' of radius @xmath341 enclosing mean mass @xmath342 : @xmath343 according to the press - schechter ansatz , the regions of the initial density field with @xmath344 at time @xmath5 have collapsed into objects of mass at least @xmath345 . it is easy to see that this would be true for a single spherical density perturbation ; the genius is applying this to the entire density field . under this model , the fraction of mass in the universe contained in clumps more massive than @xmath345 is @xmath346 on the left - hand side , @xmath347 is the number density of clumps of mass in @xmath348 ; the integral gives the total mass density in clumps more massive than @xmath345 . dividing by @xmath349 gives the mass fraction . on the right - hand side , @xmath350 is the probability that a randomly chosen point in the linear density field has @xmath351 at time @xmath5 , where @xmath345 defines the smoothing scale of @xmath352 . for a gaussian random field of density fluctuations , this probability is simply @xmath353 where erfc is the complementary error function and @xmath354 is the variance of the smoothed density . from equation ( [ massfrac ] ) we can deduce @xmath355 by differentiating with respect to @xmath345 . now , as @xmath356 , @xmath357 grows without bound for cdm models with logarithmically diverging power on small scales , so that @xmath358 . half of the mass initially is in overdense regions , half is in underdense regions @xmath359 which in linear theory can never collapse . however , this is plainly unphysical , as simulations show that virtually all of the mass , including that in initially underdense regions , accretes onto dense clumps . thus , press and schechter suggested doubling @xmath355 so that all of the mass resides in dense clumps . thus , their formula for the mass distribution of clumps is @xmath360 although realistic gravitational collapse is far more complex than imagined in this derivation , equation ( [ psdist ] ) gives surprisingly good agreement with clump masses measured in cosmological n - body simulations [ @xcite ] . the press - schechter model and its generalizations have proved to be a very useful tool for comparing ab initio theories ( power spectrum and cosmological model ) with observations of galaxies and galaxy clusters . these lecture notes have presented a few of the theoretical elements useful for understanding the microwave background and structure formation . cosmology is rapidly changing with the advent of precision measurements of cmb anisotropy , with the observation of galaxies at @xmath361 , and with large new redshift surveys . theoretical research has become more phenomenological , with a focus on providing the framework for interpreting and analyzing the new data . despite this trend or perhaps because of the high quality of new data theoretical cosmology still requires further development of the basic processes of structure formation . either way , i hope that the material presented in these lectures will be useful to cosmology graduate students . i would like to thank sergei bashinsky for help with the two - fluid solutions . i am grateful to the organizers , the other speakers , and the students of the cosmology 2000 summer school for the fruitful , stimulating atmosphere in lisbon . special thanks are due to the organizers for their wonderful hospitality . this work was supported by nsf grant ast-9803137 . [ 1 ] e. bertschinger , in cosmology and large scale structure , proc . les houches summer school , session lx , ed . r. schaeffer , j. silk , m. spiro , and j. zinn - justin , elsevier science , amsterdam ( 1996 ) 273 . [ 2 ] c .- p . ma and e. bertschinger , 455 ( 1995 ) 7 . [ 3 ] e. m. lifshitz , j. phys . ussr 10 ( 1946 ) 116 . [ 4 ] c.w . misner , k.s . thorne , and j.a . wheeler , gravitation , freeman , san francisco ( 1973 ) . [ 5 ] gravity probe b website , http://einstein.stanford.edu/. [ 6 ] j.m . bardeen , 22 ( 1980 ) 1882 . [ 7 ] ya.b . zeldovich , 5 ( 1970 ) 84 . [ 8 ] p.j.e . peebles and r.h . dicke , 154 ( 1968 ) 891 . [ 9 ] u. seljak and m. zaldarriaga , 469 ( 1996 ) 437 . [ 10 ] w. hu , u. seljak , m. white , and m. zaldarriaga , 57 ( 1998 ) 3290 . [ 11 ] p. schneider , j. ehlers , and e.e . falco , gravitational lenses , springer - verlag , berlin ( 1992 ) . [ 12 ] u. seljak , 463 ( 1996 ) 1 . [ 13 ] p.j.e . peebles and j.t . yu , 162 ( 1970 ) 815 . [ 14 ] r.k . sachs and a.m. wolfe , 147 ( 1967 ) 73 . [ 15 ] j.a . peacock , cosmological physics , cambridge univ . press ( 1999 ) . [ 16 ] s. bashinsky and e. bertschinger , astro - ph/0012153 . [ 17 ] j.m . bardeen , j.r . bond , n. kaiser , and a.s . szalay , 304 ( 1986 ) 15 . [ 18 ] e.r . harrison , 1 ( 1970 ) 2726 . [ 19 ] h. lin et al . , 471 ( 1996 ) 617 . [ 20 ] m. tegmark , m. zaldarriaga , and a.j.s . hamilton , astro - ph/0008167 . [ 21 ] e. bertschinger , astro - ph/9506070 and http://arcturus.mit.edu / cosmics/. [ 22 ] p.j.e . peebles , the large - scale structure of the universe , princeton univ . press ( 1980 ) . [ 23 ] e. bertschinger , 36 ( 1998 ) 599 . [ 24 ] t. abel , g.l . bryan , and m.l . norman , 540 ( 2000 ) 39 . [ 25 ] c .- p . ma and e. bertschinger , 434 ( 1994 ) l5 . [ 26 ] w.h . press and p. schechter , 187 ( 1974 ) 425 . [ 27 ] s. cole and c. lacey , 281 ( 1996 ) 716 .

the balitsky , fadin , kuraev and lipatov ( bfkl ) equation @xcite is meant to describe the energy dependence of certain high energy hard processes . for the bfkl equation to be applicable certain conditions must be met : ( i ) the process , or part of a process , to be described should have only a single hard transverse momentum scale . processes having more than one hard scale may have significant contribution from the dokshitzer , gribov , lipatov , altarelli and parisi ( dglap ) equation @xcite which governs scale dependence in qcd . certain observables ( e.g. @xcite ) are known to satisfy this criterion though they have not yet been well measured . ( ii ) the rapidity of the hard process , related to the centre of mass energy @xmath2 and the hard transverse momentum scale @xmath0 by @xmath3 , should be large . the bfkl equation includes all terms @xmath4 in the coupling but neglects factors of @xmath5 not accompanied by factors of @xmath6 . there is a vigorous effort @xcite to systematically include the first non - leading terms , an @xmath5 unaccompanied by a @xmath6 and this should improve the accuracy and reliability of the bfkl equation . the bfkl equation should describe high energy processes having a single hard scale in a limited range of rapidity . when @xmath6 becomes too large , unitarity corrections become important and slow the rate of growth with energy . at even higher values of energy one expects parton densities to become so large that the whole perturbative qcd picture breaks down leading to a new regime of weak coupling but high density non - perturbative qcd . the goal in small - x physics is to understand , experimentally and theoretically , the onset of bfkl behaviour , its modification due to unitarity corrections as one proceeds to higher energies and finally the nature of the transition to the high density ( saturation ) regime at even higher energies . from a theoretical point of view the simplest process to study is high energy heavy quarkonium - heavy quarkonium scattering . if the onium is sufficiently heavy the radius of the onium is small enough to set the single hard scale of the process . any understanding of bfkl dynamics and beyond gained in studying heavy onium collisions should be applicable to the more experimentally accessible single hard scale processes mentioned above . in the large @xmath7 limit of qcd a dipole picture of high energy heavy onium - heavy onium scattering has been developed which is equivalent to the usual bfkl description @xcite . in this dipole picture the light - cone wave function of a heavy onium is viewed as a collection of colour dipoles . the dipoles are made up of the quark part of a gluon ( or the heavy quark itself ) and the antiquark part of a different gluon ( or the heavy antiquark ) . high energy elastic scattering in the bfkl approximation proceeds by two - gluon exchange between a right - moving dipole in the right - moving onium and a left - moving dipole in the left - moving onium . perhaps the main advantage of the dipole picture is that the equation governing the number and distribution of dipoles in the light - cone wave function of the onium is a branching equation which allows monte carlo methods to be used . in a recent paper @xcite , one of us has studied high energy onium - onium scattering from energies where the bfkl approximation is valid into the higher energy regime where unitarity limits , for a fixed impact parameter of the scattering , are reached . in the centre of mass system unitarity corrections are simply the independent scattering of two or more dipoles of the right - moving onium with two or more dipoles of the left - moving onium . because of the large number of ( small ) dipoles in an onium wave function unitarity corrections become important when the dipoles are relatively dilute in the onium so that independent multiple scatterings are the natural leading corrections to the bfkl approximation . the purpose of the present paper is to study two related questions , the probability of producing an unusually large number of particles in an onium - onium collision and the nature of gluon ( dipole ) evolution when dipole densities become so dense that saturation effects due to dipole - dipole interactions become important . at a given energy dipole densities in an onium wave function are largest not in a typical configuration but rather in rare high multiplicity fluctuations . thus attempts to measure saturation will naturally focus on the high multiplicity tail of particle production . one can not directly study particle production in the dipole formalism . in the dipole picture the @xmath8 light - cone wave function is constructed . this is sufficient to describe an elastic scattering , but it is not sufficient to describe an inelastic event where one must follow the time evolution of the wave function to a large positive time when particles ( or jets ) are produced . however , using the abramovskii , gribov , kancheli ( agk ) cutting rules @xcite one can relate particle production and forward elastic scattering involving one or more dipole scatterings . while there has been no detailed proof of the agk rules in qcd it is likely that they are correct and we assume that this is the case in the present paper . we calculate @xmath9 the cross section for @xmath0 cut pomerons when @xmath0 takes values far above average . the multiplicity of produced particles , in the central unit of rapidity , should be proportional to @xmath0 . we begin by studying a toy model which obeys _ dipole _ evolution but where the absence of transverse dimensions makes analytic solutions to the multiplicity distribution possible . our main result for the toy model is given in eq . ( [ eq : toysigk ] ) where @xmath9 is seen to decrease exponentially in @xmath10 at large @xmath0 , a tail much longer than that given by kno scaling @xcite . we have not been able to solve the large @xmath0 behaviour for the @xmath0 cut pomeron contribution to a zero impact parameter collision in qcd , @xmath11 . however , monte carlo results obtained using oedipus ( onium evolution , dipole interaction and perturbative unitarisation software ) @xcite indicate a very similar @xmath0-dependence to that found in the toy model with an @xmath12 dependence at large @xmath0 . the same dependence , but with different values of @xmath13 and @xmath14 , also well describes onium - onium collisions integrated over impact parameter the fact that the high multiplicity tail extends so far encourages one to believe that high parton density systems can be produced at reasonable accelerator energies if the right signal is found to trigger on the interesting physics associated with saturation . in section 3 , we turn to a study of saturation @xcite , beginning with defining measures of saturation . saturation effects show up as a modification of the simple branching picture of dipoles in constructing the square of the onium wave function . thus , a non - independence of the dipoles in an onium wave function is a measure of saturation . however , it is not known how to estimate the interaction of dipoles in an onium wave function directly . one expects that the interaction of a slightly right - moving dipole with higher rapidity right - moving dipoles in a right - moving onium should be similar to that of a slightly left - moving dipole with the right - moving onium . this latter interaction is just the scattering amplitude for a simple dipole with an onium state and is straightforward to calculate . thus one measure of saturation of a light - cone onium wave function is the interaction probability with an opposite moving elementary dipole . this measure of saturation is however time consuming to evaluate . a more efficient measure is the _ overlap _ that a given dipole in the onium wave function has with all the other dipoles in the wave function . happily , it turns out that these two measures are strongly correlated so that the simpler overlap measure can be used . the toy model discussed earlier also furnishes a useful model of saturation . saturation is expected to slow the growth of the number of dipoles . by comparing the cross section calculated in frames where the two incoming particles share the incoming energy differently , a consistency condition , requiring that the cross section be frame independent , gives sufficient information to determine dipole branching as a function of rapidity . while at low energy the rate of parton evolution grows exponentially in rapidity , as in bfkl evolution , that rate reaches a constant ( saturates ) at high rapidity as shown in eq . ( [ eq : toyrsn ] ) . the toy model can be used to study the intriguing , and still not well understood , relationship between unitarity limits and saturation . in qcd the centre of mass onium - onium scattering , at a fixed impact parameter , reaches the unitarity limit at energies well below that at which saturation effects begin to appear . however , if onium - onium scattering is viewed in a frame where , say , the left - moving onium has a small rapidity , the scattering looks exactly like the measure we have earlier discussed for saturation since now the left - moving onium is simply a single dipole . thus the multiple scatterings in the centre of mass system must show up as saturation effects in a system where one of the onia carries almost all the energy . this can be most sharply seen in the toy model by considering a toy onium scattering on a toy nucleus consisting of @xmath15 onia . here it is found that saturation effects found earlier are exactly what is necessary to make _ onium - nucleus _ scattering the same in either the frame where the onium is at rest or in the frame where the nucleus is at rest . in qcd it is not possible to do such complete analytical calculations , however , we have found a phenomenological modification of dipole branching eq . ( [ eq : rlmdlstrn ] ) which makes onium - onium and onium - nucleus scattering reasonably frame independent . one particularly sharp way to see the necessity of saturation effects and their frame dependence is to compare the s - matrix , at zero impact parameter , for onium - onium scattering in the centre of mass frame with that obtained in a laboratory frame at very high rapidities . at rapidities large enough so that the s - matrix has become very small the evaluation of the s - matrix neglecting saturation is determined by the very low multiplicity fluctuations of the wave function . we find , on theoretical grounds and from numerical simulations , that @xmath16 . however , without saturation effects @xmath17 is a factor of 2 larger in the lab frame than in the centre of mass . saturation effects will slow the growth of dipoles in the laboratory frame so that the typical wave function configuration agrees with the centre of mass calculation coming from rare fluctuations . this is exactly what happens in the toy model as can be seen by comparing eqs . ( [ eq : toysn ] ) and ( [ eq : toysotyp ] ) . at large impact parameters , @xmath18 , we find that the property of the centre of mass frame , that saturation corrections set in much later than unitarity effects , starts to break down . this is brought about by several factors . at large @xmath18 , the amplitude has a significant contribution from asymmetric onium - onium configurations , for example where one onium has a only central dense cluster of small dipoles ( of order of the onium size ) : for there to be an interaction the second onium must have small dipoles at the relevant large impact parameter , where they are likely to be much more dilute . this kind of scattering looks similar to a lab frame collision , and hence the saturation and unitarity corrections will be similar . a second point is that one might expect a distribution of large dipoles which looks relatively dilute , to have only small saturation corrections . however , the sequence of evolution to produce this distribution will often have passed through a stage with a dense collection of small dipoles , whose evolution would be significantly altered by saturation effects , reducing the production of large dipoles in the first place . overall , in the centre of mass frame , at large impact parameters , these effects combine to give saturation corrections of the same order as the unitarity corrections . fortunately , the total scattering cross section is not too strongly affected by these large impact parameters , and it is still reasonable to neglect saturation effects in its calculation . it is most useful to start from the agk cutting rules @xcite as expressed in @xcite . the cross section , @xmath9 involving the exchange of exactly @xmath0 cut pomerons can be expressed as follows : @xmath19 where @xmath20 is the amplitude for the exchange of @xmath21 pomerons . in this notation , the total cross section @xmath22 is @xmath23 one knows from previous work @xcite that the terms of the sum in eq . ( [ eq : sigtotf ] ) diverge as a factorial , and therefore so will the terms in eq . ( [ eq : sigkf ] ) . the reason for this divergence is that rare configurations of the onia contribute increasingly large amounts to the amplitudes for higher pomeron exchange . the solution is to sum multiple pomeron amplitudes before averaging over onium configurations . applying this method here one obtains the differential cross section for obtaining @xmath0 cut pomerons . the @xmath21 pomeron amplitude for a fixed impact parameter @xmath24 between the two onia ( of sizes @xmath25 and @xmath26 , evolved to rapidities @xmath27 and @xmath28 ) is : @xmath29 where @xmath30 is the two gluon exchange interaction amplitude between a pair of configurations @xmath31 and @xmath32 which occur with probabilities @xmath33 and @xmath34 from onia situated at @xmath35 and @xmath36 . substituting this into eq . ( [ eq : sigkf ] ) gives @xmath37 the cross section for obtaining no cuts is the difference between the unitarised cross section and @xmath38 , which comes out as : @xmath39 in the work that follows , @xmath27 will be set to @xmath40 , so that we will be examining the number of cuts at a central rapidity . as was the case for the unitarised amplitude @xcite , the evaluation of eq . ( [ eq : sigkp ] ) is very difficult to do analytically , because it requires a detailed understanding of the distribution of dipole configurations for an onium . however some of the features of multiple cut cross sections can be extracted by examining a `` toy '' model which has no transverse dimensions @xcite . the main feature of the toy model is that the probability of obtaining a large number @xmath14 of dipoles has an exponential distribution , @xmath41 , where @xmath42 is the mean number of dipoles ( proportional to @xmath43 for an onium evolved to a rapidity @xmath27 ) . one can show that in the case with transverse dimensions , the probability of obtaining a large density of dipoles also has a distribution which is an exponential in the density @xcite . it is this similarity which leads to the correspondence between the toy model and the case with two transverse dimensions . in the toy model , the cross section for an interaction with @xmath0 cut pomerons is @xmath44 where the interaction amplitude between two dipoles is defined to be @xmath45 ( equivalent to @xmath30 above ) . there are two limits which can be taken . let @xmath46 be the amplitude for exchange of one pomeron . if @xmath47 , then unitarisation effects ( the exponential factor ) are unimportant , and the evaluation of the cuts cross sections is very similar to the evaluation of the multiple pomeron amplitudes @xcite : @xmath48 the opposite limit is when unitarisation effects are important , @xmath49 ( and also @xmath50 ) . approximating the sums by integrals , and changing variables to @xmath51 and @xmath52 gives @xmath53 performing the @xmath54 integration using the saddle point approximation , it is clear that the integral is dominated by a region where @xmath55 and @xmath14 are similar . performing the second integration gives @xmath56 the unusual feature of this result is the square root dependence on @xmath0 in the exponential . given that the number of cuts is related to the final state multiplicity , one might have expected the usual kno type @xcite exponential distribution . the origin of the @xmath1 behaviour is that , in the region of configuration space contributing to the @xmath0-cut amplitude , the factor @xmath57 is the most rapidly varying and fixes the dominant values of @xmath58 . the probabilities associated with the relevant configurations just fix the normalisation , with @xmath59 . one is most likely to find a similarity between the toy model results and the realistic situation , by examining the differential cut cross sections , which eliminates two of the transverse degrees of freedom . in @xcite correspondence between the toy model and monte carlo results for zero impact parameter has already been established for the multiple pomeron exchange amplitudes . the main difference was that where the quantity @xmath21 occurred in an expression relating to @xmath21-pomeron exchange , it had to be replaced with @xmath60 where @xmath14 was a constant which originated from the details of the dynamics in the transverse dimensions . one might expect a similar effect when looking at cut cross sections , and one can examine the differential cross section for small impact parameters for behaviour of the form @xmath61 \label{eq : sqrtfit}\ ] ] where @xmath13 , @xmath62 , @xmath14 and @xmath31 are constants to be determined . the first stage is to ensure that one really does have @xmath63 . by defining @xmath64 and plotting @xmath65 ( essentially the ratio of first and second derivatives ) one can eliminate @xmath13 and @xmath62 . figure [ fig : rbc_v_sqrt ] shows this quantity as a function of @xmath0 , determined using the oedipus monte carlo simulation @xcite . for @xmath0 sufficiently large , the linear behaviour is very clear , and the slope corresponds to @xmath66 . the value of the intercept with the @xmath0 axis yields @xmath14 , which has a value of @xmath67 . this is somewhat larger than the corresponding value found for multiple pomeron exchange ( @xmath68 , with @xmath69 ) . the difference can be explained by arguing that the multiple pomeron and multiple cut cross sections are sensitive to the dipole distributions ( and the manner in which they differ from an exponential ) in different ways . the other property to examine is the variation of @xmath13 with @xmath6 . one clearly can not expect @xmath13 to be exactly @xmath70 , however it might be reasonable for it to be proportional to @xmath70 . table [ tbl : avf ] gives @xmath13 for two values of @xmath6 , indicating that to within the accuracy of the monte carlo results , this proportionality does hold . .parameterisations of cut distributions at central impact parameters . [ cols="^,^,^,^,^",options="header " , ] the dependence of the distributions on @xmath14 is relatively weak which is why it is not determined more accurately . table [ tbl : avf ] therefore confirms that for the zero impact parameter cross section the distribution scales as suggested by the toy model . one can argue that events with @xmath0 cut pomerons will have a multiplicity proportional to @xmath0 . in an experimental situation , one might be able to determine the cross sections for various ranges of multiplicities , but this will correspond to cut cross sections integrated over impact parameter , for which there will not necessarily be any correspondence with the toy model . however figure [ fig : tc_v_sqrt ] shows that a fit of the form eq . ( [ eq : sqrtfit ] ) @xmath71\ ] ] works remarkably well . the value of @xmath14 turns out to be considerably smaller than for eq . ( [ eq : sqrtfit ] ) : this is associated with the fact that at larger values of @xmath18 the value of @xmath14 decreases . it is not clear why the @xmath72 behaviour also applies to the integrated cross section . the results do however suggest that the variation of the coefficient @xmath13 with the rapidity is not as simple as in the toy model ( or in the fixed impact parameter case ) . the numerical results for the variation with rapidity of the fraction of the cross section coming from different numbers of cuts is shown in figure [ fig : allcuts_frac ] . just as high numbers of pomeron exchange tend to be dominated by small impact parameters , one can expect events with large numbers of cuts also to be relatively central , because they are dominated by dense configurations , which arise most commonly at small impact parameters . this is illustrated in figure [ fig : cmltv_cts ] , which shows @xmath73 , the fraction of the @xmath0-cut cross section which comes from impact parameters larger than @xmath74 , for various numbers of cuts , as a function of @xmath74 . one sees for example that about half the one - cut cross section is coming from a region @xmath75 , whereas the first half of the 5-cut cross section comes from @xmath76 . there is of course a limit to how central the many cut cross section can become , since any cross section is bound to be spread out over a region at least the size of the onia . the no cut cross section is also quite central : like the many cut cross sections , it is particularly sensitive to dense configurations . note that all these cross sections have significant tails at large @xmath18 . this is to be contrasted with the 5-pomeron curve , where the tail dies off quickly : the difference arises because of the unitarity corrections that are associated with multi - cut cross sections but not multi - pomeron contributions , and which limit the relative contributions from very dense configurations . in the calculations performed to determine the unitarity corrections @xcite , it has been assumed that saturation of the wave function can be neglected . the justification for doing this is that in a situation where both wave functions are evolved to @xmath40 , unitarisation corrections will set in when @xmath77 , while saturation corrections for a particular dipole will be of the order @xmath78 . two problems can arise with this argument . the first relates to the question of whether @xmath79 is small enough . in the calculations presented here , @xmath80 , so when unitarisation is important one might expect @xmath81 . since the conditions for saturation can only be guessed to within a factor of order 1 , one is potentially close to the onset of saturation corrections as well . secondly , the argument is based on the principle that the mean dipole density will be a good gauge of the saturation corrections . however in the case of unitarisation corrections , it has been shown @xcite that the use of mean interactions can be very misleading . in this section we will perform a more sophisticated analysis of the importance of saturation corrections , relative to unitarity corrections . we will also present work on understanding wavefunction evolution once saturation becomes a large effect . it is first necessary to develop a measure of saturation . two methods will be used here . firstly one can probe a wave function with an ( unevolved ) onium . the two gluon exchange amplitude between the wave function and the onium indicates the local density dipoles of size similar to the onium and might be similar to the interaction that a dipole in the wave function would have with its neighbours . since the dipole - dipole interaction has spikes wherever two sources overlap , one should average the interaction over a region of roughly the onium size . the choice of the size of the region over which to average is one of the main uncertainties with this method . it also has the disadvantage of being relatively slow , because one needs to determine dipole - dipole interactions at many points to determine the saturation over the whole extent of the wave function and if one wants the saturation associated with dipoles of different sizes , one must use probes of different sizes . the second method is more arbitrary , but much faster to compute . each dipole has an `` overlap '' associated with it . for each pair of dipoles within the onium wave function which are separated by less than a certain distance ( which will be a function of the dipole sizes ) the quantity @xmath82 is added to the overlap for each of the two dipoles ( where the smaller dipole has size @xmath83 and the larger , size @xmath84 ) . this function is loosely based on the fact that the total two - gluon interaction amplitude between a pair of dipoles ( moving fast relative to each other ) has the form @xmath85 and that it is dominated by a region where the two dipoles are close in impact parameter . this method has the advantage that one obtains a value of the saturation for each dipole . comparing the two methods one finds that there is a good ( linear ) correlation between them , as can be seen in figure [ fig : op_histo ] which shows the frequency distribution for configurations with various values of @xmath86 ( the maximum interaction with the probe ) and @xmath87 ( the maximum overlap ) . one can carry out other comparisons , in particular , considering the location in impact parameter of the maximum saturation , and one again finds that the two methods give very similar results . in addition , if one examines other values of rapidity , one finds that the factor relating @xmath86 and @xmath87 stays approximately the same . the degree of saturation is obviously just related to the density of dipoles . therefore the distribution of @xmath86 should be related to the probability of obtaining large densities , which has been shown to be exponential @xcite , the slope of the exponential being proportional to the mean density of dipoles . instead of searching the whole relevant region of impact parameter for the location of @xmath86 , one finds the location of the dipole corresponding to @xmath87 and bases the search for @xmath86 around that point . the probe size is always the same as the onium size . one finds , as expected , that @xmath86 also shows an exponential behaviour , with a slope varying in proportion to the local density of dipoles . one also finds that the maximum saturation is usually close to the original onium which is also unsurprising since that is where the evolution has the longest rapidity range to produce a large number of dipoles . while these properties are of general interest , one of the main reason for studying saturation is to see whether it is likely to affect the onium - onium cross section . so one needs to know the typical values of @xmath86 and also the correlation between @xmath86 for a pair of onium configurations and the interaction cross section for that pair of configurations . the most appropriate value of rapidity to examine is @xmath88 , since this is the rapidity for which the mean 1-pomeron amplitude at @xmath89 , @xmath90 , is equal to 1 ( the unitarity bound ) . consider first the ` unweighted ' curve of figure [ fig : sig_nw ] : this corresponds to the probability distribution of @xmath86 in onium - onium collisions with a total rapidity @xmath88 . the rapidity for each onium is @xmath91 . @xmath86 is now defined to be the larger of the @xmath86 values for the two onium configurations . one sees that typical values of saturation are @xmath92 . the exponential probability distribution for large saturation is relatively clear . more relevant though are the distributions weighted with the onium - onium cross section for each configuration pair : the weighted distributions correspond to the fraction of the cross section coming from configuration pairs with a certain @xmath86 , and one sees that the typical saturation has increased to be in the range @xmath93 . there are two main reasons why configuration pairs with larger cross sections are associated with larger saturation . firstly if one of the onia has a high density of dipoles in some region , then the interaction of that region with a second onium will be enhanced . a second possibility , which we will consider in more detail later , is that configurations which lead to a large interaction ( e.g. those with large dipoles ) are most likely to be produced through branching sequences which pass through a high density stage . there remains the problem of assessing the effect that this degree of saturation would have . one can assume that saturation will start to become important when @xmath94 , but this is a very approximate condition , being uncertain to within at least a factor of two . this factor of two is critical : if the condition for saturation is @xmath95 then there will be no problem with saturation at @xmath88 , on the other hand if it is @xmath96 then saturation could significantly alter the results at this rapidity . if saturation is important then one wants to know how it will affect the evolution : for example the saturation might be large in one region of impact parameter , but the bulk of the cross section might come from different regions , not affected by saturation . the next sections address these and other issues . one approach to the problem is again from the toy model , using the condition that the scattering amplitude must be independent of the frame in which calculation is performed . equivalently , it should nt matter which of the two onia one chooses to evolve . starting with a state containing @xmath14 dipoles , the rate of branching into a state containing @xmath97 dipoles is defined to be @xmath98 , while the interaction between two states with @xmath14 and @xmath55 dipoles respectively is @xmath99 . if the evolution is partitioned such that for an increase @xmath100 in total rapidity , the onium with @xmath14 dipoles evolves by @xmath101 and the one with @xmath55 dipoles evolves by @xmath102 , then the evolution of the amplitude @xmath103 will be @xmath104 + ( 1-\xi ) r_m \left[f(n , m+1 ) - f(n , m ) \right ] \label{eq : toyinv1p}\ ] ] in the case of one pomeron exchange , one has @xmath105 and @xmath106 ( where @xmath107 sets the strength of the interaction ) , giving @xmath108 { \ensuremath{({\ensuremath{\alpha_{\mathcal{p}}}}-1)}}f { \ensuremath{\alpha_s}}^2 nm\ ] ] which is clearly independent of @xmath54 . if for @xmath99 one uses the form which includes unitarity corrections , @xmath109 $ ] , one obtains the following expression for the evolution of the interaction : @xmath110 e^{-{\ensuremath{\alpha_s}}^2 mn } \label{eq : toyinv}\ ] ] where the superscript @xmath111 indicates that saturation is being taken into account . the condition that the growth of the cross section be independent of @xmath54 , as well as one s knowledge of the branching rate for small @xmath14 ( where saturation should be unimportant ) leads to the following expression for the rate of branching : @xmath112 note that though the interaction with a second onium was used in the determination of this branching rate , the result is independent of the state of the second onium . to see what effect saturation will have on the final probability distribution of dipoles , one can use the following arguments . first take the case without saturation , in the continuum limit for @xmath14 . the probability distribution satisfies @xmath113 changing variables to @xmath114 , @xmath115 and @xmath116 , one has @xmath117 whose solution ( from our knowledge that @xmath118 $ ] ) is @xmath119 in the case with saturation one has @xmath120 by an appropriate change of variables , @xmath121 , @xmath115 and @xmath122 $ ] , one can obtain an equation identical to eq . ( [ eq : qnu ] ) @xmath123 because , in the limit of small @xmath14 and @xmath124 , @xmath125 and @xmath126 , the equation for @xmath127 will have the same initial conditions as that for @xmath128 and therefore their solutions will be the same . translating this back to the variables @xmath129 and @xmath14 , one has @xmath130 . \label{eq : pssln}\ ] ] for @xmath131 this approximates to the solution without saturation . the main feature of this solution though is that at large @xmath27 , instead of there being an exponential distribution in @xmath14 with mean @xmath132 , one now has a solution which has a maximum at @xmath133 , \label{eq : strnmaxn}\ ] ] and which dies off quickly beyond that . the solution eq . ( [ eq : pssln ] ) breaks down because the continuum limit used in its derivation is no longer valid ( @xmath129 varies too rapidly over a range @xmath134 ) . numerical solution shows though that there is still a rapid fall off beyond the maximum , having roughly a poisson distribution , since the branching rate @xmath135 becomes independent of @xmath14 . ] one effect of this modified probability distribution is to increase the value of the @xmath136-matrix at very large rapidities . if one calculates the @xmath136-matrix in a centre of mass frame ( rapidity @xmath6 divided equally between the two onia ) neglecting saturation one obtains @xcite : @xmath137 . \label{eq : toy_cm_ns}\ ] ] note that this is much larger than the value of the @xmath136-matrix that one obtains from typical configurations ( namely @xmath138 ) . this difference arises because the @xmath136-matrix is dominated by a small fraction of configurations which have few dipoles . calculating the @xmath136-matrix including saturation ( done in the lab frame , to reduce sensitivity to the use of the continuum limit , though the result is of course independent of frame ) , one obtains @xmath139 . \label{eq : toy_cm_st}\ ] ] when @xmath140 , eqs . ( [ eq : toy_cm_ns ] ) and ( [ eq : toy_cm_st ] ) give the same result , consistent with the idea that saturation should not be important in the centre of mass frame for that energy range . however for @xmath141 , eqs . ( [ eq : toy_cm_ns ] ) and ( [ eq : toy_cm_st ] ) do differ ( the saturated case has a logarithmic dependence on energy , absent from the calculation without saturation ) , because saturation then matters in the evolution of the onium wave functions , even in the centre of mass frame . while this approach works well for the toy model , in the real case , an analogous method will turn out to be too complex to be useful . nevertheless one can also make use of the effect of a change of frame on the average ( unitarised ) amplitude . the simplest pair of frames to examine would be centre of mass and lab frames for onium - onium scattering . however one finds that , except for logarithmic factors , the @xmath136-matrix behaves in a very similar way in the two . consider instead collisions between a toy onium and a toy `` nucleus '' consisting of @xmath15 onia . two frames will be used : a lab frame @xmath142 where the onium is at rest , and a lab frame @xmath143 where the nucleus is at rest . at high energies ( neglecting saturation ) , in frame @xmath143 , with the onium being evolved , the matrix is approximately @xmath144 except for logarithmic terms ( and the factor @xmath15 ) this has the same leading energy behaviour as the centre of mass onium - onium @xmath136-matrix , and is similarly dominated by those rare configurations of the onium which have few dipoles ( the common configurations have such a large number of dipoles that their contribution to the @xmath136-matrix can be neglected ) . in the @xmath142 frame , it is the nucleus which it is evolved . assuming for the sake of the argument that the constituent onia evolve independently ( consistent with our neglect of saturation ) , configurations with few dipoles will be much rarer , because they require that none of the @xmath15 constituent onia should have evolved much . accordingly , when one calculates the @xmath136-matrix , one finds @xmath145 which is much smaller than the result in the @xmath143 frame . one way of restoring the correspondence is by requiring that saturation alter the probability distribution of the number of dipoles so that the value of the @xmath136-matrix for a _ typical _ configuration of the nucleus is now @xmath146 equivalently the typical number of dipoles must be of the order of @xmath147/ ( f{\ensuremath{\alpha_s}}^2)$ ] . note that this is just the result of eq . ( [ eq : strnmaxn ] ) ( except for the @xmath148 term which is a result of the use of nuclei ) . the advantage of examining the effect of frame changes on the average @xmath136-matrix is that one can transfer the arguments fairly straightforwardly to the case of scattering with transverse dimensions . the real case , including transverse dimensions , is very much more complicated , to a large extent because the dipole picture breaks down at high dipole densities and one needs to take into account quadrupoles , hexapoles , and other more complicated colour structures @xcite . this section will deal with various approaches that one can nevertheless try in order to deduce more information about the effects of saturation in the real case , still using the requirement of frame independence . it is simple to write down the equation which must be satisfied for frame independence to hold . this will be the equivalent ( with transverse dimensions ) of the independence of eqs . ( [ eq : toyinv1p ] ) and ( [ eq : toyinv ] ) on @xmath54 . let @xmath31 be a particular gluon configuration and @xmath149 be the set of all colour configurations with an extra gluon added at transverse position @xmath25 , with @xmath150 being any particular one of these . in the dipole approximation ( large @xmath7 ) , a particular @xmath150 corresponds to the gluon having originated from a particular dipole . define also @xmath151 to be the interaction between configurations @xmath31 and @xmath32 ( moving in opposite directions ) when they have a relative separation in impact parameter of @xmath24 . then for the scattering to be frame independent ( or equivalently for it not to matter which onium is evolved ) the rate of evolution @xmath152 must satisfy @xmath153 in the case of of the one - pomeron approximation for the scattering amplitude @xmath103 , and the normal dipole evolution kernel for @xmath74 , if one can show that eq . ( [ eq : realinv ] ) holds for a single dipole in each onium , then because dipoles in each onium interact ( and evolve ) independently of one another eq . ( [ eq : realinv ] ) automatically holds for any collection of dipoles in each onium . the fact that eq . ( [ eq : realinv ] ) holds for the 1-pomeron case is a result of the conformal invariance of the dipole - dipole interaction and of the dipole evolution kernel . the details are given in the appendix . the effect of including multi - pomeron interactions is to connect the interaction of different dipoles within the same onium , and as a result dipoles no longer necessarily evolve independently of one another : this is the breakdown of the dipole picture of evolution which necessitates that one take into account quadrupoles and more complicated terms . unfortunately it has not so far been possible to solve eq . ( [ eq : realinv ] ) to obtain detailed information on the effects of saturation . one can nevertheless use frame independence to deduce some limited information . we will first examine the effect of saturation on central impact parameters . the argument will be along similar lines to the toy - model approach discussed at the end of the last section , except that now it will just be sufficient to consider simple onium - onium scattering in lab ( @xmath154 ) and centre of mass ( @xmath155 ) frames . it has been shown in @xcite that there is a close correspondence between toy model and real results for the probability distribution of the number of dipoles when this number is high compared to the mean . however the arguments used here depend on the probabilities of having unusually low dipole densities , where the toy model is a poor approximation to the real case . in a lab frame one finds that the @xmath136-matrix is approximately @xmath156 instead of @xmath157 in the toy model . qualitatively , the difference can be understood as coming from the extra dynamics associated with the diffusion in dipole sizes . in the toy model , the probability of having no evolution is just the exponentially suppressed probability of a single dipole not branching over a rapidity @xmath6 . in the real case the probability ( @xmath158 ) of a dipole of size @xmath159 not producing any new dipoles larger than a small size @xmath160 is @xmath161 the extra element which arises is in the choice of @xmath160 , since in the rapidity interval @xmath6 one must also require that none of the dipoles produced below size @xmath160 branch back up into ones of size @xmath159 . forcing the probability of producing an extra large dipole from the small dipoles to be small ( approximately @xmath162 ) , one obtains the condition @xmath163 which leads to the observed dependence of the @xmath136-matrix on @xmath164 ( since the @xmath136-matrix is dominated by the configurations with small interactions and therefore few dipoles ) . the behaviour in the centre of mass frame will be similar , except that the condition for a small interaction will be that both onia should be unevolved , so @xmath165 = \exp\left[-c\frac{(y - y_0')^2}{2 } \right]\ ] ] the value of @xmath166 is different from @xmath167 because it depends on the interaction between the two onia , the details of which are different in the lab and centre of mass frames . figure [ fig : smat_lgdrv ] shows results from the monte carlo simulation , plotting the derivative of @xmath168 . the clear straight line behaviour at the higher values of @xmath6 is the signal of the @xmath164 term in the exponent . fits to the slopes of the derivatives show exactly the predicted factor of two between the lab and centre of mass frames . for @xmath169 , the value of @xmath17 is @xmath170 . if one performs similar calculations for onium - nucleus collisions , one expects ( and sees ) very similar behaviour : in the @xmath143 frame , the slope is identical to the onium - onium lab frame results , and the @xmath142 slope is twice the @xmath143 slope ( where the nucleus consists of two onia ) . since , for onium - onium collisions , the asymptotic value of the @xmath136-matrix is much smaller in the lab frame than in the centre of mass frame , one concludes that saturation in the lab frame will increase the @xmath136-matrix , and so that the _ typical _ value of the @xmath136-matrix , including saturation will be @xmath171 very different from the typical value without saturation , @xmath172 $ ] . this behaviour of the @xmath136-matrix can related to the typical fields in an evolved onium or nucleus and this is done in @xcite . a second approach to saturation in the real case is intended to give some information on what happens at non - zero impact parameters . first one can compare the unitarised amplitude for the lab and centre of mass frames as a function of @xmath18 ( the amplitudes have been multiplied by @xmath173 to indicate which regions contribute most to the total cross section ) . this is shown , used for the unitarised interaction from a pair of configurations which give an interaction @xmath107 with one pomeron exchange , breaks down when the number of interacting dipoles in either of the onia is small and also when the fields in either onium become too strong . ] in figure [ fig : unit_nosat ] . the slight mismatch between the 1-pomeron curves at moderate @xmath18 is due to statistical error ( the amplitude at larger values of @xmath18 comes from very rare configurations ) . as one expects , due to the neglect of saturation , the unitarity corrections in the lab frame are significantly smaller than in the centre of mass frame . note in particular though that whereas at small impact parameters the difference between the two frames is only a small fraction of the unitarisation corrections , at large impact parameters , most of the unitarity corrections present in the centre of mass frame are absent in the lab frame . this suggests that in the lab frame , saturation effects will in fact be considerably larger than the unitarity corrections at large @xmath18 . one way of investigating this further is to try to include some simulation of the effects of saturation . the problem is that one does nt know how to include saturation effects . our approach will be to use a very simple model of saturation which retains the basic properties of normal dipole evolution and where saturation changes only the rate of dipole branching , not the spatial distribution . this is not expected to give accurate quantitative information on evolution with saturation , only to produce some understanding of the types of effects which are liable to be important if one does take saturation into account properly . in analogy with the toy model , we will use a rate of branching , @xmath174 , of a dipole @xmath175 of size @xmath159 into two , of sizes @xmath83 ( the smaller ) and @xmath84 ( the larger ) including saturation , of @xmath176 and where @xmath177 is the normal rate of branching ( i.e.without saturation ) . the overlap measure of saturation , @xmath178 , is used because it is much faster to calculate , and during the monte carlo evolution , it will be the calculation of the saturations for each dipole which will be the limiting factor . since the functional form of this saturation is taken from the toy model , one does not know what normalisation @xmath179 to use for the overlaps . the procedure used for choosing @xmath179 will be to find a value which gives a frame independent result for @xmath180 ( @xmath18 is the relative impact parameter between the colliding onia ) at a single value of rapidity ( which is always possible ) , and then to see whether this also works at other values of @xmath18 and rapidity . this has been done for @xmath181 , where we take @xmath182 . reexpressing this in terms of the a more natural quantity , the probe interaction @xmath183 , this corresponds to a modification of the branching rate by a factor @xmath184 , which seems reasonable . figure [ fig : unit_sat ] shows the scattering amplitudes for centre of mass and lab frames , including saturation for @xmath88 . the one - pomeron and unitarised centre of mass results are also plotted for reference . the procedure has worked insofar as the results in the two frames agree for small @xmath18 for the second rapidity . at larger @xmath18 there is still a small discrepancy between the two frames : the lab frame result is too low there is a little bit too much saturation at large @xmath18 this discrepancy is however small compared to the overall effect of saturation at those values of @xmath18 . the same is found at all other values of rapidity ( including @xmath181 ) . other implementations of saturation ( i.e. with different forms for eq . ( [ eq : rlmdlstrn ] ) ) , give very similar results , as long as the effect of saturation on branching rates is linear for small overlaps . as a second check on this form of saturation , one can examine onium - nucleus collisions . as well as the two lab frames discussed earlier we examine also a `` centre of mass '' frame ( or more accurately , a frame in which the mean number of central dipoles is the same for the onium and the nucleus ) . neglecting saturation one expects the @xmath142 frame to have the lowest unitarity corrections ( since with the nucleus containing all the rapidity , it will have the largest saturation corrections ) , followed by the @xmath143 and cm frames . figure [ fig : us_nclr ] shows this ordering of the amplitudes . once saturation is included ( using exactly the same procedure as for onium - onium collisions ) , the three frames give very similar amplitudes , though again there is a small discrepancy at larger @xmath18 . it would be interesting to examine how this implementation of saturation affects the typical @xmath136-matrix for small impact parameters , and whether this is in accord with the arguments presented earlier . however the asymptotic @xmath185 behaviour sets in only for relatively large rapidities ( one needs @xmath186 ) , and calculation of the saturation corrections at these rapidities is extremely time consuming ( because the calculation time is proportional to the square of the number of dipoles ) and therefore not feasible . considering the difference between the amplitude in the centre of mass frame with and without saturation , at small @xmath18 , saturation seems to have a fairly small effect on the centre of mass amplitude . this is as expected from the basic argument that each onium has half the rapidity , and should be well away from saturation when unitarity corrections start to set in . at large @xmath18 , however , the effect of saturation is bigger , and in fact of the same order as the unitarisation correction ( which relates to the observation for figure [ fig : unit_nosat ] that at larger @xmath18 there is a big difference between the centre of mass and lab frame results , indicating that saturation must have a significant effect at large @xmath18 ) . the dynamics involved in producing the effects saturation at large @xmath18 are complex : because the unitarisation corrections are small , the densities of dipoles must be small , and one might be misled into thinking that saturation corrections should unimportant . one reason why this argument fails is that there are a variety of ways in which large @xmath18 interaction can occur . a large dipole can be produced early in the evolution , which can then evolve to produce many other dipoles of similar size and the final density of dipoles will be indicative of the relevant saturation corrections ( these evolution paths tend to also have significant unitarity corrections ) . another evolution path involves only small dipoles for most of the rapidity and a branching to larger dipoles at a late stage . the evolution of the small dipoles may have been subjected to significant saturation ; but once the branching occurs to large dipoles , unitarity corrections and any further saturation corrections ( at the large scale ) are no longer affected by what is happening on the small scale . the history of saturation can therefore have an important effect on the evolution , even though the diluteness of the final configuration might make one think one could neglect saturation . this is illustrated in the upper part of figure [ fig : strn_pths ] : at two values of rapidity ( @xmath187 and @xmath188 ) , the maximum dipole size , @xmath189 is determined . the maximum overlap , @xmath87 is then calculated for all dipoles with size @xmath17 , where @xmath190 , in effect giving the saturation on the scale of the largest dipole . the figure shows the ratio @xmath191 , the increase in saturation ( on the largest scale ) , as a function of @xmath192 , the increase in the largest dipole size . the main point is that when there is a large increase in dipole size , the ratio of the saturations drops to below one , confirming that if one looks only at the final configuration , the history of the saturation can be hidden . a similar reduction of saturation with increasing rapidity can probably also take place for the small dipoles produced from large ones ( since the small ones will be more dilutely spread out ) . a second effect is shown in the lower part of figure [ fig : strn_pths ] : the production of a large dipole tends to be a rare event . by a simple probabilistic argument , the larger the number of dipoles on a small scale , the greater the likelihood that at least one of them will branch to give a large dipole , which means that evolution paths that produce a large dipole are more likely to include an intermediate high saturation stage . the lower plot of figure [ fig : strn_pths ] demonstrates this by showing the maximum saturation ( again on the largest scales ) at @xmath193 as a function of the ratio of the largest scales at @xmath188 and @xmath193 . as the ratio of the scales increases so does the maximum intermediate saturation . finally in the centre of mass frame , one has the possibility of situations where one of the onia is small and moderately dense , while the other has a dilute distribution of small dipoles ( the only ones which will interact with the small dense onium ) at large @xmath18 . the saturation corrections in the evolution of the small dense onium will then be of the same order as the unitarity corrections between the two at large @xmath18 onia , i.e. there is an asymmetry between the two onia ( and arguments for neglecting saturation when unitarity corrections set in , rely on the symmetry between the two onia ) . this is distinct from the @xmath89 case where moderate densities of dipoles in _ each _ onium led to strongly unitarised amplitude . the conclusions therefore , from the simple simulation of saturation effects , and from the other arguments presented above , is that at large impact parameters , it is not necessarily safe to neglect saturation corrections compared to unitarity corrections , and further that large dipoles , even when dilute , may have been produced via a high saturation stage . this might be of some concern when determining the corrections to the total cross section , which has a significant component coming from moderate to large @xmath18 . however , to the extent that it is possible to gain any quantitative information from our simple simulation of saturation , one finds that in the total cross section , saturation corrections ( up to @xmath194 ) are no more than a third of the unitarity corrections . we are grateful to b.r . webber for many useful discussions . one of the authors ( am ) wishes to thank peter landshoff and bryan webber for their kind hospitality and pparc for a visiting fellowship during his visit to cambridge , in june of 1995 , where this work began . this section demonstrates the invariance of the one - pomeron scattering amplitude on the choice of frame . we show that it is invariant for a single dipole in each onium . because each dipole evolves independently , and because its interaction with the other onium is independent of the other dipoles in the onium to which it belongs , this will imply that the onium - onium scattering amplitude is invariant under changes of frames at the one - pomeron level . let the first dipole have sources at @xmath195 and @xmath196 , and the second dipole have sources at @xmath197 and @xmath198 ( with the @xmath199 being complex variables , and @xmath200 their complex conjugates ) . the condition for invariance of the scattering amplitude on the division of a small amount of rapidity between them is , eq . ( [ eq : realinv ] ) , which can be written as has the property of mapping @xmath205 , @xmath206 , @xmath207 and @xmath208 , thus converting the lower integral in eq . ( [ eq : onedipinv ] ) to the upper integral ( and vice - versa ) . since the integrands are invariant under conformal transformations , this proves the equality , and therefore the invariance of the one - pomeron onium - onium scattering amplitude on the frame in which the calculation is performed , if the evolution is carried out using the dipole kernel and the interaction is by two gluon exchange .

in [ 18 ] tierney and vogel for any functor @xmath0 , where @xmath1 is a category with finite limits and with a projective class @xmath2 , and @xmath3 is an abelian category , have constructed derived functors and investigated relationships of their theory with other theories of derived functors . namely , they have shown that if @xmath1 is abelian and @xmath4 is additive , then their theory coincides with the classical relative theory of eilenberg - moore [ 7 ] , whereas if @xmath1 is abelian and @xmath4 is an arbitrary functor , then it gives a generalization of the theory of dold - puppe [ 6 ] . besides , they showed that their derived functors are naturally isomorphic to the cotriple derived functors of barr - beck [ 3 ] if there is a cotriple in @xmath1 that realizes the given projective class @xmath2 . the key point in the construction of the derived functors by tierney and vogel is that , using @xmath2-projective objects and simplicial kernels , for every @xmath5 from @xmath1 a @xmath2-projective resolution can be constructed , which is an @xmath5-augmented pseudosimplicial object in @xmath1 and which for a given @xmath5 is unique up to a presimplicial homotopy ( according to the comparison theorem for @xmath2-projective resolutions ) . a natural question arises about constructing a cubical analog of the theory of tierney and vogel . exactly this is the purpose of this paper . before turning to the content of the paper itself , let us say few words about cubical objects and techniques . as it is well known , simplicial methods are developed for long time and are successfully used in algebra and topology . with less success , but still also cubical techniques have been developed , which was initiated on one hand by the systematic use of singular cubes in the singular homology theory of topological spaces ( see , for example , [ 14 ] ) , and , on the other hand , by the papers of kan [ 12 , 13 ] which have related cubical sets to homotopy theory . further research ( see e. g. [ 1,2,4,5,8 - 11,16,17,19 ] ) has also shown that the cubical approach is interesting and important . cubical objects have a number of advantages compared to the simplicial ones . for example , a ( pre)cubical homotopy is given by a single morphism in each dimension , whereas a ( pre)simplicial homotopy requires `` many '' morphisms . on the other hand , significant disadvantages of the cubical technique with respect to the simplicial one are also apparent . for example , a cubical group , in fact even a cubical abelian group , can fail to satisfy the kan condition [ 16,17 ] . in the context of elimination of these and other disadvantages of the cubical theory , of extreme importance are cubical objects with connections introduced by brown and higgins [ 4 ] . these objects are cubical objects with extra degeneracies , called connections . to stress importance and naturality of cubical objects with connections it suffices to name e. g. the following three facts . the singular cubical complex of any topological space has naturally defined connections . next , tonks in [ 19 ] has shown that any cubical group with connections satisfies the kan condition . finally , brown and higgins [ 5 ] have recently proved that the category of cubical objects with connections in an abelian category is equivalent to the category of nonnegative chain complexes in the same category . in this paper we introduce pseudocubical objects with pseudoconnections , obtained from the brown - higgins structure of a cubical object with connections by suitably relaxing their identities . projective precubical resolutions which we are using to construct cubical derived functors possess pseudodegeneracies and pseudoconnections and this fact is essentially used in the construction of derived functors defined by us . now let us list the contents of the paper by sections . in section [ prelim ] we recall the notions of presimplicial , pseudosimplicial and simplicial objects , precubical , cubical objects and cubical objects with connections , their morphisms , and the respective augmented versions of these notions . we also recall the definitions of normalization functors in the simplicial setting by moore , and in the cubical setting by wiatek [ 16 ] . furthermore we recall the kan cubical sets and their homotopy groups . for any presimplicial object @xmath6 in an abelian category @xmath1 the normalized chain complex of @xmath6 , denoted by @xmath7 in this paper , is a chain subcomplex of the unnormalized chain complex of @xmath6 , denoted by @xmath8 . the well known moore theorem says that if @xmath6 is a simplicial object in @xmath1 , then this inclusion is a chain homotopy equivalence ( in fact this is valid for any pseudosimplicial object in @xmath1 ) . on the other hand , for any precubical object @xmath9 in an abelian category @xmath1 , one has the canonical inclusion @xmath10 of chain complexes @xmath11 and @xmath12 in @xmath1 which are cubical analogs of @xmath7 and @xmath8 ( see [ 16 , 17 ] ) respectively , and which functorially depend on @xmath9 . in section 2 we introduce pseudocubical objects with pseudoconnections and prove that @xmath13 is a chain homotopy equivalence for any pseudocubical object @xmath9 with pseudoconnections ( in particular for any cubical object @xmath9 with connections ) . thus we obtain a cubical analog of the moore theorem . then we establish some results which are not needed later on ; however they seem interesting by themselves . let us mention some of them . we indicate an alternative proof of the aforementioned brown - higgins equivalence . furthermore , we point out that this equivalence is realized by the above functor @xmath14 . next , we show that if @xmath15 is a kan cubical group , then @xmath16 , the @xmath17-th homotopy group of @xmath15 , coincides with @xmath18 , the @xmath17-th homology group of @xmath19 ( note that @xmath19 and @xmath18 are defined for any cubical group @xmath15 as well ) . using this and the cubical analog of the moore theorem together with the aforementioned result of tonks , we get that @xmath16 is naturally isomorphic to @xmath20 for any cubical abelian group @xmath15 with connections . let @xmath1 be a category with finite limits and a projective class @xmath2 , @xmath3 an abelian category , and @xmath0 an arbitrary ( covariant ) finctor . in section 3 we construct left cubical derived functors @xmath21 , @xmath22 , @xmath23 , as follows . first we show how to build for any object @xmath5 from @xmath1 an @xmath5-augmented @xmath2-projective precubical resolution , denoted @xmath24 , by means of cubical kernels and @xmath2-projective objects . then we prove the comparison theorem which in particular says that @xmath24 is unique up to precubical homotopy for a given @xmath5 . furthermore , it is shown that any @xmath2-projective precubical resolution @xmath24 is in fact an augmented pseudocubical object with pseudoconnections . we define @xmath25 , @xmath23 , to be the @xmath17-th homology of @xmath26 . besides the comparison theorem , the fact that @xmath27 has pseudodegeneracies is crucial in proving that @xmath25 are well - defined and functorially depend on @xmath5 . this contrasts with the construction of the derived functors by tierney and vogel which does not use existence of pseudodegeneracies in @xmath2-projective presimplicial resolutions ( pseudodegeneracies of @xmath2-projective presimplicial resolutions are essentially used when the theory of tierney and vogel is compared with other theories of derived functors ) . further , we define @xmath28 , @xmath23 , to be the @xmath17-th homology group of @xmath29 . now this construction essentially uses the fact that @xmath27 is a pseudocubical object with pseudoconnections , i. e. , this is crucial for proving that @xmath30 are well - defined and functorial in @xmath5 . the cubical analog of the moore theorem proved in the previous section shows that in fact there are isomorphisms @xmath31 , @xmath23 , which are natural in @xmath5 and @xmath4 . suppose @xmath1 is an abelian category with a projective class @xmath2 , @xmath3 an abelian category , and @xmath0 an additive ( covariant ) functor . then one constructs , with respect to @xmath2 , the left derived functors @xmath32 ( @xmath23 ) of @xmath4 in the sense of eilenberg - moore [ 7 ] . on the other hand , since any abelian category admits finite limits , we can build @xmath2-projective precubical resolutions , and therefore can construct the cubical left derived functors @xmath33 , @xmath23 . in section 4 , using once again the cubical analog of the moore theorem , we prove that if @xmath2 is closed [ 7 ] or , more generally , is closed with respect to retracts , then there are isomorphisms @xmath34 , @xmath35 , @xmath23 , which are natural in @xmath5 and @xmath4 . we begin with the following known definitions . a presimplicial object @xmath6 in a category @xmath1 is a family of objects @xmath36 together with face @xmath1-morphisms @xmath37 a morphism @xmath38 between presimplicial objects in a category @xmath1 is a family of @xmath1-morphisms @xmath39 which commute with the face operators . if @xmath9 and @xmath40 are ( pseudo)simplicial objects , then the @xmath1-morphisms @xmath41 must commute with the face and ( pseudo)degeneracy operators . a precubical object @xmath9 in a category @xmath1 is a family of objects @xmath42 together with @xmath1-morphisms @xmath43 satisfying @xmath44 the @xmath1-morphisms @xmath45 and @xmath46 are called face operators . a cubical object @xmath9 in a category @xmath1 is a family of objects @xmath47 together with @xmath1-morphisms @xmath48 as above and @xmath49 which satisfy @xmath50 where @xmath51 . the @xmath1-morphisms @xmath52 are called degeneracy operators . [ [ 4 ] ] a cubical object @xmath9 in a category @xmath1 is said to have connections if there are given @xmath1-morphisms @xmath53 satisfying @xmath54 the singular cubical complex @xmath55 of a topological space @xmath9 is a cubical object with connections in the category of sets . the connections @xmath56 are defined by @xmath57^n\to x\right)(t_1,t_2, ... ,t_{n+1})=f(t_1, ... ,t_{i-1},\max(t_i , t_{i+1}),t_{i+2}, ... ,t_{n+1}).\ ] ] a morphism @xmath38 between precubical objects in a category @xmath1 is a family of @xmath1-morphisms @xmath39 which commute with the face operators . if @xmath9 and @xmath40 are cubical objects , then the @xmath1-morphisms @xmath41 must commute with the face and degeneracy operators ; and if @xmath9 and @xmath40 are cubical objects with connections , then the @xmath1-morphisms @xmath41 must commute with the faces , degeneracies and connections . for any category @xmath1 , let us denote by @xmath58 the category of presimplicial objects in @xmath1 , by @xmath59 the category of precubical objects in @xmath1 , by @xmath60 the category of cubical objects in @xmath61 , and by @xmath62 the category of cubical objects with connections in @xmath1 . let @xmath1 be an abelian category , and @xmath63 the category of non - negatively graded chain complexes in @xmath1 . we essentially use the normalization functor @xmath64 of wiatek [ 16 , 17 ] which is constructed as follows . if @xmath65 and @xmath66 is a precubical morphism , then define @xmath67 of less importance for us is the functor @xmath68 which is defined for arbitrary abelian category @xmath1 by @xmath69 ( see [ 16 , 17 ] ) . let @xmath1 be again an abelian category . recall the definition of the moore normalization functor @xmath70 assume @xmath6 , @xmath71 and @xmath72 is a presimplicial morphism . then @xmath73 one also has the functor @xmath74 assigning to @xmath75 its unnormalized chain complex . more precisely , @xmath76 an augmented precubical ( resp . presimplicial ) object in a category @xmath1 is a precubical ( resp . presimplicial ) object @xmath9 in @xmath1 together with an object @xmath35 and an @xmath1-morphism @xmath77 satisfying @xmath78 ( resp . @xmath79 ) . such an object is denoted by @xmath80 . a morphism between @xmath80 and @xmath81 is a morphism @xmath82 between @xmath9 and @xmath40 together with an @xmath1-morphism @xmath83 satisfying @xmath84 . denote the category of augmented precubical ( resp . presimplicial ) objects in @xmath1 by @xmath85 ( resp . @xmath86 ) , by @xmath87 the category of augmented pseudosimplicial objects in @xmath1 , and by @xmath88 the category of augmented simplicial objects in @xmath1 . assume @xmath1 is abelian . then a nonnegative chain complex in @xmath1 augmented over @xmath35 is a nonnegative chain complex @xmath9 in @xmath1 together with an @xmath1-morphism @xmath77 satisfying @xmath89 . denote the category of augmented nonnegative chain complexes in @xmath1 by @xmath90 . for any abelian category @xmath1 , the functor @xmath91 , respectively the functors @xmath92 , @xmath93 extend in an obvious way to @xmath94 respectively @xmath95 let @xmath1 be an abelian category . using the shifting functor @xmath96 ( assigning to @xmath97 the nonnegative chain complex @xmath98 defined by @xmath99 , @xmath100 , @xmath101 , @xmath102 ) , we get functors @xmath103 and @xmath104 the restrictions of which to the category of augmented pseudosimplicial objects are used in the next section . next recall [ 12 ] the definition of the homotopy groups of a kan cubical set . a cubical set @xmath9 is said to be kan if for any @xmath105 , any @xmath106 and any collection of @xmath107 elements @xmath108 , @xmath109 , @xmath110 , @xmath111 , satisfying @xmath112 there exists @xmath113 such that @xmath114 let @xmath9 be a cubical set , @xmath115 and @xmath108 , @xmath102 , @xmath109 , @xmath110 . we write @xmath116 iff @xmath117 , @xmath109 , @xmath110 . suppose now that @xmath118 is a kan cubical set with a basepoint @xmath119 . let @xmath120 and define an equivalence relation @xmath121 on @xmath122 @xmath123 by @xmath124 or equivalently , by @xmath125 [ 12 , theorem 6 ] . denote @xmath126 , @xmath127 . let @xmath128,[y]\in\pi_n(x,\psi)$ ] @xmath129=cl_\sim(x))$ ] , @xmath102 . since @xmath9 is kan , there is @xmath130 with @xmath131 clearly , @xmath132 . define @xmath128\bl[y]=[\pa_1 ^ 1z]$ ] . this definition depends only on the equivalence classes of @xmath133 and @xmath134 and , with this multiplication , @xmath135 , @xmath102 , is a group , called the @xmath17-th homotopy group of @xmath118 . finally let us note that in the text we will freely make use of the freyd - mitchell embedding theorem ( see e. g. [ 20 , p.25 ] ) when applying various results in the literature about modules to objects in general abelian categories . let @xmath15 be a cubical group and define @xmath136 then : 1 . @xmath137 , @xmath127 . 2 . @xmath138 , @xmath139 . 3 . @xmath140 is a normal subgroup of @xmath141 and of @xmath142 , @xmath127 . let @xmath143 . by definition @xmath144 consequently , @xmath145 . hence ( a ) is proved . . then @xmath146 . using this we get @xmath147 . hence ( b ) is proved . suppose @xmath148 and @xmath149 . obviously one has @xmath150 and @xmath151 . this proves ( c ) . thus , for any cubical group @xmath15 , @xmath152 is a chain complex of ( not necessarily abelian ) groups . we call @xmath19 the moore chain complex of @xmath15 . it is obvious that @xmath19 and its homology groups @xmath153 functorially depend on @xmath15 . obviously , in the construction of @xmath19 one may replace @xmath15 by a precubical object @xmath9 in an abelian category ( in [ 16 ] @xmath19 is introduced for cubical objects in an abelian category ) . in this case we redefine the differential @xmath154 on @xmath11 by @xmath155 . then @xmath11 is a chain subcomplex of @xmath12 and one has a natural monomorphism @xmath156 . + 4 mm now let us introduce pseudocubical objects and pseudocubical objects with pseudoconnections . + 4 mm a pseudocubical object @xmath9 in a category @xmath1 is a family of objects @xmath42 together with face @xmath1-morphisms @xmath157 for @xmath51 . we say that a pseudocubical object @xmath9 in a category @xmath1 has pseudoconnections if there are given @xmath1-morphisms @xmath53 which satisfy @xmath158 morphisms of pseudocubical objects and pseudocubical objects with pseudoconnections are defined in an obvious way , just as for cubical objects and cubical objects with connections . for a category @xmath1 , let us denote by @xmath159 the category of pseudocubical objects with pseudoconnections in @xmath1 . let @xmath1 be an abelian category . define the functor @xmath160 as follows : @xmath161 on the other hand , we have the functors @xmath162 and @xmath163 from @xmath87 to @xmath164 . one can easily see that @xmath165 let @xmath15 be a simplicial object in an abelian category @xmath1 . then the normalized chain complex @xmath166 of @xmath15 is a chain subcomplex of the unnormalized chain complex @xmath167 of @xmath15 . the moore theorem says that this inclusion is a chain homotopy equivalence . the proof of this theorem [ 15 , p.94 ] does not use the simplicial identity : @xmath168 , @xmath169 . therefore one may assert that for any augmented pseudosimplicial object @xmath6 in an abelian category @xmath1 , the natural monomorphism @xmath170 is a chain homotopy equivalence . replacing now @xmath6 by @xmath171 , where @xmath9 is a pseudocubical object with pseudoconnections in @xmath1 , and using @xmath172 and @xmath173 , we get a cubical analog of the moore theorem : [ normtheo ] let @xmath9 be a pseudocubical object with pseudoconnections in an abelian category . then the natural monomorphism @xmath156 is a chain homotopy equivalence and , therefore , @xmath174 is an isomorphism for all @xmath23 . in fact @xmath156 has a natural homotopy inverse @xmath175 with @xmath176 ( see the proof of theorem 22.1 of [ 15 ] ) . note that the natural monomorphism @xmath177 need not be a chain homotopy equivalence for a cubical object @xmath9 with connections in an abelian category . as is known the prime examples of pseudosimplicial objects are the projective resolutions used by tierney and vogel to define their derived functors . in the next section we construct cubical analogs of tierney - vogel s projective resolutions ( in order to define our cubical derived functors ) and show that they have pseudodegeneracies and pseudoconnections . the material in the rest of this section is not needed in what follows ; however we believe it is interesting for its own sake . suppose that @xmath9 is a cubical object with connections in an abelian category @xmath1 . then @xmath171 is an augmented simplicial object . denote by @xmath178 the chain subcomplex of @xmath12 generated by the images of the degeneracies of @xmath171 . since the degeneracy morphisms of @xmath171 are defined by @xmath179 , @xmath102 , @xmath180 , one has @xmath181 besides , @xmath182 and @xmath183 . let @xmath9 be a cubical object with connections in an abelian category @xmath1 . then : 1 . @xmath184 and hence @xmath11 is isomorphic to @xmath185 . the canonical projection @xmath186 is a chain homotopy equivalence . as @xmath171 is an augmented simplicial object , one has @xmath187 ( where @xmath188 is the chain subcomplex of @xmath189 generated by the degenerate elements of @xmath171 ) , and the canonical projection @xmath190 is a chain homotopy equivalence ( see corollary 22.2 and 22.3 of [ 15 ] ) . but @xmath191 , @xmath192 and @xmath193 . hence ( a ) and ( b ) hold . suppose @xmath9 is a cubical set with connections . an n - cube @xmath194 is said to be folded if there exists @xmath195 such that @xmath196 for some @xmath197 ( see [ 1 ] ) . when we define the singular cubical homology @xmath198 of a topological space @xmath4 , it is necessary to factor out the degenerate singular cubes . it easily follows from theorem 2.5 that the folded singular cubes can be ignored in addition when we calculate @xmath198 . let @xmath1 be an abelian category . restricting @xmath14 and @xmath199 to the category @xmath62 , and @xmath200 to the category @xmath88 , we get the following commutative diagram @xmath201 ^ -{m}\ar[rd]_-{\ol{n}}&&ch_{\ge 0}(\sa)\\ & as(\sa)\ar[ru]_-{\wh{i}}}\ ] ] in which , by the dold - kan theorem , @xmath200 is an equivalence of categories . moreover , the functor @xmath202 is also an equivalence . the proof is very similar to the proof of the dold - kan theorem [ 20 , p.270 ] . thus we have got an alternative proof of the result by brown and higgins [ 5 ] about the equivalence of the categories @xmath62 and @xmath164 . furthermore , we have pointed out the functor @xmath203 which realizes this equivalence . in what follows , for pointed kan cubical monoids @xmath15 , we will always take the unit 1 of @xmath204 as basepoint , and denote @xmath205 by @xmath16 . [ moorekan ] let @xmath15 be a kan cubical monoid and suppose that @xmath105 . then : \(a ) for any @xmath206 , @xmath128\bullet[y]=[xy]$ ] . \(b ) @xmath16 is abelian . to prove this proposition , we need the following well known [ twoopers ] let @xmath207 and @xmath208 be binary operations on a set @xmath209 . assume that they have units and satisfy @xmath210 for all @xmath211 . then these operations coincide and @xmath212 is an abelian monoid . the proofs of propositions 17.2 and 17.3 of [ 15 ] ) . using the monoid structure on @xmath15 , a binary operation may be defined on @xmath16 by @xmath128*[y]=[xy]$ ] . this is well - defined . indeed , if @xmath206 then @xmath213 , and if @xmath214 and @xmath215 , @xmath216 , i. e. , @xmath217 then @xmath218 i.e. , @xmath219 . next , assume that @xmath128,[y],[u],[v]\in\pi_n(g)$ ] . by the kan condition , there are @xmath220 with @xmath221 whence @xmath222 then , by the definitions of @xmath223 and @xmath208 , we can write @xmath224\bl[y]\right)*\left([u]\bl[v]\right)=[\pa_1 ^ 1w_1]*[\pa_1 ^ 1w_2][\pa_1 ^ 1w_1\cdot\pa_1 ^ 1w_2]=\\ = [ xu]\bl[yv]=\left([x]*[u]\right)\bl\left([y]*[v]\right).\end{gathered}\ ] ] besides , @xmath128\bl[1]=[x]=[1]\bl[x]$ ] and @xmath128*[1]=[x]=[1]*[x]$ ] . consequently , we conclude , by lemma [ twoopers ] , that @xmath207 and @xmath208 coincide and @xmath16 is abelian . [ kanpimoore ] for any _ kan _ cubical group @xmath15 , @xmath225 let us use the following notations : @xmath226 it is evident that @xmath227 for all @xmath127 . therefore , in view of proposition [ moorekan ] , it suffices to show that @xmath228 if and only if @xmath229 , @xmath230 . suppose that @xmath206 and @xmath228 . then @xmath231 this gives @xmath128*[y^{-1}]=[xy^{-1}]=[1]$ ] , whence @xmath128=[y]$ ] , i. e. , @xmath228 . any cubical group with connections is kan [ 19 ] . combining this with theorem [ normtheo ] and proposition [ kanpimoore ] , we get for any cubical abelian group @xmath15 with connections , @xmath16 is naturally isomorphic to @xmath20 for all @xmath23 . let @xmath1 be a category and @xmath232 a sequence of @xmath1-morphisms , @xmath102 . a cubical kernel of the sequence @xmath233 is a sequence @xmath234 of @xmath1-morphisms such that * @xmath235 for @xmath236 , @xmath237 ; * * if @xmath238 is any other sequence satisfying identities @xmath239 for @xmath236 , @xmath237 , then there exists a unique @xmath1-morphism @xmath240 with @xmath241 , @xmath242 , @xmath243 . it immediately follows from the definition that cubical kernels are unique up to isomorphism if they exist . suppose @xmath1 has finite limits and let @xmath244 ^ -{f_1 ^ 0}\ar@<-0.5ex>[r]_-{f_1 ^ 1}&b}$ ] be a pair of @xmath1-morphisms . consider the diagram @xmath245_-{f_1 ^ 1}\ar[r]^-{f_1 ^ 0}&b\\ a\ar[u]^-{f_1 ^ 0}\ar[d]_-{f_1 ^ 1}&&a\ar[u]_-{f_1 ^ 0}\ar[d]^-{f_1 ^ 1}\\ b&a\ar[l]_-{f_1 ^ 1}\ar[r]^-{f_1 ^ 0}&b.}\ ] ] by assumption , we have the limit diagram @xmath245_-{f_1 ^ 1}\ar[r]^-{f_1 ^ 0}&b\\ a\ar[u]^-{f_1 ^ 0}\ar[d]_-{f_1 ^ 1}&k\ar[l]_-{k_2 ^ 1}\ar[u]^-{k_1 ^ 0 } \ar[r]^-{k_2 ^ 0}\ar[d]_-{k_1 ^ 1}&a\ar[u]_-{f_1 ^ 0}\ar[d]^-{f_1 ^ 1}\\ b&a\ar[l]_-{f_1 ^ 1}\ar[r]^-{f_1 ^ 0}&b,}\ ] ] i.e. the sequence @xmath246 is a cubical kernel of the pair @xmath247 ^ -{f_1 ^ 0}\ar@<-0.5ex>[r]_-{f_1 ^ 1}&b}.\ ] ] clearly , in fact , one has if @xmath1 admits finite limits , then cubical kernels exist in @xmath1 for any sequence @xmath248 and any @xmath102 . let @xmath2 be a class of objects of @xmath1 . recall [ 7 ] that an @xmath1-morphism @xmath83 is said to be @xmath2-epimorphic iff @xmath249 is surjective for all @xmath250 . also recall that @xmath2 is called a projective class if for each @xmath35 there exists a @xmath2-epimorphism @xmath251 with @xmath250 . let @xmath2 be a projective class in @xmath1 , @xmath252 ^ -\pa&a}$ ] an augmented precubical object over @xmath35 , and suppose that @xmath1 has finite limits . by proposition 3.2 , we have a factorization @xmath253 ^ -{\pa_1 ^ 0}\ar@<-.9ex>[rr]_-{\pa_{n+1}^1 } \ar[dr]_-{e_{n+1 } } & { { } ^{{}^{\vdots}}}&\ x_n\ \ar@<1.5ex>[rr]^-{\pa_1 ^ 0}\ar@<-.9ex>[rr]_-{\pa_n^1}\ar[dr]_-{e_n } & { { } ^{{}^{\vdots}}}&**[r]x_{n-1}\ \cdots&\ \ x_1\ar@<.5ex>[rr]^-{\pa_1 ^ 0}\ar@<-.5ex>[rr]_-{\pa_1 ^ 1}\ar[dr]_-{e_1 } & & x_0\ar[r]^-{\pa}&a \\ & k_{n+1}\ar@<1.2ex>[ur]^-{k_1 ^ 0}\ar@{}[ur]|-\ddots\ar@<-1.5ex>[ur]_-{k_{n+1}^1 } & & k_n\ar@<1.2ex>[ur]^-{k_1 ^ 0}\ar@{}[ur]|-\ddots\ar@<-1.5ex>[ur]_-{k_n^1 } & & & k_1\ar@<.5ex>[ur]^-{k_1 ^ 0}\ar@<-.5ex>[ur]_-{k_1 ^ 1}}\ ] ] where @xmath254 ^ -{k_1 ^ 0}\ar@<-0.5ex>[r]_-{k_1 ^ 1}&x_0}$ ] is a kernel pair of @xmath255 ^ -\pa&a}$ ] , and-1em @xmath256 a cubical kernel of @xmath257 for @xmath258 . we say that * @xmath252 ^ -\pa&a}$ ] is @xmath2-projective iff each @xmath259 ; * @xmath252 ^ -\pa&a}$ ] is @xmath2-exact iff @xmath154 and @xmath260 @xmath261 are @xmath2-epimorphic ; * @xmath252 ^ -\pa&a}$ ] is @xmath2-projective resolution of @xmath5 iff it is @xmath2-projective and @xmath2-exact . obviously , if @xmath1 is a category with finite limits and a projective class @xmath2 , then each @xmath35 has a @xmath2-projective resolution . moreover , the following comparison theorem shows that such a resolution is unique up to precubical homotopy equivalence . [ precm ] let @xmath252 ^ -\pa&a}$ ] be @xmath2-projective and @xmath262 ^ -{\pa'}&a'}$ ] be @xmath2-exact . then any @xmath1-morphism @xmath83 can be extended to a precubical morphism @xmath263 ^ -\pa\ar[dd]_-{\ol{f}}&&a\ar[dd]^-f\\ & & & & \\ x'\ar[rr]^-{\pa ' } & & a'}\ ] ] over @xmath264 @xmath265i . e. , @xmath266 and @xmath264 form a morphism of augmented precubical objects@xmath267 . furthermore , any two such extensions are precubically homotopic . that is , if @xmath268 are two extensions of @xmath264 , then there exist @xmath1-morphisms @xmath269 , @xmath127 , such that @xmath270 we construct the extension @xmath271 and show its uniqness up to precubical homotopy by induction on @xmath17 . since @xmath272 is @xmath2-projective and @xmath273 is @xmath2-epimorphic , there is @xmath274 with @xmath275 . next , one has @xmath276 . therefore @xmath277 and @xmath278 for a uniquely defined @xmath279 . as @xmath280 is @xmath2-projective and @xmath281 is @xmath2-epimorphic , there exists @xmath282 with @xmath283 , and we have @xmath284 and @xmath285 . thus @xmath286 and @xmath287 are constructed . inductively , suppose given @xmath1-morphisms @xmath288 for @xmath289 so that @xmath290 , @xmath291 , @xmath243 . then @xmath292 , @xmath236 , @xmath237 . hence there is a unique @xmath293 such that @xmath294 , @xmath242 , @xmath243 . since @xmath295 is @xmath2-projective and @xmath296 is @xmath2-epimorphic , there exists @xmath297 with @xmath298 . then we have @xmath299 , @xmath300 , @xmath243 . this completes the inductive step and proves the existence of @xmath266 . now suppose @xmath301 is another extension of @xmath83 . we want to construct @xmath302 with @xmath303 , @xmath304 , @xmath127 , and @xmath305 , @xmath306 , @xmath110 . for @xmath307 consider diagram @xmath308{\ x_0}\ar@{-->}[ddll]_-{h_0}\ar@{-->}[dddl]_-{q_0 } \ar[rr]^-\pa\ar@<1.2ex>[dd]^-{g_0}\ar@<.2ex>[dd]_-{f_0}&&a\ar[dd]^-f\\ \\ x'_1\ar[dr]_-{e_1}&&**[r]{\ x'_0}\ar[rr]^-{\pa'}&&**[r]a'\;\;.\\ & * * [ r]k'_1\ar@<-1ex>[ur]^-{k_1 ^ 0}\ar@<-2ex>[ur]_-{k_1 ^ 1 } & & } \ ] ] as @xmath309 , there exists @xmath310 such that @xmath311 and @xmath312 . next , since @xmath272 is @xmath2-projective and @xmath313 is @xmath2-epimorphic , there is @xmath314 with @xmath315 . this and the two previous equalities give @xmath316 and @xmath317 . thus @xmath318 is constructed . inductively , suppose given @xmath319 with the required properties . @xmath320x_n\ar@{-->}[ddll]_-{h_n}\ar@{-->}[dddl]_-{q_n } \ar@<1.5ex>[rr]^-{\pa_1 ^ 0}\ar@<-1ex>[rr]_-{\pa_n^1 } \ar@<1.4ex>[dd]^-{g_n}\ar@<-0.2ex>[dd]_-{f_n } & { { } ^{{}^{\vdots}}}&**[r]x_{n-1 } \ar@<-.2ex>[dd]_-{f_{n-1}}\ar@<1.4ex>[dd]^-{g_{n-1 } } \ar@<0.7ex>[ddll]_{h_{n-1}}\\ \\ * * [ r]x'_{n+1}\ar[dr]_-{e_{n+1}}&&**[r]x'_n \ar@<1.5ex>[rr]^-{\pa_1 ^ 0}\ar@<-1ex>[rr]_-{\pa_n^1 } & { { } ^{{}^{\vdots}}}&**[r]x'_{n-1}\ .\\ & * * [ r]k'_{n+1 } \ar@<1.1ex>[ur]^-{k_1 ^ 0}\ar@{}[ur]|-\ddots\ar@<-1.6ex>[ur]_-{k_{n+1}^1 } } \ ] ] define @xmath321 , @xmath242 , @xmath110 , as follows : @xmath322 by the induction assumption , @xmath323 that is @xmath324 for @xmath236 , @xmath325 . therefore @xmath326 , @xmath242 , @xmath110 , for a uniquely defined @xmath327 . since @xmath328 is @xmath2-projective and @xmath329 is @xmath2-epimorphic , there is @xmath269 with @xmath330 . now we have @xmath331 this finishes the inductive step and completes the proof of the theorem . the following theorem is crucial for constructing our cubical derived functors . [ resdeg ] suppose that @xmath332 ^ -{\pa_1 ^ 0}\ar@<-.9ex>[rr]_-{\pa_{n+1}^1 } \ar[dr]_-{e_{n+1 } } & { { } ^{{}^{\vdots}}}&\ p_n\ \ar@<1.5ex>[rr]^-{\pa_1 ^ 0}\ar@<-.9ex>[rr]_-{\pa_n^1}\ar[dr]_-{e_n } & { { } ^{{}^{\vdots}}}&**[r]p_{n-1}\ \cdots&\ \ p_1\ar@<.5ex>[rr]^-{\pa_1 ^ 0}\ar@<-.5ex>[rr]_-{\pa_1 ^ 1}\ar[dr]_-{e_1 } & & p_0\ar[r]^-{\pa}&a \\ & k_{n+1}\ar@<1.2ex>[ur]^-{k_1 ^ 0}\ar@{}[ur]|-\ddots\ar@<-1.5ex>[ur]_-{k_{n+1}^1 } & & k_n\ar@<1.2ex>[ur]^-{k_1 ^ 0}\ar@{}[ur]|-\ddots\ar@<-1.5ex>[ur]_-{k_n^1 } & & & k_1\ar@<.5ex>[ur]^-{k_1 ^ 0}\ar@<-.5ex>[ur]_-{k_1 ^ 1}}\ ] ] is a @xmath2-projective resolution of @xmath35 . then : \(a ) @xmath27 has pseudodegeneracy operators , i.e. , there exist @xmath333 , @xmath127 , @xmath242 , satisfying @xmath334 as an immediate consequence we have [ psec ] any @xmath2-projective resolution @xmath335 ^ -{\pa}&a}$ ] is an augmented pseudocubical object with pseudoconnections . \(a ) as @xmath336 ^ -{k_1 ^ 0}\ar@<-0.5ex>[r]_-{k_1 ^ 1}&p_0}$ ] is a kernel pair of @xmath337 , there is @xmath338 such that @xmath339 and @xmath340 . on the other hand , since @xmath341 is @xmath2-projective and @xmath313 is @xmath2-epimorphic , @xmath342 for some @xmath343 and we have @xmath344 and @xmath345 . thus @xmath343 is constructed . inductively , suppose given @xmath343 , @xmath346 with the required properties . fix @xmath347 , @xmath348 , and define @xmath349 , @xmath242 , @xmath110 , by @xmath350 using the induction assumption , one checks that @xmath351 for @xmath352 , @xmath353 . then there exists @xmath354 such that @xmath355 , @xmath242 , @xmath110 . since @xmath356 is @xmath2-projective and @xmath329 is @xmath2-epimorphic , there is @xmath357 with @xmath358 . according to this , for @xmath359 and @xmath110 , one has @xmath360 thus we have constructed @xmath361 with the desired properties . \(b ) define @xmath362 by @xmath363 , @xmath364 , @xmath365 and @xmath366 . one verifies that @xmath367 , @xmath110 . hence there exists @xmath368 such that @xmath369 , @xmath370 , @xmath110 . as @xmath371 is @xmath2-projective and @xmath372 is @xmath2-epimorphic , @xmath373 for some @xmath374 , and we have @xmath375 next , assume that @xmath374 , @xmath376 with the required properties are constructed , and for any fixed @xmath347 , @xmath109 , define @xmath377 , @xmath242 , @xmath110 , as follows : @xmath378 by the induction assumption , @xmath379 , @xmath352 , @xmath353 . consequently , there is @xmath380 such that , @xmath381 , @xmath242 , @xmath110 . since @xmath356 is @xmath2-projective and @xmath329 is @xmath2-epimorphic , there exists @xmath382 with @xmath383 then , for @xmath359 and @xmath110 , one has @xmath384 thus we have constructed @xmath385 satisfying the desired properties . let @xmath1 be an abelian category . it is shown in [ 16 ] that the functor @xmath386 sends cubically homotopic morphisms to chain homotopic morphisms . the proof , which we repeat here because of the completeness , shows that in fact one has [ prechn ] let @xmath9 be a pseudocubical object and @xmath387 a precubical object in an abelian category @xmath1 , and @xmath388 precubical morphisms . if @xmath264 and @xmath389 are precubically homotopic @xmath265see theorem @xmath390 , then @xmath391 and @xmath392 are chain homotopic . let @xmath393 be a precubical homotopy from @xmath264 to @xmath389 , and let @xmath394 denote the canonical monomorphism from @xmath395 to @xmath328 , @xmath127 . one checks that @xmath396 , @xmath348 , @xmath127 . consequently , we have morphisms @xmath397 clearly , @xmath398 further , for all @xmath102 , we have @xmath399 hence @xmath400 is a chain homotopy from @xmath391 to @xmath401 . this proposition together with theorem 2.4 implies [ prechm ] let @xmath9 and @xmath387 be pseudocubical objects with pseudoconnections in an abelian category @xmath1 , and @xmath388 precubical morphisms . if @xmath264 and @xmath389 are precubically homotopic , then @xmath402 and @xmath403 are chain homotopic . we are now ready to introduce cubical derived functors . let @xmath1 be a category with finite limits and a projective class @xmath2 , @xmath404 an abelian category , and @xmath405 an arbitrary ( covariant ) functor . we construct the left cubical derived functors @xmath406 , @xmath127 , of @xmath4 as follows . if @xmath35 , choose ( once and for all ) a @xmath2-projective resolution @xmath407 and define @xmath408 theorem 3.3 and propositions 3.4(a ) and 3.6 show that the objects @xmath25 are independent ( up to natural isomorphism ) of the resolution chosen ( if @xmath409 is a second @xmath2-projective resolution and @xmath410 a precubical morphism extending @xmath411 , then @xmath412 are isomorphisms ) . moreover , by the same statements , it is immediately clear that @xmath21 are defined on morphisms and are functors from @xmath1 to @xmath404 . besides , functoriality in the variable @xmath4 is obvious . similarly , in view of theorem 3.3 , corollary 3.5 and proposition 3.7 , one can define functors @xmath413 by @xmath414 it follows from theorem 2.4 that in fact there are isomorphisms @xmath415 which are natural in @xmath5 and in @xmath4 . as one sees the construction of @xmath21 essentially uses [ resdeg](a ) , and similarly @xmath416 essentially uses [ psec ] . this contrasts with the fact that the construction of the derived functors by tierney and vogel does not use existence of pseudodegeneracies in @xmath2-projective presimplicial resolutions . on the other hand , as shown in [ 16 ] , the functor @xmath417 ( see section 1 ) sends precubically homotopic morphisms of precubical objects to chain homotopic morphisms ( cf . propositions [ prechn ] and [ prechm ] ) . this together with theorem [ precm ] allows us to conclude that one does not need proposition [ resdeg](a ) to prove that the functors @xmath418 where @xmath27 is a @xmath2-projective precubical resolution augmented over @xmath35 , are correctly defined . but in this way one obtains `` bad '' derived functors by the following reason . one can easily see that @xmath419 for any @xmath250 and any @xmath23 . thus in general the higher ( @xmath139 ) derived functors @xmath420 do not vanish on @xmath2-projectives , i. e. , the crucial property of derived functors is not satisfied . in particular , there is no chance for @xmath420 to coincide with the classical derived functors for additive functors @xmath4 between abelian categories . at the same time , the derived functors @xmath21 and @xmath416 certainly vanish on @xmath2-projectives for @xmath139 , and in particular coincide with the classical derived functors for additive @xmath4 , as we will show below in theorem [ coinc ] . suppose that @xmath1 is an abelian category and @xmath405 an additive functor . our aim is to compare @xmath421 @xmath123 with the derived functors of @xmath4 in the sense of eilenberg - moore [ 7 ] . let @xmath1 be an abelian category with a projective class @xmath2 and suppose that @xmath252 ^ -{\pa}&a}$ ] is a precubical object in @xmath1 augmented over @xmath35 . if @xmath252 ^ -{\pa}&a}$ ] is @xmath2-exact , then its augmented moore chain complex @xmath422 is @xmath2-exact in the sense of eilenberg - moore @xmath423 $ ] . that is , the sequence of abelian groups @xmath424 where @xmath425 and @xmath426 @xmath261 , is exact for any @xmath250 . evidently , @xmath427 , @xmath127 . besides , since @xmath77 is @xmath2-epimorphic , @xmath428 is surjective . let @xmath429 be an @xmath1-morphism with @xmath430 and consider the diagram @xmath431_-{g'}\ar@{-->}@`{(-60,-20)}[dddl]_-\varphi\ar@<-1ex>[dd]_-{g}\ar@<1ex>[dd]^-0\\ \\ x_1\ar@<0.8ex>[rr]^-{\pa_1 ^ 0}\ar@<-0.9ex>[rr]_-{\pa_1 ^ 1 } \ar[dr]_-{e_1}&&x_0\ar[r]^-\pa&a,\\ & k_1\ar@<1ex>[ur]^-{k_1 ^ 0}\ar@<-1ex>[ur]_-{k_1 ^ 1 } } \ ] ] where @xmath336 ^ -{k_1 ^ 0}\ar@<-0.5ex>[r]_-{k_1 ^ 1}&x_0}$ ] is a kernel pair of @xmath77 , and @xmath313 is @xmath2-epimorphic . as @xmath432 , there exists a unique @xmath433 such that @xmath434 and @xmath435 . but @xmath436 for some @xmath437 , and we have @xmath438 and @xmath439 . it then follows from the construction of @xmath11 that @xmath440 , where @xmath441 denotes the inclusion @xmath442 and @xmath443 is a uniquely defined @xmath1-morphism from @xmath444 to @xmath445 . clearly , @xmath446 . thus the sequence is exact at @xmath447 . now assume that @xmath139 and @xmath448 is an @xmath1-morphism with @xmath449 . denote the inclusion @xmath450 by @xmath451 and consider the diagram @xmath431_-{f'}\ar@{-->}@`{(-100,-20)}[dddl]_-\varphi\ar@<-1.4ex>[dd]_-{\varphi_1 ^ 0 } \ar@<1ex>[dd]^-{\varphi_{n+1}^1}&&\\ & & { { } ^{{}^{\dots}}}\\ x_{n+1}\ar@<1.6ex>[rr]^-{\pa_1 ^ 0}\ar@<-.9ex>[rr]_-{\pa_{n+1}^1 } \ar[dr]_-{e_{n+1}}&{{}^{{}^{\vdots } } } & x_n\ar@<1.6ex>[rr]^-{\pa_1 ^ 0}\ar@<-.9ex>[rr]_-{\pa_n^1}&{{}^{{}^{\vdots } } } & x_{n-1}\\ & k_{n+1}\ar@<1.1ex>[ur]^-{k_1 ^ 0}\ar@<-1.9ex>[ur]_-{k_{n+1}^1}\ar@{}[ur]|-\ddots } \ ] ] where @xmath452 is a cubical kernel of @xmath453 , @xmath329 is @xmath2-epimorphic , and @xmath454 one checks that @xmath455 consequently , there exists a unique @xmath456 such that @xmath457 , @xmath242 , @xmath110 . since @xmath329 is @xmath2-epimorphic and @xmath250 , one has @xmath458 for some @xmath459 . but then @xmath460 from this we conclude , by the construction of @xmath11 , that @xmath461 for a uniquely defined @xmath1-morphism @xmath462 . clearly , @xmath463 , whence @xmath464 . let @xmath1 be an abelian category with a projective class @xmath2 and suppose that @xmath252 ^ -\pa&a}$ ] is a pseudocubical object with pseudoconnections in @xmath1 augmented over @xmath35 . if @xmath465 is @xmath2-exact , then the augmented chain complex @xmath466 ^ -\pa&a}$ ] is @xmath2-exact in the sense of eilenberg - moore . for any pseudocubical object @xmath9 in an abelian category @xmath1 , one has @xmath1-morphisms @xmath467 defined by @xmath468 it is immediate that @xmath469 , @xmath139 , @xmath109 . therefore , by the construction of @xmath12 , each @xmath470 factors as @xmath471 ^ -{\sg_n^x}\ar[dr]_-{\tau_n^x}&&x_n\\ & n_n(x)\ar@{^{(}->}[ur]_-{\nu_n^x}}\ ] ] where @xmath472 is the inclusion . one can easily see that @xmath473 , i.e. , @xmath474 is a retraction for each @xmath17 ( see [ 16 , 17 ] ) . this together with corollary 3.5 and proposition 4.2 gives [ retr ] let @xmath1 be an abelian category with a projective class @xmath2 which is closed with respect to retracts . if @xmath335 ^ -\pa&a}$ ] is a @xmath2-projective precubical resolution of @xmath35 , then the augmented chain complex @xmath475 ^ -\pa&a}$ ] is a @xmath2-projective resolution of @xmath5 in the sense of eilenberg - moore @xmath423 $ ] . now let @xmath1 be an abelian category with a projective class @xmath2 , @xmath404 an abelian category , and @xmath405 an additive ( covariant ) functor . then one constructs , with respect to @xmath2 , the classical left derived functors @xmath476 , @xmath127 . on the other hand , since any abelian category admits finite limits , we can build @xmath2-projective precubical resolutions , and therefore can construct the cubical left derived functors @xmath406 , @xmath127 . let @xmath478^\pa&a}$ ] be a @xmath2-projective precubical resolution of @xmath35 . by proposition 3.4(a ) , @xmath27 has pseudodegeneracy operators . then , as noted above , we have @xmath1-morphisms @xmath479 , @xmath480 with @xmath481 and @xmath482 . the latter gives @xmath483 . hence @xmath484 is an epimorphism and @xmath485 is a monomorphism . since @xmath4 is additive , @xmath486 . consider the commutative diagram @xmath487 ^ -{\nu_n^{t(p)}}\\ t(p_n)\ar[ur]^-{\tau_n^{t(p)}}\ar[dr]_-{t(\tau_n^p)}\ar[rr]^-{t(\sg_n^p)\sg_n^{t(p)}}&&t(p_n)\;\;\;\\ & t(n_n(p))\;\;.\ar[ur]_-{t(\nu_n^p ) } } \ ] ] as @xmath488 and @xmath489 are epimorphisms , @xmath490 . but @xmath491 is an inclusion and @xmath492 is a monomorphism . hence @xmath493 and @xmath494 is an isomorphism for each @xmath17 . one can easily see that these isomorphisms commute with the differentials and induce natural isomorphisms on the homologies . in view of proposition [ retr ] , @xmath475 ^ -\pa&a}$ ] is a @xmath2-projective resolution of @xmath5 in the sense of eilenberg - moore . consequently , @xmath495 . on the other hand , by the definition of @xmath21 , thus we have isomorphisms @xmath497 which are natural in @xmath5 and in @xmath4 . there remains an open question here . we do not know whether the cubical derived functors introduced by us coincide with the tierney - vogel derived functors in full generality , or at least , with the dold - puppe derived functors in the particular case of an arbitrary ( not necessarily additive ) functor on an abelian category .

roughly half of the known short - period black - hole binaries have low - mass companions ( lee , brown & wijers 2002 ) . however , it has been realized for more than a decade now that such systems are difficult to form ( see the discussion in podsiadlowski , rappaport & han 2003 [ prh ] and further references therein ) . the problem is that , in order to produce a short - period system ( i.e. , @xmath4 hrs ) with a low - mass donor star , the progenitor system has to pass through a common - envelope ( ce ) phase where the binary s period is reduced from a typical period of several years to less than a few days . in the standard model for ce evolution ( paczyski 1976 ) , this requires that the orbital energy released in the spiral - in process is sufficient to eject the massive envelope of the primary . for a low - mass companion ( i.e. , @xmath5 ) , this is energetically difficult , if not impossible , in particular considering the large binding energies of the massive envelope of the primaries ( dewi & tauris 2001 ; prh ) . this problem has led to the suggestion of a number of more exotic formation scenarios for low - mass black - hole binaries , e.g. , involving a triple scenario ( eggleton & verbunt 1986 ) or the formation of the companion from a disrupted envelope ( podsiadlowski , cannon & rees 1995 ; prh ) . alternatively , the low - mass black - hole binaries could descend from intermediate - mass systems ( justham , rappaport & podsiadlowski 2006 ) , as is the case for the majority of low - mass neutron - star binaries ( pfahl , rappaport & podsiadlowski 2003 ) . in this paper , we present a new mechanism for the ejection of the common envelope : `` explosive common - envelope ejection '' , involving nuclear rather than orbital energy , which can be highly efficient in ejecting a massive envelope even if the companion is a relatively low - mass star . a particularly interesting black - hole binary is the well studied system gro j165540 ( nova scorpii 1994 ) . it has an orbital period of 2.6d and a black hole with a mass @xmath6 ( beer & podsiadlowski 2002 ) . israelian et al . ( 1999 ) claimed that the secondary in this system has been highly enriched with the products of explosive nucleosynthesis ( e.g. , mg , si , s , ti ) produced in the supernova explosion that produced the black hole . based on the actual abundance ratios , podsiadlowski et al . ( 2002 ) found some tentative evidence that these are better explained by an energetic supernova explosion , a hypernova , with a typical ejecta energy of @xmath7ergs ( i.e. , 10 times the energy in a ` typical ' supernova ) . this may suggest that the formation of the black hole in this system could have been accompanied by a long - duration gamma - ray burst ( lgrb ) . however , in this case it is puzzling why the black - hole progenitor would have been rapidly rotating as required in the collapsar model for lgrbs ( woosley 1993 ; macfadyen & woosley 1999 ) since the core of the primary should have been spun down rather than have been spun up during its evolution ( heger , woosley & spruit 2005 ) . as we will show later in this paper , in the case of explosive ce ejection , this problem does not arise , possibly linking the formation of compact black - hole binaries having low - mass secondaries to lgrbs . this process of explosive ce was discovered in a systematic study of the slow merger of massive stars ( ivanova 2002 ; ivanova & podsiadlowski 2003 ; ivanova & podsiadlowski 2010 ) , where it was found that , in some cases in the late stage of the spiral - in process , hydrogen - rich material could be mixed into the helium - burning shell leading to a thermonuclear runaway which released enough energy to eject both the hydrogen and the helium envelope ( see figure 1 for a schematic representation ) , possibly explaining why to date all supernovae associated with lgrbs appear to be type ic supernovae ( i.e. , supernovae without hydrogen and helium in their spectrum ; cf . podsiadlowski et al . 2004 ) . in this paper , we will first discuss the numerical method in section 2 , and the physics of the process of explosive ce ejection in section 3 . in section 4 we apply it to the formation of low - mass black - hole binaries and lgrbs , and end with a broader discussion in section 5 . to model the merging of the two stars , we used a modified one - dimensional henyey - type stellar evolution code ( based on kippenhahn , weigert & hofmeister 1967 ) , which has recently been updated ( podsiadlowski , rappaport & pfahl 2002 ) . it uses opal opacities ( rogers & iglesias 1992 ) , supplemented by molecular opacities from alexander & ferguson ( 1994 ) at low temperatures and a mixing - lenth parameter @xmath8 . following the calibration by schrder , pols & eggleton ( 1997 ) and pols et al . ( 1997 ) , we generally assume 0.25 pressure scale heights of convective overshooting , unless stated otherwise . in the calculation of the spiral - in and merger phase we followed the angular - momentum transport in the envelope of the primary , using prescriptions similar to those given by heger & langer ( 1998 ) , where the source of the angular momentum is the orbital decay of the immersed binary . this is important for determining the averaged distribution of the heating sources inside the common envelope . the envelope heating is provided by the dissipation of the kinetic energy of the differentially rotating shells due to viscous friction . note that this energy is not necessarily released in the same place where the angular momentum has been deposited initially . for modelling the latter , we included a prescription that simulates the frictional energy input during the spiral - in phase of the companion star similar to the prescription given by meyer & meyer - hofmeister ( 1979 ) ( also see podsiadlowski 2001 ) . once the immersed secondary fills its own critical potential surface ( i.e. , its equivalent `` roche lobe '' defined by the effective potential of the core - secondary pair ) within the common envelope , we simulate the mass transfer guided by the detailed stream core simulations presented in ivanova , podsiadlowski & spruit ( 2002 ) . we use their equation ( 28 ) to determine the depth to which the stream penetrates the hydrogen - exhausted core of the primary and vary the parameter @xmath9 to take into account uncertainties in the modelling of the stream core impact . defines the stream penetration efficiency and depends on the amount of entropy that is generated in the stream core impact : a larger @xmath9 corresponds to the case where the ambient medium near the core has a larger pressure and temperature gradient resulting in the generation of stronger shocks and faster dissipation of the stream . ] in each time step , we determine how much mass is lost from the secondary and deposit this material with the appropriate entropy at the bottom of the stream . in this phase , the mass - transfer rate from the secondary is still determined by the frictional angular - momentum loss the immersed binary experiences , and reaches up to 100m@xmath10yr@xmath3 in the final phase . since hydrogen is deposited in layers with temperatures @xmath11k , an extended nuclear reaction network , including all important weak interactions , is included using the reaclib library ( thielemann , truran & arnould 1986 ) . this library provides the data for reaction rates which can be applied for a wide range of temperatures ( @xmath12k ) for all elements up to @xmath13kr . for heavier elements , the library has been updated for neutron - capture cross - sections and half - lifes of beta - decays for the s - elements from @xmath14rb up to @xmath15bi . the data for these cross - sections were taken from the work of beer , voss and winters ( 1992 ) and have been interpolated using the method of minimal squares to fit the formulae for reaction rates used in reaclib . the data for the half lives of the beta - decays were taken from newman ( 1978 ) . we note that , in explosive phases , the nuclear timescales of many reactions are shorter than the convective turnover timescales in convective regions , and thus the assumption of homogeneous mixing in the convective layers is not valid . to treat the time - dependent nuclear burning in convective zones , we therefore use a modified version of the two - stream formalism developed by cannon ( 1993 ) . to explore the conditions for explosive common - envelope ejection , we performed a detailed numerical study of the merger of a 18@xmath17 primary with secondaries in the range of 15@xmath17 . in all cases , we assumed that the spiral - in phase started when the primary had completed helium core burning and was ascending the red - supergiant branch for the second time ( so - called case c mass transfer ; lauterborn 1970 ) . figure 2 shows the results of a typical calculation which illustrates the evolution from the initial spiral - in phase to the point at which the envelope is ejected . to test the effects of convective overshooting , we used models with 0.25 pressure scale heights of convective overshooting and models without convective overshooting , where the former produces a h - exhausted core of 6.8@xmath17 , 1.3@xmath17 larger than the latter . to examine the uncertainties due the modelling of the stream core interaction , we use two different parameters for the entropy generation in equation ( 28 ) of ivanova et al . ( 2002 ) , @xmath18 for low - entropy generation and @xmath19 for high - entropy generation . we note that , in the present study , we did not model how rotation of the core affects the penetration depth ( see section 5.3.2 of ivanova et al . [ 2002 ] ) and how this changes during the stream core interaction phase . we do not expect that this would change our conclusions significantly . at the end of the spiral - in , when the secondary starts to fill its roche lobe within the common envelope , the secondary itself is located partially within the outer convective zone ( ocz ) of the primary , although the inner lagrangian point , @xmath20 , is within the radiative zone . the layer of envelope material at the position of the secondary has expanded because of the frictional energy input from the spiral - in and has a density roughly two orders of magnitude less than before the spiral - in , with a typical range from about @xmath21 to about @xmath22 . the spiral - in does not affect the structure of the primary s core ( the co core plus the surrounding he shell ) , it only affects the structure of the radiative hydrogen zone between the ocz and the he shell ( @xmath23 ) , since the upper part of this zone has been heated and may have become convective . for example , in the case of a 5@xmath17 donor inside an 18@xmath17 primary , the total mass of the ocz is increased by @xmath24 , while , in the case of a @xmath25 donor , it is only increased by @xmath26 . it should be noted that , in the first case , the initial @xmath23 is 0.13@xmath17 for our standard case with convective overshooting . without convective overshooting , it would be @xmath27 . the corresponding radial sizes of this zone ( @xmath28 ) are @xmath29 cm and @xmath30 cm for the case with and without convective overshooting , respectively . once the stream leaves the secondary , the depth of its penetration depends on the initial entropy of the stream material and on how much this entropy is modified during the stream s penetration into the primary s core by shocks ( ivanova et al . the latter is a function of the density contrast between the stream and the surrounding matter and depends on how strongly the medium itself is stratified . a larger density gradient in the primary s material near the stream leads to more entropy generation in the stream . the stream will stop its fall when the ambient pressure becomes comparable to the stream pressure . in the test cases , the penetration depth is two times smaller in the model with large entropy generation ( with @xmath19 ) than in the model with @xmath18 . the distance to the he shell from @xmath20 is also larger . for a secondary that is more massive than @xmath31 , the stream is unable to penetrate into the he shell since the density contrast between the stream and the ambient medium is lower and the stream entropy higher . in the layer where the hydrogen - rich stream penetrates into the helium - dominated layer ( see the middle and right panels of figure 2 ) , the stream entropy is higher than that of the surrounding material . as a consequence , both the entropy and the hydrogen mass fraction @xmath32 have negative gradients . this leads to the formation of a local convective zone that rapidly expands outwards . as the stream continues to hit the core , it penetrates progressively deeper into the core for several reasons : first , the entropy of the donor material decreases as material from deeper inside the star is transferred . second , the outer layer of the he shell starts to expand , reducing the density in the outer part of the he shell . third , the mass transfer rate steadily goes up , and , fourth , the @xmath20 point moves closer to the helium core . in the case where the initial stream does not penetrate deeply enough into the he - rich layer and is unable to create a negative entropy gradient in the ambient medium , a convective zone still develops in the region where the stream is being dissolved , but more slowly ( the stream s entropy is still higher than the ambient medium and stream energy is being deposited , further increasing the entropy ) . this convective zone propagates inwards as the stream penetrates deeper . we found that in some models this zone can connect with the ocz ; however , it either never penetrates deeply enough to connect to the convective helium - burning shell , or it starts efficient steady hydrogen burning , leading to a drop of the temperature in the he convective shell . based on these simulations , it seems that the main condition for experiencing explosive ce ejection ( ecee ) is that the stream is able to penetrate deeply enough into the he - rich zone from the very beginning such that both a negative entropy and a negative composition gradient are created . this favours models with convective overshooting and low - entropy generation . it also favours lower - mass secondaries with lower entropies . for example , for @xmath33 , after leaving @xmath20 , the stream can penetrate as much as 2.5@xmath34 of the core , while for @xmath35 it stops further from the primary center , passing through only 1@xmath34 . an alternative criterion is that @xmath23 is small : @xmath36 . based on our test calculations , we conclude that the conditions for a successful explosive common - envelope ejection are satisfied for secondary masses up to 3@xmath17 in the models with convective overshooting . a lower limit on the secondary mass is roughly given by 1@xmath17 where the energy released in the spiral - in does not provide enough energy to lead to an expansion of the envelope ( i.e. lead to a `` slow '' merger ) ; in that case the merger is expected to occur on a dynamical timescale . with these constraints to guide us , we find that the ecee conditions are found for primaries with masses up to 40@xmath17 . if the minimum mass for black - hole formation is 25@xmath17 , this leads to co cores in the range of 6.513@xmath17 ( using our standard convective overshooting parameter ) . the amount of mass transferred from the secondary before the ce ejection also depends on the secondary s mass : it is about 1m@xmath1 for a 3@xmath17 secondary , 0.8@xmath17 for a 2@xmath17 secondary and 0.20.3@xmath17 for a 1@xmath17 secondary ( but this depends somewhat on the stream entropy parameter ) . while the initial spiral - in and the early merger phase take place on a timescale of @xmath37yr , once the h - rich stream connects to the helium - core - burning shell , the evolution accelerates . the overall duration of the explosive phase is @xmath38yr , although 90% of the energy is released in the last few days . at the end of our calculations , the outer parts of the core expand with a velocity exceeding the local escape velocity , and typically 0.030.06@xmath17 of stream material has been burned explosively . the nuclear energy that has been released ( @xmath39 ergs ) exceeds the binding energy of both the he - rich layer and the common envelope by about a factor of 2 . by the time the envelope is ejected , we find that the co core has been moderately spun - up by the accretion of angular momentum and transport into the core during the pre - explosive core - accretion phase . the characteristic specific angular momentum of the core is @xmath40@xmath41s@xmath3 , at the low end of what is required in the collapsar model for lgrbs . however , further spin - up is possible , since even after the ejection of the envelope , mass transfer is likely to continue and the system is close enough for tidal spin - up to operate.h ) , the core can be spun up significantly on a timescale of @xmath42yr ( also see brown et al . however , unlike the case considered by detmers et al . ( 2008 ) , the remaining lifetime of the primary after case c mass transfer is short enough that wind mass loss will not cause significant widening of the system which would then tidally spin down the star again . ] indeed , despite the ejection of the helium shell , the system does not widen appreciably , if at all , because there is still enough frictional angular - momentum loss to keep the secondary in roche - lobe contact . this suggests that the system will continue to transfer mass after the ce ejection . this mass transfer occurs on a thermal timescale since the secondary is highly out of thermal equilibrium , in fact has a radius smaller than its equilibrium radius , and will try to expand towards a new thermal - equilibrium stage . the remaining evolution of the primary s co core to core collapse is expected to be of order a few @xmath43 to at most @xmath44yr ( in the 18@xmath17 simulation it was 3000yr ) . we note that this time is significantly longer than it would have taken the core to core collapse without the ejection of the he layer because of the signifiant core expansion and cooling . in order to assess the importance of ecee events for lgrbs and low - mass black - hole binaries , we first estimate in this section the expected rate of these events and then present detailed binary population synthesis results that simulate the properties of the resulting black - hole ( bh ) binaries . the cumulative rate for ecee events as a function of the minimum primary mass required . the black curves assume a maximum primary mass of 40@xmath45 whilst the grey curves take a maximum primary mass of 30@xmath45 . the solid curves assume a flat initial mass - ratio distribution ( @xmath46 ) , the dotted curves use @xmath47 . the light grey shading indicates a plausible range for the minimum mass of a single star which can form a black hole at solar metallicity . for these estimates we assumed that 1% of binaries have an orbital period that leads to case c mass transfer . the actual rate could be higher by an order of magnitude at low metallicity , and the minimum mass for bh formation could be lower . , width=317 ] using the results discussed in the previous section , it is straightforward to estimate the fraction of binaries that experience explosive ce ejection . this depends mainly on the secondary mass and the condition that the systems experience case c mass transfer . figure [ fig : imftimesimrd ] shows the cumulative rate for ecee events as a function of the minimum primary mass for different assumptions about the maximum mass of the primary and different mass - ratio distributions . here , we assume that 1% of all binaries experience case c mass transfer at solar metallicity . the primary masses @xmath48 are chosen from an initial mass function @xmath49 ( kroupa , tout & gilmore 1993 ) . the rates are nomalised to a core - collapse sn rate of 1 per century ( i.e. , assume that one star more massive than @xmath50 is formed per century in a typical galaxy ) . this probably underestimates the core - collapse supernova rate in the milky way ( cappellaro , evans & turatto 1999 ) . we also assumed that all massive stars form in binaries ( kobulnicky & fryer 2007 ) , with a distribution of initial separations @xmath51 that is flat in @xmath52 between 3 and @xmath53 . a significant source of uncertainty in the ecee rates is the initial mass - ratio distribution @xmath54 , where @xmath55 is the mass ratio and @xmath56 is the mass of the secondary . in figure [ fig : imftimesimrd ] we consider a distribution where @xmath57 and one where @xmath58 . in using the ecee mechanism to produce short - period black - hole x - ray binaries , we require the primary to form a bh . the light grey shading in figure [ fig : imftimesimrd ] indicates the expected minimum mass for black - hole formation , which is generally believed to be in the range of 2025@xmath17 ( see , e.g. , fryer & kalogera 2001 and further references in prh ) . black - hole formation is also generally assumed to be a requirement for the formation of lgrbs , though we note that rapid rotation at core - collapse for lower mass cores may lead to the formation of magnetars ( e.g. , thompson & duncan 1993 ; akiyama et al . magnetar formation has also been proposed as a channel for lgrbs ( e.g. , wheeler , hflich & wang 2000 ; burrows et al . 2007 ; uzdensky & macfadyen 2007 ; bucciantini et al . 2008 ) , which would increase the potential lgrb event rate from this proposed mechanism . because of the steepness of the primary mass function , the upper limit of 40@xmath45 indicated by our hydrodynamic calculations is of lesser importance to the rates . the fraction of binary systems that experience case c mass transfer is strongly dependent on the metallicity and the assumptions about the wind - loss rate from the primary star ( see justham & podsiadlowski 2010 ) , but for z=0.02 , 1% is a reasonable estimate . on the other hand , if the nieuwenhuijzen & de jager ( 1990 ) wind - loss rates are correct , only primary stars with masses @xmath59 can experience case c mass transfer . this mass goes up to @xmath60 if the nieuwenhuijzen & de jager rate is reduced by a factor @xmath61 , to take into account that the the empirical rates are not corrected for the effects of wind clumping ( e.g. , moffat & carmelle 1994 ; fullerton , massa & prinja 2006 ) . however , as figue 3 shows , the uncertainty in this upper mass limit is not a significant contribution to the uncertainty in the ecee rate . hence , if we only allow primary masses of @xmath62 to produce lgrbs through the ecee channel , and take @xmath63 , we obtain a lgrb rate of @xmath2 gal@xmath3 yr@xmath3 . this is consistent , though at the lower end , of recent lgrb rate estimates ( e.g. , podsiadlowski et al . there is some evidence that lgrbs prefer lower metallicity ( e.g. , fruchter et al . 2006 ; wolf & podsiadlowski 2007 ; modjaz et al . 2008 ) . this is consistent with the ecee scenario , since the rate of case c mass transfer increases with decreasing metallicity , by up to at least a factor of 10 at extremely low metallicity ( @xmath64 ; justham & podsiadlowski 2010 ) . given the uncertainties in these estimates , we conclude that the ecee channel could account for at least a significant subset of the lgrb rate and possibly all of it . ecee links lgrbs with short - period black - hole x - ray binaries . wijers ( 1996 ) and romani ( 1998 ) estimated the galactic population of short - period bh xrbs to be at least @xmath65 . for a birthrate of @xmath2 gal@xmath3 yr@xmath3 , this requires a plausible mean system lifetime of @xmath66gyr . this assumes that the majority of systems will not be disrupted by the supernova explosion and will reach roche - lobe contact in a reasonable time . in the following subsection we simulate the properties of the predicted x - ray binary population from this channel . for the population synthesis calculation , we randomly selected a set of @xmath67 binaries with initial primary masses @xmath68 between 25 and 40@xmath69 and initial secondary masses @xmath70 between 1 and 3@xmath69 . we again adopted an initial mass function @xmath71 and a flat mass - ratio distribution . after the ce ejection , the secondary is assumed to be filling its roche lobe , although in reality it may be slightly underfilling it because of the mass loss associated with the ce ejection . the core mass of the primary @xmath72 can be related to the initial primary mass according to the simple estimate @xmath73 ( hurley et al . this yields post - ecee remnant masses in the range of 9.317.5@xmath17 for initial primary masses between 25 and 40@xmath17 . the secondary , having lost mass during the ce phase , leaves a donor star of mass @xmath74 , where @xmath75 ( in accordance with the results in sect . the orbital period and component masses are assumed to be unchanged until the core explodes as a supernova . after the supernova , the core becomes a bh ; we determine the mass of the bh using the results of fryer & kalogera ( 2001 ) , as formulated in belczynski et al . ( 2008 ) . mass loss from the system and any kick imparted to the black hole during the supernova will change the orbital period and eccentricity ( see , e.g. , brandt & podsiadlowksi 1995 ; kalogera 1996 ) . we assume that the supernova kicks are randomly oriented , and sample the kick distributions assuming that it can be described by a maxwellian distribution with a velocity dispersion @xmath76 ( for a discussion of supernova kicks in the formation of black holes , see brandt , podsiadlowski & sigurdsson 1995 ; prh ) . we explored a range of values for @xmath77 , including zero kick velocity and kick velocities with the same kick _ momentum _ as found for single neutron stars ( hobbs et al . 2005 ) . if the system is not disrupted in the supernova , we assume that it circularises whilst conserving angular momentum . we also allow the system to survive even if the roche - lobe radius of the donor star is as small as 90% of the zams radius for the post - ecee donor mass to take into account the fact that the secondary is likely to be still undersized and expanding at this stage , as indicated by our ecee calculations . we assume that a short period of mass transfer driven by this expansion will increase the orbital period of the system until the star is in equilibrium , and assume , for simplicity , that the mass lost by the donor during that phase is negligible . the donor star is now assumed to be still essentially unevolved . we weight the output of this algorithm by the initial mass function and initial mass - ratio distribution . this produces a probability distribution for the formation of post - supernova black - hole binaries . in order to simulate the population of black - hole x - ray binaries , we combine these formation probabilities with a grid of binary evolution sequences . to follow the x - ray binary phase , we calculated a grid of 936 binary evolution sequences using an updated version of eggleton s stellar evolution code ( e.g. , eggleton 1971 ; pols et al . 1995 ) and a metallicity of 0.02 . this grid was equally spaced in 13 donor masses ( 0.8 to 2.0@xmath69 ) , 13 accretor masses ( 5 to 17@xmath45 ) , and 12 values for the initial period . the first model is placed into contact at the start of its evolution , with an orbital period @xmath78 . the second begins in a binary with @xmath79=1.1 , and the remaining ten orbital periods are equally spaced in @xmath80 between 0.1 and 1.0 ( where @xmath81 is the orbital period at which the donor is filling its roche lobe when unevolved ) . is a function of donor mass , accretor mass and donor radius . due to angular - momentum loss , the models with initial @xmath82 have shorter periods by the time they reach contact . ] the evolution of the systems is driven by angular - momentum loss . we include losses due to both gravitational wave radiation and magnetic braking , adopting the formalism of ivanova & taam ( 2003 ) for secondaries with @xmath83 ( see also ivanova & kalogera 2006 ) . we consider only systems which reach contact before the end of the main sequence and have a maximum age for each system of 13 gyr ; we varied this maximum age and found it to have a relatively minor effect . the formation probabilities produced by our population synthesis are all distributed among the eight members of the three - dimensional grid nearest to them , i.e. to a cube of gridpoints around each system , using trilinear interpolation to determine the weights attributed to each grid point . each sequence in our binary evolution grid then posesses a formation likelihood , and the grid sequences are binned and combined in proportion to these formation probabilities . within each sequence , the statistical weight of each evolutionary timestep is proportional to the duration of that timestep . this leads to a ` steady - state ' x - ray binary population , assumed to be representative of the distribution in the milky way at the current epoch . orbital period distribution ( upper panel ) and distribution of effective temperatures of the secondaries for two population synthesis simulations . the thick grey curves assume a supernova kick dispersion of @xmath84 , and the solid black curves correspond to no kick . both simulations assume a flat initial mass ratio distribution ( @xmath85 ) ( changing this to @xmath58 makes little difference to the shape of the distribution , but changes the overall normalisation ) . in the top panel , the dashed curve represents a histogram of the orbital periods of known systems ( also marked with arrows ) . the normalisation of the histogram of observed systems is arbitrary . the shapes of the period distributions are reasonably well matched . the lower panel shows the median effective temperature ( solid curves ) and upper and lower quartiles ( broken curves ) of the populations as a function of orbital period . we also mark the maximum effective temperatures consistent with the spectral types of the known systems ( crosses with arrows).,width=317 ] figure [ fig : teffperioddists ] shows a selection of our population synthesis calculations , which reproduces the shape of the observed orbital - period distribution fairly well , for @xmath86 and for no kick . however , the donor temperatures do not easily match the observations , unless either our model effective temperatures or our conversion from spectral type to effective temperatures is incorrect . note that these donor temperatures give a significantly better match to the observations than systems which initially contained intermediate - mass donors would provide ( see , e.g. , justham et al . 2006 ) , and that this apparent mismatch in effective temperature seems to be a problem for all models , unless the donor stars in short - period bh binaries are pre - main sequence stars ( ivanova 2006 ) . hence this problem with effective temperatures appears to be a generic problem for models of these systems . in this paper , we have presented a new mechanism for the ejection of a common envelope where the energy source is not orbital energy but nuclear energy . this provides a new channel to produce plausible progenitors of short - period black - hole binaries and long - duration grbs . we have also demonstrated that this scenario may be able to explain both the origin and the main properties of short - period black - hole binaries , such as the period distribution , although the distribution of spectral types is still not fully satisfactory . two of the most attractive features of the explosive ce ejection scenario are that ( 1 ) it leads to the ejection of both the hydrogen and the helium layer , explaining why all lgrb supernovae to - date have been classified as type ic supernovae , and ( 2 ) this ejection occurs late in the evolution of the star ; hence the progenitor will only experience a short wolf - rayet phase , in which it will not be spun down significantly by wind mass loss . this may also explain why extended wolf - rayet wind bubbles are not being found around lgrbs ; e.g. , in the case of grb 021004 , van marle , langer & garcia - segura ( 2005 ) found that the wolf - rayet phase had to last less than @xmath87yr ; this is consistent with the explosive ce - ejection scenario , since it always requires late case c mass transfer . this also predicts that such wolf - rayet bubbles are terminated by a dense shell from the ejected common envelope . our rate estimate for this channel ( @xmath88yr@xmath3 ) implies that this can produce a significant fraction of all lgrbs ; this rate should be higher at lower metallicity , because case c mass transfer is expected to be more common at lower metallicity . unlike some of the single - star progenitor models for lgrbs ( yoon & langer 2005 ; woosley & heger 2006 ; yoon , langer & norman 2006 ) , lgrbs may occur even at solar metallicity , but they are expected to be more common at low metallicity . indeed , there is some evidence now that lgrbs can also occur in super - solar host galaxies ( see , e.g. , levesque et al . the different metallicity biases may provide a possible way of distinguishing between these two different scenarios . explosive ce ejection also operates for lower - mass primaries that are expected to produce neutron stars rather than black holes . it is tempting to associate these with rapidly rotating neutron stars and possibly magnetars , for which sn 2006aj , which was associated with the x - ray flash grb 060218 , may provide an observed example in nature ( mazzali et al . 2006 ) . ni gratefully acknowledges support from the nserc of canada and from the canada research chairs program . sj is partially supported by the national science foundation of china under grant numbers 10903001 and 10950110322 , and by the chinese postdoc fund ( award number 20090450005 ) . akiyama s. wheeler j. c. , meier d. l. , lichtenstadt i. , 2003 , apj , 584 , 965 alexander d. r. , ferguson j. w. 1994 , apj , 437 , 879 beer h. , voss , f. , winters , r. r. , 1992 . apjs , 80 , 403 beer m. e. , podsiadlowski ph . , 2002 , mnras , 331 , 351 belczynski k. , et al . , 2008 , apjs , 174 , 223 brandt w. n. , podsiadlowski ph . 1995 , mnras , 274 , 461 brandt w. n. , podsiadlowski ph . , sigurdsson s. , 1995 , mnras , 277 , l35 brown g. e. , lee c .- h . , wijers r. a. m. j. , lee h. k. , israelian g. , bethe h. a. , 2000 , newa , 5 , 191 brown g. e. , lee c .- h . , moreno mndez e. , 2007 , apj , 671 , l41 bucciantini n. , quataert e. , arons j. , metzger b.d . , thompson t. a. , 2008 , mnras , 383 , l25 burrows a. , dessart l. , livne e. , ott c. d. , murphy j. , 2007 , apj , 664 , 416 cannon r. , 1993 , mnras , 263 , 817 cappellaro e. , evans r. , turatto m. 1999 , a&a , 351 , 459 detmers r. g. , langer n. , podsiadlowski , ph . , izzard r. g. , 2008 , a&a , 484 , 831 dewi j. , tauris t. , 2001 , in podsiadlowski ph . , rappaport , s. , king a. r. , dantona f. , burderi l. , eds , asp conf . 229 , evolution of binary and multiple star systems . soc . pac . , san francisco , p. 225 eggleton p. p. , 1971 , mnras , 151 , 351 eggleton p. p. , verbunt f. , 1986 , mnras , 220 , 13 foellmi c. , dall t. h. , depagne e. , 2007 , a&a , 464 , 61 fruchter a. s. , et al . , 2006 nat 441 , 463 fryer c. l. , kalogera v. , 2001 , apj , 554 , 548 fullerton a. w. , massa d. l. , prinja r. k. , 2006 , apj , 637 , 1025 gonzlez hernndez j. i. , rebolo r. , israelian g. , 2008 , a&a , 478 , 203 heger a. , langer n. , 1998 , a&a , 334 , 210 heger a. , woosley s. e. , spruit h. c. , 2005 , apj , 626 , 350 hobbs g. , lorimer d. r. , lyne a. g. , kramer m. , 2005 , mnras , 360 , 974 hurley , j. r. , pols , o. r. , tout , c. a. , 2000 , mnras , 315 , 543 israelian g. , rebolo r. , basri g. , casares j. , martin e. l. , 1999 , nat , 401 , 142 ivanova n. , 2002 , dphil thesis ( oxford ) ivanova n. , 2006 , apj , 653 , l137 ivanova n. , kalogera v. , 2006 , apj , 636 , 985 ivanova n. , podsiadlowski ph . , 2003 , in w. hillebrandt , b. leibundgut , eds , from twilight to highlight : the physics of supernovae . springer , berlin , p. 19 ivanova n. , podsiadlowski ph . , 2010 , in preparation ivanova n. , podsiadlowski ph . , spruit h. , 2002 , mnras , 334 , 819 ivanova n. , taam r. e. , 2003 , apj , 599 , 516 justham s. , rappaport s. , podsiadlowski ph . , 2006 , mnras , 366 , 1415 justham s. , podsiadlowski ph . , 2010 , in preparation kalogera v. , 1996 , apj , 471 , 352 kippenhahn r. , weigert a. , hofmeister e. , 1967 , in alder b. , fernbach s. , rothenberg m. , eds , methods in computational physics , vol . 7 . academic , new york , p. 129 kobulnicky h. a. , fryer c. l. , 2007 , apj , 670 , 747 kroupa p. , tout c. a. , gilmore g. , 1993 , mnras 262 , 545 lauterborn d. , 1970 , a&a , 7 , 150 lee c .- h . , brown g. e. , wijers r. a. m. j. , 2002 , apj , 575 , 996 levesque e. m. , kewley l. j. , graham j. f. , fruchter a. s. , 2010 , apjl , 712 , l26 macfadyen a. i. , woosley s. e. , 1999 , apj , 524 , 262 mazzali p. , et al . , 2006 , nat , 442 , 1018 meyer f. , meyer - hofmeister e. , 1979 , a&a , 78 , 167 modjaz m. , et al . , 2008 , aj , 135 , 1136 moffat a. f. j. , carmelle r. , 1994 , apj , 421 , 310 newman , m. , 1978 . apj , 219 , 676 nieuwenhuijzen h. , de jager c. , 1990 , a&a , 231 , 134 paczyski b. 1976 , in eggleton , p. p. , mitton s. , whelan j. , eds , structure and evolution in close binary systems . dordrecht , reidel , p. 75 pfahl e. , rappaport s. , podsiadlowski ph . , 2003 , apj , 597 , 1036 podsiadlowski ph . , 2001 , in podsiadlowski ph . , rappaport s. , king a. r. , dantona f. , burderi l. , eds , asp conf . voll 229 , evolution of binary and multiple star systems . , san francisco , p. 239 podsiadlowski ph . , cannon r. c. , rees m. j. , 1995 , mnras , 274 , 485 podsiadlowski ph . , mazzali p. a. , nomoto k. , lazzati d. , cappellaro e. , 2004 , apj , 607 , l17 podsiadlowski ph . , nomoto k. , maeda k. , nakamura t. , mazzali p. , schmidt b. , 2002 , apj , 567 , 491 podsiadlowski ph . , rappaport s. , han z. , 2003 , mnras , 341 , 385 ( prh ) podsiadlowski ph . , rappaport s. , pfahl e. , 2002 , apj , 565 , 1107 pols o. r. , tout c. a. , eggleton p. p. , han z. , 1995 , mnras , 274 , 964 pols o. r. , tout c. a. , schrder k .- , eggleton p. p. , manners j. , 1997 , mnras , 289 , 869 rogers f. j. , iglesias c. a. 1992 , apjs , 79 , 507 romani r. w. , 1998 , a&a , 333 , 583 schrder k .- , pols o. r. , eggleton p. p. 1997 , mnras , 285 , 696 siess l. , livio m. 1999a , mnras , 304 , 925 siess l. , livio m. 1999b , mnras , 308 , 1133 thielemann f .- k . , truran j. w. , arnould m. , 1986 , in advances in nuclear astrophysics , frontieres , gif - sur - yvette , p. 525 thompson c. duncan r. c. , 1993 , apj , 408 , 194 uzdensky d. a. , macfadyen a. i. , 2007 , apj , 669 , 546 van marle a. j. , langer n. , garcia - segura g. , 2005 , a&a , 444 , 837 wheeler j. c. , yi i. , hflich p. , wang l. , 2000 , apj , 537 , 810 wijers r. a. m. j. , 1996 in wijers r. a. m. j. , davies m. b. , tout c. a. , eds , evolutionary processes in binary stars . kluwer , dordrecht , p. 327 wolf c. , podsiadlowski ph . , 2007 , mnras , 375 , 1049 woosley s. e. , 1993 , apj , 405 , 273 woosley s. e. , heger a. , 2006 , apj , 637 , 914 yoon s .- c . , langer n. , 2005 , a&a , 443 , 643 yoon s .- c . , langer n. , norman c. , 2006 , a&a , 460 , 199 yungelson l. r. , lasota j .- p . , nelemans g. , dubus g. , van den heuvel e. p. j. , dewi j. , portegies zwart s. , 2006 , a&a , 454 , 559

recently a program has been proposed to study neutrino masses @xcite . this program is based on the following assumptions . * assumption 1 : * there are three right - handed dirac neutrinos in addition to the three known left - handed neutrinos . * assumption 2 : * the lepton mass matrices have the same structure as the quark mass matrices , as described by lehmann _ _ @xcite . in carrying out this program , one of the technically most difficult aspects is to treat the mikheyev - smirnov - wolfenstein ( msw ) effect for the sun @xcite . it is the purpose of this paper to discuss this problem . the difficulty arises in the following way . on the basis of the super - kamiokande data on atmospheric neutrinos @xcite , the muon neutrino @xmath0 oscillates with another neutrino which is not the electron neutrino @xmath1 . from assumption 1 , this neutrino must be the tau neutrino @xmath2 . in other words , @xmath0 and @xmath2 are coupled significantly . again from assumption 1 , the electron neutrinos @xmath1 from the sun must therefore oscillate with either @xmath0 or @xmath2 . since @xmath0 and @xmath2 are coupled significantly , the electron neutrinos @xmath1 from the sun must oscillate with _ both _ @xmath0 _ and _ @xmath2 . it is accordingly necessary to understand the msw effect with all three generations of neutrinos @xmath1 , @xmath0 and @xmath2 . while there is an extensive literature on the msw effect for two generations of neutrinos @xcite , relatively little is known for the case of three generations @xcite . one of the major differences between these two cases is the following . for two generations , the msw effect occurs predominantly in the vicinity of a particular density in the sun . for this reason , the important parameter is the rate of change of the density at this particular density . in contrast , for three generations , the important contributions to the msw effect can come from several regions . when the parameters , such as the neutrino masses , vary , these regions not only move but also merge and separate , features that are absent for the much simpler case of two generations . there are many different ways to formulate the problem of the msw @xcite effect in the sun for three generations of neutrinos . these different formulations are of course closely related to each other ; it has been found that , at least for the present program @xcite , the following one is most appropriate . in a current basis , the charged lepton currents are @xmath3 where @xmath4 \quad { \rm and } \quad \ell=\left[\tighten\matrix { \ell_1 \cr \ell_2 \cr \ell_3\cr}\right]\ ] ] denote the neutrino and charged lepton fields . the neutrino mass matrix @xmath5 @xcite , from the vacuum expectation value of the higgs field in the neutrino - higgs coupling , is @xmath6 let @xmath7 be the momentum of the neutrino , the value of which is taken to be much larger than the neutrino masses , which are the eigenvalues of @xmath5 . typically , @xmath7 is in the range from 0.4 mev/@xmath8 to 10 mev/@xmath8 , while the neutrino masses are believed to be no more than about 1 ev ; hence this assumption is well satisfied . under this assumption , it is @xmath9 that enters in the msw effect . let @xmath10\ ] ] be real and symmetric . note that @xmath11 are not defined , only @xmath12 . of course the eigenvalues of this matrix @xmath9 are the squares of the neutrino masses . in terms of this @xmath9 , the msw effect for neutrino oscillations is described by the coupled ordinary differential equation @xcite @xmath13 = \left ( \left[\tighten\matrix { d(r ) & 0 & 0\cr 0 & 0 & 0\cr 0 & 0 & 0\cr}\right ] + \frac{1}{2p } \left[\tighten\matrix { m^2_{11 } & m^2_{12 } & m^2_{13}\cr m^2_{21 } & m^2_{22 } & m^2_{23}\cr m^2_{31 } & m^2_{32 } & m^2_{33 } \cr}\right ] \right ) \left[\tighten\matrix { \phi_1(r)\cr \phi_2(r)\cr \phi_3(r ) \cr}\right],\ ] ] where @xmath14 with @xmath15 the fermi weak - interaction constant and @xmath16 the electron density at the point @xmath17 . so far this formulation is general . in view of the observation in sec . i that , for three generations of neutrinos , important contributions to the msw effect can come from several values of @xmath17 , we propose to choose a suitable function @xmath18 . on the one hand , this @xmath18 should describe approximately the electron density in the sun ; on the other hand , this @xmath18 should be sufficiently simple so that eq . ( [ eq : schr-1 ] ) can be solved explicitly . this choice is made in the following way . first , since most neutrinos are produced by nuclear reactions @xcite near the center of the sun , the simplifying assumption is made that the neutrinos are produced at the center of the sun . this has the consequence that the @xmath17 in eqs . ( [ eq : schr-1 ] ) and ( [ eq : define - d ] ) is the radial distance from the center of the sun . this is the reason for the notation @xmath17 . secondly , the density of the sun as a function of the radial distance is fairly accurately known @xcite , as shown by the solid curve in fig . 1 , where the vertical axis is the density in g/@xmath19 in a logarithmic scale , and the horizontal axis is @xmath20 . the radius of the sun is @xmath21 from this solid curve in fig . 1 , it is seen that the solar density is approximately exponential , as shown by the dotted line of fig . 1 . this fit is reasonably good , to an accuracy of roughly 15% , for @xmath22 in other words , @xmath18 is taken to be given by @xmath23 [ fig : rho ] ( 12,7.5 ) ( 2.0,1.0 ) from the dotted line of fig . 1 , the following values are obtained @xmath24 the problem is therefore to solve the msw differential equation , for @xmath25 , @xmath26 = \frac{1}{2p } \left[\tighten\matrix { 2pd(0)e^{-r / r_0}+m^2_{11 } & m^2_{12 } & m^2_{13}\cr m^2_{21 } & m^2_{22 } & m^2_{23}\cr m^2_{31 } & m^2_{32 } & m^2_{33 } \cr}\right ] \left[\tighten\matrix { \phi_1(r)\cr \phi_2(r)\cr \phi_3(r ) \cr}\right]\ ] ] with the boundary conditions @xmath27 = \left[\tighten\matrix { 1\cr 0\cr 0\cr}\right].\ ] ] the differential equation ( [ eq : schr-2 ] ) can be solved explicitly in terms of generalized hypergeometric functions . this is to be carried out in the present section . indeed , this is the underlying reason for the choice of the exponential function ( [ eq : d - exponential ] ) . let @xmath28 with @xmath29\sim -8.07.\ ] ] then ( [ eq : schr-2 ] ) is @xmath30 = \left[\tighten\matrix { e^{-u}+{\displaystyle \frac{r_0}{2p}}\,m^2_{11 } & { \displaystyle \frac{r_0}{2p}}\,m^2_{12 } & { \displaystyle \frac{r_0}{2p}}\,m^2_{13}\cr \noalign{\vskip4pt } { \displaystyle \frac{r_0}{2p}}\,m^2_{21 } & { \displaystyle \frac{r_0}{2p}}\,m^2_{22 } & { \displaystyle \frac{r_0}{2p}}\,m^2_{23}\cr \noalign{\vskip4pt } { \displaystyle \frac{r_0}{2p}}\,m^2_{31 } & { \displaystyle \frac{r_0}{2p}}\,m^2_{32 } & { \displaystyle \frac{r_0}{2p}}\,m^2_{33 } \cr}\right ] \left[\tighten\matrix { \phi_1\cr \phi_2\cr \phi_3 \cr}\right].\ ] ] note that @xmath31 corresponds to @xmath32 in view of the form of eq . ( [ eq : schr-3 ] ) , it is convenient to diagonalize the @xmath33 matrix @xmath34.\ ] ] let @xmath35 be the angle of rotation that diagonalizes this @xmath33 matrix , i.e. , @xmath36 \frac{r_0}{2p } \left[\tighten\matrix { m^2_{22 } & m^2_{23}\cr m^2_{32 } & m^2_{33 } \cr}\right ] \left[\tighten\matrix { \cos\theta_0 & \sin\theta_0\cr -\sin\theta_0 & \cos\theta_0 \cr}\right ] = \left[\tighten\matrix { \omega_2 & 0\cr 0 & \omega_3 \cr}\right]\ ] ] with @xmath37 the reason for this eq . ( [ eq : omega - order ] ) is that , if @xmath38 , then the problem reduces to neutrino oscillations with only two neutrinos . define also @xmath39 = \frac{r_0}{2p } \left[\tighten\matrix { \cos\theta_0 & -\sin\theta_0\cr \sin\theta_0 & \cos\theta_0 \cr}\right ] \left[\tighten\matrix { m^2_{12}\cr m^2_{13 } \cr}\right]\ ] ] and @xmath40 = \left[\tighten\matrix { \cos\theta_0 & -\sin\theta_0\cr \sin\theta_0 & \cos\theta_0 \cr}\right ] \left[\tighten\matrix { \phi_2(u)\cr \phi_3(u ) \cr}\right];\ ] ] then eq . ( [ eq : schr-3 ] ) becomes @xmath41 = \left[\tighten\matrix { \omega_1+e^{-u } & \chi_2 & \chi_3\cr \chi_2 & \omega_2 & 0\cr \chi_3 & 0 & \omega_3 \cr}\right ] \left[\tighten\matrix { \psi_1(u)\cr \psi_2(u)\cr \psi_3(u ) \cr}\right].\ ] ] if @xmath42 or @xmath43 is zero , then once again this reduces to neutrino oscillations with two generations . by reversing the signs of @xmath44 and/or @xmath45 if necessary , it follows that , without loss of generality @xmath46 let @xmath47 , @xmath48 and @xmath49 be the eigenvalues of the @xmath50 matrix @xmath51;\ ] ] then these three @xmath52 s are the squares of the neutrino masses multiplied by @xmath53 . note that @xmath54 for cases of interest @xcite , the @xmath55 will typically be large , @xmath56 . this is of course not a question of units , but rather a reflection of the physical situation of having a large number of oscillations within the solar radius . the interlacing property implies that , by renumbering the @xmath52 s if necessary , @xmath57 again assuming that there is no reduction to two - neutrino oscillation . the following properties are easily verified : with fixed @xmath58 , @xmath59 , @xmath42 and @xmath43 , * @xmath60 * as @xmath61 , @xmath62 * as @xmath63 , which is physically not possible , @xmath64 a third - order ordinary differential equation can be obtained from eq . ( [ eq : schr-4 ] ) as follows . define @xmath65 , @xmath66 and @xmath67 such that @xmath68 these @xmath65 , @xmath66 and @xmath67 are of course not unique . it then follows from eqs . ( [ eq : schr-4 ] ) and ( [ eq : chi - positive ] ) that @xmath69 therefore , @xmath70 in particular , @xmath71 define this function of eq . ( [ eq : psi22-psi33 ] ) to be @xmath72 , then it follows that @xmath73 substitution into eq . ( [ eq : psi - i ] ) then expresses @xmath74 , @xmath75 and @xmath76 in terms of the single function @xmath72 : @xmath77 furthermore , this @xmath72 is unique because of ( [ eq : omega - order ] ) to obtain the third - order ordinary differential equation for @xmath72 , it only remains to substitute ( [ eq : psi1 - 2 - 3 ] ) into the first equation of ( [ eq : schr-4 ] ) : @xmath78\psi.\end{aligned}\ ] ] from the definition of @xmath47 , @xmath48 and @xmath49 , this can be rewritten as @xmath79 to put this equation into the form of the differential equation for the generalized hypergeometric function , let @xmath80 then the differential equation ( [ eq : schr-5 ] ) for @xmath72 is @xmath81\psi=0.\ ] ] this is the differential equation for the generalized hypergeometric function @xmath82see , for example , p. 184 of @xcite . three linearly independent solutions of this third - order differential equation ( [ eq : schr-6 ] ) are @xmath83 , \nonumber \\[5pt ] \psi^{(2 ) } & = & k_2e^{-i\mu_2u } { } _ 2f_2\left[\left . \tighten\matrix { -i(\omega_2-\mu_2 ) , & -i(\omega_3-\mu_2)\cr 1-i(\mu_1-\mu_2 ) , & 1-i(\mu_3-\mu_2 ) \cr } \right|ie^{-u } \right ] , \nonumber \\[5pt ] \psi^{(3 ) } & = & k_3e^{-i\mu_3u } { } _ 2f_2\left[\left . \tighten\matrix { -i(\omega_2-\mu_3 ) , & -i(\omega_3-\mu_3)\cr 1-i(\mu_1-\mu_3 ) , & 1-i(\mu_2-\mu_3 ) \cr } \right|ie^{-u } \right],\end{aligned}\ ] ] where @xmath84 , @xmath85 and @xmath86 are arbitrary non - zero constants . for the convenience of numerical calculations , we make the following choice : @xmath87^{-1/2 } , \nonumber \\ k_2&=&\left[(\omega_2-\mu_2)(\omega_3-\mu_2)(\mu_1-\mu_2)(\mu_3-\mu_2 ) \right]^{-1/2 } , \nonumber \\ k_3&=&\left[(\omega_2-\mu_3)(\omega_3-\mu_3)(\mu_1-\mu_3)(\mu_2-\mu_3 ) \right]^{-1/2}.\end{aligned}\ ] ] the function @xmath82 can be defined by the following series expansion emerges as the solution when one of the @xmath88 s equals one of the @xmath89 s , i.e. , for the case of two generations . ] @xcite @xmath90 = \sum_{k=0}^\infty \frac{(\alpha_1)_k(\alpha_2)_k}{(\rho_1)_k(\rho_2)_k}\ , \frac{z^k}{k!},\ ] ] where the pochhammer symbol is defined as @xmath91 since the taylor series of @xmath82 about @xmath92 , eq . ( [ eq : f22-series ] ) , is always convergent , the differentiations specified in eqs . ( [ eq : psi1 - 2 - 3 ] ) can be carried out term by term on these taylor series . thus these derivatives of @xmath93 can be easily expressed in terms of @xmath82 with parameters slightly modified . the result is @xmath94 where @xmath95 , \nonumber \\[5pt ] \psi_1^{(2)}&=&k_2(\mu_2-\omega_2)(\mu_2-\omega_3)e^{-i\mu_2u } { } _ 2f_2\left[\left . \tighten\matrix { 1-i(\omega_2-\mu_2 ) , & 1-i(\omega_3-\mu_2)\cr 1-i(\mu_1-\mu_2 ) , & 1-i(\mu_3-\mu_2 ) \cr } \right|ie^{-u } \right ] , \nonumber \\[5pt ] \psi_1^{(3)}&=&k_3(\mu_3-\omega_2)(\mu_3-\omega_3)e^{-i\mu_3u } { } _ 2f_2\left[\left . \tighten\matrix { 1-i(\omega_2-\mu_3 ) , & 1-i(\omega_3-\mu_3)\cr 1-i(\mu_1-\mu_3 ) , & 1-i(\mu_2-\mu_3 ) \cr } \right|ie^{-u } \right],\end{aligned}\ ] ] @xmath96 , \nonumber \\[5pt ] \psi_2^{(2)}&=&k_2\chi_2(\mu_2-\omega_3)e^{-i\mu_2u } { } _ 2f_2\left[\left . \tighten\matrix { -i(\omega_2-\mu_2 ) , & 1-i(\omega_3-\mu_2)\cr 1-i(\mu_1-\mu_2 ) , & 1-i(\mu_3-\mu_2 ) \cr } \right|ie^{-u } \right ] , \nonumber \\[5pt ] \psi_2^{(3)}&=&k_3\chi_2(\mu_3-\omega_3)e^{-i\mu_3u } { } _ 2f_2\left[\left . \tighten\matrix { -i(\omega_2-\mu_3 ) , & 1-i(\omega_3-\mu_3)\cr 1-i(\mu_1-\mu_3 ) , & 1-i(\mu_2-\mu_3 ) \cr } \right|ie^{-u } \right],\end{aligned}\ ] ] @xmath97 , \nonumber \\[5pt ] \psi_3^{(2)}&=&k_2\chi_3(\mu_2-\omega_2)e^{-i\mu_2u } { } _ 2f_2\left[\left . \tighten\matrix { 1-i(\omega_2-\mu_2 ) , & -i(\omega_3-\mu_2)\cr 1-i(\mu_1-\mu_2 ) , & 1-i(\mu_3-\mu_2 ) \cr } \right|ie^{-u } \right ] , \nonumber \\[5pt ] \psi_3^{(3)}&=&k_3\chi_3(\mu_3-\omega_2)e^{-i\mu_3u } { } _ 2f_2\left[\left . \tighten\matrix { 1-i(\omega_2-\mu_3 ) , & -i(\omega_3-\mu_3)\cr 1-i(\mu_1-\mu_3 ) , & 1-i(\mu_2-\mu_3 ) \cr } \right|ie^{-u } \right].\end{aligned}\ ] ] in eqs . ( [ eq : psi1 - 2 - 3-b ] ) , the constants @xmath98 , @xmath99 and @xmath100 are determined by the boundary conditions eqs . ( [ eq : boundary - cond ] ) . for the exponential matter distribution ( [ eq : d - exponential ] ) , this gives completely the solution of eq . ( [ eq : schr-1 ] ) , which describes the msw effect in the sun . we recall that by eq . ( [ eq : u0 ] ) , @xmath101 . thus , the prefactors @xmath102 are proportional to the exponentials describing propagation in vacuum , @xmath103 . in fact , as @xmath104 , all these functions approach a simple limit : @xmath105=1.\ ] ] while this solution ( [ eq : psi1 - 2 - 3-b ] ) is specifically for three generations of neutrinos , generalization , if desired , to @xmath106 generations is completely straightforward , leading to the generalized hypergeometric function @xmath107 . even for three generations , @xmath82 is a complicated function . from eq . ( [ eq : u0 ] ) , corresponding to the center of the sun , @xmath108 is about 3000 ; for arguments of this order of magnitude , it is not practical to calculate the various @xmath82 of eqs . ( [ eq : psi1])([eq : psi3 ] ) using the taylor series expansion . sections iv and v are devoted to the issue of how these @xmath82 can be calculated . of the nine @xmath82 that appear in eqs . ( [ eq : psi1])([eq : psi3 ] ) , three of them are evaluated approximately in sec . iv . in sec . v , the remaining six @xmath82 are then expressed in terms of another set of generalized hypergeometric functions @xmath109 . these @xmath109 have convenient integral representations , useful for numerical evaluation . the approach described in the rest of this paper is the best we have found , but there may well be other methods that are superior . in this section , we study the three @xmath82 that appear in the expressions of @xmath111 , @xmath112 and @xmath113 as given by eqs . ( [ eq : psi1])([eq : psi3 ] ) . define @xmath114 then by ( [ eq : interlace ] ) @xmath115 and the three @xmath82 of interest are @xmath116 , \nonumber \\[5pt ] _ 2f_2(2)&= & _ 2f_2\left[\left . \tighten\matrix { i\rho_2 , & 1+i\rho_4\cr 1+i\rho_1 , & 1+i\rho_3 \cr } \right|ie^{-u } \right ] , \nonumber \\[5pt ] _ 2f_2(3)&= & _ 2f_2\left[\left . \tighten\matrix { 1+i\rho_2 , & i\rho_4\cr 1+i\rho_1 , & 1+i\rho_3 \cr } \right|ie^{-u } \right].\end{aligned}\ ] ] let @xmath117 be considered in some detail ; the treatments of @xmath118 and @xmath119 are similar . from the definition of @xmath82 by its taylor series , which is always convergent , it is straightforward to verify , by closing the contour of integration in the right half - plane , that @xmath120 & & \mbox{}\times\int_{\gamma - i\infty}^{\gamma+i\infty } \frac{\gamma(-s)\gamma(1+i\rho_2+s)\gamma(1+i\rho_4+s ) } { \gamma(1+i\rho_1+s)\gamma(1+i\rho_3+s ) } \left(e^{-i\pi/2}e^{-u}\right)^s ds\end{aligned}\ ] ] where @xmath121 this integral is to be evaluated approximately for @xmath122 large and negative . except for an indentation near @xmath123 , the contour of integration can be taken along the imaginary axis . it is therefore convenient to let @xmath124 the stirling formula @xmath125 takes the form @xmath126 when @xmath127 is purely imaginary . using ( [ eq : stirling ] ) , the integrand in ( [ eq : f21-contour ] ) is @xmath128 & = & \frac{\gamma(-i\tau)\gamma(1+i(\tau+\rho_2))\gamma(1+i(\tau+\rho_4 ) ) } { \gamma(1+i(\tau+\rho_1))\gamma(1+i(\tau+\rho_3))}\ , e^{\pi\tau/2}e^{i|u|\tau } \nonumber \\[5pt ] & \sim&\left(\frac{2\pi}{-i\tau}\right)^{1/2 } \frac{[i(\tau+\rho_2)]^{1/2}[i(\tau+\rho_4)]^{1/2 } } { [ i(\tau+\rho_1)]^{1/2}[i(\tau+\rho_3)]^{1/2 } } \ , e^{\pi\theta_1(\tau)/2}\,e^{i\theta_2(\tau)},\end{aligned}\ ] ] with @xmath129 and @xmath130 the condition @xmath131 for stationary phase gives @xmath132 eq . ( [ eq : stat - phase ] ) together with the fact that @xmath133 is a non - decreasing function of @xmath134 implies that the relevant point of stationary phase occurs when @xmath134 is positive . let @xmath135 be the unique positive root of the cubic equation @xmath136 then @xmath135 is the point of stationary phase that gives the leading contribution to @xmath117 . the result is , with all gamma functions replaced by the corresponding stirling formula ( [ eq : stirling-0 ] ) , @xmath137^{1/2}\ , e^{i[\rho_1(\ln\rho_1 - 1)-\rho_2(\ln\rho_2 - 1 ) + \rho_3(\ln\rho_3 - 1)-\rho_4(\ln\rho_4 - 1 ) ] } \nonumber \\ & & \mbox{}\times \left[1+\frac{\tau_0}{\tau_0+\rho_1}-\frac{\tau_0}{\tau_0+\rho_2 } + \frac{\tau_0}{\tau_0+\rho_3}-\frac{\tau_0}{\tau_0+\rho_4}\right]^{-1/2 } e^{i\theta_2(\tau_0)},\end{aligned}\ ] ] for @xmath122 negative and large . similarly , @xmath138 \label{eq : f22-stirling-3 } _ 2f_2(3 ) & \sim&\frac{\rho_4}{\tau_0+\rho_4}\ , _ 2f_2(1).\end{aligned}\ ] ] these are the desired results . the next step is to study the @xmath82 that appear in the expressions ( [ eq : psi1])([eq : psi3 ] ) for @xmath141 , @xmath142 and @xmath143 with @xmath144 . it should be emphasized that , while the treatment in the preceding sec . iv involves approximations , what is to be carried out in this section is exact . since exact manipulations are usually more straightforward , the description will be relatively more brief in this section . consider the third - order ordinary differential equation @xmath145f=0.\ ] ] a comparison with eq . ( [ eq : schr-6 ] ) shows that @xmath146 the values of @xmath147 and @xmath148 are different for the various @xmath72 s : @xmath149 the three linearly independent solutions of eq . ( [ eq : schr-7 ] ) are @xmath150 , \nonumber \\[5pt ] f^{(2)}(z)&=&z^{-\beta_2 } { } _ 2f_2\left[\left . \tighten\matrix { \alpha_1-\beta_2 , & \alpha_2-\beta_2\cr 1+\beta_1-\beta_2 , & 1+\beta_3-\beta_2 \cr } \right|z \right ] , \nonumber \\[5pt ] f^{(3)}(z)&=&z^{-\beta_3 } { } _ 2f_2\left[\left . \tighten\matrix { \alpha_1-\beta_3 , & \alpha_2-\beta_3\cr 1+\beta_1-\beta_3 , & 1+\beta_2-\beta_3 \cr } \right|z \right],\end{aligned}\ ] ] where @xmath151 is the function treated in the preceding sec . iv . on the other hand , if in eq . ( [ eq : schr-7 ] ) the independent variable is changed to @xmath152 then the differential equation is @xmath153f=0.\ ] ] in this form , two of the three linearly independent solutions of eq . ( [ eq : schr-7 ] ) are ( in terms of @xmath127 ) : @xmath154 , \nonumber \\[5pt ] \hat f^{(2)}(z ) & = & z^{-\alpha_2 } { } _ 3f_1\left[\left . \tighten\matrix { -\beta_1+\alpha_2 , & -\beta_2+\alpha_2 , & -\beta_3+\alpha_2\cr 1-\alpha_1+\alpha_2 & & \cr } \right|-z^{-1 } \right].\end{aligned}\ ] ] for a discussion of the generalized hypergeometric functions @xmath155 with @xmath156 , see for example chapter v of @xcite . it should be added parenthetically that , for the present purposes , @xmath157 , as an example , can sometimes be very small because by eqs . ( [ eq : z - u ] ) and ( [ eq : beta - mu ] ) @xmath158 thus , when @xmath47 is large , which is often the case , for numerical calculations it is convenient to replace @xmath157 by @xmath159 , and similarly for the other powers of @xmath127 multiplying the @xmath82 and @xmath109 . since eq . ( [ eq : schr-7 ] ) is a third - order linear ordinary differential equation , the two solutions as given by eq . ( [ eq : hat - f ] ) must be expressible as linear combinations of the three solutions of eq . ( [ eq : f1-f2-f3 ] ) . the result is @xmath160 & & \mbox{}\times\left [ \frac{\gamma(-\beta_2+\beta_1)\gamma(-\beta_3+\beta_1 ) } { \gamma(1-\alpha_1+\beta_1)\gamma(1-\alpha_2+\beta_1 ) } \,\frac{1}{\sin\pi(\alpha_1-\beta_1)}\ , f^{(1)}(z)\right . \nonumber \\[4pt ] & & \mbox{}+\frac{\gamma(-\beta_1+\beta_2)\gamma(-\beta_3+\beta_2 ) } { \gamma(1-\alpha_1+\beta_2)\gamma(1-\alpha_2+\beta_2 ) } \,\frac{1}{\sin\pi(\alpha_1-\beta_2)}\ , f^{(2)}(z ) \nonumber \\[4pt ] & & \mbox{}+\left.\frac{\gamma(-\beta_1+\beta_3)\gamma(-\beta_2+\beta_3 ) } { \gamma(1-\alpha_1+\beta_3)\gamma(1-\alpha_2+\beta_3 ) } \,\frac{1}{\sin\pi(\alpha_1-\beta_3)}\ , f^{(3)}(z ) \right],\end{aligned}\ ] ] @xmath161 eqs . ( [ eq : hat - f1 ] ) and ( [ eq : hat - f2 ] ) can be considered to be the formulas that express @xmath162 and @xmath163 in terms of @xmath164 , @xmath165 and @xmath151 . since @xmath151 is given by eqs . ( [ eq : f22-stirling-1])([eq : f22-stirling-3 ] ) , it only remains to obtain @xmath164 and @xmath165 , which are given by @xmath109 rather than @xmath82 . the advantage of @xmath109 over @xmath82 is due to the integral representation @xmath166 = \frac{1}{\gamma(a_1)}\,x^{a_1 } \int_0^\infty dt\ , e^{-xt}\ , t^{a_1 - 1}\ , { } _ 2f_1(a_2,a_3;b ;- t),\ ] ] where @xmath167 is the usual hypergeometric function . this eq . ( [ eq : f31-int - rep-1 ] ) follows from eq . ( 8) on p. 214 and eq . ( 1 ) on p. 215 of @xcite , and has the great advantage of not containing in the integrand any gamma function that depends on the variable @xmath168 , because such gamma functions require excessive computer time . for the @xmath167 in the integrand of ( [ eq : f31-int - rep-1 ] ) , the representation @xmath169 & & \mbox{}\times\int_{\cal p } s^{b-1}(1-s)^{c - b-1}(1+ts)^{-a}ds\end{aligned}\ ] ] is convenient , where @xmath170 is the pochhammer contour of fig . 2 . in the numerical evaluation through eqs . ( [ eq : f31-int - rep-1 ] ) and ( [ eq : f21-pochhammer ] ) , it is convenient to deform the contours of integration through the point of stationary phase . this point is discussed in more detail in the appendix . [ fig : pochhammer ] ( 8,7.0 ) ( -3.,2.0 ) the mikheyev - smirnov - wolfenstein effect in the sun has been studied and the differential equations solved for three types of neutrinos coupled through their mass matrix . under the assumption , as expressed by eq . ( [ eq : d - exponential ] ) , that the electron density can be approximated by an exponential function , the msw differential equations are solved exactly in sec . iii , especially eqs . ( [ eq : psi1 - 2 - 3-b])([eq : psi3 ] ) , in terms of the generalized hypergeometric function @xmath82 . the method used there can be immediately generalized to @xmath106 types of neutrinos coupled in the same way , leading to @xmath107 . since @xmath82 can not be considered to be a familiar function , secs . iv and v are devoted to the issue of how they can be calculated . of the nine @xmath82 that appear in eqs . ( [ eq : psi1])([eq : psi3 ] ) , three are evaluated approximately in sec . iv . in sec . v , the remaining six @xmath82 are then expressed in terms of another set of generalized hypergeometric functions , @xmath109 . these @xmath109 have convenient integral representations given by eqs . ( [ eq : f31-int - rep-1 ] ) and ( [ eq : f21-pochhammer ] ) , useful for numerical evaluation . eqs . ( [ eq : f31-int - rep-1 ] ) and ( [ eq : f21-pochhammer ] ) can also be used to get asymptotic expressions for @xmath109 and hence @xmath82 , but this development is not discussed here because it is not needed for the study of neutrino oscillations @xcite . finally , it should be mentioned that , while the calculation of sec . iv can be generalized immediately to @xmath106 types of neutrinos , there does not seem to be a similar straightforward generalization for sec . v. more precisely , while @xmath107 can be related to @xmath171 , the generalization of eq . ( [ eq : f31-int - rep-1 ] ) involves @xmath172 , which does not have an integral representation similar to eq . ( [ eq : f21-pochhammer ] ) involving a single integral . we believe that , for @xmath171 , it is necessary to use an @xmath173-fold integral to avoid having gamma functions that depend on the variables of integration . in this appendix , eqs . ( [ eq : f31-int - rep-1 ] ) and ( [ eq : f21-pochhammer ] ) are studied further . for definiteness , let @xmath147 and @xmath148 be defined by ( [ eq : psi-1-alphas ] ) , and consider the @xmath109 that appears in the first equation in ( [ eq : hat - f ] ) . use the inequality ( [ eq : interlace ] ) to define the four positive quantities @xmath174 then eq . ( [ eq : f31-int - rep-1 ] ) gives , for this @xmath109 under consideration , @xmath175 } \qquad\nonumber \\[5pt ] & & \mbox{}=\frac{1}{\gamma(1+i\xi)}\,\left(e^{i\pi/2}y\right)^{1+i\xi } \int_0^\infty dt\ , e^{-ity}\ , t^{i\xi}\ , { } _ 2f_1(1-i\eta,1+i\zeta;1+i\zeta';-t),\end{aligned}\ ] ] where , by eq . ( [ eq : z - u ] ) , @xmath176 is large for @xmath122 negative and large . the substitution of eq . ( [ eq : f21-pochhammer ] ) into eq . ( [ eq : f31-int - rep-2 ] ) then gives @xmath177}\qquad \nonumber \\[5pt ] & & \mbox{}=\frac{1}{\gamma(1+i\xi)}\,\left(e^{i\pi/2}y\right)^{1+i\xi } \frac{\gamma(1+i\zeta')e^{-i\pi(1+i\zeta ' ) } } { \gamma(1-i\eta)\gamma(i(\eta+\zeta'))4\sin\pi(1-i\eta)\sin\pi i(\eta+\zeta ' ) } \ , i,\end{aligned}\ ] ] where @xmath178 & = & \int_0^\infty dt \int_{\cal p } ds\ , ( 1+s+st)^{-1}\ , e^{-i\phi(t , s)},\end{aligned}\ ] ] with @xmath179 this @xmath180 is of course not related to the @xmath181 of eq . ( [ eq : schr-1 ] ) . for the double integral in eq . ( [ eq : i - t - s ] ) , the point or points of stationary phase are given by @xmath182 or @xmath183 and @xmath184 elimination of @xmath185 between eqs . ( [ eq : stat - cond-1 ] ) and ( [ eq : stat - cond-2 ] ) gives the cubic equation for @xmath168 , @xmath186 with @xmath187 t -\xi(\zeta'-\xi).\ ] ] when @xmath188 , the three roots of eq . ( [ eq : ell - cubic ] ) are explicitly @xmath189^{1/2 } \bigr\ } > 0 , \\ t&=&t_-=(2y)^{-1}\bigl\{-(y-\xi-\eta)-[(y-\xi-\eta)^2 + 4\xi y]^{1/2 } \bigr\ } < 0 , \\ t&=&-(\zeta-\xi)/y < 0.\end{aligned}\ ] ] since by ( [ eq : xi - eta - zeta - zetap ] ) @xmath190 we have @xmath191 it therefore follows that the cubic equation ( [ eq : ell - cubic ] ) has one and only one positive root . since the @xmath168 integral in eq . ( [ eq : i - t - s ] ) is along the positive real axis , this implies that there is exactly one relevant point of stationary phase , say ( @xmath192 ) , with @xmath193 given by this unique positive root of eq . ( [ eq : ell - cubic ] ) . furthermore , since @xmath194 it follows from ( [ eq : t0-cases ] ) that @xmath195 from eq . ( [ eq : stat - cond-1 ] ) , the corresponding @xmath196 is given by @xmath197^{-1},\ ] ] and hence @xmath198 for the numerical evaluation of this @xmath109 of eq . ( [ eq : f31-int - rep-3 ] ) , using eq . ( [ eq : i - t - s ] ) , it is convenient and efficient to choose the pochhammer contour @xmath170 to go through this value of @xmath196 , and also to deform the contour for the @xmath168 integration so that at @xmath193 it is locally the path of steepest descent . l. wolfenstein , phys . d * 17 * , 2369 ( 1978 ) and d * 20 * , 2634 ( 1979 ) ; + s. p. mikheyev and a. yu . smirnov , yad . * 42 * , 1441 ( 1985 ) [ sov . j. nucl . 42 * , 913 ( 1985 ) ] , nuovo cimento * 9c * , 17 ( 1986 ) . for an exponential density , see , for example t. kaneko , prog . phys . * 78 * , 532 ( 1987 ) ; m. ito , t. kaneko and m. nakagawa , _ ibid . _ * 79 * , 13 ( 1988 ) ; + s. toshev , phys . lett . b * 196 * , 170 ( 1987 ) ; + s. t. petcov , phys . b * 200 * , 373 ( 1988 ) ; nucl . b ( proc . suppl . ) * 13 * , 527 ( 1990 ) . see , for example : t. k. kuo and j. pantaleone , phys . lett . * 57 * , 1805 ( 1986 ) ; + c. w. kim and w. k. sze , phys . d * 35 * , 1404 ( 1987 ) ; + h. w. zaglauer and k. h. schwarzer , z. phys . c * 40 * , 273 ( 1988 ) ; + g. l. fogli , e. lisi and d. montanino , phys . d * 49 * , 3626 ( 1994 ) ; + e. torrente lujan , phys . d * 53 * , 4030 ( 1996 ) ; + t. sakai , o. inagaki and t. teshima , int . a , * 12 * , 1953 ( 1999 ) ; + t. ohlsson and h. snellman , royal inst . stockholm , preprint ( oct 1999 ) , hep - ph/9910546 + g. l. fogli , e. lisi , d. montanino and a. palazzo , preprint bari - th/365 - 99 , hep - ph/9912231 . the early papers on the nuclear reactions in the sun are : + c. f. von weizsacker , phys . zeits . ( leipzig ) * 38 * , 176 ( 1937 ) and * 39 * , 633 ( 1938 ) ; + h. a. bethe and c. l. critchfield , phys . rev . * 54 * , 248 ( 1938 ) and _ ibid . _ 862 ; + h. a. bethe , phys . * 55 * , 434 ( 1939 ) .

warped extra dimensions present an elegant solution to the hierarchy problem , where all fundamental parameters are of the order of the planck scale . the weak scale planck scale hierarchy is obtained by an exponential warp factor , which is naturally small provided the higgs field is localized towards the so - called infrared brane @xcite . if all standard model ( sm ) fields propagate in the bulk , the theory leads to the presence of kaluza klein ( kk ) modes which tend to be localized towards the ir brane and therefore couple sizably to the higgs . this in turn leads to large mixing between the heavy sm particles and their kk modes , leading to modifications of the electroweak parameters and therefore to strong constraints from electroweak precision measurements @xcite@xcite . these constraints may be weakened by the introduction of brane kinetic terms @xcite@xcite or custodial symmetries @xcite , @xcite , which allow the presence of kk modes with masses of the order of a few tev . one of the attractive features of these models is the natural explanation of the hierarchy of fermion masses by the localization of the fermion fields in the bulk @xcite@xcite . the chiral properties of the fermions are obtained by imposing an orbifold symmetry and demanding that the fields are odd or even under such a symmetry . fermion fields that are even under the orbifold symmetry at the infrared and ultraviolet branes present zero modes , which are chiral and therefore may be identified with the sm fermion fields . the localization of the zero modes is governed by the bulk mass parameter @xmath2 , with @xmath3 the curvature of the extra dimension and @xmath4 , a number of order one . while the zero modes of chiral fields with @xmath5 couple weakly to the higgs and are therefore light , the heavy sm fields are associated with bulk mass parameters @xmath6 . due to the exponential behavior of the zero mode wave functions , large hierarchies between the fermion masses are generated by small variations of the corresponding @xmath4-parameters . gauge higgs unification models identify the higgs field with the five dimensional component of the gauge fields @xcite . an extended gauge symmetry is necessary for the successful implementation of this mechanism . in particular , models based on the gauge group @xmath0 include the custodial and weak gauge symmetry via @xmath7 @xcite,@xcite @xcite . moreover , provided the @xmath8 symmetry is broken to a subgroup of @xmath9 by boundary conditions at both branes , the fifth dimensional components of @xmath10 have the proper quantum numbers to be identified with the higgs field , which is exponentially localized towards the ir brane . since the higgs originates from gauge fields , its tree level potential vanishes . in a previous work @xcite , we computed the one - loop effective potential and demonstrated that electroweak symmetry breaking , with the proper generation of third generation quark and gauge boson masses may be obtained for the same values of the bulk mass parameters that lead to agreement with precision electroweak data at the one - loop level . moreover , we showed that the higgs mass is in the range 110150 gev and that a light kk mode of the top quark , @xmath11 , appears in the spectrum , with a mass small enough so that the kk gluon may decay into it . the presence of this light kk mode has a strong impact on the phenomenology of the model @xcite . for instance , searches for the kk gluon by its decay into top - quarks @xcite,@xcite , @xcite is rendered difficult due to the presence of the additional dominant decay mode into kk top - quarks and the broadness of the kk gluon . on the contrary , the constructive interference between the qcd and kk gluon induced pair production of the top - quark kk mode allows to search for a @xmath11 for values of the masses much larger than those at reach in the case of just the qcd production cross section . in this article , we will analyze the addition of the lepton sector in the gauge higgs unification scenario . to add leptons , we will proceed in a similar way as for the quark sector . the left - handed leptons will be added in a fundamental representation of @xmath8 , with @xmath12 , while the right - handed neutrino and charged lepton fields will be added in a fundamental representation and a @xmath13 of @xmath8 also with charge @xmath14 , respectively . due to the gauge origin of the higgs field , a possible local infrared brane operator @xmath15 , which could lead to large values of the neutrino majorana masses , should come from the fifth dimensional component of the covariant derivative of the lepton fields and therefore can only naturally arise from the integration of the right - handed neutrinos , with a local ir majorana mass . indeed , since the fields associated with the right - handed neutrino zero modes are singlets under the conserved gauge groups on both the infrared and ultraviolet branes , one can always add majorana masses for these fields on the infrared and ultraviolet branes . we will therefore consider these masses and implement a see - saw mechanism for the generation of the light neutrino masses @xcite,@xcite . we will show how to incorporate these masses within the context of these models , obtain the modified profile functions and define the conditions to derive a realistic spectrum . furthermore , the introduction of an exchange symmetry @xcite , which is preserved in the bulk , yields a natural dark matter candidate in the spectrum , that may be identified with the odd kk modes of the right - handed neutrino . this exchange symmetry requires the identification of the bulk mass parameter of the dark matter candidate with the one of the right - handed neutrinos , establishing a connection between neutrino masses and the relic dark matter density . we will show that , if the odd fermions are assumed to be dirac particles , the predicted relic density is the correct one for dirac fermion masses of the order of 1 tev . in the majorana case , somewhat smaller masses are allowed , and the results depend on the relative values of the ultraviolet and infrared majorana masses . we will show that in both the dirac and majorana cases , direct detection experiments will efficiently probe the existence of the proposed dark matter candidates . the article is organized as follow : in section 2 we discuss the properties of leptons in scenarios of gauge higgs unification . in section 3 we discuss the generation of charged lepton and neutrino masses . in section 4 we analyze the possibility of incorporating dark matter via an exchange bulk symmetry , analyzing the couplings and the annihilation diagrams , as well as the direct detection of these dark matter candidates . we reserve section 5 for our conclusions . some technical details are given in the appendix . the goal of the current work is to add the lepton sector into the minimal gauge higgs unification model described above , including right handed neutrinos , with brane majorana masses . we will proceed in a similar way as in the quark case @xcite , and discuss the general question of charged lepton and neutrino mass generation . similar to the quark sector , we let the sm @xmath16 lepton doublet containing the left - handed charged leptons and neutrinos , @xmath17 and @xmath18 , arise from a @xmath19 of @xmath20 , where the subscript refers to the @xmath21 charge . the right handed neutrino , @xmath22 will be included as the singlet component in a fundamental representation of @xmath8 , while the right - handed charged lepton , @xmath23 , is placed in a @xmath24 of @xmath8 analogously to the @xmath25 . one may also include brane mass terms connecting different multiplet components , as well as new brane majorana masses for the right handed neutrino @xmath22 : @xmath26 the right handed neutrino , @xmath22 , in principle could be identified with the singlet right - handed component in the same multiplet as the left - handed leptons . however , in order to naturally suppress lepton flavor violation effects and maintain agreement with precision electroweak measurements , the left - handed leptons should be localized towards the uv brane @xcite,@xcite . the zero modes of the corresponding right - handed multiplet will therefore be localized towards the ir brane . this implies that even with a natural scale for the brane masses @xmath27 , the exponential suppression of the wave function at the uv brane would lead to a effective majorana mass for @xmath22 which is much smaller than the planck scale . this , in turn , after the implementation of the see - saw mechanism , leads to too large values for the observed neutrino masses . therefore , as we will discuss below , it will prove to be necessary to have the left - handed leptons in a different multiplet as the right - handed neutrinos . if @xmath22 belongs to the same multiplet as the left - handed leptons : @xmath28 @xmath29 alternatively , the right - handed neutrino can be incorporated in a different multiplet from the left - handed lepton one . the two multiplet assignments are as follows : @xmath30 @xmath31 where we show the decomposition under @xmath32 , and explicitly write the @xmath33 charges . the @xmath34s are bidoublets of @xmath32 , with @xmath35 acting vertically and @xmath1 acting horizontally . the @xmath36 s and @xmath37 s transform as @xmath38 and @xmath39 under @xmath32 , respectively , while @xmath40 and @xmath41 are @xmath32 singlets . the superscripts , @xmath42 , label the three generations . we also show the parities on the indicated 4d chirality , where @xmath43 and @xmath44 stands for odd and even parity conditions and the first and second entries in the bracket correspond to the parities in the uv and ir branes respectively . let us stress that while odd parity is equivalent to a dirichlet boundary condition , the even parity is a linear combination of neumann and dirichlet boundary conditions , that is determined via the fermion bulk equations of motion as discussed below . the boundary conditions for the opposite chirality fermion multiplet can be read off the ones above by a flip in both chirality and boundary condition , for example @xmath45 . in the absence of mixing among multiplets satisfying different boundary conditions , the sm fermions arise as the zero - modes of the fields obeying @xmath46 boundary conditions . the remaining boundary conditions are chosen so that @xmath47 is preserved on the ir brane and so that mass mixing terms , necessary to obtain the sm fermion masses after ew symmetry breaking , can be written on the ir brane . consistency of the above parity assignments with the original orbifold @xmath48 symmetry at the ir brane was discussed in appendix b of ref . the three families will behave similarly , and therefore , we will drop the family indices and concentrate only on one lepton family . large mixing angles in the lepton sector can be naturally obtained while suppressing lepton flavor changing neutral currents if the left - handed leptons have similar bulk mass parameters , @xmath49 @xcite,@xcite , and in the following we will assume them to be equal . we will return to this issue in section [ sec : flavor ] . the zero modes of the leptons are too light and too weakly coupled to the higgs boson to affect the higgs potential in any significant way . the lepton kk modes may be coupled more strongly to the higgs , but their gauge invariant mass is much larger than the higgs induced one and therefore they tend to contribute only weakly to both the higgs potential and to precision electroweak observables . one can add localized brane mass terms to the lagrangian in both the one and the two multiplet cases : @xmath50 - \big[m_{ir}\delta(x_5-l)-m_{uv}\delta(x_5)\big]n_r n_r~;\label{localizedmasses_1}\\ \mathcal{l}_{2 } & = & -2\delta(x_{5}-l ) \big [ \bar{n}^\prime_l m_{l_1 } n_r + \bar{l}_{1l } m_{l_2 } l_{3r } + \mathrm{h.c . } \big ] - \big[m_{ir}\delta(x_5-l)-m_{uv}\delta(x_5)\big]n_r n_r~.\label{localizedmasses}\end{aligned}\ ] ] with the introduction of the brane mixing terms , the different multiplets are now related via the equations of motion . the fermions , like the gauge bosons , can be expanded in their kk basis : @xmath51 solving the equations of motion in the presence of @xmath52 becomes complicated , as the different modes are mixed . however , 5d gauge symmetry relates these solutions to solutions with @xmath53 @xcite , with @xmath54 , the gauge transformations that removes the vev of @xmath52 : @xmath55 .\ ] ] the form of the bulk profile functions at @xmath56 is given in appendix a. the boundary masses lead to a redefinition of the effective boundary conditions for the fermion fields at the branes . let s first analyze the case of a dirac boundary mass term on the infrared brane , @xmath57 , involving fields from different multiplets @xmath58 , @xmath59 for the different multiplets . quite generally , we shall call the profile of the left - handed field and the right - handed field participating in the mass term @xmath60 and @xmath61 , respectively . the right handed component , @xmath62 with profile function @xmath63 and the left - handed component @xmath64 with profile function @xmath65 have dirichlet boundary conditions on the brane , and therefore @xmath66 . the equations of motion of the fields are affected by the localized masses , which induce a discontinuity on the odd - parity profile functions at the infrared brane . indeed , keeping only the relevant terms , the integration of the equation of motion leads to @xmath67 therefore , we obtain : @xmath68 and , similarly for the @xmath65 @xmath69 eq . ( [ limit ] ) and ( [ limit2 ] ) can now be reinterpreted as the new boundary conditions for the profiles at the ir brane . analogously , one can analyze the effect of the majorana boundary mass , @xmath70 , where @xmath71 or @xmath72 . let s take the specific case of the field @xmath22 , with a profile function @xmath73 . its chiral partner will have a profile function @xmath65 having an odd parity profile on both branes . the equation of motion in the presence of both the majorana masses and the dirac mass term @xmath74 leads to the following relationship @xmath75dx_5.\ ] ] where the minus and plus signs are associated with the boundary conditions at @xmath76 , and @xmath77 , respectively . the odd parity at the branes then implies a dirichlet boundary condition for the function @xmath78 , which as before will present a discontinuity at the brane . for the ir boundary conditions we obtain : @xmath79 for the uv boundary condition , instead : @xmath80 eq . ( [ limit2_ir]-[limit2_uv ] ) can now be reinterpreted as the new boundary condition for the profiles at the branes . the generalization of these expressions to the general case is straightforward . the wave functions defined in appendix a , @xmath81 and @xmath82 are associated with dirichlet boundary conditions at the ultraviolet brane for the left - handed and right - handed fields , respectively . in the presence of ultraviolet majorana masses , however , the boundary conditions for the singlet component of the fundamental multiplets of @xmath8 read @xmath83 where @xmath84 refers to the particular multiplet under analysis . in order to derive the profile function of these fields , let us first redefine the functions @xmath85 where @xmath86 . with this redefinition , these functions satisfy the naive normalization condition , @xmath87 the general solution for @xmath88 is given by : @xmath89 where @xmath90 is the associated particle mass . defining @xmath91 @xmath92 satisfy the simple equation of motion , @xmath93 now , using eq . ( [ flfr ] ) , one obtains @xmath94 the second derivative functions may be replaced by means of the equation of motion of the fermion fields , namely @xmath95 we therefore see that @xmath96 reduces to : @xmath97 rewriting these in terms of the @xmath98 rather than @xmath99 , with @xmath100 , we obtain @xmath101 to solve for @xmath102 in terms of @xmath103 , we need to use the uv boundary condition induced by the uv majorana mass for @xmath22 , eq . ( [ uv_cond ] ) : @xmath104 therefore , with the coefficients @xmath102 and @xmath103 as calculated above , the singlet functions become @xmath105 in the case of a single multiplet containing the left- and right - handed neutrinos , and @xmath106 in the case of two multiplets , where @xmath107 and @xmath108 . therefore , the fermion multiplets with @xmath56 take the form @xmath109\\ \\ \end{array } & & f_{3,r}(x_{5},0)=\left[\begin{array}{c } c_{11}s_{-m_{3}}\\ c_{12}s_{-m_{3}}\\ c_{13}s_{-m_{3}}\\ c_{14}s_{-m_{3}}\\ c_{15}s_{-m_{3}}\\ c_{16}s_{-m_{3}}\\ c_{17}s_{-m_{3}}\\c_{18}s_{-m_{3}}\\ c_{19}s_{-m_{3}}\\ c_{20}\dot{s}_{m_{3}}\end{array}\right]\end{aligned}\ ] ] in the case of a single multiplet containing the left - handed and right - handed neutrinos , and @xmath110\\ \\ \\ f_{2,l}(x_{5},0)=\left[\begin{array}{c } c_{6}\dot{s}_{-m_{2}}\\ c_{7}\dot{s}_{-m_{2}}\\ c_{8}\dot{s}_{-m_{2}}\\ c_{9}\dot{s}_{-m_{2}}\\ f^5_{2,l}\end{array}\right ] \end{array } & & f_{3,r}(x_{5},0)=\left[\begin{array}{c } c_{11}s_{-m_{3}}\\ c_{12}s_{-m_{3}}\\ c_{13}s_{-m_{3}}\\ c_{14}s_{-m_{3}}\\ c_{15}s_{-m_{3}}\\ c_{16}s_{-m_{3}}\\ c_{17}s_{-m_{3}}\\c_{18}s_{-m_{3}}\\ c_{19}s_{-m_{3}}\\ c_{20}\dot{s}_{m_{3}}\end{array}\right]\end{aligned}\ ] ] in the two multiplet case , where the @xmath111 are normalization constants . applying the boundary conditions at @xmath112 , taking into account the mass mixing terms from eqs . ( [ localizedmasses_1 ] ) and ( [ localizedmasses ] ) and using the procedure defined in eqs . ( [ limit])([limit2_ir ] ) , we derive the conditions on the lepton wave functions @xmath113 in the presence of the higgs field . in the case of only one multiplet containing both the left - handed and right - handed neutrinos one gets the following conditions at the ir brane : @xmath114 in the two multiplets , instead , one obtains : @xmath115 where the superscripts denote the vector components . this defines a system of linear equations for the normalization constants @xmath111 . asking that the determinant of the functional coefficients of this system vanishes in order to get a non - trivial solution @xcite , one obtains the following relations : @xmath116-m_{l_2}^2\dot{\tilde{s}}_{-m_1}\sin\left[\frac{\lambda h}{f_h}\right]^2=0&\label{lepton}\\ & & \nonumber\\ & 2\left[-m_{l_2}^4\tilde{s}_{m_1}\tilde{s}_{-m_3}^2\left[\dot{\tilde{s}}_{-m_1}(\tilde{s}_{m_1}-e^{2c_1kl}m_{uv}\dot{\tilde{s}}_{-m_1})+m_{ir}(1-\tilde{s}_{m_1}\tilde{s}_{-m_1}+e^{2c_1kl } m_{uv}\tilde{s}_{-m_1}\dot{\tilde{s}}_{-m_1})\right]\right . & \nonumber\\ & -m_{l_2}^2\tilde{s}_{-m_3}\left[2\tilde{s}_{m_1}^2\tilde{s}_{-m_1}-\tilde{s}_{m_1}\left[1+m_{ir}\tilde{s}_{-m_1}\dot{\tilde{s}}_{m_1}-e^{2c_1kl } m_{uv}\tilde{s}_{-m_1}(2m_{ir}\tilde{s}_{-m_1}-\dot{\tilde{s}}_{-m_1})\right]\right.&\nonumber\\ & \left .- m_{ir}\dot{\tilde{s}}_{m_1}^2\dot{\tilde{s}}_{-m_1}-e^{2c_1kl } m_{uv}(m_{ir}\tilde{s}_{-m_1}+\dot{\tilde{s}}_{m_1}\dot{\tilde{s}}_{-m_1}^2)\right]\dot{\tilde{s}}_{-m_3}&\nonumber\\ & \left.-\tilde{s}_{-m_1}\left[\dot{\tilde{s}}_{m_1}(\tilde{s}_{m_1}-m_{ir}\dot{\tilde{s}}_{m_1})+e^{2c_1kl } m_{uv}(1-\tilde{s}_{m_1}\tilde{s}_{-m_1}+m_{ir}\tilde{s}_{-m_1}\dot{\tilde{s}}_{m_1})\right]\dot{\tilde{s}}_{-m_3}^2\right]&\nonumber\\ & + ( m_{l_2}^2m_{ir}\tilde{s}_{-m_3}+\dot{\tilde{s}}_{-m_3})\left[m_{l_2}^2\tilde{s}_{-m_3}\left[\tilde{s}_{m_1}+e^{2c_1kl } m_{uv}\dot{\tilde{s}}_{-m_1}(\tilde{s}_{m_1}\tilde{s}_{-m_1}-\dot{\tilde{s}}_{m_1}\dot{\tilde{s}}_{-m_1})\right]\right.&\nonumber\\ & + \left.\left[e^{2c_1kl } m_{uv}\tilde{s}_{-m_1}+\dot{\tilde{s}}_{m_1}(\tilde{s}_{m_1}\tilde{s}_{-m_1}-\dot{\tilde{s}}_{m_1}\dot{\tilde{s}}_{-m_1})\right]\dot{\tilde{s}}_{-m_3}\right]\sin\left[\frac{\lambda h}{f_h}\right]^2=0&\label{neutrino1}\end{aligned}\ ] ] in the case of a singlet neutrino multiplet , and @xmath117 ^ 2=0&\label{exotic3}\\ & & \nonumber\\ & 2\tilde{s}_{m_3}\left[m_{l_2}^2\tilde{s}_{-m_3}\dot{\tilde{s}}_{-m_1}+\tilde{s}_{-m_1}\dot{\tilde{s}}_{-m_3}\right]-m_{l_2}^2\dot{\tilde{s}}_{-m_1}\sin\left[\frac{\lambda h}{f_h}\right]^2=0&\label{lepton2}\\ & & \nonumber\end{aligned}\ ] ] @xmath118&\nonumber\\ & \tilde{s}_{-m_1}\left[m_{l_1}^2\tilde{s}_{m_1}(1-\tilde{s}_{m_2}\tilde{s}_{-m_2}+e^{2c_2kl } m_{uv}\tilde{s}_{-m_2}\dot{\tilde{s}}_{-m_2})-\dot{\tilde{s}}_{m_1}\left[\dot{\tilde{s}}_{-m_2}(\tilde{s}_{m_2}-e^{2c_2kl } m_{uv}\dot{\tilde{s}}_{-m_2})\right.\right.&\nonumber\\ & \left.\left.\left.+m_{ir}(1-\tilde{s}_{m_2}\tilde{s}_{-m_2}+e^{2c_2kl } m_{uv}\tilde{s}_{-m_2}\dot{\tilde{s}}_{-m_2})\right]\right]\dot{\tilde{s}}_{-m_3}\right]&\nonumber\\ & + \left[m_{l_2}^2\tilde{s}_{-m_3}\left[-2m_{l_1}^2\tilde{s}_{m_1}\dot{\tilde{s}}_{-m_1}-\tilde{s}_{m_2}\dot{\tilde{s}}_{-m_2}+e^{2c_2kl } m_{uv}\dot{\tilde{s}}_{-m_2}^2\right.\right.&\nonumber\\ & \left.+m_{ir}(-3 + 2\tilde{s}_{m_1}\tilde{s}_{-m_1}+\tilde{s}_{m_2}\tilde{s}_{-m_2}-e^{2c_2kl } m_{uv}\tilde{s}_{-m_2}\dot{\tilde{s}}_{-m_2})\right]&\nonumber\\ & \left.+\left[2m_{ir}\tilde{s}_{-m_1}\dot{\tilde{s}}_{m_1}-m_{l_1}^2 ( 1 + 2\tilde{s}_{m_1}\tilde{s}_{-m_1}-\tilde{s}_{m_2}\tilde{s}_{-m_2}+e^{2c_2kl } m_{uv}\tilde{s}_{-m_2}\dot{\tilde{s}}_{-m_2})\right]\dot{\tilde{s}}_{-m_3}\right]\sin\left[\frac{\lambda h}{f_h } \right]^2&\nonumber\\ & + \left[m_{l_2}^2 m_{ir}\tilde{s}_{-m_3}+m_{l_1}^2\dot{\tilde{s}}_{-m_3}\right]\sin\left[\frac{\lambda h}{f_h } \right]^4=0&\label{neutrino2}\end{aligned}\ ] ] in the case of two multiplets . in the above , for simplicity , we did not write the dependence on @xmath119 and @xmath90 and furthermore , we have used the crowian : [ e.cr ] -_m(x_5,z)_-m(x_5,z ) + _ m(x_5,z ) _ -m(x_5,z ) = 1 . the roots of the above equations define the values of @xmath90 corresponding to the masses of the lepton zero modes and kk modes in the presence of the higgs fields . since the charged lepton masses are given by the mixing of the first and third multiplet via @xmath120 , the expression determining its mass is formally the same for the case in which the right - handed neutrino is in the same multiplet as the left - handed one as for the two neutrino multiplet case ( eqs . ( [ lepton ] ) and ( [ lepton2 ] ) ) . additionally , since the lepton masses are much smaller then @xmath121 , one can use an expansion of @xmath81 for small values of @xmath122 . as we shall show , the approximate functions we derive in this way agree very well with the full numerical investigation we carried out . we shall concentrate on values of @xmath123 , which are preferred to cancel flavor violation effects in a natural way and provide agreement with precision electroweak data @xcite,@xcite . at small values of @xmath90 compared to @xmath121 , one can express the function @xmath81 in the form @xcite : @xmath124 using this in eq . ( [ lepton ] ) , we can solve for the mass : @xmath125\sqrt{2 \ ; ( \frac{1}{2}-c_1)(c_3 ^ 2-\frac{1}{4})}}{\sqrt { \left[(\frac{1}{2}-c_3)(e^{2(c_1-\frac{1}{2})kl}-1)-m_{l_2}^2 ( c_1-\frac{1}{2})(e^{2(c_3-\frac{1}{2})k}-1)\right ] ( e^{2(\frac{1}{2}+c_3)kl}-1)}}.\ ] ] if @xmath126 and @xmath127 , this reduces to : @xmath128\sqrt{2 \ ; ( c_1-\frac{1}{2})(c_3+\frac{1}{2})}\ ] ] for @xmath126 and @xmath129 , instead , @xmath130\sqrt{2 \ ; ( c_3 ^ 2-\frac{1}{4})}\ ] ] finally , for the case @xmath126 and @xmath131 : @xmath132\sqrt{2 \;(\frac{1}{2}-c_1)(c_3+\frac{1}{2})}\ ] ] where we have assumed that @xmath133 . we note that in the above , the lepton masses depend at most linearly on @xmath120 . as shown in fig . [ figure : lepmass ] , the above relations are verified by our numerical results . realistic lepton masses may be obtained for e.g. for @xmath134 and @xmath135 , @xmath136 and @xmath137 for the case of the tau , muon and electron , respectively . if a common value of @xmath138 for the three generations is demanded , as explained above , and for values of @xmath120 of order one , as chosen in fig . [ figure : lepmass ] , the value of @xmath138 is restricted to be in the range @xmath139 . larger values of @xmath138 become incompatible with the heavier charged lepton masses . [ 1 ] the neutrino masses are analyzed in a similar manner . first we look at the case in which the left - handed and right - handed neutrinos belong to the same muliplet case . for @xmath126 , @xmath140 ^ 2}{m_{ir}}\ ] ] from eq . ( [ 1mn2 ] ) we see that values of @xmath141 would be necessary in order to get the correct values for the neutrino masses . however , values of @xmath142 are strongly constrained in order to reproduce the proper charged lepton masses . indeed , as we emphasized at the end of last section , the proper values of @xmath143 and @xmath144 masses may not be obtained for @xmath141 . therefore , we conclude that if all @xmath138 s are about the same , as preferred to obtain large flavor mixing naturally without inducing large lepton flavor changing effects @xcite , two multiplets are required in order to obtain the correct lepton spectrum . in the two multiplet case , the dependence of the neutrino masses on the mixing with the third multiplet through @xmath120 is always exponentially suppressed . therefore , we shall set @xmath145 in the following approximate expressions . the approximate mass expressions , for * @xmath126 and @xmath146 : + @xmath147 ^ 2}{m_{ir}}\ ] ] * @xmath126 and @xmath148 : + @xmath149 ^ 2}{m_{uv}}\ ] ] * @xmath126 and @xmath150 : + @xmath151 ^ 2}{m_{uv } + m_{ir}}\ ] ] where in eq . ( [ 2mn3 ] ) we have assumed @xmath152 . in the linear regime , @xmath153 , these neutrino masses become proportional to the square of the higgs vev , and show the characteristic see - saw behavior governed by the brane majorana masses . from the above expressions we see that it will only be possible to generate the correct order of the neutrino masses when @xmath154 . moreover , for @xmath155 , the values of @xmath138 are such that the correct heavy lepton masses can not be generated . these conclusions are verified in our numerical work . we present the relevant parameter space in the @xmath156 plane leading to the correct order of the neutrino masses in fig . [ fig.n1 ] . the width of the bands for the different masses corresponds to varying @xmath121 and the different brane masses in the range indicated in fig . [ fig.n1 ] . as indicated by the above expressions , we were not able to numerically find any solutions for @xmath157 . finally , although positive values of @xmath158 are not represented in fig . [ fig.n1 ] , the neutrino masses become independent of @xmath158 for @xmath159 , and therefore the values of @xmath138 are the same as for @xmath160 . [ 0.8 ] in the above , we have not discussed the problem of flavor . it is well known that , if the effective yukawa couplings have an anarchic structure , large flavor violating effects are induced , which may only be suppressed by pushing the kk masses to values above 10 tev , excluding any possible phenomenology of warped extra dimensional models at the lhc , as well as any possible dark matter candidate coming from the kk modes ( see , for example ref . @xcite as well as ref . @xcite for an alternative approach to this question ) . the problem stems in part from the fact that the yukawa couplings and the bulk mass parameters are not diagonalized in the same basis , and therefore the quark mass eigenstates have flavor violating couplings with the gluon kk modes . a possible solution to this problem is to demand an alignment between the bulk mass parameters and the yukawa couplings , as has been proposed in ref . flavor violation in the lepton sector can also be suppressed by a similar alignment @xcite,@xcite . this is equivalent to demanding that the bulk mass parameters obey the following relationships : @xmath161 where @xmath162 and @xmath163 are the effective charged lepton and neutrino yukawa couplings and @xmath164 , @xmath165 , @xmath166 and @xmath167 are numerical constants and the @xmath168 are now matrices where the off - diagonal terms mix the different generations . the gauge - higgs unification structure introduced above demands a redefinition of the above equations , since no explicit yukawa coupling has been written . as can be seen from eqs . ( [ c3g0.5lc1])([c3l0.5 ] ) , and eqs . ( [ 2mn1])([2mn3 ] ) , the role of the yukawa coupling is now being played by the boundary masses @xmath74 and @xmath120 . hence , the above equations must be rewritten as @xmath169 if @xmath170 , the charged lepton masses would be diagonalized in the same basis as the bulk mass parameters , inducing minimal flavor violation in the lepton sector . in this case , all flavor violation will be associated with the charged currents , leading to values of the lepton flavor violation processes consistent with experiment for kk masses as low as a few tev . as emphasized above , large mixing angles may naturally arise within this framework , if all left - handed zero modes localization parameters take equal values , namely when @xmath171 @xcite . the results of refs . @xcite,@xcite show that it is possible to impose a flavor symmetry in this class of models such that no large flavor violation occurs . in this work , we will assume that this , or a similar @xcite,@xcite,@xcite flavor protection exists , and will postpone a more detailed analysis of this question for future work . dark matter in warped extra dimensions was first introduced in ref . @xcite within a framework which solves the proton stability problem in unification scenarios . the introduction of a kk parity in warped extra dimensions , leading to a stable dark matter candidate , was further proposed in ref . @xcite . in this work , we shall proceed in a different way : following ref . @xcite , we shall introduce an additional exchange @xmath48 symmetry under which all the lepton multiplets introduced so far would be even . one can then define extra fermion multiplets , that will be chosen as the `` odd '' partners of the lepton multiplets . if this symmetry is preserved , the lightest odd particle ( lop ) will be stable , and therefore can be considered as a possible dark matter candidate . in the framework of ref . @xcite the equality of the even and odd mass parameters was enforced by giving the original fermions , whose even and odd combinations form the even and odd fields , different charges under an extended @xmath172 gauge symmetry . since in our case the leptons are neutral under @xmath21 this property does not hold . additionally , contrary to ref . @xcite , we shall assume that the quark and gauge boson multiplets do not have odd partners . even though , the structure of our model does not require the equality of the bulk masses for the odd and even fields , for simplicity , we shall assume that the bulk mass parameters are identified with each other and are controlled by the requirement of obtaining the correct small neutrino masses via the see - saw mechanism . our assumption is equivalent to requiring that there are no off - diagonal bulk mass parameters mixing the original fields for which the @xmath48 exchange symmetry holds . in order to explore this possibility , we shall identify the multiplet containing the dark matter candidate with the odd partners of the second lepton multiplet containing the right - handed neutrinos . as has been shown in the previous section , this demands values of @xmath173 , and therefore we shall require the bulk mass parameter of the dark matter candidate to be in this range . the exchange symmetry , introduced in ref . @xcite allows arbitrary boundary masses between even fields , necessary for obtaining the proper lepton masses , as well as between the odd fields . boundary masses mixing odd and even fields are , instead , forbidden by this symmetry . on the other hand , the boundary conditions for the even and odd fields are independent of each other . the main link between even and odd fields is through the identification of the bulk mass parameters . for simplicity , we shall choose the boundary conditions of the odd partners of the chiral first and third lepton multiplets , containing the left - handed and right - handed charged leptons , to be the same as the one of the even biodoublet components , ( - + ) and ( + - ) , respectively . for values of the bulk masses @xmath174 and @xmath175 , these fields will be relatively heavy , with masses about a few times @xmath121 , and decoupled from the higgs . for the second multiplet , the odd fields , denoted by @xmath176 will have the same boundary conditions as @xmath177 for the even bidoublets . however , since the fifth component does not contain a zero mode , it must have different boundary conditions on the ir and uv branes . that leaves us with two options for the left handed component of this field : @xmath178 or @xmath179 . the goal here is to consider the possibility of a neutral odd lepton , mainly singlet under the @xmath180 symmetry , as a dark matter candidate . the singlet and the doublet states mix via their interactions with the higgs field , which will act as a small perturbation to their masses . in order to make the coupling to the higgs effective and to split the doublet and singlet masses , we shall choose the singlet right - handed field to have the same boundary conditions on the ir brane as the bidoublet left - handed field for at least one of the three odd partners of the second multiplets . therefore , the boundary conditions for the odd multiplet containing the lop are chosen to be : @xmath181 regarding the other two generations of odd partners of the second multiplets , for simplicity , we will choose their singlet states to have the opposite boundary conditions from the one presented in eq . ( [ oddmult ] ) . this would force their masses to be heavy for @xmath182 , ensuring that the multiplet with the boundary conditions given by eq . ( [ oddmult ] ) , @xmath176 , would generate the lop . in addition , small dirac boundary masses may be included , which would allow a small mixing between the odd multiplets inducing decays of the heavier generation odd states to the lop , through the weak gauge bosons and the higgs boson . even in the case of a very small mixing , due to the large mass differences , the lifetime of these heavy odd partners would naturally be very short , and therefore these heavier odd multiplets would not contribute to the lop relic density in any relevant way . finally , in order to estimate the dark matter density , we shall restrict ourselves to the first level of odd kk modes , since they give the dominant contribution to the annihilation cross section . this is due not only to their relatively small masses with respect to the heavier modes in the tower , but also due to their larger couplings to the lop . we have checked that the inclusion of the second kk level leads to a very small modification of the annihilation cross section and therefore of the freeze - out temperature and the predicted relic density . let us finally comment that in the region consistent with the proper dark matter density , the mixing between the singlet and doublet particles is small and these particles lead to only a small contribution to the higgs effective potential and the precision electroweak observables . similar to the standard model fields discussed in the previous section , the boundary conditions , eq . ( [ oddmult ] ) , lead to a set of equations which determine the masses of our odd multiplet fermions . for the kk modes that couple to the higgs boson , in the case of vanishing majorana masses for the odd fields , we find the following condition @xmath183 ^ 2+\dot{\tilde{s}}_{m_2}\dot{\tilde{s}}_{-m_2}=0.\ ] ] [ 0.65 ] the solutions for the lop are plotted in fig . [ fig.cm1d ] . the behavior of the lop mass may be understood from the @xmath56 limit . in this limit , the singlet and doublet states do nt mix and the singlet becomes the lightest odd particle for @xmath184 , while the doublet becomes the lop for @xmath185 . at @xmath186 and @xmath187 , the singlet and doublet masses are degenerate . when @xmath188 the singlet and doublet states mix and their masses are split by the higgs v.e.v . only the lightest state mass is plotted in fig . [ fig.cm1d ] . in the presence of the higgs , at @xmath160 , the lop is an equal admixture of the singlet and doublet state . as we move away from @xmath186 , the roots of the determinant will start splitting into two clearly spaced masses . for @xmath189 , the lighter mass is mostly a singlet state and the heavier one is mostly a doublet state . since these are dirac particles , the positive and negative roots of the determinant are equal . in the above , we have not considered the impact of majorana mass terms that could in principle be written for this multiplet , both on the ir and the uv branes , as was done for the even neutrinos , and would modify the couplings to the higgs boson . including the majorana masses for the odd multiplet , the equation determining the odd lepton masses is given by @xmath190 ^ 2 = 0&\nonumber\\ & & \end{aligned}\ ] ] [ 0.75 ] the effect of introducing the majorana mass terms can be seen in the different negative and positive masses as one moves away from @xmath191 . due to the different behavior of the functions @xmath192 and @xmath98 , by inspection , we can see that the positive and negative roots of eq . ( [ odd_maj ] ) will no longer be equal . the two dirac states have been split into four majorana states . these states can still be recognized as two mostly singlet and two mostly doublet states by comparing their masses to the charged states . in the following , we will sometimes refer to these states as singlet or doublet , where it should be understood that these are not really the original states , but those mixed by the higgs . without the mixing , the coupling between the singlet - singlet and the doublet - doublet states and the higgs would vanish . therefore , we expect the singlet - singlet coupling to be suppressed compared with the coupling of the mostly singlet state to the mostly doublet states . additionally , looking at fig . [ fig.cm1muv ] we see that as expected , the majorana masses do nt effect the mass of the lop when the lop is mainly a doublet state ( purple circles for @xmath159 ) , as the convergence of the different curves corresponding to the different values of @xmath193 and @xmath194 clearly show . the behavior of the lop mass as well as its coupling to the higgs may be studied by looking at the roots of eq . ( [ odd_maj ] ) . using the small @xmath90 expansion for @xmath195 , we obtain @xmath196 ^ 2}{e^{2 c_2 k l}m_{ir_o}+4\frac{(\frac{1}{2}+c_2)}{(\frac{1}{2}-c_2)}m_{uv_o}}~ , \label{maj_app}\ ] ] which is valid for the case in which at least one of the majorana masses is non - vanishing and @xmath184 , in which the singlet becomes the lop . this clearly shows that when only one of the majorana masses is non - zero , we get a see - saw effect governing the lop mass . this behavior is confirmed in our numerical analysis plotted in fig . [ fig.cm1muv ] . observe that if only the ultraviolet mass is non - vanishing , the mass of the lop ( which is mainly a left - handed singlet ) is exponentially suppressed , unless @xmath197 . indeed , in the limit of vanishing infrared majorana masses , eq . ( [ maj_app ] ) reduces to @xmath198 ^ 2 } { 2 m_{uv_o}}~. \label{maj_appmuv}\ ] ] as can be seen from eq . ( [ maj_appmuv ] ) , the higher operator coupling of the lop to the higgs is also suppressed for @xmath184 . we will show that the annihilation cross section for a mainly singlet state is sufficiently enhanced only when the s - channel higgs diagram becomes sizable and therefore , unless @xmath197 , the dark matter density becomes very large compared to the experimentally observed value . as both @xmath193 and @xmath194 are turned on , we see an abrupt change in the behavior of the mass spectrum for @xmath189 , which becomes independent of the exact value of @xmath194 and only depends on @xmath121 and @xmath193 , @xmath199 the lop mass in this case is of the order of @xmath121 , and does not show an explicit dependence on the higgs vaccum expectation value . indeed , as can be seen from eq . ( [ maj_app ] ) , the effective coupling to the higgs for @xmath200 is exponentially suppressed . therefore , as happens in the case of vanishing @xmath193 , a good dark matter candidate may only be obtained for values of @xmath201 . when @xmath150 , the mass can be approximated by : @xmath202 ^ 2}{e^{2 c_2 k l}m_{ir_o}+4m_{uv_o}}\ ] ] we see that for @xmath203 , for values of @xmath52 in the linear regime and very small values of @xmath158 , we can get a cancelation resulting in very small lop masses . this behavior is clearly portrayed in fig . [ fig.cm1muv ] . finally , let us analyze the case @xmath204 and @xmath205 . as @xmath193 is turned on , it strongly modifies the spectrum with respect to the dirac case for negative values of @xmath158 . in this case , the lop becomes mostly a right - handed singlet and its mass is given by @xmath206 ^ 2 } { m_{ir_o}}~. \label{maj_appmir}\ ] ] as seen in eq . ( [ maj_appmir ] ) , the lop mass in this case is , again , of the order of the weak scale but with an explicit higgs v.e.v . dependence induced by a higher order operator coupling with a characteristic scale of the order of the kk masses . therefore the coupling of the lop to the higgs becomes sizable for kk masses of the order of the tev scale , allowing for the possibility of a dark matter candidate for @xmath184 . in the following section , we shall perform a more precise quantitative analysis of the masses and couplings associated with the annihilation cross section of the singlet state for both the majorana and dirac cases . to calculate the couplings of the majorana and dirac states with the higgs , the profile function of the odd leptons which couple to the higgs bosons need to be computed . the mass eigenstate profile functions are given in terms of combinations of the profile functions without the higgs . in the particular case of the neutral odd leptons these are admixtures of the neutral states belonging to the bidoublet and the singlet state , with normalization coefficients @xmath207 , @xmath208 and @xmath209 . the boundary conditions determine the coefficients @xmath208 and @xmath209 as functions of @xmath207 . the fermion profile functions in the presence of the higgs are given by : @xmath210\right ) -c_3 \left(1 - \cos\left[\frac{\lambda h}{f_h}\right]\right)\right)\right . \nonumber\\ & & -\left . \sqrt{2 } ( \tilde{s}_{m_2 } - e^{2 c_2 k x } m_{uv_o } \dot{\tilde{s}}_{-m_2})~ c_5 \sin\left[\frac{\lambda h}{f_h}\right]\right)\label{fl2}\\ & & \nonumber\\ f_l^3(h)&=&\frac{1}{2 } e^{\frac{1}{2 } ( 1 - 2 c_2 ) k x } \left(e^{2 c_2 k x } \dot{\tilde{s}}_{-m_2}\left(c_3 \left(1 + \cos\left[\frac{\lambda h}{f_h}\right]\right ) - c_2 \left(1 - \cos\left[\frac{\lambda h}{f_h}\right]\right)\right)\right.\nonumber\\ & & -\left . \sqrt{2 } ( \tilde{s}_{m_2 } - e^{2 c_2 k x } m_{uv_o } \dot{\tilde{s}}_{-m_2})~ c_5 \sin\left[\frac{\lambda h}{f_h}\right]\right)\label{fl3}\\ & & \nonumber\\ f_l^5(h)&=&\frac{1}{2 } e^{\frac{1}{2 } ( 1 - 2 c_2 ) k x } \left(2 ( \tilde{s}_{m_2 } - e^{2 c_2 k x } m_{uv_o } \dot{\tilde{s}}_{-m_2 } ) ~c_5 \cos\left[\frac{\lambda h}{f_h}\right ] + \sqrt{2 } e^{2 c_2 k x } \dot{\tilde{s}}_{-m_2 } ( c_2 + c_3 ) \sin\left[\frac{\lambda h}{f_h}\right]\right)\nonumber\\ & & \label{fl5}\end{aligned}\ ] ] @xmath211\right)-c_3 \left(1 - \cos\left[\frac{\lambda h}{f_h}\right]\right)\right)\right . \nonumber\\ & & + \left . \sqrt{2 } ( e^{2 c_2 k x } m_{uv_o } \tilde{s}_{-m_2 } - \dot{\tilde{s}}_{m_2})~ c_5 \sin\left[\frac{\lambda h}{f_h}\right]\right)\label{fr2}\\ & & \nonumber\\ f_r^3(h)&=&\frac{1}{2 } e^{\frac{1}{2 } ( 1 - 2 c_2 ) k x } \left(e^{2 c_2 k x } \tilde{s}_{-m_2}\left ( c_3\left(1 + \cos\left[\frac{\lambda h}{f_h}\right]\right ) - c_2 \left(1 - \cos\left[\frac{\lambda h}{f_h}\right]\right)\right)\right.\nonumber\\ & & + \left . \sqrt{2 } ( e^{2 c_2 k x } m_{uv_o } \tilde{s}_{-m_2 } - \dot{\tilde{s}}_{m_2})~ c_5 \sin\left[\frac{\lambda h}{f_h}\right]\right)\label{fr3}\\ & & \nonumber\\ f_r^5(h)&=&\frac{1}{2 } e^{\frac{1}{2 } ( 1 - 2 c_2 ) k x } \left(2 ( \dot{\tilde{s}}_{m_2}-e^{2 c_2 k x } m_{uv_o } \tilde{s}_{-m_2})~ c_5 \cos\left[\frac{\lambda h}{f_h}\right ] + \sqrt{2 } e^{2 c_2 k x } \tilde{s}_{-m_2 } ( c_2 + c_3 ) \sin\left[\frac{\lambda h}{f_h}\right]\right)\nonumber\\ & & \label{fr5}\end{aligned}\ ] ] where the functions @xmath192 and @xmath98 are functions of @xmath212 and the masses @xmath90 of the odd fermions . [ 0.7 ] [ 0.7 ] [ 0.7 ] for the doublet and singlet states mixed by the higgs , a non - trivial solution may be only obtained if the following relations are fulfilled . @xmath213_l}{\tilde{s}_{m_2}-e^{2 c_2 k l } m_{uv_o}\dot{\tilde{s}}_{-m_2}};\label{c5 } .\\ & & \nonumber\end{aligned}\ ] ] for the neutral leptons which decouple from the higgs , instead , the following relations are fulfilled : @xmath214 this implies that only the symmetric combination of neutral bidoublet states with coefficients given by eq . ( [ c3 ] ) and ( [ c5 ] ) , couple to the higgs . moreover , the normalization coefficient @xmath207 may be computed by demanding well normalized functions , namely , @xmath215 the above definition is appropriate in the majorana case , in which the left - handed components of the fermions acquire contributions from both the original left - handed and the ( charge conjugate ) right - handed modes , eqs . ( [ fl2])([fr5 ] ) . in the dirac case , the left - handed and right - handed functions acquire equal normalizations and therefore the proper factor @xmath207 is equal to the one computed above divided by @xmath216 . in the following , we will keep the above definition of @xmath207 for both the majorana and dirac cases and take care of the proper @xmath216 factors explicitly . to calculate the couplings , we also need the higgs profile and normalization : @xmath217 defining @xmath218 : @xmath219 - \left[f_r^{2*}(m_i)+ f_r^{3*}(m_i)\right ] f_l^5(m_j)\right]\ ] ] the left - right couplings of the higgs with the different states , @xmath220 , in the majorana and dirac cases can be written as : @xmath221 the factor of @xmath222 in the dirac coupling is due to the definition of the @xmath207 factor discussed above . observe that , while in the majorana case @xmath223 , there is no such relation in the dirac case . the different couplings are plotted in figs . [ fig.hcd ] [ fig.hcm5 ] . for the dirac case and @xmath224 , represented in fig . [ fig.hcd ] , small values of the masses are obtained for smaller values of @xmath158 . the left - right couplings of the singlet and doublet states , @xmath225 acquire large values for negative values of @xmath158 . this stems from the fact that for this case , the left - handed singlet component is localized towards the ir brane . as @xmath158 goes to 0 , the localization effects become less pronounced and this coupling starts getting suppressed . the @xmath226 couplings have the opposite behavior to the cross couplings . for @xmath158 negative , the mass difference between the singlet and the doublet state is large , while their mixing is small . since the self - couplings of the mass eigenstates are induced by the product of the singlet and doublet components of these states , they become very suppressed . however , as @xmath158 goes to zero , the mass eigenvalues become symmetric and antisymmetric combinations of the singlet and doublet states , and the self couplings of the mass eigenstates become large , while the cross couplings tend to zero . in the majorana case with @xmath227 , although the quantitative values are not the same , the @xmath228 and @xmath229 couplings behave similarly to the @xmath230 couplings , since @xmath231 and @xmath232 are the two mostly singlet states , which are split due to the non - zero @xmath193 . when both the majorana masses are non - zero , we see an abrupt change in the behavior of the couplings . as mentioned before , for @xmath184 the self - coupling of the lightest state is exponentially suppressed . this behavior is clearly demonstrated in the higgs dependent part of the approximation for the lop given in eq . ( [ maj_app ] ) . the couplings of the lop to the mostly doublet states , however , continue to be large . [ 0.7 ] [ 0.7 ] [ 0.7 ] the @xmath233 couplings to the lepton sector are defined in a similar manner to the couplings with the quark sector @xcite,@xcite . however , in the lepton case @xmath12 and the neutral states that couple to the higgs have @xmath235 , implying that @xmath236 . therefore , the @xmath233 , and all its kk modes , as well as the neutral components of the @xmath1 gauge bosons do nt have any couplings with any two of these states . however , the orthogonal neutral state in the bidoublet , which does not couple to the higgs has @xmath237 and @xmath238 . hence , an off - diagonal coupling exists between this mode , the neutral states that couple to the higgs and the @xmath233 . the @xmath239 couple the charged fermions with the neutral components . in component form , the coupling is between @xmath240 and @xmath241 . the profile functions and their normalization coefficients for the neutral components were given in eqs . ( [ fr5 ] ) ( [ c3 ] ) . the charged fermions and the neutral component of the bidoublet which do nt couple to the higgs state are both governed by the same five dimensional wave - function , namely : @xmath242 these fermion masses are given by the roots of @xmath243 and since the majorana masses do nt influence them , they are always dirac states . further , it can be shown trivially that the @xmath244 coupling to the lop and the positive charged state is equal to the coupling of the lop to @xmath245 and the negative charged states . in the annihilation cross - section , we will only be interested in the couplings between the two charged fermions and the lop . we will denote these couplings by @xmath246 . the expression for these couplings is given in appendix b. similarly , for the @xmath233 we will only be interested in the couplings between the lop and the @xmath247 state , the neutral bidoublet component that does not mix with the higgs , and we will denote these couplings by @xmath248 . the gauge boson couplings are plotted in figs . [ fig.gcd ] [ fig.gcm5 ] . again we see that the dirac and the majorana , @xmath227 couplings behave in a similar way . the coupling of the mostly singlet state to the charged or neutral fermion is obtained through the mixing with the bidoublet states . as discussed before , this mixing is small , for @xmath200 , and increases for larger values of @xmath158 . as @xmath158 approaches 0 , in the dirac case , we expect that since the mixing is maximal , the @xmath234 couplings should approach the neutrino - lepton sm coupling , @xmath249 , reduced by a factor @xmath250 , due to the mixing of the singlet with the doublet state , times another factor @xmath250 due to the projection of the neutral doublet state on the @xmath251 partner of each of the charged fields . this behavior is clearly seen in fig . [ fig.gcd ] , where only values of @xmath184 are plotted and increasing values of @xmath231 are associated with larger values of @xmath158 . for @xmath200 , the left - handed states which are located towards the infrared brane couple more strongly to the charged states than the right - handed states . moreover , the couplings of the @xmath233 and the @xmath234 become proportional to each other , with a coefficient of proportionality governed by @xmath252 , namely @xmath253 the additional factor of @xmath216 that appears between sm couplings of neutrinos to the @xmath233 and @xmath254 , is not present in this case . in the case of zero ultraviolet majorana mass but non - vanishing @xmath193 , depicted in fig . [ fig.gcm0 ] , the behavior is similar to the dirac case but the couplings are reduced due to the larger singlet components of the majorana particles . also , there is a sizable reduction of the left - handed couplings due to the larger right - handed component of the majorana state . observe that as @xmath194 is turned on , for @xmath184 , the couplings to the gauge bosons become independent of @xmath158 . finally , as @xmath158 becomes positive , the couplings increase , due to the larger bidoublet component of the lop . we will denote the neutral states mixed by the higgs by @xmath255 , where @xmath84 , or @xmath256 for the two dirac or four majorana states respectively , where @xmath257 labels the states in increasing order of their absolute masses . @xmath258 will denote the charged fermions and , as said before , @xmath247 will denote the bidoublet neutral fermion which does not couple to the higgs . the @xmath259 is the lop , our dark matter candidate . ignoring co - annihilation effects , we consider the following five dominant processes for @xmath260 annihilation : @xmath261 and @xmath262 ( observe that due to the cancelation of the z coupling to the states that couple to the higgs , the @xmath263 annihilation channel is suppressed ) . the feynman diagrams contributing to each of these processes are shown in figs . [ fig.nnttf ] [ fig.nnwwm ] . the virtual @xmath255 exchanges in these diagrams should be understood to be summed over @xmath257 , where @xmath257 as noted above runs over the appropriate index depending on whether we are considering the dirac or the majorana case . the @xmath264 in the following formulae is the relative velocity between the initial particles in the center of mass frame . @xmath265 , @xmath266 , @xmath267 and @xmath268 are the couplings of the higgs to the top , the @xmath239 and @xmath233 bosons , and itself , which were discussed in refs . @xcite and @xcite . [ 0.3 ] due to the cancelation of the coupling of @xmath259 to the @xmath233 , the annihilation into fermion pairs proceeds via an s - channel higgs interchange , and is therefore proportional to the corresponding fermion mass . therefore , only the top contributes in a relevant way . the dirac and majorana cross - sections are given by the same formula , but the higgs coupling should be understood to be the one appropriate for each case . assuming @xmath270 , we obtain : @xmath271 [ 0.5 ] the annihilation into higgs pairs proceeds via an s - channel higgs interchange diagram , which is subdominant , and the t - channel interchange of the neutral odd fermions . the result , in the limit @xmath273 , is given by : @xmath274\label{hhdirac}\end{aligned}\ ] ] where we have taken the couplings to be real . the sum runs over the two dirac states labeled by their masses , @xmath231 and @xmath275 . in the case of real couplings , the majorana cross - section can be simply seen from the above with the replacement @xmath276 , and the indices now run over the four majorana states : @xmath277\label{hhmaj}\end{aligned}\ ] ] the annihilation into the @xmath234 and @xmath233 gauge bosons also proceeds via the s - channel interchange of a higgs , plus the t - channel interchange of the charged fermion @xmath258 and the neutral fermion @xmath247 , respectively . in the formula below , the label @xmath280 corresponds to either the @xmath239 or @xmath233 gauge bosons , and @xmath281 for the @xmath282 cross - section and @xmath283 for the @xmath284 case . the diagrams contributing to the process in the dirac case are given in fig . [ fig.nnwwf ] . for the majorana case , two additional diagrams contribute , and are given in fig . [ fig.nnwwm ] . using the properties of the majorana couplings , one can demonstrate that these new diagrams are equal to the ones associated to the cross diagrams in the amplitudes for the annihilation into the @xmath234 ( due to the interchange of the fermion of opposite charge ) and @xmath233 gauge bosons . therefore , the cross - section is given by the same formula for both the dirac and majorana cases , but with appropriate couplings , and a factor @xmath285 for the majorana case , and @xmath286 for the dirac case . although in the numerical analysis the full annihilation cross section was used , for simplicity , we will only quote the cross - section for the longitudinal modes , in the limit @xmath287 : @xmath288\right]\end{aligned}\ ] ] [ 0.5 ] [ 0.7 ] these cross - sections are plotted in figs . [ fig.csd][fig.csm5 ] . for negative @xmath158 ( smaller values of @xmath231 ) , we observe an interesting correlation between the annihilation cross sections into @xmath234 , @xmath233 and higgs pairs . we see a dominance of the longitudinal modes for this range of values of @xmath158 and the magnitudes of the @xmath239 , @xmath289 and @xmath290 cross - sections obey the @xmath291 behavior expected due to the goldstone equivalence theorem . for larger values of @xmath158 , the bidoublet component of the lop increases and the transverse components of the gauge bosons are no longer subdominant in their contribution to the annihilation cross section . our extensive numerical and analytic study showed that for negative @xmath158 , the major contributions to the @xmath239 and @xmath289 cross - sections are due to the s - channel higgs exchange . in the @xmath290 cross - section , this is matched by the contribution from the virtual exchange of the @xmath255 in the t - channel . therefore , as emphasized before , for @xmath184 , a sizable annihilation cross section may only be obtained when the higgs coupling to the lop becomes of order one . [ 0.65 ] [ 0.65 ] [ 0.65 ] we shall follow the standard formalism for the calculation of the thermal dark matter density @xcite , @xcite . in calculating the annihilation cross - sections , we used the non - relativistic approximation for the initial particles . the cross - section used in calculating the dark matter density is the sum of all the different contributions in the previous sections and will be denoted by @xmath292 , and @xmath293 as usual . the relative velocity is related to the freeze - out temperature : @xmath294 the non - relativistic expansion of the thermal annihilation cross section may be expressed as @xmath295 the dark matter density is then given by @xmath296 where @xmath297 , @xmath298 gev , @xmath299 , @xmath300 , @xmath301 , @xmath302 is the degrees of freedom of our dark matter candidate and @xmath303 or @xmath304 to account for the antiparticles for the dirac and majorana case respectively . we take @xmath302 in the dirac case since we compute the density of the particle and antiparticle separately . the factor @xmath303 in the relic density then accounts for the duplication of the density from both the @xmath259 particle and the antiparticle in this case . [ 0.65 ] to check the veracity of our calculations , we extensively studied the limit of the majorana case reducing to the dirac case as @xmath193 and @xmath194 go to 0 . as the majorana masses go to 0 , the two singlet masses start getting degenerate in mass , and the @xmath305 and the @xmath306 cross - sections become equal . to properly analyze this limit , then , we must take co - annihilation between the lightest majorana sates into account @xcite . one can check that the coupling of the higgs to each of the degenerate lop majorana states , @xmath307 , becomes equal to the one of the higgs to the lop in the dirac case . moreover , the cross coupling of the higgs , @xmath308 , vanishes identically in the limit of vanishing majorana masses . further simple relations exist between the higgs couplings to fermions in the majorana and dirac cases . one can check that due to these relations the annihilation cross section into higgs states in the majorana case @xmath309 become the same as the @xmath310 cross section in the dirac case . the same happens in the case of annihilation into fermions . in the case of gauge bosons the situation is more complicated . as noted in calculating the couplings , the gauge boson couplings in the majorana case are reduced by a factor @xmath250 . this implies that for the @xmath239 and @xmath289 , the interference between the t and s - channel diagrams is reduced by @xmath311 and the t - channel diagrams by @xmath312 compared to the dirac cross - section case . these factors are exactly compensated for by the extra diagrams that contribute in the majorana case ( fig . [ fig.nnwwm ] ) , and therefore one obtains that the equality of annihilation cross sections defined above for the higgs final states extends to all final states . we also verified that the @xmath313 annihilation cross - section is exactly 0 in this limit . including co - annihilation between the two majorana states @xcite , the effective degrees of freedom are then 4 , and the effective cross - section is @xmath314 , where @xmath315 is the annihilation cross section between @xmath259 and its antiparticle in the dirac case . therefore , in this limit the dark matter density due to the dirac and majorana cases in the limit @xmath316 is exactly the same , as expected . we plotted the values of @xmath158 and @xmath231 leading to the correct dark matter density in both the dirac and the majorana cases in fig . [ fig.cm1all ] . we restricted the values of @xmath317 tev , for which the sm higgs couplings are close to their sm values . this provides a lower bound on the lop mass and on the value of @xmath158 . the bands in this figure represent the relic density uncertainty . we see that in the dirac case , we can only obtain the correct dark matter density for values of @xmath318 tev @xmath319 tev and @xmath320 , consistent with the exchange symmetry . the value of @xmath121 is correlated with the value of @xmath231 , and is constrained to be in the range @xmath222 tev @xmath321 tev for the above range of masses . in the majorana case , for @xmath227 , we are able to obtain the correct dark matter density for a larger range of @xmath231 values , which extend from about 3 tev up to the lowest values of about @xmath322 gev , and for the range of negative values of @xmath323 . the values of @xmath231 are , again , correlated with the values of @xmath121 , which is in the range @xmath324 tev @xmath325 tev . for @xmath326 , instead , we see that we can only obtain the correct dark matter density for @xmath327 . due to the effect of the majorana masses , the singlet lop state mass is still significantly smaller than the doublet mass in this case and therefore co - annihilation effects may be ignored . the values of @xmath158 obtained for the case of vanishing ultraviolet majorana masses are fully compatible with the identification of the lop with the odd partner of one of the right - handed neutrinos . indeed , as can be seen from fig . [ fig.n1 ] , for values of @xmath3280.7 , the proper neutrino masses are obtained for values of @xmath329 , for which a proper dark matter candidate may be obtained , with a mass @xmath330 tev @xmath331 tev . the situation is different for non - vanishing values of the ultraviolet majorana mass @xmath194 . in this case , a proper dark matter candidate may only be obtained for values of @xmath332 . such values of @xmath158 are incompatible with the exchange symmetry if the @xmath138 s of the three generations are approximately the same , as assumed in this article . therefore , a proper dark matter candidate would demand that the majorana mass at the uv brane is either zero or smaller than @xmath333 , since in such a case , according to eq . ( [ maj_app ] ) , masses of the order of the weak scale and couplings to the higgs of order one would be obtained . observe that the difficulty in obtaining a proper dark matter candidate for @xmath334 stems from the fact that we have assumed equal bulk mass parameters for the fermions interchanged by the @xmath48 exchanged symmetry introduced in refs . although this is an attractive possibility , since it allows a connection between the dark matter properties and the neutrino masses , this does not need to be the case . values of @xmath332 would be allowed in the more general case , or for a more general discrete symmetry . alternatively , if the value of @xmath138 of the left - handed leptons was allowed to be different for the three generations , values of @xmath332 would be consistent with values of @xmath138 compatible with the relatively small electron mass . the annihilation cross section for the odd neutrinos becomes of the proper size for sizable values of the coupling of the odd neutrino to the higgs boson , @xmath335 @xmath336 , with larger couplings associated with larger lop masses . such large couplings induce a relatively large scattering cross section of the odd lepton with nuclei that may be probed at direct dark matter detection experiments . more quantitatively , the spin - independent elastic scattering cross - section for an odd lepton scattering off a heavy nucleus is : _ si = ( z f_p + ( a - z ) f_n)^2 where @xmath337 and @xmath338 is the mass of the nucleus . the factors @xmath339 are given by f_p , n & = & ( _ q = u , d , s f_t_q^(p , n ) + f_tg^(p , n ) _ q = c , b , t ) m_p , n + a_u , d & = & - [ adu : eq ] , where the quark form factors are @xmath340 and @xmath341 @xcite . hence , we find that the contribution is f_p , n & & -m_p , n ( f_t_u^p , n + f_t_d^p , n + f_t_s^p , n + f_tg^p , n ) + & & - 0.2 m_p where we have neglected the differences between the proton and the neutron mass and have used the fact that the neutron and proton @xmath342 factors are relatively similar . assuming that the mass of the odd neutrino is much larger than that of the nucleus we have @xmath343 and _ si & & a^2 f_p^2 + & & , [ sigma_si : eq ] where @xmath344 is the neutrino - nucleon spin - independent cross - section . from eq . ( [ sigma_si : eq ] ) , the spin - independent cross - section scales as @xmath345 and therefore direct dark matter detection experiments like cdms can put strong constraints on regions of small @xmath346 and large @xmath347 . as discussed above , in the majorana case , an odd neutrino with a mass of about 700 gev and a coupling to the higgs of about 0.35 leads to an acceptable dark matter density . the spin independent cross section obtained in such a case for a higgs mass , @xmath348 gev , is about @xmath349 @xmath350 . the current limit coming from cdms , from a combination of the ge data and under the standard assumptions of local dark matter density distribution , is about 2.5 @xmath351 @xmath350 @xcite . the xenon experiment puts a slightly weaker limit for this range of lop masses @xcite . therefore , the predicted spin independent cross section is only a factor of a few lower than the current experimental limits . for larger masses , of about 1 tev , the higgs couplings grows to about 0.38 and the predicted cross section is therefore about @xmath352 @xmath350 , while the cdms bound is about two times larger . in the dirac case , the couplings are about fifty percent larger than in the majorana case and therefore the predicted cross section for a mass @xmath353 tev is slightly above the cdms reported bound . one would be able to conclude that the dirac case for @xmath354 gev is therefore disfavored . there are , however , uncertainties of order of a few associated with the local dark matter density distribution and the nuclear form factors which should be taken into account before ruling out a specific model . it is expected that both the xenon and cdms experiments will further improve their sensitivity by about an order of magnitude by the end of 2009 @xcite , @xcite . therefore , even considering possible uncertainties associated with the local density and the nuclear form factors , the minimal model discussed in this article should be probed by these experiments in the near future . let us comment that the mass of the @xmath259 particle may be in the appropriate range to provide an explanation of the anomalous excess in electrons and positrons observed by the pamela @xcite and atic @xcite experiments . however , since in these model these particles decay mostly into higgs and gauge bosons , if these particles are distributed throughout the halo of the galaxy , an excess of positrons of the size observed by these experiments will need a large boost factor enhancement and would probably lead to an unobserved excess of antiprotons @xcite,@xcite . in this article , we have considered the question of incorporating the charged and neutral leptons in a gauge - higgs unification scenario based on the gauge group @xmath0 in warped extra dimensions . these models are attractive since the @xmath355 subgroup of @xmath8 incorporates in a natural way the weak gauge group as well as an appropriate custodial symmetry group in order to suppress large contributions to the precision electroweak observables . moreover , the fifth dimensional components of @xmath10 gauge bosons have the right quantum numbers to play the role of the higgs doublet responsible for the breakdown of the electroweak symmetry . we have shown that , similar to the quark sector , the leptons can be incorporated by including the left - handed zero modes in a fundamental representation of @xmath8 and the right - handed charged leptons in a @xmath13 of @xmath8 . the model includes right - handed neutrinos which are singlets under the @xmath356 group and can therefore acquire localized majorana masses on the ir and uv branes . the simple inclusion of the right - handed neutrinos in the same multiplet as the charged leptons , fails to produce the correct lepton masses . the correct charged lepton and neutrino masses may be obtained from a three multiplet structure similar to the quark case . the bulk mass parameters @xmath138 , @xmath158 and @xmath357 of the left - handed leptons , right - handed neutrinos and right - handed charged leptons , respectively acquire values @xmath358 0.7 , @xmath359 and @xmath360 , where larger negative values of @xmath357 correspond to the first generation leptons . we have further investigated the possibility of incorporating a dark - matter candidate by including an exchange symmetry , under which all sm leptons multiplets are even , and which ensures the stability of the lightest odd lepton partner . we therefore analyzed the possibility of associating the dark matter with the lightest neutral components of the odd leptons , transforming in the fundamental representation of @xmath8 . we have shown that these neutral components have interesting properties . the neutral leptons that couple to the higgs do not have self couplings to the @xmath233-boson . however , these neutral states couple to the orthogonal combination of neutral states in the bidoublet and to the @xmath233 , as well as to the charged leptons and the @xmath254-gauge boson . we computed the couplings of the neutral odd leptons to the gauge bosons and the higgs in a functional way and computed the dominant contributions to the annihilation cross section into higgs , neutral and charged gauge bosons and fermions ( top - quark ) final states . we considered the cases in which the majorana masses for the neutral leptons vanish in both branes ( dirac case ) as well as the case in which at least one of them is non - vanishing . by doing that , we have shown that in the dirac case , a proper dark matter candidate may be obtained for masses of about 1 tev to 2 tev and localization parameter @xmath329 in agreement with the exchange symmetry . if only the majorana mass in the infrared brane is non - vanishing , one obtains lower values of the required odd - lepton masses , which may be of about @xmath361 gev to @xmath362 tev , and a range for the localization parameter @xmath363 . finally , in the case that the ultraviolet majorana mass is different from zero , the self - couplings of the lop with the higgs is exponentially suppressed for @xmath184 and becomes non - vanishing only for @xmath197 , for which a proper dark matter may be obtained with a mass similar to the one obtained when only @xmath364 is different from zero . this last possibility is incompatible with the exchange symmetry and the proper neutrino masses if a common values of @xmath138 is assumed for the three families . the collider signatures of these models have been previously discussed in the literature . the odd leptons introduced in this article will be hard to detect at collider experiments , since the masses of the charged and neutral non - lop odd leptons are above a few tev , and they have relatively weak interactions . in all cases , a proper dark matter candidate is obtained for values of the self - coupling of @xmath259 to the higgs of about @xmath365 @xmath336 , with larger higgs couplings corresponding to larger lop masses , for which the cross section of the dark matter with nuclei becomes sizable . we have computed the dark matter cross section with nuclei and have shown that these models will be probed by the cdms and xenon direct dark matter detection experiments in the near future . we would like to thank shri gopalakrishna , howard haber , eduardo ponton , jose santiago and tim tait for useful discussions and comments . work at anl is supported in part by the us doe , div . of hep , contract de - ac02 - 06ch11357 . fermilab is operated by fermi research alliance , llc under contract no . de - ac02 - 07ch11359 with the united states department of energy . we would like to thank the aspen center for physics and the kitpc , china , where part of this work has been done . in the @xmath187 gauge , we redefine @xmath366 and we write our vector - fermionic fields in terms of chiral fields . we can kk decompose the fermionic chiral components as , @xmath367 where @xmath368 is normalized by : @xmath369 therefore the profile function for the zero mode fermion corresponds to @xmath370 . from the 5d action , concentrating on the free fermionic fields , we can derive the following first order coupled equations of motion for @xmath371 , @xmath372 we see from eq . ( [ 1order ] ) that we can redefine @xmath373 and relate the opposite chiral component of the same vector - like field by @xmath374 . for the left handed field having dirichlet boundary conditions on the uv brane , we can derive a second order equation for the chiral component @xmath375 : @xmath376 \tilde{f}_{l , n } = 0\ ] ] the solution of which we shall call @xmath377 , with boundary conditions @xmath378 , @xmath379 . similarly , if the right - handed field fulfills dirichlet boundary conditions on the uv - brane , we can redefine @xmath380 and then relate the opposite chirality via @xmath381 . we can further write the equation of motion for @xmath382 : @xmath383 \tilde{f}_{r , n } = 0.\ ] ] we shall correspondingly denote the solution to this equation with @xmath384 , fulfilling the boundary conditions @xmath385 , @xmath386 . the solution to eq . ( [ feom ] ) is given by @xcite : @xmath387.\ ] ] the solution for eq . ( [ feom1 ] ) , @xmath82 , is given by eq . ( [ sm ] ) with the replacement @xmath388 . following the notation of ref . @xcite , the @xmath239 boson profile functions are given by : @xmath389 c^g_{\hat{1 } } + \frac{1}{\sqrt{2}}\sin\left[\frac{\lambda h}{f_h}\right ] c^g_{1_r}\right)-\frac{1}{\sqrt{2 } } c(x_5 ) \sin\left[\frac{\lambda h}{f_h}\right]c^g_{1_l}\label{f1}\\ & & \nonumber\\ & & \nonumber\\ f^{1_l}_g(h)&= & \frac{1}{2 } \left(s(x_5)\left(\left(\cos\left[\frac{\lambda h}{f_h}\right]-1\right)c^g_{1_r}+\sqrt{2}\sin\left[\frac{\lambda h}{f_h}\right ] c^g_{\hat{1 } } \right)+c(x_5)\left(1 + \cos\left[\frac{\lambda h}{f_h}\right]\right)c^g_{1_l}\right)\nonumber\\ & & \nonumber\\ & & \label{f5}\\ & & \nonumber\\ f^{1_r}_g(h)&= & \frac{1}{2 } \left(s(x_5)\left(\left(\cos\left[\frac{\lambda h}{f_h}\right]+1\right)c^g_{1_r}-\sqrt{2}\sin\left[\frac{\lambda h}{f_h}\right ] c^g_{\hat{1 } } \right)+c(x_5)\left(1- \cos\left[\frac{\lambda h}{f_h}\right]\right)c^g_{1_l}\right)\nonumber\\ & & \nonumber\\ & & \label{f8}\end{aligned}\ ] ] the normalization coefficients @xmath390 , in terms of @xmath391 are given by : @xmath392_l c(l)^\prime s(l)^\prime+c_h h e^{2 k l}\sin\left[\frac{\lambda h}{f_h}\right]_l\left(c(l)^\prime s(l)+c(l)s(l)^\prime\right)}{\sqrt{2}s(l)^\prime\left(c_h h e^{2 k l}\cos\left[\frac{\lambda h}{f_h}\right]_l s(l)+2\sin\left[\frac{\lambda h}{f_h}\right]_l s(l)^\prime\right)}\label{c1g}\\ c^g_{1_r}&= & -c^g_{1_l}\frac{c(l)^\prime}{s(l)^\prime}\label{c8g}\end{aligned}\ ] ] the five dimensional weak coupling is defined as @xmath393 . in terms of these , the dirac couplings for @xmath245 and @xmath233 , ( denoted by @xmath280 ) , are given by : @xmath394 again , due to our choice of normalization for the coefficient @xmath207 , the factor @xmath216 , coming from the definition of the @xmath234 fields , is not present in this expression . the majorana couplings are given in terms of the dirac couplings , @xmath395 with @xmath396 given in eq . ( [ wlop ] ) . s. chang , j. hisano , h. nakano , n. okada and m. yamaguchi , phys . d * 62 * , 084025 ( 2000 ) [ arxiv : hep - ph/9912498 ] . t. gherghetta and a. pomarol , nucl . phys . b * 586 * , 141 ( 2000 ) [ arxiv : hep - ph/0003129 ] ; c. csaki , j. erlich and j. terning , phys . d * 66 * , 064021 ( 2002 ) [ arxiv : hep - ph/0203034 ] . j. l. hewett , f. j. petriello and t. g. rizzo , jhep * 0209 * , 030 ( 2002 ) [ arxiv : hep - ph/0203091 ] . h. davoudiasl , j. l. hewett and t. g. rizzo , phys . d * 68 * , 045002 ( 2003 ) [ arxiv : hep - ph/0212279 ] . m. s. carena , e. ponton , t. m. p. tait and c. e. m. wagner , phys . d * 67 * , 096006 ( 2003 ) [ arxiv : hep - ph/0212307 ] . m. s. carena , a. delgado , e. ponton , t. m. p. tait and c. e. m. wagner , phys . d * 68 * , 035010 ( 2003 ) [ arxiv : hep - ph/0305188 ] . m. s. carena , a. delgado , e. ponton , t. m. p. tait and c. e. m. wagner , phys . d * 71 * , 015010 ( 2005 ) [ arxiv : hep - ph/0410344 ] . k. agashe , a. delgado , m. j. may and r. sundrum , jhep * 0308 * , 050 ( 2003 ) [ arxiv : hep - ph/0308036 ] . k. agashe , r. contino , l. da rold and a. pomarol , phys . b * 641 * , 62 ( 2006 ) [ arxiv : hep - ph/0605341 ] . m. carena , e. ponton , j. santiago and c. e. m. wagner , nucl . b * 759 * , 202 ( 2006 ) [ arxiv : hep - ph/0607106 ] . n. s. manton , nucl . b * 158 * , 141 ( 1979 ) ; y. hosotani , phys . b * 126 * , 309 ( 1983 ) ; h. hatanaka , t. inami and c. s. lim , mod . a * 13 * , 2601 ( 1998 ) [ arxiv : hep - th/9805067 ] ; i. antoniadis , k. benakli and m. quiros , new j. phys . * 3 * , 20 ( 2001 ) [ arxiv : hep - th/0108005 ] ; m. kubo , c. s. lim and h. yamashita , mod . a * 17 * , 2249 ( 2002 ) [ arxiv : hep - ph/0111327 ] ; c. csaki , c. grojean and h. murayama , phys . d * 67 * , 085012 ( 2003 ) [ arxiv : hep - ph/0210133 ] ; n. haba , m. harada , y. hosotani and y. kawamura , nucl . b * 657 * , 169 ( 2003 ) [ erratum - ibid . b * 669 * , 381 ( 2003 ) ] [ arxiv : hep - ph/0212035 ] ; c. a. scrucca , m. serone and l. silvestrini , nucl . b * 669 * , 128 ( 2003 ) [ arxiv : hep - ph/0304220 ] ; g. burdman and y. nomura , nucl . b * 656 * , 3 ( 2003 ) [ arxiv : hep - ph/0210257 ] . a. d. medina , n. r. shah and c. e. m. wagner , phys . d * 76 * , 095010 ( 2007 ) [ arxiv:0706.1281 [ hep - ph ] ] . m. carena , a. d. medina , b. panes , n. r. shah and c. e. m. wagner , phys . d * 77 * , 076003 ( 2008 ) [ arxiv:0712.0095 [ hep - ph ] ] . k. agashe , a. belyaev , t. krupovnickas , g. perez and j. virzi , arxiv : hep - ph/0612015 . b. lillie , l. randall and l. t. wang , arxiv : hep - ph/0701166 . b. lillie , j. shu and t. m. p. tait , arxiv:0706.3960 [ hep - ph ] . s. j. huber and q. shafi , phys . b * 583 * , 293 ( 2004 ) [ arxiv : hep - ph/0309252 ] . c. csaki , a. falkowski and a. weiler , jhep * 0809 * , 008 ( 2008 ) [ arxiv:0804.1954 [ hep - ph ] ] . s. casagrande , f. goertz , u. haisch , m. neubert and t. pfoh , jhep * 0810 * , 094 ( 2008 ) [ arxiv:0807.4937 [ hep - ph ] ] . c. csaki , c. delaunay , c. grojean and y. grossman , jhep * 0810 * , 055 ( 2008 ) [ arxiv:0806.0356 [ hep - ph ] ] . c. csaki , a. falkowski and a. weiler , arxiv:0806.3757 [ hep - ph ] . j. santiago , jhep * 0812 * , 046 ( 2008 ) [ arxiv:0806.1230 [ hep - ph ] ] . k. agashe and g. servant , phys . * 93 * , 231805 ( 2004 ) [ arxiv : hep - ph/0403143 ] . k. agashe , a. falkowski , i. low and g. servant , jhep * 0804 * , 027 ( 2008 ) [ arxiv:0712.2455 [ hep - ph ] ] . g. b. gelmini , p. gondolo and e. roulet , nucl . b * 351 * , 623 ( 1991 ) . m. srednicki and r. watkins , phys . b * 225 * , 140 ( 1989 ) . m. drees and m. nojiri , phys . d * 48 * , 3483 ( 1993 ) [ arxiv : hep - ph/9307208 ] . j. r. ellis , a. ferstl and k. a. olive , phys . b * 481 * , 304 ( 2000 ) [ arxiv : hep - ph/0001005 ] . z. ahmed _ et al . _ [ cdms collaboration ] , arxiv:0802.3530 [ astro - ph ] . j. angle _ et al . _ [ xenon collaboration ] , phys . lett . * 100 * , 021303 ( 2008 ) [ arxiv:0706.0039 [ astro - ph ] ] . o. adriani _ et al . _ , arxiv:0810.4995 [ astro - ph ] . j. chang _ et al . _ , nature * 456 * , 362 ( 2008 ) . see , for example , m. cirelli , m. kadastik , m. raidal and a. strumia , arxiv:0809.2409 [ hep - ph ] . j. hall and d. hooper , arxiv:0811.3362 [ astro - ph ] .

in the last years a considerable interest in density - embedding methods has led to a widespread development of subsystem approaches @xcite within density functional theory ( dft ) @xcite . in particular , the frozen density embedding ( fde ) method @xcite has emerged as a popular tool to treat intermolecular interactions in a simple and potentially exact embedding scheme . in the fde approach the total electronic density of a system is partitioned into a set of subsystem densities ( e.g. , for two subsystems @xmath0 and @xmath1 : @xmath2 ) . the total electronic energy is then written as @xmath3 = e[\rho_{\text{a}}+\rho_{\text{b } } ] = t_{\text{s}}[\rho_{\text{a } } ] + t_{\text{s}}[\rho_{\text{b } } ] + t_{\text{s}}^{{\text{nadd}}}[\rho_{\text{a}},\rho_{\text{b } } ] + \\ \nonumber & & + \frac{1}{2}\int\int\frac{\left(\rho_{\text{a}}({\mathbf{r}}_1)+\rho_{\text{b}}({\mathbf{r}}_1)\right ) \left(\rho_{\text{a}}({\mathbf{r}}_2)+\rho_{\text{b}}({\mathbf{r}}_2)\right)}{{\left|{\mathbf{r}}_1-{\mathbf{r}}_2\right|}}\,d{\mathbf{r}}_1d{\mathbf{r}}_2 + \\ \nonumber & & + \int \left(v_{\text{ext}}^{\text{a}}({\mathbf{r}}_1)+v_{\text{ext}}^{\text{b}}({\mathbf{r}}_1)\right ) \left(\rho_{\text{a}}({\mathbf{r}}_1)+\rho_{\text{b}}({\mathbf{r}}_1)\right)d{\mathbf{r}}_1 + \\ \label{e1 } & & + e_{\text{xc}}[\rho_{\text{a } } ] + e_{\text{xc}}[\rho_{\text{b } } ] + e_{\text{xc}}^{\text{nadd}}[\rho_{\text{a}},\rho_{\text{b}}]\ , \end{aligned}\ ] ] where @xmath4 is the noninteracting kinetic energy functional , the external potential was partitioned as @xmath5 , and the superscript @xmath6 denotes the nonadditive energy contributions to the kinetic and exchange - correlation ( xc ) energies . the minimization of the energy of eq . ( [ e1 ] ) with respect to one of the subsystem electron densities ( e.g. , @xmath7 ) , keeping the other one fixed and using the expansion of the electron density in terms of ( auxiliary ) kohn - sham orbitals @xmath8 , leads to the kohn - sham equations with constrained electron density ( ksced ) @xcite @xmath9({\mathbf{r } } ) + v_{\text{emb}}[\rho_{\text{a}};\rho_{\text{b}}]({\mathbf{r}})\right)\psi^{\text{a}}_i = \varepsilon_i^{\text{a}}\psi_i^{\text{a}}\ ] ] with @xmath10({\mathbf{r } } ) & = & \int\frac{\rho_{\text{a}}({\mathbf{r}}_1)}{{\left|{\mathbf{r}}-{\mathbf{r}}_1\right|}}\,d{\mathbf{r}}_1 + v_{\text{ext}}^{\text{a}}({\mathbf{r } } ) + \frac{\delta e_{\text{xc}}[\rho_{\text{a}}]}{\delta\rho_{\text{a}}({\mathbf{r } } ) } \\ \label{e3_vks } & & \\ \nonumber v_{\text{emb}}[\rho_{\text{a}};\rho_{\text{b}}]({\mathbf{r } } ) & = & v_{\text{ext}}^{\text{b}}({\mathbf{r } } ) + \int\frac{\rho_{\text{b}}({\mathbf{r}}_1)}{{\left|{\mathbf{r}}-{\mathbf{r}}_1\right|}}\,d{\mathbf{r}}_1 + \\ \label{e3 } & + & \frac{\delta t_{\text{s}}^{{\text{nadd}}}[\rho_{\text{a}},\rho_{\text{b } } ] } { \delta\rho_{\text{a}}({\mathbf{r } } ) } + \frac{\delta e_{\text{xc}}^{{\text{nadd}}}[\rho_{\text{a}},\rho_{\text{b } } ] } { \delta\rho_{\text{a}}({\mathbf{r } } ) } \ . \end{aligned}\ ] ] the ksced of eq . ( [ e2 ] ) provide the solution for the ground state of subsystem @xmath0 subject to the interaction with the frozen density of the subsystem @xmath1 . if the ksced of both subsystems are considered in a freeze - and - thaw procedure @xcite , the full variational solution for the total system , equivalent to the usual kohn - sham solution , is obtained , except for approximations eventually included in the nonadditive interaction terms . henceforth , the acronym fde wil be used to refer to this approach . so far , numerous applications of the fde - based method , using the ksced approach , have been presented in literature to study the interaction of non - covalently bound ( or weakly interacting ) molecular complexes or solvated molecules @xcite . in this cases in fact the nonadditive kinetic energy term can be efficiently approximated by means of semilocal approximations @xcite . however , because in the conventional formulation the use of ksced is limited to local / semilocal xc functionals , most studies have focused on hydrogen - bond and dipole - dipole systems , since these interactions are dominated by electrostatic and polarization effects that can be described relatively well at the semilocal level of theory @xcite . recently the method has been also extended to the generalized kohn - sham ( gks ) framework @xcite , so that hybrid functionals can be considered as well . in this approach , named gks - fde , the nonadditive kinetic and xc terms in eq . ( [ e3 ] ) are replaced by the nonadditive direct and residual interactions @xmath11 and @xmath12 which include respectively the orbital - dependent and the explicitly density - dependent kinetic and xc terms . the presence of nonlocal and/or orbital - dependent terms makes however the direct implementation of the method rather involved . therefore , a practical computational scheme was proposed where these terms are approximated with local functionals @xcite . thus , in practice in the gks - fde method the electronic structure is evaluated using a nonlocal exchange operator for the intra - subsystem xc effects but with an embedding potential evaluated at the local / semilocal level of theory . a similar approximation was also proposed for the extention of the ksced to the use of orbital - dependent optimized effective potentials @xcite . the gks - fde method was tested for a variety of hydrogen bond and dipole - dipole interactions and proved to be a very good approach for embedded simulations with hybrid functionals @xcite . in particular , with respect to fde - based calculations employing semilocal xc functionals , it enables a general reduction of the embedding errors , because of the smaller overlap of subsystem densities at the hybrid level , and it allows a better description of subsystem properties due to the removal of most of the self - interaction error ( sie ) @xcite . furthermore , it makes it possible to extend the application of fde - based techniques to treat systems that are typically poorly described by semilocal xc functionals . an example of the latter is given by ground - state charge - transfer complexes ( also called donor - acceptor complexes ) . these systems derive in fact a considerable portion of their stabilization energy from an electron charge - transfer between two centers . thus , unlike for hydrogen bonds or dipole - dipole interactions , where electrostatic and polarization effects dominate , the interaction in charge - transfer complexes is largely determined by orbital interactions . as a consequence , the description of ground - state charge - transfer complexes is very challenging for conventional semilocal xc functionals because of the sie inherent in these methods . on the other hand , a good performance is generally obtained for hybrid functionals , especially when rather large portions of hartree - fock exchange are included @xcite . similar results also concern the charge transfer between organic molecules and metal substrates @xcite . due to the difficult application of dft - based methods using semilocal xc functionals to these systems , in the last years only few fde investigations treated ground - state charge - transfer complexes and related problems @xcite . some of them @xcite made use of complicated or _ ad hoc _ reparametrized kinetic energy functionals in order to obtain improved kinetic embedding potentials and improve the description of the system . however , these kinetic functionals can not be routinely applied in fde calculations for large systems . on the other hand , an extensive investigation of different kinetic energy functionals @xcite showed that most popular kinetic energy approximations perform similarly and reasonably well for complexes displaying small / moderate charge transfer , so that it appears reasonable to consider in such applications the same kinetic energy approximation as those commonly used for hydrogen bond and dipole - dipole interactions . the role of the xc functionals in embedding calculations dealing with charge - transfer systems , instead , was not given much attention so far @xcite . in this paper we cover this issue and demonstrate how hybrid functionals can be practically utilized within an embedding scheme to obtain an improvement of the performance in the description of embedded systems for a benchmark of charge - transfer complexes . the ksced @xcite and gks - fde @xcite approaches are implemented in the turbomole program package @xcite , version 6.4 . all calculations were performed using the _ fde _ script , which uses a freeze - and - thaw procedure @xcite to guarantee the full relaxation of the embedded ground - state electron density of both subsystems , until dipole moments of the embedded subsystems converged to @xmath13 au . a supermolecular def2-tzvppd @xcite basis set was employed in all calculations to expand the subsystem electron densities . a monomolecular ( or too small ) basis set is in fact insufficient to carry on fde calculations with good accuracy , resulting in basis - set errors larger than the fde errors and the differences between the methods . the def2-tzvppd basis set adds diffuse basis functions to the def2-tzvpp @xcite basis set , thus granting an accurate representation for the electron densities even at the relatively large bonding distances characteristic of the systems under consideration . very accurate integration grids were employed to minimize numerical errors . additional details about our implementation and computational procedure are reported in refs . . in our investigation we considered a set of twelve representative ground - state charge - transfer complexes ( c@xmath14h@xmath15-f@xmath14 , nh@xmath16-f@xmath14 , c@xmath14h@xmath14-clf , hcn - clf , nh@xmath16-cl@xmath14 , h@xmath14o - clf , nh@xmath16-clf , hcn - nf@xmath16 , hnc - nf@xmath16 , hf - nf@xmath16 , clf - nf@xmath16 , and c@xmath14h@xmath15-cl@xmath14 ) . the geometries and the reference binding energies were taken from refs . , except for the c@xmath14h@xmath15-cl@xmath14 complex ( for this latter case we optimized the geometry at the mp2/aug - cc - pvtz level of theory @xcite and calculated the reference binding energy , in analogy with ref . @xcite as @xmath17 , where @xmath18 is the binding energy computed at the ccsd(t)/aug - cc - pvqz level of theory @xcite and @xmath19 is the difference between the mp2 binding energies computed at the complete basis set limit ( 56-extrapolation @xcite ) and with an aug - cc - pvqz basis set ; the resulting binding energy is in very good agreement to that reported in ref . ) . the calculations were performed using conventional semilocal xc functionals ( blyp @xcite , pbe @xcite ) as well as hybrid methods ( pbe0 @xcite and bhlyp @xcite ) . all calculations were performed using dft - d3 dispersion corrections @xcite . to compute the embedding nonadditive interactions , the nonadditive noninteracting kinetic energy term was approximated using the gga functional revapbek @xcite . in the case of hybrid xc functionals also the nonadditive exchange term was approximated : the pbe and the b88x @xcite exchange functionals were employed for pbe0 and bhylp , respectively ; the resulting methods are indicated as pbe0/pbe and bhlyp / blyp . in the case of semilocal xc functionals instead , no approximation for the nonadditive exchange contribution to the embedding potential is required : the same exchange functional is used for the embedding potential , subsystem and conventional supermolecule ks calculations . the quality of embedded densities was investigated by considering the absolute deviation of plane - averaged embedding density , defined as @xmath20 where @xmath21 is the electron density obtained from a ( generalized ) ks calculation on the whole system . a quantitative measurement of the absolute error associated with a given embedded density was then obtained by computing the embedding density error @xmath22 with @xmath23 being the number of electrons . in the evaluation of @xmath24 only valence densities were considered to avoid numerical problems related to the high value assumed by the electron density in the core regions and in consideration of the fact that the core density is not important for the determination of chemical and physical properties of interaction between the subsystems , which are of interest here . the net charge on the @xmath25-th subsystem ( monomer charge ) was calculated for each method by considering a grid - based atoms - in - the - molecule partitioning @xcite of the electron density into atomic basins @xmath26 as @xmath27 with @xmath28 the nuclear charge of the @xmath29-th atom and @xmath30 the number of atoms in the @xmath25-th subsystem . for two interacting subsystems @xmath31 and thus we define the charge transfer as @xmath32 . reference values for the charge transfer were evaluated using relaxed mp2 densities computed with an aug - cc - pvtz basis set . finally , to inspect the quality of the embedding energies we considered the error defined as @xmath33 = e^{\text{emb}}[\rho_{\text{a}},\rho_{\text{b } } ] - e^{\gks}[\rho^{\gks}]\ , \ ] ] where @xmath34 is the embedding energy , computed using the embedding densities into eq . ( [ e1 ] ) , and @xmath35 is the energy resulting from a conventional ( generalized ) ks calculation on the full system . further analysis was also performed considering , according to ref . , the error decomposition @xmath36 & = & \delta t[\rho_a,\rho_b,\rho^\gks]+ \delta d[\rho,\rho^\gks ] + \\ \label{eq : enerror } & & + \delta x^s[\rho_a,\rho_b,\rho^\gks]\ , \end{aligned}\ ] ] where the error on embedding energy is partitioned into kinetic ( @xmath37 ) , exchange ( @xmath38 ) , and relaxation ( @xmath39 ) contributions . for full details see ref . . in this section we report and analyze the results of the fde - based calculations on the charge - transfer complexes . we evaluate the performance of hybrid and semilocal xc functionals ( bhlyp , pbe0 , blyp , and pbe ) , elucidating what are the roles of the exact exchange , of the geometry and of the charge transfer between the subsystems . in tab . [ tab : enerr ] we report the binding energies obtained from different embedding methods ( @xmath40 ) and the corresponding results from conventional dft calculations on the whole supermolecular system ( @xmath41 ) . in addition , we report the deviation of these results from accurate reference values ( @xmath42 and @xmath43 , respectively ) and the error on embedding energies as defined in eq . ( [ eq_emberr ] ) . we note that the embedding error on the total energy is exactly equivalent to the difference of the binding energies computed respectively from embedding and conventional supermolecular calculations ( i.e. the dft calculation on the whole system ) , since the contributions from isolated subsystems exactly cancel out . moreover , all binding energies reported in tab . [ tab : enerr ] are dispersion - corrected and , because we use a supermolecular basis set , the basis set superposition error for the total - system calculation is treated in the same way in both dft and fde - based methods ( boys - bernardi correction @xcite ) . in tab . [ tab : enerr ] we also report in the last lines the mean signed error ( mse ) , the mean absolute error ( mae ) , and the mean absolute relative error ( mare ) with respect to the reference binding energies . [ cols="<,>,^,>,>,>,>,>,^,>,>,>,>,>",options="header " , ] @xmath44 ccsd(t)-f12b / vtz - f12 , ref . . + @xmath45 w1 benchmark calculations , ref . . + @xmath46 ccsd(t ) extrapolated to cbs limit , this work . a first inspection of the results , focusing the attention on @xmath41 and @xmath43 , confirms the finding , well established in literature @xcite , that hybrid functionals outperform semilocal ones for the calculation of interaction energies of ground - state charge - transfer complexes . in fact , bhlyp with a mare of 16.63% is the most accurate method . however , when interaction energies from embedding calculations are considered ( i.e. @xmath40 ) , all methods yield rather similar performances with mares in the range 30 - 37% ( for a discussion of the issues related to the comparison of embedding binding energies with experimental or theoretical references see e.g. refs . this suggests that a strong error cancellation occurs for embedding calculations using semilocal functionals , while the cancellation is much less in the case of hybrid approaches . the reason for this behavior can be traced back to the fact that , in general , conventional semilocal functionals overestimate significantly the binding energy of charge - transfer complexes @xcite ( mse is largely positive ) while , on the other hand , when used in methods based on the fde theory they show usually a tendency to produce too low interaction energies ( @xmath47 is negative in most cases ) . on the contrary , hybrid functionals are quite accurate for the calculation of the interaction energies of charge - transfer complexes and , at the same time , corresponding embedding calculations reproduce with good accuracy the parent supermolecular gks calculations . concerning the error on embedding energies ( @xmath47 ) , we note that the methods using hybrid xc functionals definitely outperform the ones employing semilocal xc approximations , yielding average deviations more than twice smaller . this improvement can be partially reconduced , in analogy to what shown in refs . , to a smaller overlap of the subsystem densities ( i.e. to a smaller sie ) at the hybrid level , that leads to a reduction of the error on embedding energy . this effect is best recognized if we consider a set of systems where only one subsystem is varying , as for nh@xmath16 interacting with cl@xmath14 , clf , and f@xmath14 respectively . in this case the value of @xmath47 is increasing from 2.6 to about 7 mha when the cl atoms are replaced by f atoms and this increase nicely correlates to the fact that the fluorine atom is affected by a larger self - interaction error than chlorine , so that it is plagued by a stronger overestimation of the diffuseness of the electron density at semilocal level of theory . however , in general it is not possible to find a trend valid for all the systems due to the different details of the interaction when both subsystems are changed . in fact , despite the inclusion of exact exchange in the calculations may help to reduce the errors on embedding energies by lowering the effective overlap between the two subsystems , this is not the only effect determining the accuracy of the gks - fde energy calculations . in particular for hybrid methods still a subtle error compensation between kinetic , relaxation , and exchange contributions occurs . thus , to shed light on the different aspects determining the embedding energy error we can recur to the embedding energy decomposition introduced in ref . . [ tab : errdecom ] we report the energy error contributions @xmath48 and @xmath38 for all the hybrid methods ( bhlyp and pbe0 ) and system considered . for each energy error contribution we report the mse , mae and mare . the contributions due to kinetic ( @xmath37 ) and relaxation ( @xmath39 ) effects are reported summed together because both terms yield very large values ( 100 - 200% of the binding energies ) and with opposite sign , so they only contribute to the total error through a strong error cancellation . lcc systems & bhlyp & pbe0 + & + nf@xmath16-hcn & -0.4 & -0.4 + c@xmath14h@xmath15-f@xmath14 & 2.6 & 3.4 + nf@xmath16-hnc & -0.3 & -0.2 + c@xmath14h@xmath15-cl@xmath14 & 0.3 & 1.1 + nh@xmath16-f@xmath14 & 3.7 & 5.5 + nf@xmath16-clf & 1.1 & 1.8 + nf@xmath16-hf & 0.6 & 0.8 + c@xmath14h@xmath14-clf & 2.2 & 3.3 + hcn - clf & 0.4 & 1.2 + nh@xmath16-cl@xmath14 & 0.9 & 2.3 + h@xmath14o - clf & 0.8 & 1.9 + nh@xmath16-clf & 2.9 & 4.3 + mae & 1.35 & 2.19 + mse & 1.23 & 2.09 + mare & 39.19% & 58.65% + sdre & -33.51% & -12.97% + & + nf@xmath16-hcn & 1.0 & -0.0 + c@xmath14h@xmath15-f@xmath14 & -0.4 & -1.5 + nf@xmath16-hnc & 0.5 & -0.3 + c@xmath14h@xmath15-cl@xmath14 & -0.4 & -1.6 + nh@xmath16-f@xmath14 & -1.2 & -2.5 + nf@xmath16-clf & 0.4 & -0.9 + nf@xmath16-hf & 0.2 & -0.6 + c@xmath14h@xmath14-clf & -0.5 & -1.7 + hcn - clf & 0.3 & -1.2 + nh@xmath16-cl@xmath14 & -1.0 & -2.1 + h@xmath14o - clf & 0.0 & -1.5 + nh@xmath16-clf & -2.7 & -3.0 + mae & 0.72 & 1.41 + mse & -0.32 & -1.41 + mare & 19.10% & 34.10% + xde & -0.41 & -1.28 + xdre & -6.79% & -28.83% + the analysis of the @xmath48 contributions shows that the use of the subsystem electron densities from hybrid calculations can provide a significant improvement of the semilocal embedding energies ( the maes are 1.35 and 2.19 to be compared with 2.54 and 2.61 ) , although the errors are still larger than the global ones reported in tab . [ tab : enerr ] ( maes of 0.94 and 0.90 for bhlyp and pbe0 , respectively ) . this improvement is determined exclusively from the reduced overlap displayed by the densities obtained from hybrid calculations , since here no exchange approximation is considered . to make a quantitative evaluation of the effect it is possible to use the semilocal differential relative error ( sdre ) statistical indicator , defined as @xmath49 with the sum extend over all the @xmath23 systems reported in the table . a negative value of sdre indicates that using subsystem densities from a given hybrid calculation improves on average the calculation of kinetic and relaxation energy contributions over the use of gga densities . using the sdre indicator it is thus possible to compare energetic error terms with the same explicit level of approximation , i.e. the semilocal kinetic energy approximation and the density relaxation error @xcite . looking at the values of sdre for the two hybrids considered in the table , it can be seen that in both cases these methods outperform the corresponding semilocal ones ( negative values of sdre ) . this result reflects the good quality of the electron densities computed with hybrid embedding methods ( see next section ) and evidences once more the importance of the density overlap in this context . in fact , the larger improvement is obtained for bhlyp ( sdre of -33.51% ) due to its higher content of exact exchange with respect to pbe0 ( sdre of -12.97% ) . considering the exchange contributions to the embedding error ( @xmath38 ) we note immediately that , for both methods , they are in general of opposite sign and of the same order of magnitude as the @xmath48 terms . hence , an important error cancellation effect occurs in all cases . this effect can be measured by considering the exchange differential error ( xde ) and the exchange differential relative error ( xdre ) statistical indicators @xcite @xmath50 which provide ( by increasingly negative values ) an estimation of the improvement due to the error cancellation effect induced by the additional approximation on the exchange nonadditive embedding terms . for bhlyp and pbe0 the xde and xdre values are negative , indicating , as expected , an error cancellation between the kinetic - relaxation and the exchange contributions . however , the xde is three times larger for pbe0 than for bhlyp , so in the former case a much more effective error cancellation occurs . this explains why pbe0 has the smaller energy error @xmath47 in tab . [ tab : enerr ] , despite it has kinetic - relaxation errors significantly larger than bhlyp . in this section we consider the ability of different embedding methods to reproduce the electron density of supermolecular charge - transfer systems . this is an important test for embedding approaches and provides direct insight into the quality of the embedding potential @xcite . to discuss this issue we report in fig . [ denserr ] the plot of the absolute deviation of the plane - averaged densities ( @xmath51 ; see eq . ( [ drho ] ) ) for the c@xmath14h@xmath15-f@xmath14 complex , as an example . for other systems similar plots are obtained ( not reported ) . absolute deviation from the reference supermolecular electron density of plane - averaged embedding densities obtained from different embedding approaches , for the c@xmath14h@xmath15-f@xmath14 complex . the filled circles on the @xmath52-axis denote the atoms positions . ] the figure shows that , unlike for hydrogen bond systems @xcite , which are prototypical examples for fde - based studies , the density errors are quite different for different methods , especially if hybrid and gga approaches are compared . this trend can be well captured by considering the integrated errors on embedding densities ( @xmath24 ; see eq . ( [ eq : xi ] ) ) , as resulting from different methods , which are reported in tab . [ tab : dens_err ] . lcccc & & + systems & bhlyp & pbe0 & blyp & pbe + nf@xmath16-hcn & 0.13 & 0.24 & 0.29 & 0.29 + c@xmath14h@xmath15-f@xmath14 & 1.58 & 2.75 & 6.94 & 6.35 + nf@xmath16-hnc & 0.41 & 0.49 & 0.58 & 0.58 + c@xmath14h@xmath15-cl@xmath14 & 3.34 & 4.31 & 5.83 & 5.77 + nh@xmath16-f@xmath14 & 2.24 & 4.36 & 9.84 & 9.59 + nf@xmath16-clf & 0.64 & 0.99 & 1.64 & 1.73 + nf@xmath16-hf & 0.60 & 0.74 & 0.96 & 0.94 + c@xmath14h@xmath14-clf & 3.14 & 4.32 & 6.03 & 6.02 + hcn - clf & 1.76 & 2.33 & 3.13 & 3.21 + nh@xmath16-cl@xmath14 & 3.98 & 5.48 & 7.45 & 7.60 + h@xmath14o - clf & 2.37 & 3.41 & 4.89 & 5.06 + nh@xmath16-clf & 8.07 & 9.37 & 11.27 & 11.15 + + mae & 2.35 & 3.23 & 4.90 & 4.86 + these data , along with the proportionality of the @xmath53 profiles reported in fig . [ denserr ] ( the embedding errors in the density have a similar spatial distribution ) , show that for all systems the errors on embedded densities are directly correlated to the amount of exact exchange included in the calculations ( see also next section ) . in fact , the best results are found for bhlyp , which includes 50% of hartree - fock exchange and has a mae of only 2.35 , comparable to that found in the case of hydrogen - bond and dipole - dipole interactions @xcite . on the contrary , methods based on gga xc functionals yield significantly larger errors , more than twice as large as the bhlyp ones . we note that such large variations of the embedding density error with the method are rather unusual and are not encountered in the case of hydrogen bond and dipole - dipole complexes @xcite . this indicates the special relevance of the inclusion of exact exchange for the treatment of charge - transfer complexes via fde - based methods and fully agrees with the analysis made in the previous section about the role of relaxation effects for the energy errors of different methods . beside the total quality of the embedding density , in the present context another quantity of great interest is the net charge transfer predicted by different methods for the complexes under examination . the results for the various approaches , together with reference mp2 values are reported in tab . [ tab : ct ] . lrccrccrc systems & @xmath54 & @xmath55 & @xmath56 & @xmath54 & @xmath57 & @xmath56 & @xmath54 & mp2 + & & & & & & + nf@xmath16-hcn & & 0.017 & 0.011 & & 0.012 & 0.010 & & 0.009 + c@xmath14h@xmath15-f@xmath14 & & 0.082 & 0.014 & & 0.029 & 0.015 & & 0.018 + nf@xmath16-hnc & & 0.007 & 0.007 & & 0.006 & 0.006 & & 0.005 + c@xmath14h@xmath15-cl@xmath14 & & 0.093 & 0.044 & & 0.061 & 0.045 & & 0.053 + nh@xmath16-f@xmath14 & & 0.105 & 0.018 & & 0.032 & 0.015 & & 0.026 + nf@xmath16-clf & & 0.035 & 0.013 & & 0.015 & 0.009 & & 0.013 + nf@xmath16-hf & & 0.025 & 0.013 & & 0.018 & 0.012 & & 0.018 + c@xmath14h@xmath14-clf & & 0.102 & 0.051 & & 0.067 & 0.049 & & 0.052 + hcn - clf & & 0.050 & 0.028 & & 0.033 & 0.025 & & 0.034 + nh@xmath16-cl@xmath14 & & 0.122 & 0.060 & & 0.064 & 0.044 & & 0.061 + h@xmath14o - clf & & 0.072 & 0.034 & & 0.041 & 0.028 & & 0.042 + nh@xmath16-clf & & 0.217 & 0.209 & & 0.163 & 0.126 & & 0.177 + & & & & & & & & + mse & & 0.035 & -0.001 & & 0.003 & -0.011 & & + mae & & 0.035 & 0.007 & & 0.005 & 0.011 & & + mare & & 118.07% & 18.64% & & 17.50% & 25.63% & & + & & & & & & + nf@xmath16-hcn & & 0.016 & 0.011 & & 0.013 & 0.011 & & 0.009 + c@xmath14h@xmath15-f@xmath14 & & 0.074 & 0.013 & & 0.039 & 0.013 & & 0.018 + nf@xmath16-hnc & & 0.007 & 0.007 & & 0.006 & 0.006 & & 0.005 + c@xmath14h@xmath15-cl@xmath14 & & 0.093 & 0.044 & & 0.074 & 0.044 & & 0.053 + nh@xmath16-f@xmath14 & & 0.101 & 0.017 & & 0.052 & 0.015 & & 0.026 + nf@xmath16-clf & & 0.037 & 0.013 & & 0.024 & 0.011 & & 0.013 + nf@xmath16-hf & & 0.025 & 0.013 & & 0.020 & 0.013 & & 0.018 + c@xmath14h@xmath14-clf & & 0.101 & 0.050 & & 0.081 & 0.049 & & 0.052 + hcn - clf & & 0.051 & 0.028 & & 0.040 & 0.027 & & 0.034 + nh@xmath16-cl@xmath14 & & 0.125 & 0.061 & & 0.089 & 0.050 & & 0.061 + h@xmath14o - clf & & 0.074 & 0.035 & & 0.054 & 0.030 & & 0.042 + nh@xmath16-clf & & 0.221 & 0.206 & & 0.191 & 0.162 & & 0.177 + & & & & & & & & + mse & & 0.035 & -0.001 & & 0.015 & -0.006 & & + mae & & 0.035 & 0.006 & & 0.015 & 0.007 & & + mare & & 113.80% & 19.26% & & 48.22% & 21.00% & & + @xmath44 the statistical indicators are computed using the error @xmath58 with @xmath59ks , gks , emb . the table shows the well - known tendency of the gga xc functionals to overestimate the charge transfer @xcite , so that blyp and pbe yield for @xmath57 mares well above 100% , while much better results ( mare of 17.5% ) are obtained by including a large fraction of exact exchange , as in bhlyp . however , a different situation is observed when embedding calculations are considered . these have , in fact , a marked tendency towards an underestimation of the charge transfer , compared to the corresponding supermolecular calculation on the whole system , probably related to the excessive repulsive character of the approximated embedding potential . this behavior is also much more pronounced for gga methods than for hybrids . as a consequence , the bhlyp results are slightly worsened compared to the reference mp2 ones when embedding is used , while the semilocal methods display a strong improvement . in order to shine light on this finding , we consider that embedding densities computed at the semilocal level of theory , when used to compute the energy , provide higher energies for the complexes with respect to corresponding densities from supermolecular calculations ( semilocal functionals stabilize the energy of charge - transfer complexes by overestimating the charge transfer to compensate for the absence of long - range interactions @xcite ) . because , on the other hand , the isolated systems are not affected by this problem , this leads in general to a reduction of the computed binding energy ( see tab . [ tab : enerr ] ) , thus effectively compensating for the tendency of the semilocal xc functionals to over - bind . this explains the fact that in tab . [ tab : enerr ] , for semilocal functionals , @xmath42 has a much smaller mae and mare than @xmath43 . it is worth noting that the same considerations also apply to the hybrid approaches , but in these cases the effect is smaller . in fact , pbe0 binding energies are slightly improved ( on average ) when computed via an embedding approach , while bhlyp results end up to suffer from a slight underbinding . in previous sections we evidenced the important role of the exact exchange for the description of embedding energies and densities , by comparing the performance of fde - based methods using semilocal xc functionals with that of gks - fde approaches using hybrid functionals based on exactly the same semilocal approximation . to push this investigation one step further here we consider more generally the family of one - parameter hybrid functionals @xmath60 where @xmath61 is the hartree - fock exchange , @xmath62 is a parameter , and @xmath63 is the lee - yang - parr correlation energy functional @xcite . ( [ eq : family ] ) reduces to blyp for @xmath64 , to bhlyp for @xmath65 , and was already used to study the role of exact exchange in embedding calculations of hydrogen - bond and dipole - dipole interacting systems in ref . . [ fig : alpha ] reports , as a function of @xmath62 , the errors on the embedding energies for selected systems as well as the individual contributions @xmath66 and @xmath38 . for most of the systems the error @xmath47 is found to decrease with @xmath62 until @xmath67 , and then to slightly increase for larger values of the parameter . this behavior is a consequence of an error cancellation between the @xmath66 contribution ( middle panel ) , which starts from a rather large positive value and decreases with @xmath62 , and the @xmath38 term ( right panel ) , which shows instead an evident parabolic shape with minima located around @xmath68 and provides mostly negative contributions for @xmath69 . this behavior is quite different from the one observed for dipole - dipole and hydrogen - bond interactions ( see figure i of ref . ) , which is generally characterized by positive contributions of @xmath38 , increasing with @xmath62 . the present finding thus shows that for charge - transfer systems a more delicate balancing must be found between the need to increase the exact - exchange contribution , in order to reduce the density overlap , and the necessity to obtain the correct description of long - range xc interactions in the hybrid functional . errors on the embedding density for selected systems as computed with the family of functionals of eq . ( [ eq : family ] ) . in all calculations we used the revapbek functional for the non - additive kinetic energies and a supermolecular def2-tzvppd basis set . ] the family of one - parameter hybrid functionals can be also be employed to study the effect of the exact exchange on the description of embedded densities . the result of this study was already partially anticipated in previous sections and it is clearly shown in fig . [ adens ] : by increasing the amount of exact exchange the error on the embedding density is monotonically decreased . as mentioned in previous discussions and in ref . this behavior is mainly due to the reduction of the overlap of the subsystem densities consequent to the reduction of the sie . of course , the effect is somehow attenuated for very large values of the parameter @xmath62 because of the increased importance of the semilocal approximation used for the exchange term in the gks - fde scheme in these cases , as shown in right panel of fig . [ fig : alpha ] . finally , we can consider the ability of methods including different amounts of exact exchange to correctly reproduce the charge transfer . we thus report in fig . [ ct_alpha ] the difference @xmath70 between the charge transfer computed for each value of @xmath62 , with or without the embedding treatment , and the reference values obtained from mp2 relaxed densities . the plot shows that considering full gks calculations the charge transfer is generally overestimated at the gga level and better agreement with the reference is only found when a significative amount of hartree - fock exchange is considered . on the other hand , for embedding calculations the gks charge transfer can be only reproduced when exact exchange is included , while calculations using a semilocal xc potential yield a strong underestimation with respect to the corresponding supermolecular calculation . in fact , the charge transfer computed from embedding methods appears to be almost independent from the percentage of exact exchange included in the xc functional . to understand these results we must consider that within fde - based methods the number of electrons on each subsystem is fixed a priori , so that no real charge transfer between them can occur . the charge transfer must be instead mimicked by a strong polarization of the two subsystem densities , in such a way that their sum will give the same spatial distribution as the electron density of the full system . the role of the embedding potential is thus to provide a polarization of the subsystems , while each subsystem must respond properly to this perturbation . in the case of semilocal functionals they are well known to provide a poor description of polarization @xcite , thus a poor description of charge transfer must be expected . on the contrary , hybrid functionals have been shown to give a good response to polarization perturbations within the gks - fde scheme @xcite , thus they can provide a good reproduction of the charge transfer between the subsystems . difference between the charge transfer computed with functionals defined by eq . ( [ eq : family ] ) and the reference mp2 value . in the left panel we report the results of gks calculations on whole systems ; in the right panel the corresponding embedding results are shown . in all calculations we used the revapbek functional for the non - additive kinetic energies and a supermolecular def2-tzvppd basis set . ] in an analysis of the performance of embedding methods for the simulation of ground - state charge - transfer complexes it is obviously interesting to study in details what is the relationship between the outcome of the calculations and the magnitude of the charge transfer . however , this study can not be performed straightforwardly by a simple comparison of different systems displaying different values of the charge transfer because in this way also many other details of the interaction are changed at the same time . for this reason in this section we prefer to focus the attention on a single system , namely the linear hcn - clf complex , and force a modulation of the charge transfer through the application of an external constant electric field directed along the axis of the complex . in fig . [ field ] we report the error on embedding energy ( @xmath47 ) , the error on embedding density ( @xmath24 ) , and the error on the computed net charge on the dihalogen monomer clf ( @xmath71 ) as a function of the charge transfer induced ( at the gks level ) by the applied external field for the two functionals bhlyp and blyp . the corresponding electric field is reported in the inset of fig . [ field ] . errors on embedding energy ( @xmath47 ) , embedding density ( @xmath24 ) , and the computed signed charge transfer on the dihalogen monomer clf ( @xmath71 ) as a function of the charge transfer induced ( at the gks level ) by the applied external field , for bhlyp and blyp . the inset reports the external electric field used to generate different charge transfer values . ] the figure shows that both functionals display the same behavior with the induced charge transfer , with all errors increasing as the charge transfer grows . in fact , the minima of the embedding errors are approximately located at zero charge transfer . the energy error decomposition ( not reported ) confirms that the increase of the energy error is essentially due to a large increase in the kinetic and relaxation contributions . moreover , the lowest panel of fig . [ field ] shows , in agreement with our analysis of previous sections , that the increase of charge transfer is in linear correspondence with an increase of the inaccuracy in its evaluation by embedding methods . this implies , as also shown by a comparison of the two upper panels of fig . [ field ] with the lower one , that the embedding errors are correlated ( almost linearly ) with the inaccuracy of embedding approaches to correctly describe the amount of charge transfer . all previous investigations have been carried out at reference geometries . however , for charge - transfer complexes large differences in the equilibrium geometry can be expected for different methods . in particular , hybrid xc functionals are able to produce equilibrium structures very close to the reference ones but semilocal xc approximations lead to significantly shorter intermolecular distances @xcite . as shown in ref . the geometry and the intermolecular distance play an important role in determining the accuracy of embedding calculations . it is therefore interesting , in the present context , to assess this issue for the limits represented by the structures optimized at the semilocal blyp and hybrid bhlyp level . to this end we consider in fig . [ geom ] the error on embedding energy ( @xmath47 ) , the error on embedding density ( @xmath24 ) , and the error on the computed monomer charge ( @xmath71 ) at different geometries interpolating linearly between the blyp and the bhlyp optimized geometries of the complex c@xmath14h@xmath15f@xmath14 ( similar results are found for other systems ) . the intermolecular distance between the f@xmath14 monomer and the plane of the ethylene is @xmath72 and @xmath73 for blyp and bhlyp , respectively . errors on embedding energy ( @xmath47 ) , embedding density ( @xmath24 ) , and the computed net charge on the ethylene monomer ( @xmath71 ) at different geometries interpolating between the blyp and bhlyp equilibrium geometry of the c@xmath14h@xmath15-f@xmath14 complex . in the inset the embedding error decomposition for the bhlyp case is reported . ] the figure shows the important role of the geometry for all the embedding results . this finding can be understood considering that at small intermolecular distances , i.e. at the blyp geometry , the overlap between the electron densities of the two subsystems rapidly grows . moreover , because at reduced distances between the subsystems the orbital interactions are enhanced , also the charge transfer is highly increased , together with the error in its evaluation . as a consequence , large errors can be expected for the embedded densities , due to the well - known limitations of nonadditive kinetic potentials for strongly overlapping densities @xcite . similarly , large errors are found for the embedding energies . in this case an energy decomposition of the bhlyp embedding error ( see inset of fig . [ geom ] ) further confirms that the increase of the error is essentially due to a large increase of the relaxation - kinetic term , while only a minor role is played by the exchange term @xmath38 . nevertheless , the latter being always opposite in sign to the relaxation - kinetic contribution provides a moderate error cancellation which improves the final performance of the hybrid method . the results of fig . [ geom ] indicate in conclusion that , for the simulation of charge - transfer complexes via embedding approaches , the semilocal methods may display serious problems related to the use of relaxed structures ( e.g. in geometry optimizations or molecular dynamics ) , beside the limitations already highlighted in the previous sections . therefore , the hybrid approaches appear definitely more robust in such applications . in previous sections we showed that only embedding approaches making use of hybrid xc functionals can reproduce the ground - state charge transfer of the investigated complexes with reasonable accuracy , while a strong underestimation occurs when semilocal approaches are used . this fact has important consequences on the accuracy of embedding calculations , that nevertheless are partially , or even mostly , mitigated by error compensation effects . the improper description of the charge transfer can have however more subtle effects , especially on the one - electron spectrum of the subsystems . this is an important issue , since the molecular orbitals of the embedded subsystems can be used as the basis for an analysis of the properties of the interacting subsystems and/or as input quantities in response calculations @xcite . density of states ( dos ) for both gks and hybrid bhlyp ( upper panel ) and the semilocal blyp ( bottom panel ) embedding calculations on the charge - transfer complex nh@xmath16-f@xmath14 . the solid black line indicates gks dos of the whole system ; the red dashed and solid turquoise lines indicate dos from embedding calculations on the monomers nh@xmath16 and f@xmath14 respectively . in the insets the gks dos ( thick black line ) is compared with the sum of the two subsystems doss ( dashed magenta line ) . energy values are in ha . ] the driving force for ground - state charge transfer in molecular complexes is a sizable difference in the chemical potential between the constituting ( isolated ) subsystems , so that , when the subsystems start to interact , a fraction of the electron density must be moved from one fragment to another to equilibrate it . the final chemical potential of the complex is an average between those of the initial non - interacting subsystems and it is determining the offset for the kohn - sham potential , thus de facto fixing the energy of the molecular orbitals . in an ideal exact fde - based embedding calculation the embedding potential shall be able , by definition , to yield for the embedded subsystems the same chemical potential as for the full complex and , of course , shall induce a polarization able to mimic the exact charge - transfer . as a side consequence , the single - particle spectrum of the subsystems will be also correct ( as compared to the corresponding supermolecular calculation ) . similar conclusions apply for any approximated embedding calculation having sufficient accuracy . on the other hand , if the embedding procedure fails to reproduce the correct charge transfer ( i.e. correct with respect to the calculation on the full system ) , the polarization exerted by the embedding potential on the two subsystems will be insufficient , thus the chemical potential of the two fragments will be rather different from the one of the full system ( too large in one subsystem and too low in the other ) . consequently , the molecular orbitals of the two subsystems will lay at too high ( for one subsystem ) and too low energies ( for the other ) . the two cases described above fit very well to the embedding calculations based on hybrid and semilocal functionals , respectively . this is nicely shown in fig . [ fig : dos ] where , as an example , we report the density of states ( dos ) of the nh@xmath16-f@xmath14 complex compared to that of the two embedded subsystems as resulting from bhlyp and blyp embedding calculations . the comparison is in this case rather fair , since the two fragments composing the nh@xmath16-f@xmath14 complex interact only weakly so that the molecular orbitals of the full system can be easily attributed to either of the constituent subsystems . the figure shows very clearly that bhlyp embedding calculations yield subsystems orbitals which match well the ones from the full system ( within 0.01 ev ) . instead , the orbitals from the blyp embedding calculations are systematically shifted with respect to the ones of the corresponding supermolecular calculation , with f@xmath14 orbitals shifted at lower energies and nh@xmath16 ones shifted at higher energies , consistently with the underestimation of charge transfer from nh@xmath16 to f@xmath14 . finally , we recall in favor of hybrid calculations that even supermolecular calculations using semilocal xc functionals provide a poor description of the single - particle spectrum , while better results can be only obtained by reducing the sie @xcite . we applied methods based on the frozen density embedding theory to study the electronic properties of ground - state charge - transfer complexes and to clarify the ability of such approaches to treat accurately this class of systems and interactions . this topic is of particular interest due to the growing popularity of embedding approaches in the field of computational chemistry and the importance of charge - transfer complexes in many chemical applications . however , to date , embedding calculations have been mainly presented for hydrogen - bond and dipole - dipole interactions , while only few applications dealt with systems where charge transfer plays a significant role @xcite . our work clarifies some of the motivations for this lack of embedding calculations concerning charge - transfer systems , showing that standard embedding approaches , making use of semilocal xc functionals , display several drawbacks in this context . moreover , despite they can generally yield interaction energies of good quality ( with respect to accurate references ) , this is due to a strong error cancellation effect and does not reflect a property of the corresponding kohn - sham supermolecular calculation . for these reasons and because semilocal xc functionals are known to perform rather poorly for charge - transfer complexes even in conventional kohn - sham supermolecular calculations , the use of such approaches does not appear appropriate . the introduction of the gks - fde method mitigates different problems in the fde - based calculations and thus provides a valuable tool to handle embedding calculations of charge - transfer systems . in fact , in the present work we showed the relevance of exact exchange contributions to reduce the embedding errors and improve the overall performance of the calculations , through a reduction of the coulomb self - interaction error and a better description of subsystems electronic properties . in conclusion , fde - based approaches making use of hybrid xc functionals appear as reliable and efficient methods to handle electronic structure calculations of systems characterized by charge - transfer interactions . this suggest the possibility of future studies in this direction making use of these tools to investigate charge - transfer interactions in complex environments . this work was partially funded by the european research council ( erc ) starting grant fp7 project dedom , grant no . the authors thank turbomole gmbh for providing the turbomole program package and m. margarito for technical support .

compton scattering in quasi - spherical accretion flows is thought to be responsible for the x - ray spectra of many accreting compact objects . for example , the hard x - ray spectra of weakly - magnetic accreting neutron stars can be accounted for if a geometrically thick , hot medium surrounds the neutron star and inner accretion disk ( see , e.g. , psaltis , lamb , & miller 1995 ) . the spectral signature of isolated neutron stars accreting from the interstellar medium , which is required for their positive identification , is determined mainly by the efficiency of compton upscattering of thermal photons emitted from the stellar surfaces ( see , e.g. , zampieri et al . moreover , compton scattering plays a major role in the formation of the high - energy spectra of advection - dominated accretion flows , which may be the relevant mode of accretion onto galactic and supermassive black holes ( see , e.g. , esin et al . 1997 ) . compton scattering in static and moving media , even in plane - parallel or spherical symmetry , couples strongly all three independent phase - space coordinates of the photons , namely their energy , direction of propagation , and spatial coordinate . for this reason , monte carlo methods have been used extensively in calculating comptonization spectra , because of their easy implementation and natural treatment of the diverse length scales in an accretion flow ( see , e.g. , laurent & titarchuk 1999 for a recent study ; methods and results based on monte carlo treatments have been reviewed by pozdnyakov , sobol , & sunyaev 1983 and will not be discussed here ) . alternatively , compton scattering can be studied by solving the energy - dependent radiative transfer equation or its moments . implementing such a method is more difficult and often relies on a number of approximations for describing the interaction of radiation with matter ( see , however , poutanen & svensson 1996 and hsu & blaes 1998 for complete solutions in static media ) . however , at the same time it is substantially faster and offers significant advantages when the radiative transfer problem is coupled to a solution of the fluid dynamics , as well as in treating steep spectra and processes that do not conserve photons . over the last twenty five years several authors have derived the photon kinetic or radiative transfer equation and their moments for compton scattering in static and moving media ( see , e.g. , kompaneets 1957 ; babuel - peyrissac & rouvillois 1969 ; pomraning 1973 ; chan & jones 1975 ; payne 1980 ; blandford & payne 1981a ; thorne 1981 ; fukue , kato , & matsumoto 1985 ; madej 1989 , 1991 ; titarchuk 1994 ) . the moment equations derived by these various sets of authors have been widely used in studies of comptonization by static media ( see , e.g. , katz 1976 ; shapiro , lightman , & eardly 1976 ; sunyaev & titarchuk 1980 ) or by strong shocks and accretion flows onto compact objects ( see , e.g. , blandford & payne 1981b ; payne & blandford 1981 ; lyubarskij & sunyaev 1982 ; colpi 1988 ; riffert 1988 ; mastichiadis & kylafis 1992 ; titarchuk & lyubarskij 1995 ; turolla et al . 1996 ; titarchuk , mastichiadis , & kylafis 1997 ) , either under the diffusion approximation or with closure relations that were specified a priori . most of the above analyses of compton scattering in moving media were performed in the system frame and to first order in the electron bulk velocity . on the other hand , schmid - burgk ( 1978 ) , zane et al . ( 1996 ) , and titarchuk & zannias ( 1998 ) have solved the general - relativistic radiative transfer equation for compton scattering without the need for specifying a priori closure relations . in paper i ( psaltis & lamb 1997 ) of this series , the time - dependent photon kinetic and radiative transfer equations were derived , as well as their zeroth and first moments that describe absorption , emission , and spontaneous and induced compton scattering in static and moving media , correcting various errors in the literature . the system - frame equations that were derived are valid to first order in @xmath2 and @xmath3 , and to _ second _ order in @xmath1 , where @xmath4 and @xmath5 are the electron mass and temperature , @xmath6 is the photon energy , and @xmath1 is the bulk velocity of the electrons in units of the speed of light ; the fluid - frame equations that were derived are valid for arbitrary values of the bulk velocity @xmath1 . using these equations it was argued in paper i that the effects of comptonization by the bulk electron velocity that are described by the terms that are second - order in @xmath1 are qualitatively different than and can become at least as important as the effects described by the terms that are first - order in @xmath1 , even when @xmath1 is small . as a result , these second - order terms should generally be retained ( see also yin & miller 1995 ) . in this paper , i describe a numerical algorithm for the solution of time - independent radiative transfer problems in systems with spherical symmetry . even though the natural reference frame for solving transport problems is the one comoving with the flow , i work in the system frame , which can be any inertial frame in which the central object is at rest . this is important for the fast algorithm described below , because the moment equations in the system frame are always parabolic ( as shown in 3 ) in contrast to the fluid - frame equations ( turolla , zampieri , & nobili 1995 ; smit , cernohorsky , & dullemond 1997 ; see also krner & janka 1992 ) . moreover , in problems of mass accretion onto a compact star , the boundary conditions are often more easily implemented in the system frame . i also neglect any general relativistic effects , which can be shown to introduce only small quantitative corrections to the emerging spectra ( papathanassiou & psaltis 2000 ) , and truncate the transfer equation keeping only terms up to second order in @xmath1 . the algorithm used is a generalization of the method of variable eddington factors ( mihalas 1980 ; see also mihalas 1978 , p. 157158 ) . it is therefore accurate for systems of arbitrary optical depth and does not require a priori specification of closure relations . it is based on the iterative solution of both the transfer equation and the systems of its first two moments and hence the validity and accuracy of the solution can be verified explicitly at the end of each run . the method also requires only a small number of iterations in order to converge to the solution and is therefore ideal for future extensions to time - dependent and multi - dimensional transport problems . in 2 , i summarize the results from paper i and introduce the notation . in 3 , i discuss the mathematical character of the system of equations and in 4 , i present the numerical algorithm . in 5 , i describe the results of numerical solutions of the transfer equations for a variety of situation . finally , in 6 , i briefly summarize the implications of the above for comptonization spectra in static and moving media . _ units_throughout this paper i set @xmath7 , where @xmath8 is planck s constant , @xmath9 is boltzmann s constant , and @xmath10 is the speed of light . i also normalize all the spatial coordinates to the inner radius of the flow , the electron temperature @xmath5 and photon energy @xmath6 to the electron rest mass @xmath4 , and the electron density @xmath11 to its value at the inner radius of the flow . finally , i normalize the emission and absorption coefficients @xmath12 and @xmath13 to the inverse of the electron scattering mean - free path in the thomson limit . _ the electron gas_let @xmath14 be the system - frame three - velocity of a given electron . i define the local bulk velocity @xmath15 of the electrons measured in the system frame as @xmath16 where the sharp brackets indicate the average over the local electron velocity distribution . i also assume that the electron velocity distribution in the fluid frame is a relativistic maxwellian and therefore @xmath17 where i have neglected terms of order @xmath18 and higher . _ the radiation field_i describe the radiation field at any position in the system frame with coordinate vector @xmath19 using the monochromatic specific intensity @xmath20 , where @xmath21 is the direction of propagation and @xmath6 is the photon energy . here i consider only unpolarized radiation and hence i have suppressed the dependence of the specific intensity on polarization . because of the assumed spherical symmetry of the problem , the radiation field depends only on the radial distance from the center of symmetry ( i.e. , the @xmath22component of the vector @xmath19 ) , on the angle @xmath23 between the radial direction and the direction of propagation ( i.e. , @xmath24 ) , and on the photon energy @xmath6 . i define the first five moments of the monochromatic specific intensity as @xmath25 the non - zero components of the first five moments of the specific intensity for systems with spherical symmetry are given in appendix a. _ the transfer equation and its moments_following mihalas ( 1980 ; see also mihalas 1978 ) , i solve the radiative transfer equation in a plane that contains the center of symmetry of the system , between the inner and outer radii @xmath26 and @xmath27 of the flow , using the coordinates @xmath28 and @xmath29 . the radiative transfer equation along a ray of constant impact parameter @xmath30 is @xmath31 where the signs ` @xmath32 ' and @xmath33 correspond to the equations for the outgoing and incoming rays respectively and i have neglected the effects of induced compton scattering . in equation ( [ rteq ] ) i have used the optical depth along the ray defined as @xmath34 dz$ ] , where @xmath35 at the outer boundary and @xmath36 is the angle - integrated thomson scattering cross section . @xmath37 is the generalized source function for absorption , emission , and compton scattering in systems with spherical symmetry and is given in appendix b. the equations for the moments of the specific intensity can be derived either directly from equation ( [ rteq ] ) or from equations ( 34 ) and ( 40 ) of paper i. written in compact form and using the dimensionless quantities defined above , the moment equations are term reported in paper i ( psaltis & lamb 1997 ) contained a typo that has been corrected here . ] @xmath38 where @xmath39 , @xmath40 , @xmath41 , and @xmath42 . when the electron density is uniform , then the dimensionless quantity @xmath43 is equal to the electron scattering optical depth in the radial direction measured from @xmath27 . the coefficients in equations ( [ 0th3 ] ) and ( [ 1st ] ) are given in appendix c. _ boundary conditions_the boundary conditions for both the transfer equation and the system of its moments depend on the specific problem under study . for the model problems discussed in 5 , i shall assume that the flow is not illuminated from the outside , i.e. , @xmath44 at the outer boundary , and specify the radiation flux at the inner boundary as @xmath45 for solving the moments of the radiative transfer equation i shall specify , for all photon energies @xmath6 , the first moment of the monochromatic specific intensity at the inner radius , i.e. , @xmath46 as well as the ratio of the first to the zeroth moment of the monochromatic specific intensity at the outer radius @xmath47 the quantity @xmath48 is not known a priori but is calculated in a self - consistent way in the iterative method described below . i shall also set at all radii @xmath49 and @xmath50 which reflect the fact that the photon occupation number and the radiation flux are finite quantities . in the method of variable eddington factors , the system of partial differential equations ( [ 0th3 ] ) and ( [ 1st ] ) is solved iteratively with the radiative transfer equation ( [ rteq ] ) , given a set of eddington factors computed from the previous iteration . the boundary conditions and the method used for the solution of the system depends on the character of the partial differential operator . for example , the system of moment equations in the frame _ comoving _ with the flow ( or the local lorentz frame when gravitational effects are taken into account ) may be hyperbolic or elliptic , depending on the velocity field ( turolla et al . 1995 ) . in this section i follow the procedure outlined by ames ( 1992 , pp . 812 ) to show that the _ system - frame _ , energy dependent moment equations always form a parabolic system of partial differential equations , thus simplifying the implementation of the numerical algorithm . i define the quantities @xmath51 and @xmath52 , which together with the partial differential equations ( [ 0th3 ] ) and ( [ 1st ] ) , and the equations for the differentials @xmath53 , @xmath54 , @xmath55 , and @xmath56 , form an @xmath57 system of algebraic equations for the eight derivatives @xmath58 \cdot \left [ \begin{array}{c } \partial_\e j\\ \partial_{\tau_r } j\\ \partial_\e h^r\\ \partial_{\tau_r } h^r\\ \partial_\e l\\ \partial_{\tau_r } l\\ \partial_\e m^r\\ \partial_{\tau_r } m^r\\ \end{array } \right ] & = & \qquad\qquad\qquad\qquad\nonumber\end{aligned}\ ] ] @xmath59\;.\ ] ] equating the determinant of the coefficient matrix to zero , i obtain the characteristic equation for the system of equations , @xmath60 the coefficients @xmath61 and @xmath62 are proportional to the electron temperature and the square of the bulk velocity ( see appendix c ) and hence for the problems under consideration they are never zero . therefore , for a finite electron energy @xmath6 , the system of partial differential equations ( [ 0th3 ] ) and ( [ 1st ] ) has only one multiple , real characteristic direction , i.e. , @xmath63 , and hence is always parabolic . the difference with respect to the moment equations in the comoving frame arises from the presence of second order energy derivatives in the differential operator . equation ( [ rteq ] ) is an integro - differential equation for the specific intensity of the radiation field and therefore any numerical solution of this equation is not trivial to obtain . on the other hand , the system of equations ( [ 0th3 ] ) and ( [ 1st ] ) depends on two variable eddington factors and an unknown outer boundary condition , which can be computed only when the solution of equation ( [ rteq ] ) for the specific intensity is obtained . here , i employ a numerical algorithm that is a generalization of the method of variable eddington factors ( mihalas 1980 ; see also mihalas 1978 , pp . 157158 ) in order to solve iteratively equation ( [ rteq ] ) for the specific intensity of the radiation field and the system of partial differential equations ( [ 0th3 ] ) and ( [ 1st ] ) for its zeroth and first moments . this algorithm has been proven to be very efficient and gives a solution to the complete transfer equation that is limited only by numerical accuracy , independent of the optical depth of the medium . in this method , i first solve the moments of the transfer equation using an initial guess for the variable eddington factors and for the outer boundary condition . i then compute the generalized source function using the calculated moments of the specific intensity , solve the radiative transfer equation , and use this solution to update the eddington factors . i repeat the whole procedure until convergence is achieved . the moments of the transfer equation depend only on radius and photon energy , when the variable eddington factors are known . therefore , i discretize the domain of solution and the differential operators of equations ( [ 0th3 ] ) and ( [ 1st ] ) on a two dimensional mesh of @xmath64 grid points over the variable @xmath43 and @xmath65 grid points over the photon energy @xmath6 . the mesh of discrete grid points over the variable @xmath43 must resolve the rapid change of the variable eddington factors with optical depth near both boundaries of the domain of solution . in order to achieve this , i define the mesh points to be equidistant in the quantity @xmath66 where @xmath67 is the total optical depth ( and not the quantity @xmath43 ) in the radial direction measured from @xmath27 , @xmath68 is the maximum radial optical depth in the medium , and @xmath69 is a parameter that allows a combination of a logarithmic grid near the boundaries with a linear grid in the interior . because the spectra that emerge from comptonizing media are usually power laws , i use a logarithmic mesh of discrete grid points over photon energy @xmath6 . the system of moment equations involve differentiation over two density - like quantities , the radiation energy density @xmath70 and the first eddington factor @xmath71 , and two flux - like quantities , the radiation energy flux @xmath72 and the second eddington factor @xmath73 . i discretize density - like quantities in a shell - centered fashion and flux - like quantities on the grid points . let @xmath74 be the value of a physical quantity at the @xmath75th grid point in the variable @xmath43 and the @xmath76th grid point in photon energy . away from the boundaries , i differentiate density - like quantities over the variable @xmath43 on the grid points as @xmath77 and flux - like quantities as @xmath78 in an explicit or fully implicit scheme . in both equations i approximate the first derivative of a physical quantity @xmath79 with respect to photon energy by @xmath80 and the second derivative with respect to photon energy by @xmath81 where the index ` @xmath75 ' in the quantity @xmath74 may also be fractional , corresponding to a shell center , as needed by an implicit differencing scheme over the variable @xmath43 . the energy differencing scheme ( [ 1stder])([2ndder ] ) can not be used directly in differencing the radiation energy density and flux in equations ( [ 0th3 ] ) and ( [ 1st ] ) because it does not guarantee particle conservation when only scattering is taken into account ( chang & cooper 1970 ) . this is important when the energy spectrum has reached quasi - equlibrium and small numerical errors lead to significant deviation from particle conservation ( see eq . [ 10 ] in chang & cooper 1970 ) . a better differencing scheme can be deduced by examining the properties of the differential operator at that limit . the solution of the moment equations reaches quasi - equlibrium when the photon mean free path is very small compared to any characteristic length scale in the system . in that limit , @xmath82 , and when only processes that conserve photon number are operating , the moment equation of zeroth order becomes @xmath83\ ; , \label{cons}\ ] ] where @xmath11 is the dimensional electron density . the photon number flux is simply equal to @xmath84 and integrating equation ( [ cons ] ) over photon energy gives @xmath85 i.e. , photon conservation . following chang & cooper ( 1970 ) , i define a generalized photon current @xmath86 as @xmath87 and write the zeroth and first moment equations as @xmath88 where the coefficient are given in appendix d. differencing equations ( [ 0th4])([1st4 ] ) according to the scheme ( [ back])([2ndder ] ) results in strict photon number conservation in the limit of small photon mean free paths . a final point with the energy differencing is related to the requirement that the photon energy density is everywhere positive . this introduces a restriction on the spacing of the energy grid , depending on the differencing scheme . for the scheme used above it can be shown that the requirement for a positive photon energy density results in the restriction ( cf . . [ [ cons ] ] and chang & cooper 1970 , eq . [ 14 ] ) @xmath89 which can be easily met . i solve the system of equations ( [ 0th4 ] ) and ( [ 1st4 ] ) for the zeroth and first moments of the specific intensity of the radiation field , @xmath70 and @xmath72 , in @xmath64 grid points in optical depth and @xmath90 grid points in photon energy . therefore the system of equations has @xmath91 unknowns . in all the interior grid points ( @xmath92 ; @xmath93 ) i solve both equations ( [ 0th4 ] ) and ( [ 1st4 ] ) using the differencing scheme of equations ( [ back])([2ndder ] ) . for the inner spatial boundary , i.e. , for @xmath94 and for all @xmath76 , i solve equations ( [ inboundmom ] ) and ( [ 1st4 ] ) , whereas for the outer spatial boundary , i.e. , for @xmath95 and for all @xmath76 , i solve equations ( [ outboundmom ] ) and ( [ 0th4 ] ) . finally , at both energy boundaries , i.e. , for all @xmath75 and for @xmath96 , i set both the zeroth and first moments of the specific intensity equal to zero , according to boundary conditions ( [ ein ] ) and ( [ eout ] ) . the resulting system of equations is linear in both the zeroth and first moments of the specific intensity . defining the quantity @xmath97 as @xmath98 the system of equations takes the form shown in figure 1 . the matrix of coefficients is a sparse matrix and can be solved efficiently with the biconjugate gradient method ( see , e.g. , press et al . 1992 , pg . 209 ) with a preconditioner matrix . i choose the preconditioner matrix to be the coefficient matrix , with the elements at the outer two diagonal bands set equal to zero . the preconditioner matrix can then be inverted efficiently with the lu - decomposition method ( see , e.g. , press et al . 1992 , pg . 202 ) . in this method , in order to avoid round - off errors , i scale all the rows in the coefficient matrix so that the highest value of the elements in each row is unity . depending on the complexity of the particular problem , the system of difference equations can be solved to a fractional accuracy of @xmath99 within less than fifty iterations , which can be performed in a few seconds on a current workstation . in the iterative method of the variable eddington factors , the transfer equations ( [ rteq ] ) is solved at every grid point in the @xmath100 plane for the boundary conditions ( [ inbound ] ) and ( [ outbound ] ) . because the generalized source function is calculated using the results from the previous iterations , the solution of equation ( [ rteq ] ) is just the formal solution that can be obtained analytically , i.e. , @xmath101 where @xmath102 and @xmath103 or @xmath104 when @xmath105 or @xmath106 respectively . in calculating the integrals in equation ( [ formal ] ) , i use a piece - wise linear interpolation of the source function through each shell and evaluate the energy derivatives according to equations ( [ 1stder])([2ndder ] ) . the rapid convergence of the method of variable eddington factors is preserved in the current implementation for the problem of compton scattering . this is shown in figure 2 , where the maximum fractional error in the radiation energy density is plotted for each iteration , for two uniform media with different scattering optical depths . for the model problems shown , the outer radius of the scattering medium is set to @xmath107 , its electron temperature to 10 kev , and the energy - dependence of the radiation flux at the inner boundary to be that of a blackbody with a temperature of 0.5 kev . the converge is typically faster for problems with higher optical depths , because i have used @xmath108 as an initial value for the first variable eddington factor , which is closer to the real value at the limit of high optical depth . the numerical method described in this section also allows for a consistency check between the solution of the transfer equation and the solution of the moment equations that can be performed for each individual problem . for this test , equation ( [ rteq ] ) can be solved iteratively without using the solution of the system of equations ( [ 0th3])([1st ] ) , by starting with an assumed generalized source function [ e.g. , @xmath109 , solving the transfer equation for the specific intensity , updating the generalized source function , and repeating until convergence is achieved . the moments of the specific intensity can then be obtained in this way and compared directly with the moments obtained by solving equations ( [ 0th3 ] ) and ( [ 1st ] ) . the desired accuracy of the calculations , which is affected mainly by the choice of the discrete mesh , depends on the particular problem . because the main goal of this study is the calculation of the spectra of accreting compact objects , the x - ray colors of which are observed to change by a few percent when the mass accretion rate changes by a factor of @xmath110 , i require for the solutions a fractional accuracy of @xmath111 at each grid point . in this section i solve the radiative transfer equation for a number of model problems with spherical symmetry that are related to various astrophysical systems and in particular to media around compact objects . in 5.1 , i discuss comptonization of soft x - ray photons by a static , uniform medium of hot electrons . in 5.2 , i solve the radiative transfer equation in a cold , divergence - free inflow ; although such a configuration requires fine tuning of the physical conditions around an accreting object and may not occur in nature , it is used here to provide physical understanding of the effect of comptonization by the bulk electron velocity . finally , in 5.3 , i discuss more realistic problems of spherical accretion onto and outflows from compact stars . for the first model problem , i consider a static , spherically symmetric , purely scattering medium and set the parameters to values that are typical for weakly magnetic , accreting neutron stars ( see psaltis et al . i set the outer radius of the scattering medium to @xmath107 , its electron temperature to 10 kev , its electron scattering optical depth to 4 , and the energy - dependence of the radiation flux at the inner boundary to be that of a blackbody with a temperature of 0.5 kev ; the normalization of the input flux is arbitrary because i have assumed that there are no sources of photons in the medium . this is a frequently solved problem that has been analyzed in detail by , e.g. , katz ( 1976 ) , shapiro et al . ( 1976 ) , and sunyaev & titarchuk ( 1980 ) . all these authors showed that the emerging radiation spectrum at energies much larger than the energy of the injected photons is largely independent of the details of the input spectrum and can be described by a power - law with an exponential cut - off at @xmath112 . here i solve this simple problem to study the fact that the variable eddington factors are not independent of energy even when the photon mean - free path is independent of energy and the medium is static ( see also the discussion in paper i ) . figure 3 shows the zeroth and first moments of the specific intensity of the radiation field , @xmath70 and @xmath72 , as a function of photon energy and radius . both quantities decrease overall with increasing radius because of the dilution of the radiation field . the energy dependence of the radiation energy density changes within one photon mean - free - path from the inner boundary but remains unchanged at larger radii . on the other hand , the radiation flux evolves throughout the scattering medium . this is a consequence of the fact that the system of moment equations has been solved by imposing an inner boundary condition on @xmath113 and an outer boundary condition on the ratio @xmath114 and not on @xmath70 itself . figure 4 shows the first eddington factor @xmath115 at two photon energies as well as the ratio of the energy - integrated second to zeroth moments of the specific intensity of the radiation field . the eddington factor depends on photon energy even though the scattering cross section and hence the photon mean - free - path are mostly energy independent at these low photon energies . this is because photons were injected into the medium with energies @xmath116 and on average gain energy at each scattering . as a result , photons emerging with low energy have experienced on average a smaller number of scatterings than photons with higher energy and therefore their distribution is less isotropic . figure 5 compares the emerging radiation spectrum obtained using the method of variable eddington factors with the spectra obtained using the same boundary conditions but two energy - independent prescriptions of the first eddington factor that are commonly used ; the second eddington factor does not enter the calculation when the bulk velocity of the electrons is zero . note that neither prescription is the correct diffusion approximation for compton scattering because they both neglect the energy dependence of @xmath71 on photon energy shown in figure 4 . however , the discrepancy between the self - consistent solution and the ones obtained using prescribed eddington factors is small for the prescription that depends on optical depth and can be as large as @xmath117% at high photon energies for the prescription that is independent of optical depth . this discrepancy increases when the ratio of the photon mean - free path to the smallest characteristic length scale in the system increases . in a moving medium , the photons gain energy by scattering off the fast moving electrons . as measured by an observer _ comoving with the medium _ , the photon energy density increases mostly due to the convergence of the flow , which is described by terms that are proportional to @xmath118 ( blandford & payne 1981 ) . on the other hand , the flux that is measured by an observer _ at infinity _ increases because of the systematic energy exchange between photons and electrons , which is described mostly by terms that are second - order in @xmath1 ( psaltis & lamb 1997 ) . in order to investigate the effects of comptonization by the bulk electron flow and distinguish them from the effects of the convergence of the flow , i study in this section the photon - electron interaction and the formation of spectra in fast , divergence - free flows . for the model problems shown below , i set the outer radius of the scattering medium to @xmath119 , the electron temperature to zero , and the electron velocity to the divergence - free profile @xmath120 . finally , i set the electron density to a constant value , which arises from mass conservation in the flow , and characterize each flow by the total electron scattering optical depth in the radial direction . figure 6 shows the radiation spectrum at the inner boundary , the emerging radiation spectrum when only terms that are first - order in the electron bulk velocity are taken into account , as well as the emerging radiation spectrum that is correct to second order in the electron bulk velocity , for a flow with a scattering optical depth of 2 . the terms that are first - order in the electron bulk velocity describe mainly the effects of non - relativistic doppler shift and hence produce a displacement in @xmath121 of the input spectrum towards higher photon energies ( see also psaltis & lamb 2000 ) . on the other hand , the terms that are second - order in the electron bulk velocity result in a systematic photon upscattering and produce a power - law tail at photon energies higher than the average energy of the injected photons . note here that the approximation employed here of keeping , in the scattering kernel , only terms that are first order in @xmath6 limits the validity of the calculation to energies @xmath122 kev . when the total optical depth of the scattering medium , and hence the average number of scatterings that each photon experiences , are increased , then on average photons gain more energy by scattering off moving electrons . as a result , the emerging radiation spectrum is displaced towards higher photon energies , and the power - law tail becomes flatter ( see fig . when the bulk electron velocity in the flow increases , the average energy gain per scattering increases as well , and hence the power - law tail also becomes flatter ( fig . 7b ) . as a result , scattering of photons by fast moving electrons produces power - law tails at high photon energies , even in divergence - free flows , in a way that is similar to the well understood process of thermal comptonization . i finally perform simple model calculations of accretion onto and outflows from a central object by assuming a free - fall density profile and setting everywhere in the flow the electron temperature to zero and the inward radial velocity equal to the free - fall velocity from infinity onto a compact object with a radius of 10 km and a gravitational mass of @xmath123 . i also set the outer radius of the medium to @xmath119 ; at these large radii the photon mean - free path is large compared to radius and the electron bulk velocity is small compared to the speed of light that the flow does not effect significantly the photon distribution . finally , instead of specifying the mass accretion rate onto the compact object , i specify the total electron - scattering optical depth of the flow . figure 8a shows the radiation spectrum at the inner boundary , the emerging radiation spectrum when only terms that are first - order in the electron bulk velocity are taken into account , as well as the emerging radiation spectrum that is correct to second order in the electron bulk velocity . as in the model calculation discussed in 5.2 , the terms that are first - order in the electron bulk velocity describe mostly the non - relativistic doppler shifts and produce a displacement in @xmath121 of the input radiation spectrum towards higher photon energies . on the other hand , the terms that are second - order in the electron bulk velocity systematically upscatter photons and produce a power - law tail at high photon energies . when the optical depth of the flow is increased , the systematic upscattering of photons becomes more efficient and the high - energy tail becomes flatter . figure 9 shows the same information as figure 8 but for the case of an outflow . for this model problems i have used the same parameters as in the inflow calculation but changed the sign of the velocity at all radii . in an outflow , the terms that are first - order in the electron bulk velocity produce a displacement in @xmath121 of the input radiation spectrum towards _ lower _ photon energies . on the other hand , the effect of the terms that are second - order in the electron bulk velocity is the same as in the case of inflow , i.e. , they systematically upscatter the photons and produce a tail at high photon energies . for the optical depths and bulk velocities considered here , the effect of the second - order terms nearly cancels the effect of the first - order terms and the major difference between the input and emerging spectra is the power - law tail at high photon energies . as in the case of the inflow , when the optical depth of the flow increases , the power - law tail becomes flatter , but the overall effects is significantly reduced . in the previous sections i described a method for solving numerically the radiative transfer equation that describes compton scattering in spherically symmetric , static and moving media , and applied it to a variety of cases related to flows around compact objects . i showed that the first eddington factor , which describes the degree of isotropy of the radiation field at a given point in space , may depend on photon energy , even though the photon mean - free - path is energy independent , if the photons are preferably upscattered or downscattered in energy at each scattering . i also demonstrated that the terms that are first - order in the electron bulk velocity alter the input spectrum in a way that is qualitatively and quantitatively different than the effect of the terms that are of second - order . the systematic upscattering of photons caused by bulk comptonization is described by the terms that are second - order in @xmath1 . this becomes apparent by the fact that , at each scattering with a moving electron , the average fractional energy change of a photon in the system frame is always positive , independent of the photon energy , and equal to @xmath124 ( see , e.g. , rybicki & lightman 1979 , ch . 7 ; see also the discussion in psaltis & lamb 1997 ) . if the distribution of photon residence times in the comptonizing medium drops exponentially to zero at late times , then a power - law tail is produced at high photon energies ( for total scattering optical depths @xmath125 ) , in analogy to the generation of power - law tails by thermal comptonization . the velocity at which the terms that are second - order in @xmath1 become important depends on the ratio of the photon mean - free - path @xmath126 to the shortest characteristic length scale @xmath127 in the problem . in general , when @xmath128 , i.e. , in regions close to and within the photon - trapping radius , the terms that are first- and second - order in @xmath1 become comparable in size ( yin & miller 1995 ; psaltis & lamb 1997a ) and hence the latter should not be neglected . the generalization of the method to time - independent multi - dimensional configurations is conceptually easy . however , a problem with axial symmetry requires the solution of the radiative transfer equation in five dimensions in phase space and the introduction of a large number of independent variable eddington factors . moreover , the presence of high flow velocities in multi - dimensions introduces sharp gradients in the variable eddington factors , which can not be guessed a priori , as required for the first iteration ( dykema , klein , & castor 1996 ) . even with these problems , however , the method of variable eddington factors will still offer a significant advantage , when compared to any other iterative method , which arises from the fact that the eddington factors are bound to lie over a very narrow range of values . the fast convergence of the numerical algorithm presented here makes it also ideal for future extensions to transport problems coupled to time - dependent hydrodynamics . for the case of accretion onto and outflows from compact objects , the characteristic photon diffusion timescale from a flow of dimension @xmath129 is of order @xmath130 which is significantly smaller than the corresponding @xmath131 ms dynamical timescale . as a result , for most cases the structure of the radiation field can be obtained by solving the time - independent radiative transfer problem for each snapshot of the fluid properties . given that the variable eddington factors will be only marginally different between successive timesteps , the convergence of the algorithm will be even faster than what is shown in figure 2 . i am grateful to fred lamb for his support and numerous useful discussions , to dimitri mihalas for his help on solving numerically radiative transport equations , and to d. swesty for bringing to my attention the biconjugate gradient method for solving linear systems of equations . it is also a pleasure to thank luca zampieri and feryal zel for many helpful discussions and comments and an anonymous referee for a very careful review of the manuscript . i acknowledge the support of a postdoctoral fellowship of the smithsonian institution . this research was also supported in part by nsf grant ast 93 - 15133 and nasa grant nag 5 - 2925 . * appendices * in defining the moments of the specific intensity in the system frame i shall use the quantities @xmath132\ ] ] and @xmath133\;.\ ] ] the zeroth moment of the specific intensity is then @xmath134 where @xmath22 depends on @xmath135 and @xmath30 through the relation @xmath136 . because of the assumed spherical symmetry , the only non - zero component of the first moment of the specific intensity is the @xmath22component : @xmath137 similarly , the non - zero components of the second , third , and fourth moments of the specific intensity are @xmath138 finally , i shall also use the first two eddington factors defined by @xmath139 using equation ( a8)(a12 ) of paper i , i derive the generalized source function for a spherically symmetric system : @xmath140 where @xmath141 j\\ s_2^\pm & = & \frac{3}{8 } \left\ { \left(-1 + \e\partial_\e\right ) \left [ 3 - \cos^2\theta+\left(3\cos^2\theta-1\right)f^{rr } \right]\right . j \nonumber\\ & & \quad\quad+\left.\left(1-\e\partial_\e\right ) \left[5 - 3\cos^2\theta+\left(5\cos^2\theta-3\right ) g^{rr}\right]h^r\cos\theta\right\}\\ s_3^\pm & = & \frac{3}{4}\left\ { \left[3\cos^2\theta-1- 3\left(3\cos^2\theta-1\right)f^{rr}\right ] j\right.\nonumber\\ & & \quad\quad+ 2\left[1 - 3g^{rr}+ \left(5 g^{rr}-3\right)\cos^2\theta \right]h^r \cos\theta\nonumber\\ & & \quad\quad+ \frac{1}{2}\left(-2\e \partial_\e + \e^2\partial^2_\e\right ) \left[3-\cos^2\theta+\left(3\cos^2\theta-1\right)f^{rr } \right]j\nonumber\\ & & \quad\quad+ \frac{1}{2}\left(2\e \partial_\e -\e^2\partial^2_\e\right ) \left[5 - 3\cos^2\theta+\left(5\cos^2\theta-3\right)g^{rr } \right]h^r\cos\theta\\ s_4^\pm & = & v\cos\theta i + \frac{3}{8}v\left\ { 4\cos\theta\left[2-\cos^2\theta + \left(3\cos^2\theta-2\right)f^{rr}\right]j \right.\nonumber\\ & & \quad\quad + \left[-1 - 5\cos^2\theta+\left(3\cos^2\theta- 1\right)g^{rr}\right]h^r\nonumber\\ & & \quad\quad -\cos\theta\partial_\e \left[3-\cos^2\theta+ \left(3\cos^2\theta-1\right)f^{rr}\right]j\nonumber\\ & & \quad\quad\left . + \partial_\e \left[\frac{1}{2}\left(3-\cos^2\theta\right)+ \left(3\cos^2\theta-1\right)g^{rr}\right]h^{r}\right\}\\ s_5^\pm & = & \frac{3}{4}v^2 \left\ { \left [ -5\cos^4\theta+\frac{21}{2}\cos^2\theta-1 -2\cos^2\theta\left(2-\cos^2\theta\right ) \e\partial_\e \right.\right.\nonumber\\ & & \quad\quad\quad\quad \left.+\frac{1}{4}\cos^2\theta\left(3-\cos^2\theta\right ) \e^2 \partial_\e^2\right]j\nonumber\\ & & \quad\quad+\left[4 - 10\cos^2\theta+ \left(3+\cos^2\theta\right)\e \partial_\e -\frac{1}{2}\left(3-\cos^2\theta\right ) \e^2\partial_\e^2\right]\cos\theta h\nonumber\\ & & \quad\quad + \left [ \frac{1}{2}\left(30\cos^4\theta-37\cos^2\theta+7\right ) + 2\left(1 - 3\cos^4\theta\right)\e\partial_\e \right.\nonumber\\ & & \quad\quad\quad\quad \left . + \frac{1}{4}\left(3\cos^4\theta-3\cos^2\theta+4\right ) \e^2\partial_\e^2\right]f^{rr}k^{rr}\nonumber\\ & & \quad\quad + \left [ \left(6\cos^2\theta-4\right)+\left(3\cos^2\theta-3\right ) \e\partial_\e -\frac{1}{2}\left(3\cos^2\theta-1\right)\e^2 \partial_\e^2\right]\cos\theta g^{rr } h^r \nonumber\\ & & \quad\quad \left . + \left(2\cos^2\theta-1\right)\left ( 1 + 2\e \partial_\e + \frac{1}{2}\e^2\partial_\e^2 \right)r^{rrrr}\right\}\end{aligned}\ ] ] where @xmath142 and the variable eddington factors and the moments of the specific intensity are defined as in appendix a. using equations ( 34 ) and ( 40 ) of paper i , i derive @xmath143 \nonumber\\ & & \quad\quad-\left[1+\frac{1}{2}v^2 + \left(f^{rr}+\frac{1}{2}\right)v^2 \e\partial_\e + \frac{1}{2}f^{rr}v^2 \e^2\partial^2_\e\right]\chi\\ a_2 & = & n_e \left\{\e-2t_e-\frac{2}{3}v^2 + \left[\frac{1}{10}\left(f^{rr}-\frac{1}{3}\right ) + \frac{22}{20}\e\partial_\e f^{rr}\right]v^2\right\}\\ a_3 & = & n_e \left[t_e+\frac{1}{3}v^2 + \frac{11}{20}\left(f^{rr}-\frac{1}{3}\right)v^2\right]\\ a_4 & = & -\frac{2}{\tau_0-\tau_{\rm r}}+v\chi + v \e\partial_\e \chi\\ a_5 & = & n_e v\\ c_1 & = & -\chi s_e + \frac{1}{6}v^2 s_e \left(\e\partial_\e - \e^2 \partial^2_\e\right)\chi + \frac{1}{6}v^2 \e \left(\partial_\e s_e\right ) \left(1 - 2\e\partial_\e\right ) \chi\nonumber\\ & & - \frac{1}{6 } v^2 \e^2 \left(\partial^2_\e s_e\right ) \chi\end{aligned}\ ] ] and @xmath144\nonumber\\ & & \quad\quad -\left[1+\frac{1}{2}v^2+v^2\left(g^{rr } + \frac{1}{2}\right ) \e\partial_\e + \frac{1}{2}v^2g^{rr}\e^2\partial^2_\e\right]\chi\\ b_4 & = & n_e \left[-\frac{2}{5}\e+\frac{4}{5}t_e + \frac{9}{10}v^2 -\frac{3}{10}v^2\left(1+\frac{2}{3}\e\partial_\e\right ) g^{rr}\right]\\ b_5 & = & n_e\left[-\frac{2}{5}t_e -\frac{1}{10}v^2 \left(g^{rr}+3\right)\right]\\ c_2 & = & -\frac{2}{3}v\chi s_e+ \frac{1}{3}v \chi \e\partial_\e s_e + \frac{1}{3}v s_e \e\partial_\e \chi\;,\end{aligned}\ ] ] where @xmath145 and i have suppressed the dependence of the various quantities on spatial position and photon energy . using equations ( 34 ) and ( 40 ) of paper i i derive @xmath146\chi\\ a_2 ' & = & n_e \left[\frac{1}{10}\left(f^{rr}-\frac{1}{3}\right ) + \frac{22}{20}\e\partial_\e f^{rr}\right]v^2\\ a_3 ' & = & n_e \frac{11}{20}\left(f^{rr}-\frac{1}{3}\right)v^2\\ a_4 ' & = & -\frac{2}{\tau_0-\tau_{\rm r}}+v\chi + v \e\partial_\e \chi\\ c_1 ' & = & -\chi s_e + \frac{1}{6}v^2 s_e \left(\e\partial_\e - \e^2 \partial^2_\e\right)\chi + \frac{1}{6}v^2 \e \left(\partial_\e s_e\right ) \left(1 - 2\e\partial_\e\right ) \chi\nonumber\\ & & - \frac{1}{6 } v^2 \e^2 \left(\partial^2_\e s_e\right ) \chi\;,\end{aligned}\ ] ] where @xmath145 and i have suppressed the dependence of the various quantities on spatial position and photon energy .

a variety of nonequilibrium phenomena such as open flows in hydrodynamics , thermal convection in pure fluids and binary mixtures , processes in lasers and nucleation during the first - order phase transitions , can be modelled by the complex ginzburg - landau equations . ( for review and reference , see e.g. @xcite . ) of primary importance are pulse - like solitary wave solutions , which represent localised structures widely observed in nonequilibrium systems . it is well known that in the cubic ginzburg - landau equation , @xmath2 solitary waves are unstable for all positive @xmath0 , @xmath3 and real @xmath4 . in the @xmath5 case , however , they can be stabilised by the addition of quintic terms : @xmath6 with positive @xmath7 @xcite . this example suggests that solitary pulses can be stable in a more general class of ginzburg - landau equations where the zero solution undergoes a subcritical ( rather than supercritical ) bifurcation to a flat nonzero solution @xcite . in eq.([glquintic ] ) , the terms with @xmath3 and @xmath0 account for linear homogeneous and nonhomogeneous losses , respectively , while the terms with @xmath4 and @xmath7 describe the cubic gain and quintic dissipation . the present work deals with another equation of the ginzburg - landau type exhibiting the subcritical bifurcation , viz . the _ parametrically driven _ ginzburg - landau : @xmath8 here , as in ( [ glcubic])-([glquintic ] ) , the positive @xmath0 and @xmath3 are the homogeneous loss and diffusion coefficients , respectively . the term with an asterisk ( indicating the complex conjugation ) represents the parametric driver . ( the driver s amplitude @xmath9 can be chosen positive without loss of generality . ) when @xmath10 , eq.([eq : ddcgle_bis ] ) gives the parametrically driven damped nonlinear schrdinger equation . this special case has been studied extensively ; in particular , stable solitary waves @xcite and their stable complexes @xcite were found , and their bifurcations and supercritical dynamics analysed @xcite . the objective of the present work is to advance beyond the nonlinear schrdinger limit . we will still keep @xmath11 but allow for a nonzero diffusion coefficient , @xmath0 . as we will see , even such a minimal generalisation gives rise to a new phenomenology of localised solutions which includes the multistability of pulses and pulse - front transitions . like the ginzburg - landau equations with intrinsic gain , the parametrically driven equation ( [ eq : ddcgle_bis ] ) arises in a wide range of physical applications . these include nonlinear faraday resonance in vertically vibrated layers of water @xcite and nonlinear lattices @xcite ; commensurate - incommensurate transitions in convective systems @xcite ; waves in nematic and cholesteric liquid crystals in rotating magnetic fields @xcite ; magnetisation waves in easy - plane ferromagnet exposed to microwave fields @xcite ; domain walls in the easy - axis ferromagnet near the curie temperature @xcite and in the easy - plane magnet in the stationary magnetic field @xcite ; nonlinear fiber lines with phase - sensitive amplification and mean - field models of degenerate optical parametric oscillators under continuous - wave pumping @xcite ; pulsed optical parametric oscillators with spectral filtering and lasers with intracavity parametric amplification @xcite . in most cases the models considered in literature are either purely diffusive ( @xmath12 ) @xcite or purely dispersive ( @xmath1 ) @xcite . however there are situations where it is crucial that @xmath0 be finite but nonzero . one physical phenomenon to which both diffusion and dispersion make essential contributions , is faraday resonance in strongly dissipative media such as viscous fluids and granular materials @xcite . another situation where @xmath13 , corresponds to nondegenerate optical parametric oscillators @xcite and fiber - optic telecommunication links ; in these contexts , the term @xmath14 represents spectral filtering in the spatial or temporal domain , respectively . in fact the applicability of eq.([eq : ddcgle_bis ] ) with nonzero @xmath0 and @xmath4 may happen to be even wider ; it is commonly held that this equation provides a phenomenological description to a broad range of pattern - forming systems @xcite . to study the solitary wave phenomenology introduced by taking the ( @xmath15)-term into account , we will use the diffusion - free limit ( @xmath1 ) as a starting point , and perform the numerical continuation of analytical solutions available in that case , to nonzero @xmath0 . the stability of solutions obtained in this way will also be studied numerically . we will show that switching on " the diffusion gives rise to a sequence of stable multihumped pulses occurring in the vicinity of a certain particular value of the diffusion coefficient , @xmath16 . the closer @xmath0 is to @xmath17 , the greater is the number of multihump solutions coexisting at this @xmath0 . the solutions with more than 5 or 6 humps describe a flat plateau " ( where @xmath18 is equal to the stable flat nonzero solution ) sandwiched between two fronts of opposite polarity . the paper is organised as follows . section [ flat_backgrounds ] deals mainly with spatially homogeneous solutions . in particular , we show that there is a stable flat nonzero solution for sufficiently large @xmath0 , and this uniform solution can serve as a background for solitary waves . in section [ continua ] we use perturbation - type arguments to demonstrate that both @xmath19 and @xmath20 solitons are continuable in @xmath0 and in section [ sec : adiabatic ] we construct the solutions with nonzero @xmath0 in the adiabatic approximation . the actual continuation is carried out numerically in section [ sec : continuation ] where we also examine the stability of the continued solutions . some additional insight into the bifurcation of stationary pulses is gained in section [ h_vs_c ] . finally , section [ sec : conclusion ] summarises results of this work . we close this introduction by mentioning a recent publication @xcite which was devoted to the study of the single - humped solution of eq.([eq : ddcgle_bis ] ) , by means of an averaging technique and direct numerical simulations . neither the multistability of pulses nor the pulse - front transitions were dealt with in ref.@xcite . the transformation @xmath21 casts eq.([eq : ddcgle_bis ] ) in a ` standard ' autonomous form @xmath22 where we set @xmath11 and rescaled @xmath23 so that @xmath24 . this is the representation that we are going to work with in what follows . the equation ( [ eq : ddcgle ] ) has three time - independent spatially uniform solutions , or ` flat backgrounds ' , for short . ( we do not distinguish between solutions different only in the overall sign here . ) one flat solution is @xmath25 ; it will be central for this work where we are focussing on solutions decaying to zero at infinities . the other two flat solutions are given by @xmath26 where @xmath27 note that these solutions do not depend on @xmath0 . one can easily check that the zero solution is stable as long as @xmath28 , irrespectively of the value of @xmath29 . the analysis of the flat nonzero solutions is somewhat more laborious . linearising eq.([eq : ddcgle ] ) about @xmath30 , and assuming a perturbation @xmath31 $ ] yields the dispersion relation @xmath32 where @xmath33 ( here @xmath34 stands for @xmath35 or @xmath36 , depending on which solution we are linearizing about . ) the flat solution is stable iff @xmath37 , i.e. , when @xmath38 , for all real @xmath39 . the latter condition amounts to an inequality @xmath40 where @xmath41 stands for @xmath42 . let , first , @xmath43 . since @xmath44 , the inequality ( [ inequalityfork ] ) does not hold for @xmath45 ; hence the ` low ' background @xmath46 is always unstable . let now @xmath47 , the ` high ' background . the quadratic expression in ( [ inequalityfork ] ) will be positive for all @xmath48 if either its two roots @xmath49 and @xmath50 are both real negative or complex . ( note that we can not have two real roots of opposite signs as the quadratic s constant term , @xmath51 , is always positive for @xmath52 . ) whether the roots are real or complex is determined by the discriminant of the quadratic ( [ inequalityfork ] ) which can be written as @xmath53(c - c_+)(c - c_-).\ ] ] here we have introduced @xmath54 we need to consider two cases . assume , first , that @xmath55 . in this case the discriminant is negative ( and hence , the roots @xmath56 are complex ) provided @xmath0 lies in the interval @xmath57 . on the other hand , the roots @xmath56 are real negative in this case if @xmath58 is @xmath59 and the coefficient in front of the middle term in ( [ inequalityfork ] ) is positive : @xmath60 since in the case at hand we have @xmath61 , the union of the above two ` stable ' regions , @xmath57 and @xmath62 , with the endpoints included , is simply @xmath63 . the second case is defined by inequality @xmath64 . here we have @xmath65 and the quadratic ( [ inequalityfork ] ) can not have negative real roots as the region ( [ c0 ] ) does not overlap with the region where @xmath66 . however , it can have complex roots provided @xmath67 . thus we arrive at a simple stability criterion for the flat nonzero solution @xmath68 , valid for all @xmath9 and @xmath3 : @xmath69 with @xmath70 as in ( [ eq : cpmroots ] ) . note that the stability threshold @xmath71 is always strictly positive . this implies , in particular , that the solution @xmath68 is always unstable in the case @xmath1 . ( this fact has already been mentioned in the literature @xcite . ) therefore the ` focusing ' , or ` attractive ' , damped - driven schrdinger equation ( eq.([eq : ddcgle ] ) with @xmath1 ) does not have stable flat backgrounds except the trivial one , @xmath72 . the analysis of localised solutions over flat _ nonzero _ backgrounds becomes meaningful only within the full ginzburg - landau equation , i.e. eq.([eq : ddcgle ] ) with positive @xmath0 . no less important is the question of when a flat solution can serve as an asymptotic value to a localised waveform in other words , when is spatial decay to the flat solution possible . to find the corresponding criterion , we set @xmath73 in equation ( [ eq : dispersioneqn ] ) . this results in a quadratic equation @xmath74 where @xmath75 . the spatial decay to a flat background is possible unless both roots of eq.([eq : squad ] ) , @xmath49 and @xmath50 , are positive . in the case of the @xmath46 background , the discriminant of equation ( [ eq : squad ] ) is positive while the constant term is negative , whence @xmath76 and @xmath77 . consequently , the decay to @xmath46 may occur for all values of @xmath9 , @xmath78 and @xmath0 . ( this fact is of little importance , however , since we have just shown that this background is always unstable . ) in the case of the @xmath68 solution , we _ may _ have two positive roots provided @xmath79 and @xmath80 , with @xmath81 as in ( [ c0 ] ) . following the steps in the previous subsection , one can readily show that this situation occurs only if @xmath82 , with @xmath70 as in ( [ eq : cpmroots ] ) . thus the _ stable _ constant solutions , defined by inequality ( [ flat_stab ] ) , can always serve as backgrounds to fronts and pulses . we will come across localised solutions over nonvanishing backgrounds in section [ sec : continuation ] below . finally , to examine the case of the zero background , we set both @xmath83 and @xmath73 in equation ( [ eq : dispersioneqn ] ) . this yields @xmath84 where , as before , @xmath75 . since the middle term in ( [ zero_decay ] ) has a positive coefficient , we have @xmath85 , which means that the situation where both @xmath49 and @xmath50 are positive , can never occur in this case . thus the decay to the zero background is possible for all @xmath9 , @xmath3 and @xmath0 . our last result in this section concerns the range of driving strengths which can support localised solutions in the presence of dissipation . multiplying eq.([eq : ddcgle ] ) by @xmath86 and subtracting the complex conjugate of the resulting equation gives what would be a local conservation law of the number of particles if @xmath0 , @xmath9 and @xmath3 were equal to zero : @xmath87 -2i\gamma \left| \psi \right| ^{2 } + ic\left ( \psi _ { xx}\psi ^{\ast } + \psi _ { xx}^{\ast } \psi \right ) \ , . \label{eq : ada - eq1x}\end{aligned}\ ] ] assuming that @xmath18 and @xmath88 as @xmath89 and integrating ( [ eq : ada - eq1x ] ) over the real line , we get @xmath90 dx-2c \int|\psi_x|^2 dx , \label{decay}\ ] ] where we have denoted @xmath91 the phase of the complex field @xmath18 : @xmath92 . let @xmath93 ( and remember that @xmath94 ) . in this case it follows from ( [ decay ] ) that the time derivative of @xmath95 remains negative at all times . hence as @xmath96 , @xmath97 for all @xmath98 . no stationary , time - periodic , quasiperiodic or chaotic solutions , decaying to zero as @xmath89 , can arise if @xmath93 . we will make use of this simple criterion in what follows . in the limit @xmath1 , eq.([eq : ddcgle ] ) becomes the parametrically driven damped nonlinear schrdinger ( nls ) equation , which has exact time - independent solitary wave solutions @xcite of the form @xmath99 with @xmath100 and @xmath101 as in ( [ plusminus ] ) . the solution @xmath102 is unstable for all @xmath9 and @xmath78 , while @xmath103 is stable in a certain parameter region @xcite . the purpose of this section is to show that the solitary pulse solutions @xmath104 of the nls equation persist for @xmath105 . we restrict ourselves to stationary solutions ( @xmath106 ) . writing @xmath107 with @xmath108 a constant phase to be chosen later , eq.([eq : ddcgle ] ) becomes an ordinary differential equation for @xmath109 : @xmath110 to find out whether solutions available at @xmath1 can be continued to nonzero @xmath0 , we expand @xmath111 in power series @xmath112 , substitute into eq.([eq : eqnforphi ] ) and match like powers of @xmath0 . it is convenient to choose @xmath108 to be @xmath113 in the case of the soliton @xmath103 , and @xmath114 for @xmath102 ; this choice makes @xmath115 real . matching terms linear in @xmath0 and decomposing @xmath116 into its real and imaginary part ( @xmath117 ) , yields an equation @xmath118 where the primes stand for the derivatives in @xmath98 and the operators @xmath119 are defined by @xmath120 ( the subscripts @xmath121 and @xmath122 indicate whether we are checking the continuability of @xmath103 or @xmath102 . ) according to fredholm s alternative , eq.([eq : eqnuv ] ) has bounded solutions if and only if its right - hand side is orthogonal to the kernel of the adjoint operator @xmath123 . transforming to @xmath124 and @xmath125 in the case of the @xmath103 and @xmath102 solution , respectively , the equation for the zero modes spanning @xmath126 becomes @xmath127 where @xmath128 , @xmath129 , @xmath130 and we have introduced the standard pschl - teller operators with familiar spectral properties : @xmath131 the only discrete eigenvalue of @xmath132 is @xmath133 , while the continuous spectrum occupies the semiaxis @xmath134 . therefore , the operator @xmath135 is invertible as long as @xmath136 and @xmath137 . assuming that the zero background is stable ( i.e. , assuming that @xmath138 ) , the quantity @xmath139 is indeed less than one but greater than zero . on the other hand , the @xmath140 is negative . ( an exception is the point @xmath141 , where we have @xmath142 . ) thus we conclude that @xmath143 is invertible except in the special case @xmath141 . ( however , even in this special case the operator @xmath144 is invertible on _ odd _ functions because the eigenfunction @xmath145 associated with the zero eigenvalue is even . ) consequently , @xmath146 can not be equal to zero for @xmath147 otherwise the bottom equation in ( [ eq : ker - l - dag ] ) would imply that @xmath148 , too . fortunately , the operator @xmath149 does have a zero eigenvalue , and therefore @xmath146 can be a nonzero multiple of the corresponding eigenfunction ( which is @xmath150 . ) thus in the case @xmath151 the kernel of @xmath152 is spanned by just one zero mode , namely @xmath153 on the other hand , when @xmath141 , the dimension of the kernel space is two . firstly , the zero mode ( [ eq : zerom - hneqg ] ) persists for @xmath154 as the operator @xmath155 is defined on odd functions . secondly , there is another zero mode given by @xmath156 it is obvious that the vector - function ( [ eq : zerom - hneqg ] ) , whose both components are odd functions of @xmath157 , is orthogonal to the right - hand side of eq.([eq : eqnuv ] ) which is even . hence eq.([eq : eqnuv ] ) is solvable for @xmath158 . it is also easy to check that the mode ( [ eq : zerom ] ) is _ not _ orthogonal to the right - hand side of ( [ eq : eqnuv ] ) and so the solvability condition is _ not _ satisfied for @xmath141 . thus , having started from the two nonlinear schrdinger solitons , we constructed two one - pulse solutions of the ginzburg - landau equation to within the first order in the small parameter @xmath0 : @xmath159 . consequently , we expect to be able to continue the schrdinger solitons @xmath160 into the region @xmath161 ( provided that @xmath158 . ) this expectation is born out by results displayed in section [ sec : continuation ] below , which present an outcome of the _ numerical _ continuation of @xmath104 in @xmath0 . before attempting the full - scale numerical continuation , it is instructive to construct approximate solutions . our approximation will be valid for small @xmath0 and exploit the fact that when @xmath1 , the stationary pulse - like solutions of equation ( [ eq : ddcgle ] ) have the form @xmath162 where @xmath163 and @xmath164 are constants defined by eqs.([plusminus ] ) . for @xmath0 small nonzero , approximate solutions can be sought for by assuming that @xmath18 retains the form ( [ eq : ada - soln ] ) , but @xmath165 and @xmath166 become slowly varying functions of @xmath23 . to obtain an expression for @xmath167 ( the overdot stands for the derivative in @xmath23 ) , we substitute the ansatz ( [ eq : ada - soln ] ) into ( [ eq : ada - eq1x ] ) . integrating over @xmath98 and using the boundary conditions @xmath168 as @xmath169 produces an evolution equation for the pulse s amplitude : @xmath170 where @xmath171 . an equation for the pulse s phase arises by multiplying ( [ eq : ddcgle ] ) by @xmath172 , adding with its complex conjugate , substituting ( [ eq : ada - soln ] ) for @xmath18 and integrating over @xmath98 : @xmath173 fixed points of the system ( [ eq : ada - eq1 ] ) and ( [ eq : ada - eq2 ] ) correspond to stationary solutions of ( [ eq : ddcgle ] ) . these can be easily found explicitly : eliminating @xmath174 from the stationary system @xmath175 produces a quadratic equation @xmath176 with roots @xmath177 the corresponding @xmath178 are defined by their sine and cosine in ( [ eq : stat - eq2 ] ) . it is not difficult to determine when the roots ( [ eq : ada - soln - for - a ] ) are real and positive . we assume that the zero background solution of eq.([eq : ddcgle ] ) is stable , i.e. the constant term in ( [ eq : stat - eqnfora ] ) is positive . hence if the roots are real , they are of the same sign , and this sign is opposite to that of the middle term in ( [ eq : stat - eqnfora ] ) . therefore we have two positive roots provided the discriminant @xmath179 is nonnegative , and , at the same time , the inequality @xmath180 holds true . in ( [ eq : ada - realaa ] ) we have introduced @xmath181 where the @xmath121 corresponds to @xmath182 and @xmath122 to @xmath183 . it is straightforward to verify that for small @xmath9 , @xmath184 , we have @xmath185 , while for larger @xmath9 , @xmath186 , we have @xmath187 . ( here we have made use of the inequality @xmath188 ; as proved in section [ sec : threshold ] , no stationary localised solutions exist for @xmath189 . ) therefore , the region of @xmath0 values where the roots of ( [ eq : stat - eqnfora ] ) are positive , is given by the inequality @xmath190 for all @xmath9 . thus we conclude that the adiabatic equations have two stationary points for @xmath191 , and none for @xmath192 . we complete the adiabatic analysis by classifying their stability and bifurcation . linearising equations ( [ eq : ada - eq2 ] ) about the stationary points and assuming small perturbations of the form @xmath193 and @xmath194 , yields a characteristic equation @xmath195=0.\ ] ] since the coefficient in front of the middle term in this equation is positive , either its two roots are complex with negative real parts , or we have two real roots , of which one is negative . the case when the second root is positive ( and hence the fixed point is unstable ) occurs if the constant term is negative . conversely , if the constant term is nonnegative : @xmath196 \ge 0 , \label{stabby}\ ] ] the fixed point is stable . simplifying eq.([stabby ] ) , we get the stability condition in the form @xmath197 comparing this to the expressions for @xmath198 in ( [ eq : ada - soln - for - a ] ) , we conclude that for all @xmath0 , @xmath9 and @xmath3 , the points @xmath199 and @xmath200 are a stable node and saddle , respectively . the two fixed points come into being through a saddle - node bifurcation which occurs as the diffusion coefficient @xmath0 is decreased past @xmath201 for the fixed @xmath9 and @xmath3 or , alternatively , as the driving amplitude @xmath9 is increased for the fixed dissipation coefficients @xmath0 and @xmath3 . one can easily find the bifurcation value of @xmath9 , at which two complex roots of the quadratic equation ( [ eq : stat - eqnfora ] ) converge on the positive real axis . ( here we are assuming that @xmath202 is smaller than @xmath203 . ) writing the discriminant ( [ eq : ada - realaa ] ) as @xmath204,\ ] ] the threshold driving strength is given by @xmath205 where the subscript ad " serves to remind that eq.([threshold ] ) was obtained in the adiabatic approximation . it is important to emphasise that this formula is valid only for small @xmath202 . equivalently , the expression ( [ eq : ada - soln - for - c ] ) for the turning point @xmath206 is valid only for @xmath9 close to @xmath3 . in this section we describe the bifurcation diagram obtained by the numerical continuation of the solitons @xmath207 in the parameter @xmath0 . the diagram is presented in fig.[biffy ] and displays the sobolev norm of the solution , @xmath208 as a function of @xmath0 . for stationary solutions , @xmath209 , equation ( [ eq : ddcgle ] ) reduces to an ordinary differential equation @xmath210 for the numerical continuation of solutions to eq.([stat_gle ] ) we utilised the auto97 software package @xcite . the infinite line was approximated by a finite interval @xmath211 , with @xmath212 , and the boundary conditions @xmath213 . the tolerance of the computation ( the grid norm of the difference between the right and left - hand side of ( [ stat_gle ] ) ) was set to be @xmath214 . the solid and broken curves in fig.[biffy ] represent stable and unstable branches , respectively . the stability was examined by linearising eq.([eq : ddcgle ] ) about the corresponding stationary solution . choosing the small perturbation in the form @xmath215 e^{\lambda t}$ ] , with real @xmath216 and @xmath217 , we arrive at an eigenvalue problem @xmath218 where @xmath219 and @xmath220 the @xmath221 and @xmath222 are the real and imaginary parts of the solution : @xmath223 . we solved the eigenvalue problem by expanding @xmath224 and @xmath225 over 500 fourier modes in the interval @xmath226 . the continuation in @xmath0 was performed for fixed values of @xmath9 and @xmath3 . we selected @xmath227 and @xmath228 ; for these @xmath9 and @xmath3 the nonlinear schrdinger soliton @xmath19 is stable @xcite . before proceeding to the @xmath19 soliton , however , we briefly deal with the @xmath20-case . in agreement with predictions of section [ continua ] , the soliton @xmath20 was found to persist both for @xmath229 and @xmath94 . we were in fact able to continue it indefinitely without encountering any obstacles in both directions . as @xmath230 , the width of the localised solution of ( [ stat_gle ] ) grows and the solution tends to @xmath231 , where @xmath232 stands for a pulse - shaped solution of equation @xmath233 in a similar way , as @xmath234 , the solution tends to @xmath235 . when continued to @xmath94 , the pulse remained unstable for all @xmath0 , with a single positive eigenvalue in its spectrum . ( as for the negative-@xmath0 region , all solutions there are _ a priori _ unstable against continuous spectrum perturbations with arbitrarily large @xmath236 . ) the continuation of the @xmath19 soliton proved to be more rewarding from the stability viewpoint . the corresponding bifurcation diagram is displayed in fig.[biffy ] . as with the soliton @xmath20 , the @xmath19 persists both for @xmath237 and @xmath238 in agreement with section [ continua ] . continuing into the @xmath94 region , we have found that the solution gradually changes its shape , with the hump in the imaginary part splitting into two ( see fig.[conti](a ) ) . at @xmath239 the branch turns back ( fig.[biffy ] ) . note that the turning point occurs for @xmath0 much smaller than @xmath240 , the saddle - node bifurcation point predicted by the adiabatic analysis . what is more important , the @xmath19 solution turns not into the @xmath20 ( as the adiabatic approach predicted ) but into some other solution which has two well - separated humps in the imaginary part . the discrepancy is not surprising as the adiabatic approximation is valid only for small @xmath0 , where the shape of the solution is still well reproduced by the single - hump constant - phase trial function ( [ eq : ada - soln ] ) . in the entire stretch between @xmath1 and the turning point , the solution remains stable . continuing this branch further , additional humps are added as the solution passes a sequence of turning points . each pass of a turning point results in the creation of a new hump in the middle of the solitary wave . as we move along the branch , the distance between successive turning points ( the difference between the corresponding values of @xmath0 ) becomes smaller and the new humps come with smaller amplitudes . as a result , a long plateau is formed which keeps on expanding as we continue the branch ( fig.[conti](c ) ) . the broadening plateau accounts for the vertical segment of the curve in fig.[biffy ] . if the bifurcation parameter @xmath0 is seen as a function of the sobolev norm @xmath241 ( that is , if we turn fig.[biffy ] by @xmath242 ) , the curve @xmath243 has the form of a decaying oscillation . the stability of the solution alternates at each successive turning point . these changes are due to a single real eigenvalue which moves back and forth through the origin on the real line . ( at the turning points the eigenvalue is right at the origin , of course . ) the lengths of the incursions this eigenvalue makes into the positive and negative real lines decrease with each new turning point until the eigenvalue becomes indistinguishable from zero . therefore the branch becomes ( neutrally ) stable sufficiently ` high up ' in @xmath241 in fig.[biffy ] ( i.e. for sufficiently long plateaus . ) despite the fact that negative values of the diffusivity @xmath0 are not physically meaningful , we did continue to @xmath229 in the hope that the resulting branch would reach a turning point and then return to the positive @xmath0-semiaxis ( which it did indeed do ) . the resulting branch is also shown in fig.[biffy ] . as we move into the region @xmath229 , the @xmath19 solution gradually develops into a three - humped state and when the curve returns to @xmath1 , we have a complex of the @xmath244 type , with a large separation between the individual solitons . ( this complex of three solitons of the parametrically driven nonlinear schrdinger equation was previously found in ref.@xcite . ) crossing into the @xmath94 region , the central @xmath19 soliton in the complex transforms as if it did not have the @xmath20 solitons attached to its flanks . as a result , the @xmath94-portion of the corresponding @xmath245 curve has virtually the same shape as the curve resulting from the continuation of @xmath19 to the region @xmath94 ; the only difference is that the curve emanating out of @xmath244 is shifted upwards relative to the curve emanating out of @xmath19 . similarly to the continuation of @xmath19 , the continuation of @xmath244 goes via a series of turning points , with each pass of the turning point resulting in the creation of another hump in the middle of the central region which becomes a long plateau . the lateral @xmath20 solitons are not affected by this process . the linearised spectrum is the union of the spectrum of the long pulse described in the previous paragraph and spectra of two @xmath20 solitons . in particular , it includes two positive real eigenvalues contributed by the @xmath20 s and so the entire branch resulting from the continuation of @xmath244 is unstable . we disregard it in what follows . the plateau arising in the final stage of continuation of the two branches shown in fig.[biffy ] , is nothing but an interval on the @xmath98-axis where @xmath18 equals @xmath68 , the flat nonzero solution given by eq.([flat_back ] ) . the corresponding value of @xmath0 , @xmath246 , falls within the region @xmath63 where the background @xmath68 is stable . here @xmath247 is given by eq.([eq : cpmroots ] ) ; in particular , @xmath248 . in the language of phase transitions , the long pulse shown in fig.[conti](c ) can be seen as a bubble " of one stable phase in another one . the long pulse can also be interpreted as a bound state of two fronts interpolating between different stable backgrounds , @xmath72 and @xmath52 . this intuitively - appealing idea can be put on the quantitative footing by considering the spatial decay of perturbations to the flat background @xmath68 . indeed , consider the turning point separating the second stable branch from the first unstable branch in fig.[biffy ] . ( this is a point with coordinates @xmath249 and @xmath250 . note that we are only considering branches obtained by the continuation of the soliton @xmath19 to positive @xmath0 . branches obtained by the continuation of @xmath20 to negative @xmath0 first , are disregarded here . ) it is at this point that the modulus squared of @xmath251 becomes double - humped ; before that , that is on the branch that leads to this point ( the first unstable branch ) , the function @xmath252 remained single - humped despite the double - humped imaginary part . letting @xmath228 , @xmath253 and @xmath249 we check that @xmath64 and hence , according to section [ spatially_uniform ] , the quadratic ( [ inequalityfork ] ) , ( [ eq : squad ] ) has two complex roots , @xmath254 and @xmath255 . therefore the wavenumber @xmath39 is complex as well : @xmath256 . solving equation ( [ eq : squad ] ) for @xmath41 we obtain , subsequently , @xmath257 ( we have chosen positive values for @xmath258 and @xmath259 here . ) for @xmath0 near the turning point in question , the plateau has not yet formed between the two humps and so they can be crudely thought of as two overlapping nls - like solitons with oscillations on their adjacent ( i.e. partner - facing ) tails . the bound state arises when one solitary wave is trapped in the potential well formed by the oscillatory tail of its partner . making use of the potential of interaction of two attractive nls solitons @xcite , @xmath260 ( where @xmath261 stands for the distance between the two humps ) , and taking into account that @xmath262 , we obtain a rough estimate for the separation : @xmath263 . this formula could also be derived from purely qualitative considerations ; all one needs to notice is that the absolute value of the field @xmath18 obtained from the linear superposition of two in - phase solitons is maximised ( and so the energy is minimised ) by placing one soliton at the first maximum of @xmath264 on the tail of its partner . substituting @xmath258 from ( [ kk ] ) , the approximate formula gives @xmath265 . this is in qualitative agreement with the numerically found value @xmath266 . moving further along the ( second stable ) branch , the plateau appears and the two humps can no longer be approximated by the nls - like solitons . this makes the above estimate invalid . the solution can still be regarded as a bound state of two fronts but this time , in order to calculate the characteristic width of the pulse one would need to know the full profile of the front . we conjecture that for a given @xmath9 and @xmath3 , a free - standing stationary front exists just for a single value of @xmath0 , namely @xmath267 . ( we are planning to verify this conjecture in future publications . ) on the other hand , stable and unstable bound states of fronts exist in a finite interval of @xmath0 values containing @xmath267 as an internal point . it is fitting to note here that similar pulse - to - front transformations occur also in the other system featuring the subcritical bifurcation , viz . the cubic - quintic ginzburg - landau equation with internal gain @xcite . as we mentioned in the previous section , there is a certain discrepancy between the adiabatic analysis and numerical continuation . numerically , the @xmath268 soliton was found to be continuable all the way to @xmath269 whereas the adiabatic approach predicted the existence of a turning point at @xmath270 , where the @xmath268 should have merged with the @xmath271 branch . as for the @xmath19 solution , we found that it turns into a pulse with the double - humped imaginary part ( and not into the @xmath268 branch as suggested by the adiabatic approximation . ) in order to shed some light on the possible source of the discrepancy we performed the numerical continuation of the pulse @xmath268 in @xmath9 , for several fixed values of @xmath0 . here , by the @xmath268 we mean the pulse solution which results from the continuation of the schrdinger @xmath268 soliton to positive @xmath0 , for some fixed large value of @xmath9 ( in our case for @xmath272 . ) having obtained this starting - point solution for several values of @xmath0 , we then continued it in @xmath9 , from @xmath272 to smaller @xmath9 . the stability of the arising solutions was examined by computing eigenvalues of the operator ( [ ev_problem ] ) at sample values of @xmath9 . in each case considered , the @xmath268 branch was found to turn into the @xmath271 solution as @xmath9 reached the threshold value @xmath273 . ( that is , for @xmath274 there are two branches of solutions whereas for @xmath275 , there is none ; see fig.[conti_in_h](a ) ) . the entire @xmath20 branch is unstable ; the single positive real eigenvalue moves to the negative semiaxis as the branch is continued past the turning point . continuing the arising @xmath271 branch to larger @xmath9 , we reach another turning point at @xmath276 , where the @xmath271 solution transforms into a pulse with the double - humped imaginary part . the values @xmath277 and @xmath278 are shown in fig.[conti_in_h](b ) , as functions of @xmath0 ( for the fixed @xmath279 ) . as @xmath280 , the difference @xmath281 decreases but remains nonzero . we verified this by computing @xmath277 and @xmath278 for equation ( [ c_infty ] ) which pertains to @xmath12 . in the same plot we display the function ( [ threshold ] ) which gives the adiabatic approximation to the curve @xmath282 . ( note that for small @xmath0 , there is a good agreement between numerical and approximate values but as @xmath0 grows , the two curves diverge . ) continuing the @xmath271 branch past the second turning point , the solution adds another hump in the middle of the pulse , turns back again , adds another one , and so on . ( see fig.[conti_in_h](a ) ) . a long plateau developes in the middle of the pulse , just like when it was continued in @xmath0 . similarly to the @xmath0-continuation , the turning points on the @xmath9-axis separate regions of stability from regions of instability , with the instability being caused by a single positive real eigenvalue ( which moves through @xmath283 at the turning points . ) from figure [ conti_in_h](b ) it is clear why the saddle - node bifurcation point where the @xmath271 and @xmath268 solutions would merge , did not appear in fig.[biffy ] . the reason is that fig.[biffy ] was plotted for a relatively large value of @xmath9 @xmath284 whereas according to fig.[conti_in_h](b ) , a horisontal line @xmath285 with @xmath286 can have no intersections with the saddle - node curve @xmath287 . ( here all the numbers are for @xmath279 . ) the same fig.[conti_in_h](b ) explains what semed to be a discrepancy between the adiabatic analysis and the numerical continuation of the soliton @xmath20 in @xmath0 . the numerical result that seemed to contradict the adiabatics was that for @xmath228 , @xmath20 could be continued without bounds . it is now obvious from fig.[conti_in_h](b ) that the unbounded continuation is only possible for @xmath9 greater than @xmath288 . continuing the @xmath268 soliton to positive @xmath0 for @xmath9 _ smaller _ than @xmath288 , the branch turns back ( already as the @xmath271 pulse ) after hitting the lower solid curve in fig.[conti_in_h](b ) . therefore , the pattern arising for @xmath9 close to @xmath3 actually _ is _ in qualitative agreement with the adiabatic analysis , which was expected to be valid precisely for small @xmath0 or , equivalently , for small @xmath289 differences . in this paper we studied a cubic complex ginzburg - landau equation in which linear losses and diffusion are compensated by the linear parametric drive . the nonlinear term in the equation was taken to be purely conservative . there are three stationary homogeneous solutions to equation ( [ eq : ddcgle ] ) , and we have shown that the @xmath72 solution is stable if @xmath290 , as long as @xmath29 . this stability condition coincides with the corresponding condition for the nonlinear schrdinger case ( i.e. for @xmath1 ) . on the other hand , the stability properties of the _ nonzero _ homogeneous solutions are not the same as in the @xmath1 case . indeed , unlike for @xmath1 , there is a stable flat _ solution @xmath52 for sufficiently large diffusion coefficients , @xmath291 . having established the persistence of the nls solitons @xmath19 and @xmath20 for small nonzero @xmath0 , we continued them in @xmath0 numerically . the continuation of the soliton @xmath19 yields a sequence of pulse - like solutions , separated by turning points , with increasing number of humps . the stability of these solutions changes at each turning point , so that stable multihump solutions coexist with unstable ones . it is fitting to note here that the multistability of multipulse solutions is _ not _ observed in the schrdinger limit where only the two - soliton complex was found to be stable @xcite . as @xmath292 , where @xmath293 , the solution takes the form of a long plateau ( an interval of the stable background @xmath68 ) sandwiched between two fronts . we also performed the continuation in @xmath9 , for the fixed @xmath0 . for each @xmath94 , two localised solutions are born in a saddle - node bifurcation as @xmath9 exceeds a threshold value . ( we obtained an analytic formula for the threshold in the adiabatic approximation ; it is in agreement with the numerical results for small @xmath0 . ) the subsequent continuation gives rise to a sequence of coexisting stable multihump solutions culminating in a bound state of two widely separated fronts . solitary pulses in the form of long shelves ( plateaus ) can allow easy experimental observation in physical systems described by our model . in particular , they may be employed as a natural basis for the non - return - to - zero ( nrz ) format of the data transmission in optical telecommunications . in the nrz format , the 1 and 0 bits are coded , respectively , by sending or withholding the signal within a standard time slot . a string of several 1 s looks as a long uniform pulse of an essentially arbitrary length . the stability of such pulses is crucial to maintain the fixed shape of the long array of 1 s , and to prevent the _ inter - symbol interference _ , i.e. , the blurring of empty intervals between such strings , which represent ( strings of ) 0 s ( see , e.g. ref . @xcite and references therein . ) in the case of lasers , which can also be described by the present model @xcite , the possibility of the generation of stable long pulses of arbitrary duration , i.e. , an effective _ tunability _ of the output , is an essential advantage too . we thank nora alexeeva for her advice on numerics . the first author s ( i.b.s ) work was supported by the nrf of south africa under grant no.2053723 , by the johnson bequest fund and the urc of the university of cape town . the second author ( s.c . ) was supported by the nrf and the harry crossley foundation . b.a.m . appreciates hospitality of the department of physics at the universitt erlangen - nrnberg , and a partial support from the european science foundation , through the `` @xmath294-shift '' programme . n. akhmediev and a. ankiewicz . nonlinear pulses and beams . chapman and hall , 1997 ; n. akhmediev and a. ankiewicz . solitons of the complex ginzburg - landau equation . in : s. trillo , w. torruellas ( eds . ) spatial solitons . springer series in optical sciences , vol . 82 , 2001 proceedings of the workshop on the complex ginzburg - landau equation : theoretical analysis and experimental applications in the dynamics of extended systems . edited by m. br and a. torcini . physica d * 174 * , 1 - 220 ( 2003 ) i.v . barashenkov and e.v . zemlyanaya , phys . 83 , 2568 ( 1999 ) m. bondila , i.v . barashenkov and m.m . bogdan , physica d * 87 * , 314 ( 1995 ) ; s. longhi , g. steinmeyer , and w.s . wong , j. opt . * b 14 * , 2167 ( 1997 ) ; n.v . alexeeva , i.v . barashenkov and d.e . pelinovsky , nonlinearity * 12 * , 103 ( 1999 ) ; v.s . shchesnovich , i.v . barashenkov , physica d * 164 * , 83 ( 2002 ) b. denardo , b. galvin , a. greenfield , a. larraza , s. putterman , and w. wright , phys . lett . * 68 * , 1731 ( 1992 ) ; s .- y . lou and g. huang , mod . b * 9 * , 1231 ( 1995 ) ; g. huang , s .- y . lou , and m. velarde , int . j. bif . and chaos * 6 * , 1775 ( 1996 ) ; j. yu , phys . lett . a * 259 * , 366 ( 1999 ) ; n.v . alexeeva , i.v . barashenkov and g.p . tsironis , phys . * 84 * , 3053 ( 2000 ) ; w. chen , b. hu , and h. zhang , phys . b * 65 * , 13402 ( 2002 ) a. mecozzi , w.l . kath , p. kumar , and c.g . goedde , opt . lett . * 19 * , 2050 ( 1994 ) ; s. longhi , opt . lett . * 20 * , 695 ( 1995 ) ; k. staliunas , j. mod * 42 * , 1261 ( 1995 ) ; s. longhi , j. mod . opt . * 43 * , 1089 ( 1996 ) ; g .- oppo , a.j . scroggie , and w.j . firth , phys . e * 63 * , 66209 ( 2001 ) p. coullet , j. lega , b. houchmanzadeh , and j. lajzerowicz , phys . lett . * 63 * , 1352 ( 1990 ) ; p. coullet and k. emilsson , physica * d 61 * , 119 ( 1992 ) ; b.a . malomed and a.a . nepomnyashchy , europhys . * 27 * , 649 ( 1994 ) ; c. elphick , a. hagberg , b.a . malomed and e. meron , phys . a * 230 * , 33 ( 1997 ) ; c. utzny , w. zimmermann , and m. br , europhys . lett . * 57 * , 113 ( 2002 ) ; g.j . de valcrcel , i. prez - arjona , and e. roldn , phys . lett . * 89 * , 164101 ( 2002 )

the application of the general relativity on the description of astrophysical objects has been matter of interest in the theoretical physics . exact solutions to the einstein field equations with spherical symmetry are of great relevance if systems like stars , planets or spherical clusters are considered . now , in order to model systems like galaxies , accretion disks or rotating black holes , axially symmetric solutions must be considered . for the case of galaxies , the first disks models in general relativity were proposed by bonnor and sackfield in 1968 @xcite and the second one by morgan and morgan in 1969 @xcite . these solutions increased the interest in this kind of solutions , and later , many other solution were proposed : solutions for static sources are presented in @xcite , while disks with a rotational motion ( stationary ) are presented in @xcite . the study of the gravitational field produced by an astrophysical source , from the point of view of general relativity , has been a difficult task due to the nonlinear nature of the einstein equations . the situation is not better when we try to find solutions that describe the gravitational field of a system consisting of two or more parts . examples of this kind of systems are the galaxies because they have ( in general ) a disk , a galactic halo , a bulb , and in many cases , a central black hole . the case of a disk immersed in a galactic halo is interesting because the halo is the largest and most massive part of a galaxy . this problem has already been treated by vogt and letelier in @xcite and by gutirrez , gonzlez and quevedo in @xcite . in the first paper the authors apply the displacement , cut and reflect method to some known disk - like solutions in order to generate disks with haloes , and in the second paper the authors present a new family of disk - like solutions with halo for a conformastatic spacetime in einstein - maxwell gravity . in this paper we present two families of axially symmetric thin disks of dust immersed in a spheroidal halo . the disks are obtained from solutions of the einstein equations for an axially symmetric conformastatic spacetime in which the metric tensor is characterized only by one metric function . by introducing a finite discontinuity on the first derivatives of the metric tensor , solutions with a singularity of the delta function type are obtained , so describing thin disks . the nonzero components of the energy - momentum tensor , both for the disk and the halo , are obtained from the einstein equations . in this way , the energy densities and pressures of the sources are determined . by imposing the fulfillment of all the energy conditions we obtain a constraint over the solutions , in such a way that the metric function can be properly expressed in terms of a solution of the laplace equation . by using the solution to laplace equation in cylindrical coordinates we find infinite disks and by using the solution to laplace equation in oblate spheroidal coordinates we find finite disks . in both cases we obtain particular solutions with energy densities and pressures well behaved everywhere . we also show that the masses of de disks and the haloes are finite . finally we solve the geodesic equation for circular orbits in the plane of the disk . some preliminaries results of this work have been presented in @xcite . the structure of this article is the follow : in section [ sec : sols ] we solve the einstein equations for a conformastatic metric in order to find the energy density and stress of the halo in terms of solutions to the laplace equation . in section [ sec : emtensor ] we apply the thin shell formalism to find a disk - like source that satisfy the energy conditions ( physically well behaved ) . in section [ sec : mass ] we introduce the gravitational mass , both for the disk and for the halo , and we study the dynamics of a particle moving in circular orbits in the plane of the disk in order to obtain the rotational curves , which describe the radial dependency of the speed of a particle that moves in circular paths around the center of the galaxy and in the plane of the disk . we then present , in section [ sec : kuzmin ] , a particular family of relativistic solutions of infinite thin disks with halo and we analyze the behavior of the physical properties and , in section [ sec : morgan ] , we present a second family of solutions of finite thin disks with halo and discuss the physical features of this solutions . finally , in section [ sec : concluding ] , we present a brief discussion of our main results . in order to formulate the einstein field equations for an axially symmetric static spacetime with an infinitesimally thin discoidal source which is immersed in a fluid or matter halo , we first introduce the cylindrical coordinates @xmath1 . now , we consider the spacetime as described by the conformastatic metric @xcite @xmath2 where we demand @xmath3 to be a continuos function of the @xmath4 and @xmath0 coordinates . the einstein field equations for this metric yield the energy - momentum tensor for the halo fluid @xmath5 , whose non - null components are [ e0116 ] @xmath6 now , to give a physical interpretation to the energy - momentum tensor of the halo , we write @xmath5 in a locally minkowskian reference frame whose space - like vectors are oriented in the main stresses directions . this reference frame is described by the orthonormal basis , [ e0116 ] @xmath7 where @xmath8 . by writing @xmath5 in this reference frame , we find the energy density @xmath9 the main stresses of the halo , which are given by @xmath10 and the isotropic pressure @xmath11 which is defined as the average of the main stresses . the physical quantities we have already obtained depend on the unknown metric function @xmath3 , so we need an additional equation to get @xmath3 . on the other hand , if we demand @xmath12 and @xmath13 to satisfy the energy conditions @xcite , then we get the inequality @xmath14 these conditions are important to yield physically well behaved solutions , in other words , to fulfill the minimum conditions for gravitational systems . one way to satisfy ( 7 ) is considering @xmath3 functions that are solutions to the equation @xmath15 with @xmath16 . this particular choice to fulfill the energy conditions leads us to the state equation @xmath17 for the halo . here we can see that the pressure and the energy density are linearly related by a constant @xmath18 that depends on @xmath19 value . it means that a specific kind of fluid in the halo depends on a particular choice of @xmath19 ; for example , if @xmath19 tends to infinity , then @xmath13 tends to zero and the fluid in the halo would be made of dust . if we choose @xmath19 equal to 1 , then the state equation would describe a fluid made of radiation . henceforth , we will not analyze the behavior of @xmath13 , because it is related @xmath12 via the state equation . now , in order to get solutions for the @xmath3 function , we rewrite equation ( [ eq8 ] ) as @xmath20 in such a way that solutions for @xmath3 can be obtained form solutions of the laplace equation . furthermore , as the spacetime needs to be asymptotically flat , then @xmath21 has to be @xmath22 at infinity , and we can write @xcite @xmath23 where @xmath24 es any solution of the laplace equation that vanishes at infinity . this relation allows us to write @xmath12 in terms of @xmath24 as @xmath25 it is clear that we can obtain different solutions depending on the choice of @xmath24 ; nevertheless , in the next section we will see that @xmath24 must satisfy some conditions in order to have a disk solution . in the preceding section we computed the energy - momentum tensor for the fluid in the halo ; now , we want to compute the one of the disk . to accomplish this objective we use the distributional approach @xcite and the thin shells formalism @xcite to describe the disk - like source . mathematically the disk is described by an hypersurface @xmath26 whose function is @xmath27 and its normal vector is @xmath28 . this hypersurface separates the spacetime in two regions : @xmath29 ( above @xmath26 ) and @xmath30 ( bellow @xmath26 ) . by using distributional functions , the energy - momentum tensor can be written as @xmath31 where @xmath32 is the heaviside function , @xmath33 is the dirac function , @xmath34 is the energy - momentum tensor in @xmath35 , and @xmath36 is the energy - momentum tensor in @xmath26 . now , to describe properly a surface layer it is necesary to satisfy the condition of continuity of the metric across the plane @xmath37 , which is mathematically expressed via the jump function as @xmath38=g_{\alpha\beta}|_{_{z=0^{+}}}-g_{\alpha\beta}|_{_{z=0^{-}}}=0 . \label{eq14}\ ] ] this condition can be set in terms of the solution to laplace equation through ( [ eq1 ] ) and ( [ eq10 ] ) as @xmath39 we also need to ensure that @xmath40 is discontinuous in the plane @xmath37 , i.e. @xmath41=2g_{\alpha\beta , z}|_{_{z=0^{+}}}\neq 0 . \label{eq16b}\ ] ] with the metric tensor defined in ( [ eq1 ] ) and the relation ( [ eq10 ] ) it is easy to show that the last condition is satisfied if we choose a function @xmath24 such that @xmath42 on the other hand , we can compute @xmath36 through the expression @xmath43 where @xmath44 is the part of the ricci tensor associated to @xmath26 . here it is important to note that @xmath45 by virtue of the condition ( [ eq16b ] ) . computing @xmath36 we find that its only non - null component is @xmath46 which is evaluated in @xmath47 . now , the surface energy - momentum tensor of the disk is computed from @xmath36 through the integral @xmath48 where @xmath49 . this integral yield the non - null component of @xmath50 , which we write in the tetrad ( [ e0116 ] ) to interpret it physically as the surface energy density of the disk - like source , @xmath51 this solution can be rewritten in terms of @xmath24 as @xmath52 which is evaluated in @xmath47 as in the previous section , @xmath53 needs also fulfill the energy conditions , but in this case it is easier than before because our disk is made of dust ( @xmath54 ) , i.e. the particles in the fluid are collisionless ; so all the energy conditions are satisfy if @xmath55 here we do not consider the case @xmath56 because it implies @xmath57 and @xmath58 . once the energy - momentum tensor has been specified , it is possible to compute the total mass of the system . to do it , we use the komar formulae @xcite , which is written as @xmath59 where @xmath60 is the normal vector to the space - like hypersurface , @xmath61 is the time - like killing vector , @xmath62 is the determinant of the induced metric and @xmath63 is the volume element . in our case , @xmath64 , @xmath65 , and @xmath66 . by using the non null components of @xmath5 and the relation ( [ eq10 ] ) in the komar formulae we can compute the mass of the halo as @xmath67 where @xmath68 is the usual euclidean volume element of the space . on the other hand , if we take @xmath69 in the komar formulae we find that the mass of the disk is @xmath70 finially , the total mass of the system is given by @xmath71 . now , we can also study the movement of a particle in a spacetime described by the line element ( [ eq1 ] ) . the particle is interacting with a gravitational field produced by a disk - like distribution of energy ( [ eq20s ] ) surrounded by a halo whose energy distribution is described by ( [ eq10u ] ) . now , a very simple case is the one of a particle that moves in a circular orbit in the plane of the disk ( @xmath37 ) , with velocity @xmath72 tangential to its path . this is interesting , from the physical point of view , because we have observational information for @xmath73 from spiral galaxies @xcite . the relation between the tangential velocity and the distant from the center of the galaxy ( @xmath4 ) is described by the rotational curves . for a circular path in the plane of the disk @xmath37 and @xmath4 is constant ; therefore , the velocity is described only by two components , @xmath74 where @xmath75 is the angular velocity . by imposing the normalization condition @xmath76 , we find that @xmath77 where @xmath78 is the circular velocity or tangential velocity . on the other hand , we solve the geodesic equation , @xmath79 to obtain an expression for @xmath72 in terms of the function @xmath24 . as we saw , the spacetime has time symmetry and axial symmetry , so @xmath80 and @xmath81 are not going to change in @xmath82 , so , the geodesic equation reduces to @xmath83 , and it yield the expression for the tangential velocity , @xmath84 from which we obtain the rotational curves . the energy density of the halo and the disk , the stress of the halo , the gravitational mass and the circular velocity are the physical properties that describe the disk - halo system . all these properties are written in terms of the solution to laplace equation @xmath24 . nevertheless , as we saw in sec . [ sec : emtensor ] , @xmath85 needs to satisfy the conditions ( [ eq15 ] ) and ( [ eq17u ] ) in order to define a surface energy density in the plane @xmath37 . to find a particular family of models we choose @xmath24 to be the solution to laplace equation in spherical coordinates @xmath86 . now , if we use the usual coordinates transformations @xmath87 and @xmath88 , then the condition ( [ eq17u ] ) will not be satisfied because @xmath57 . to solve this problem we use the displacement , cut and reflect method used by kuzmin ( 1956 ) and toomre ( 1963 ) @xcite to generate disk - like sources in the frame of newtonian gravity . the method is based on the transformation @xmath89 where @xmath90 is a constant . with this , the function @xmath24 takes the form @xmath91 where @xmath92 , @xmath93 , @xmath94 are arbitrary constants and @xmath95 are the legendre polynomials . the solution to the laplace equation presented in ( [ ukuz ] ) allows us to write the energy density of the halo as @xmath96^{2 } + \left[\sum_{l=0}^{n}\left(a_{l}(l+1)p_{l+1}(\frac{|z|+a}{r})\right)/r^{l+2}\right]^{2}}{\left[1+\sum_{l=0}^{n}a_{l}p_{l}(\frac{a}{r})/r^{l+1}\right]^{2(k+1)/k } } \right\}. \label{rhokuzmin}\ ] ] in a similar way , we can write the energy density of the disk as @xmath97/r_{0}^{l+2}}{\left[1+\sum_{l=0}^{n}a_{l}p_{l}(a / r_{0})/r_{0}^{l+1}\right]^{(k+1)/k } } , \label{sigmakuzmin}\ ] ] where @xmath98 . so , from the behavior of @xmath95 and the fact that @xmath99 vanishes at infinity , we can see that @xmath100 and @xmath101 are always positive , have their maximum value at the center of the system ( @xmath102 , @xmath37 ) and both vanish at infinity . now , we are going to analyze the behavior of @xmath103 and @xmath104 , and to do so we will consider only the models with @xmath105 . the solutions that describe the energy density of the halo for the first three models are given by @xmath106^{2}+[\tilde{z}(\tilde{a}_{0}\tilde{r}^{2}+3\tilde{a}_{1}\tilde{z})-\tilde{a}_{1}\tilde{r}^{2}]^2}{\tilde{r}^{2(2k-3)/k}[\tilde{r}^{3}+\tilde{r}^{2}\tilde{a}_{0}+\tilde{a}_{1}\tilde{z}]^{2(1+k)/k } } , \label{r1}\\ \nonumber\\ \tilde{\rho}_{2}&=&\frac{\tilde{r}^{2(5 - 2k)/k}\left[\tilde{a}_{0}\tilde{r}\tilde{r}^{4}+(3\tilde{a}_{1}\tilde{z}+\tilde{a}_{2})\tilde{r}\tilde{r}^{2}+\frac{5}{2}\tilde{a}_{2}(2\tilde{z}^{2}-\tilde{r}^{2})\right]^{2}}{\left[\tilde{r}^{5}+\tilde{a}_{0}\tilde{r}^{4}+\tilde{a}_{1}\tilde{r}^{2}\tilde{z}+\frac{1}{2}\tilde{a}_{2}(\tilde{z}^{2}-\tilde{r}^{2})\right]^{2(1+k)/k } } \nonumber \\ \nonumber \\ & & + \frac{\tilde{r}^{2(5 - 2k)/k}\left[(\tilde{a}_{0}\tilde{z}-\tilde{a}_{1})\tilde{r}^{4}+3\tilde{z}(\tilde{a}_{1}\tilde{z}-\tilde{a}_{2})\tilde{r}^{2}+\frac{5}{2}\tilde{a}_{2}\tilde{z}(2\tilde{z}^{2}-\tilde{r}^{2})\right]^{2}}{\left[\tilde{r}^{5}+\tilde{a}_{0}\tilde{r}^{4}+\tilde{a}_{1}\tilde{r}^{2}\tilde{z}+\frac{1}{2}\tilde{a}_{2}(\tilde{z}^{2}-\tilde{r}^{2})\right]^{2(1+k)/k } } , \label{r2}\end{aligned}\ ] ] + where @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 , @xmath112 , @xmath113 , and @xmath114\rho_{n}$ ] . in fig . [ denhalokuz ] we plot ( [ r0]-[r2 ] ) to analyze the behavior of the energy density of the halo . this figure allows us to see that the most of the energy in the halo is located at the center of the system , i.e. in the region near to @xmath115 and @xmath116 . the density presents a maximum at the center and , it tends to zero at infinity with a rate that depends on the constants values in each particular model . on the other hand , the energy density of the disk for the models @xmath105 is given by @xmath117\tilde{r}_{0}^{\frac{3 - 2k}{k}}}{\left[\tilde{r}_{0}^{3}+\tilde{a}_{0}\tilde{r}_{0}^{2}+\tilde{a}_{1}\right]^{\frac{1+k}{k}}},\label{s1}\\ \nonumber \\ \tilde{\sigma}_{2}&=&\frac{\left[(\tilde{a}_{0}-\tilde{a}_{1})\tilde{r}_{0}^{4}+3(\tilde{a}_{1}-\tilde{a}_{2})\tilde{r}_{0}^{2}+\frac{5}{2}\tilde{a}_{2}(2-\tilde{r}^2)\right]\tilde{r}_{0}^{\frac{2(5 - 2k)}{k}}}{\left[2\tilde{r}_{0}^{5}+2\tilde{a}_{0}\tilde{r}_{0}^{4}+2\tilde{a}_{1}\tilde{r}_{0}^{2}+\tilde{a}_{2}(2-\tilde{r}^2)\right]^{\frac{1+k}{k}}},\label{s2}\end{aligned}\ ] ] where @xmath118 , @xmath110 , @xmath111 , @xmath112 , @xmath113 , and @xmath119 . we plot , in fig . [ densidiskuz ] , the energy density @xmath120 for the first models with @xmath121 . in each graph we plot some curves for different values of the constants @xmath122 and @xmath19 in order to analyze different behaviors . hereafter , we introduce a convenient notation in which each curve is labeled by the set of numbers ( @xmath123 ) . it is important to clarify that the constants are chosen so that the energy density is positive ; this assure the condition ( [ eq21 ] ) . the surface energy density profiles , presented in fig . [ densidiskuz ] , go to zero at infinity from a maximum value located at the center of the disk . we can change the maximum value of @xmath124 and its decrease rate by changing the values of the constants . in the three models , the most of the energy is concentrated at the central region of the disk , but as we can see , when we increase @xmath125 , the energy is less distributed in the space , i.e. the energy density goes to zero faster . we can compute the total mass for the disk - halo system , which is defined through ( [ mh ] ) and ( [ md ] ) , for any model characterized by the integer value of @xmath125 . by doing the integrals for @xmath126 we find that , @xmath127.\label{mh0}\end{aligned}\ ] ] now , since @xmath104 and @xmath103 vanish at infinity , it is important to show that @xmath128 will converge for all @xmath125 . to do so , we start by putting ( [ md ] ) in the form @xmath129 where @xmath98 and @xmath130 with ( [ ukuz ] ) it is easy to show that @xmath131 and by using the limit comparison test @xcite we can demonstrate that @xmath132 will be finite if @xmath133 converges . now , since @xmath134 has a finite value , then the mass of the disk for every model will be finite too . to guarantee the convergence of @xmath135 we proceed in a similar manner : the integral ( [ mh ] ) , with @xmath136 can be rewritten as @xmath137 where @xmath138 and @xmath139 . it is easy to show that @xmath140 so by virtue of the limit comparison test , if @xmath141 converge , then @xmath142 will converge too and from ( [ mhproof ] ) , @xmath135 and @xmath143 will be finite . since @xmath144 has a finite value , then the mass of the halo for all models converge . @xmath145 finally , we present the circular velocity for the models we are considering . to compute the expressions for this physical quantity we use ( [ ukuz ] ) and ( [ eq33 ] ) to obtain @xmath146}{k[m^{2}+m\tilde{a}_{1}+m^{3}\tilde{a}_{0}+\frac{1}{2}\tilde{a}_{2}(3-m)]m^{3}-\tilde{r}^{2}[3\tilde{a}_{2}+3\tilde{a}_{1}m^{2}+\tilde{a}_{0}m^{4}+\frac{3}{2}\tilde{a}_{2}(3-m)m ] } , \label{v2}\end{aligned}\ ] ] + where @xmath147 , @xmath110 , @xmath111 , @xmath112 , and @xmath113 . now , we plot these expressions to see clearly the dependence of @xmath148 with the coordinate @xmath149 for a time - like particle in a circular orbit . the resulting plots are presented in fig . [ circvelkuz ] . here we choose the same constants values that we use to perform fig . [ densidiskuz ] in order to compare the rotational curve with its corresponding surface energy density profile . in the plot for @xmath126 model we see that the velocity increases very fast until a maximum value from which we distinguish clearly two kind of behavior : in the first one , the velocity decrease with @xmath149 , and in the second one it remains approximately constant . if we compare with the corresponding plot in fig . [ densidiskuz ] , we see that the second kind of behavior is associated with energy density profiles whose maximum is higher than the cases where the velocity decrease after reaching its maximum . in the plot for @xmath150 , all the curves show that the velocity remains constant after reaching a maximum , but , by comparing with fig . [ densidiskuz ] , we see that if the maximum is higher , the energy density reaches a lower value in @xmath115 . in the plot for @xmath151 we obtain different kind of behavior . of particular interest are those with two peaks because actually , the observational data shows that the velocity is not exactly constant after the first maximum , it has fluctuations qualitatively similar to those of our plots . finally , comparing with the plot for @xmath152 in fig . [ densidiskuz ] , we see that , when the rotational curves present two peaks , the corresponding energy density profiles have two peaks as well . @xmath153 @xmath154 in order to obtain a new family of finite thin disks with halo , we use oblate spheroidal coordinates ( @xmath155 ) to solve the laplace equation , because they adapt to the geometry of the source and introduce , in a natural way , a finite radius for the disk . the transformation relations with the cylindrical coordinates are @xmath156 where @xmath157 and @xmath158 . the coordinates of the disk of radius @xmath159 are @xmath160 and @xmath161 , and if we cross the disk , @xmath162 changes sign . in these coordinates , the solution to the laplace equation takes the form @xmath163 where @xmath164 are the legendre polynomials of order @xmath165 , @xmath166 , with @xmath167 the legendre function of second kind of imaginary argument @xcite , and @xmath168 are constants choosen to satisfy both the boundary condition ( [ eq14 ] ) and the condition for the allowed values for the speed of a particle in a circular trajectory , @xmath169 . according to ( [ eq15 ] ) and ( [ eq17u ] ) , to define properly a disk - like source , we want the metric tensor to be continuous and its first @xmath0-derivate to be discontinuous across the disk ; therefore , since @xmath170 is continuous across the disk and @xmath162 is discontinuous , we only need to consider even legendre polynomials . therefore , @xmath24 in ( [ uobla ] ) is formed only by the even terms of the general solution to laplace equation . to analyze the behavior of the physical quantities we have to consider particular solutions by restricting the sum in ( [ uobla ] ) . one way to do it is to consider @xmath24 to be the gravitational newtonian potential of a finite disk , where the constants @xmath168 are found by imposing the orthogonality property of the legendre polynomials to give @xcite @xmath171 where , @xmath172 , \label{eq35}\ ] ] if , @xmath173 , and @xmath174 if @xmath175 , for @xmath176 . in the above equation @xmath177 and @xmath159 are the total mass and the radius of the disk respectively . depending of the value of @xmath178 we will have a particular solution because the constants @xmath168 are restricted by @xmath179 ; so we are going to choose only the three first models ( @xmath180 ) to see the behavior of the energy densities of the disk and the halo , their masses , the state equation of the halo and the rotational curves . the energy density of the halo for the first three models are computed by using ( [ uobla ] ) in ( [ eq10u ] ) in such a way that @xmath181 , \label{eq35}\ ] ] where [ u ] @xmath182 , \\ u_{2 } & = - { \tilde m } \left [ \operatorname{acot}\xi + \frac{10 a}{7 } ( 3\eta^{2 } - 1 ) + b ( 35 \eta^{4 } - 30 \eta^{2 } + 3 ) \right ] , \\ u_{3 } & = - { \tilde m } \left [ \operatorname{acot}\xi + \frac{10 a}{6 } ( 3 \eta^{2 } - 1 ) + \frac{21 b}{11 } ( 35 \eta^{4 } - 30 \eta^{2 } + 3 ) + c ( 231 \eta^{6 } - 315 \eta^{4 } + 105 \eta^{2 } - 5 ) \right],\end{aligned}\ ] ] with @xmath183 and @xmath184 , \\ b & = \frac{3}{448 } \left [ ( 35 \xi^{4 } + 30 \xi^{2 } + 3 ) \operatorname{acot}\xi - 35 \xi^{3 } - \frac{55}{3 } \xi \right ] , \\ c & = \frac{5}{8448 } \left [ ( 231 \xi^{6 } + 315 \xi^{4 } + 105 \xi^{2 } + 5 ) \operatorname{acot}\xi - 231 \xi^{5 } - 238 \xi^{3 } - \frac{231}{5 } \xi \right].\end{aligned}\ ] ] note that @xmath185 and @xmath162 are related with the cylindrical coordinates by virtue of the transformation relations ( [ robla ] ) and ( [ zobla ] ) , so we can compute the first @xmath4-derivate of the above expressions , [ ur ] @xmath186 , \\ u_{2,r } = & \frac{15\tilde{m}\tilde{r}}{16d(1+\xi^{2 } ) } \left\ { \xi-13\eta^{2}\xi+(3 - 15\eta^{2})\xi^{3}+(1+\xi^{2 } ) [ 1 - 3\xi^{2}+3\eta^{2}(1 + 5\xi^{2 } ) ] \operatorname{acot}\xi\right\ } , \\ u_{3,r } = & \frac{35\tilde{m}\tilde{r}}{128d(1+\xi^{2})}\left\ { \xi \left [ 3 - 4 \xi^{2 } - 15 \xi^{4 } + \eta^{2 } ( 6 + 200 \xi^{2 } + 210 \xi^{4 } ) - \eta^{4}(113 + 420 \xi^{2 } + 315 \xi^{4 } ) \right ] \right . \notag \\ & \left . + 3 ( 1+\xi^{2 } ) \left [ 1 - 2 \xi^{2 } + 5 \xi^{4 } + \eta^{2}(2 - \xi^{2}-70\xi^{4})+5\eta^{4}(1 + 14\xi^{2}+21\xi^{4 } ) \right ] \operatorname{acot}\xi \right\},\end{aligned}\ ] ] and also , the first @xmath0-derivative , which are [ uz ] @xmath187 , \\ u_{2,z } = & \frac{5\tilde{m}\eta}{4d } \left\ { -9 \xi^{2 } + \eta^{2 } ( 4 + 15 \xi^{2})+ 3 \xi \left [ 1 + 3 \xi^{2 } - \eta^{2 } ( 3 + 5 \xi^{2 } ) \right ] \operatorname{acot}\xi \right\ } , \\ u_{3,z } = & -\frac{7\tilde{m}\eta}{64d } \left\ { 50 \eta^{2 } \xi^{2}(11 + 21 \xi^{2 } ) - 15 ( \xi^{2 } + 15 \xi^{4 } ) - \eta^{4 } ( 64 + 735 \xi^{2 } + 945 \xi^{4 } ) \right . \notag \\ & \left . + 15 \xi \left [ -1 + 6 \xi^{2 } + 15 \xi^{4 } - 2 \eta^{2}(3 + 30 \xi^{2 } + 35 \xi^{4 } ) + \eta^{4}(15 + 70 \xi^{2 } + 63 \xi^{4 } ) \right ] \operatorname{acot}\xi \right\}.\end{aligned}\ ] ] @xmath188 in fig . [ densihalokal ] , we show the plots of the rescaled energy density @xmath189 as a function of @xmath190 and @xmath191 . we can observe two kinds of behavior , both of them present in all the particular solutions discussed in this paper , that can be obtained by choosing properly the constant values @xmath19 , @xmath192 . the first behavior is the one presented in the plots for @xmath193 , where the energy density reaches its maximum at the center of the halo ( @xmath194 ) , and then goes to zero at infinity . it is clear , in this case , that the energy is more concentrated in the region near the disk . the second behavior is shown in the plot associated to the @xmath195 model . here we see an unusual distribution of energy ( if we think , only , in the visible matter of the halo ) , where the maximum of density is located at @xmath196 and @xmath197 , i.e. near the edge of disk , and goes to zero at infinity . we can also get analytic expressions for the energy density of the disk for the particular models considered in this section by using ( [ uobla ] ) in ( [ eq20s ] ) , so that [ s ] @xmath198^{\frac{1+k}{k}}},\label{s1}\\ \tilde{\sigma}_{2}&=&\frac{20(1-\tilde{r}^{2})^{3/2}}{\left[1+\frac{15\pi}{128}\tilde{m}a_{2}(\tilde{r})\right]^{\frac{1+k}{k}}},\label{s2}\\ \tilde{\sigma}_{3}&=&\frac{28(1-\tilde{r}^{2})^{5/2}}{\left[1+\frac{35\pi}{512}\tilde{m}a_{3}(\tilde{r})\right]^{\frac{1+k}{k}}},\label{s3}\end{aligned}\ ] ] + where @xmath199 , @xmath190 , @xmath200 , and @xmath201 in fig . [ dendiskal ] , we plot the energy density @xmath120 of the disk as a function of the normalized radius for the models @xmath180 . in the first plot we compare the energy density for the constants values @xmath202 and @xmath203 , and for different values of @xmath178 . in this plot we see that @xmath120 has a maximum at the center of the system @xmath204 , and goes to zero at the edge ( @xmath205 ) . the spatial distribution of the energy depends on the particular value of @xmath178 : for @xmath195 , we see that the energy density remains approximately constant , and starts to fall from @xmath206 ; but if we increase the value of @xmath178 , then the energy density will be more concentrated at the center , and its maximum value will be higher . the remaining plots show how , by changing @xmath19 and @xmath192 , we can get different kinds of behavior in each particular model . it is clear that we can manipulate the maximum value of the energy density and its rate of decline by varying the constants , but in all cases the behavior is very similar and the energy is concentrated in the center of the disk . now , with ( [ eq8 ] ) we can write the gravitational mass ( [ mh ] ) equivalently as @xmath207 where @xmath208 is the area of a closed surface which contains a mass @xmath177 . we choose @xmath208 to be the surface of an ellipsoid determined by @xmath185 constant . accordingly , we can define the mass of the disk throw the integral @xmath209 and the total mass of the system by @xmath210 so that , the mass of the halo is @xmath211 . + @xmath212 despite the fact that in this model the halo has infinite extension , it is possible to compute the total mass of the system through the integral given on ( [ mtobla ] ) , and show that @xmath128 converges . to do it , we first take advantage of the fact that at infinity , @xmath24 tends to zero . so , by using ( [ uobla ] ) , we find that @xmath213 now , with the recurrence relation , @xmath214 we can transform ( [ eq25 - 1 ] ) on @xmath215 as @xmath216 becomes zero at infinity , and @xmath217 goes to @xmath218 if @xmath219 for all @xmath220 , then @xmath221 , and the total mass of the disk - halo system becomes @xmath222 which depends on the radius of the disk , the first constant @xmath223 and the particular kind of fluid in the halo ( determined by the value of @xmath19 ) . with this , we proof that the total mass is finite , even though the halo has infinite extension . this proof is very general , so @xmath128 is the same for all particular models . now , for the first model ( @xmath195 ) the integral ( [ mdobla ] ) yield to @xmath224 , \label{eq36}\ ] ] where @xmath225 ; so the mass of the relativistic disk is related to the newtonian mass . the analytic expression for the circular velocity of a particle on the plane @xmath37 has two parts : one part correspond to the particle moving on the disk ( @xmath226 ) , for which we need to do @xmath160 on ( [ eq33 ] ) ; and the other part correspond to the velocity of a particle moving outside the disk ( @xmath227 ) , for which we have to do @xmath228 on the expression for @xmath229 . both parts must coincide exactly at @xmath205 . now , by using the expressions for @xmath230 and @xmath231 in ( [ eq33 ] ) , and doing the appropriate evaluation in each region we get the solutions for the circular velocity for the models we are considering , which are given by @xmath232 - 6 \tilde{m } \pi \tilde{r}^{2 } } , & 0 \leq \tilde{r } \leq 1 , \\ \\ \dfrac{6 \tilde{m } ( \tilde{r}^{2 } \operatorname{acot}j - j ) } { 6 \tilde{m } j + k ( 4 + 3 \tilde{m } j ) - 3 \tilde{m } [ 2 \tilde{r}^{2 } + k ( \tilde{r}^{2 } - 2 ) ] \operatorname{acot}j } , & \tilde{r } \geq 1 , \end{array } \right . \\ \nonumber \\ v_{c_{2}}^{2 } & = \left\ { \begin{array}{ll } \dfrac{- 60 \tilde{m } \pi \tilde{r}^{2 } l_{1}}{60 \tilde{m } \pi \tilde{r}^{2}l_{1 } + k ( 128 + 15 \tilde{m } \pi l_{2 } ) } , & 0 \leq \tilde{r } \leq 1 , \\ \\ \dfrac{60 \tilde{m } [ \tilde{r}^{2}l_{1 } \operatorname{acot}j - j l_{3 } ] } { 60\tilde{m } j l_{3 } + k [ 45\tilde{m}(\tilde{r}^{2 } - 2 ) j - 64 ] - 15 \tilde{m } ( 4 \tilde{r}^{2 } l_{1 } + k l_{2 } ) \operatorname{acot}j } , & \tilde{r}\geq 1 , \end{array } \right . \\ \nonumber \\ v_{c_{3}}^{2 } & = \left\ { \begin{array}{ll } \dfrac{-210 \tilde{m } \pi \tilde{r}^{2 } n_{1}}{210 \tilde{m } \pi \tilde{r}^{2}n_{1 } + k ( 35\tilde{m}\pi n_{2 } - 512 ) } , & 0\leq \tilde{r}\leq 1 , \\ \\ \dfrac{210 \tilde{m } [ 3 \tilde{r}^{2 } n_{1 } \operatorname{acot}j - j n_{3 } ] } { 210 \tilde{m } j n_{3 } + k ( 768 + 35 \tilde{m } j n_{4 } ) - 105 \tilde{m } ( 6 \tilde{r}^{2}n_{1 } + k n_{2 } ) \operatorname{acot}j } , & \tilde{r}\geq 1 , \end{array } \right.\end{aligned}\ ] ] where @xmath233 , @xmath234 , @xmath235 , @xmath236 , @xmath237 , @xmath238 , @xmath239 and @xmath240 . in fig . [ velcirckal ] , we show the profiles of the circular velocity as a function of @xmath149 for some constants values @xmath241 . all the constants values , in this paper , are chosen in such a way that the physical quantities are well behaved ( @xmath242 and @xmath243 ) . in the first plot , in fig . [ velcirckal ] , we compare the rotational curve in the models @xmath180 for @xmath19 and @xmath192 fixed . here we can see that if @xmath178 is higher , then the peak would be less acute , the union between the two curves mentioned in the last paragraph would be smoother and the maximum value of @xmath72 would be attained closer to the center of the disk . + in all particular solutions , the behavior of @xmath72 is the same : the velocity starts from zero and reaches a maximum that depends on the particular constants values , then the velocity falls to a value that remains about constant ; this is very important because many real rotational curves from spiral galaxies @xcite has the same behavior ( qualitatively speaking ) . an specific shape for the rotational curve can be obtained by choosing properly the values for @xmath244 and @xmath192 , e.g , in the plots for @xmath180 , the curves associated to the values @xmath245 and @xmath246 show that we can also get circular velocities that fall after reaching the maximum . @xmath247 we obtained two sets of new exact solutions to the einstein equations for an axially symmetric static spacetime from the conformastatic metric . these solutions describe the physical properties of an infinite disk of dust , i.e. without stresses , which is surrounded by a fluid or halo with an equation of state @xmath248^{-1}\rho$ ] . this equation implies different kind of fluids : radiation if @xmath249 , dust if @xmath19 tends to infinity or another kind of fluid according to the value of @xmath19 . the disk - like solutions obtained here are physically well behaved because they satisfy the energy conditions required for any source of gravitational field , and because they are asymptotically flat . the asymptotic behavior was ensured by writing an appropriate relation between the metric function @xmath3 and the auxiliary function @xmath24 . according to the solutions , we got for the halo a fluid whose main stresses are different from zero ( @xmath250 ) and they are in the same direction of the main axes . on the contrary , the solutions obtained for the disk describe a fluid with null stress . in the kuzmin - toomre disks with halo we obtained that the most of the energy density was located at the center of the system ; so it is possible to see a well - defined central region with a maximum at ( @xmath115 and @xmath116 ) ; nevertheless , as far as @xmath125 increases , the amount of constants increases too , so we have more free parameters to adjust the energy density profile to a particular behavior . for instance , we presented in fig . [ densidiskuz ] one case in which the maximum energy density is not at the center of the system . we found that the morgan - morgan relativistic thin disks models presented in section [ sec : morgan ] does not have the same freedom as the kuzmin - toomre ones . in these models we only have two parameters to adjust : @xmath19 and @xmath192 , so by varing these parameters we only change the maximum of @xmath53 , @xmath12 and @xmath72 , and their corresponding rates of decrease . we computed the total mass for the @xmath126 model of the kuzmin - toomre relativistic disks with halo , turning out that @xmath251 , and we proved that the mass of the system converges for all @xmath125 . on the other hand , in the morgan - morgan relativistic disks solutions we computed the total gravitational mass for all the family of solutions which is @xmath252 . finally , we presented the rotational curves for some particular models in both families of solutions . we saw that , for some values of the constants , it is possible to get a behavior in which the square of the circular velocity increases from @xmath115 until a radius where @xmath72 stays approximately constant . nevertheless , we got some rotational curves with two peaks in the model @xmath253 of the kuzmin - toomre relativistic solutions . this is a very interesting behavior because the observed rotational curves in spiral galaxies exhibit these kind of features @xcite . gag was supported in part by vie - uis , under grants number 1347 and 1838 , and colciencias , colombia , under grant number 8840 . omp wants to thank the financial support from the program _ jvenes investigadores _ of colciencias , colombia .

in adiabatic quantum computation ( aqc ) @xcite quantum algorithms are implemented by initializing a system in an easily prepared ground state , followed by adiabatic evolution subject to a hamiltonian whose final ground state represents the solution to a computational problem . it is known that this model is computationally equivalent to the standard circuit model of quantum computation , in the sense that each model can simulate the other with polynomial resource overhead @xcite . while error correction methods have been proposed for aqc @xcite , and arguments have been put forth that aqc is inherently insensitive to certain types of errors @xcite , unlike the circuit model it is still an open question whether aqc can be made fault tolerant subject to realistic noise models pertaining to aqc in an open system setting . indeed , even what constitutes a consistent picture of adiabatic evolution in open systems is still the subject of some debate @xcite . nevertheless , experiments in aqc using superconducting qubits have made great strides recently @xcite . the performance of adiabatic quantum algorithms is characterized by the adiabatic error , i.e. , the fidelity loss between the actual time - evolved state ( the solution of the schrdinger equation ) and the instantaneous ground state , calculated at the final time @xmath0 . since the adiabatic theorem fock , kato , messiah : book , teufel : book guarantees that in the limit of arbitrarily slow evolution the error approaches zero , one expects that a more slowly varying hamiltonian and/or a longer evolution time should result in higher fidelity or accuracy . this is the _ accuracy - time tradeoff _ in quantum adiabatic algorithms . this tradeoff is often formalized by the adiabatic condition , which states ( roughly ) that the variation rate of the hamiltonian should be @xmath1 the adiabatic error times the gap squared . a clear disadvantage of this condition is the inherent vagueness of @xmath1 , which makes it difficult to reliably quantify the evolution time _ vs _ the desired accuracy . moreover , violations of the traditional adiabatic condition have been reported @xcite , in the sense that the condition is neither necessary nor sufficient for adiabatic evolution . while these violations have been explained as being either due to inconsistent manipulations @xcite or due to resonant transitions amin , it appears that there is no single adiabatic theorem . rather , a number of different rigorous conditions have been derived , which apply under various mathematical assumptions @xcite . a rigorous condition which holds for analytic hamiltonians and exibits the explicit scaling with system size , needed for aqc resource quantification , was derived in ref . @xcite . in this work we perform a comprehensive analysis of the adiabatic error for the case of an adiabatic quantum search algorithm , in a closed system setting . we focus on quantum search not only because it is an important example of a quantum speedup @xcite , but also because it is amenable to an exact analytical treatment . indeed , rather than relying on a particular form of the adiabatic theorem , our approach is based on an exact treatment of the underlying dynamics . we calculate the adiabatic error as an explicit function of the evolution time . we work out the formal @xmath2 expansion of the error often used in rigorous treatments of the adiabatic theorem through which we provide a large-@xmath0 _ polynomial _ upper bound for the adiabatic error . we shall argue that a leading order truncation of this series expansion can result in misleading estimates for the scaling of the evolution time _ vs _ system size . we draw this conclusion on the basis of a careful study of the adiabatic error , showing that it may exhibit a short-@xmath0 _ exponential _ decay , which is hidden in the formal polynomial expansion . we show that this short - time exponential decay heralds the correct scaling for the evolution time _ vs _ system size ( a quadratic speedup over classical search ) , avoiding the overestimation that results from the long - time polynomial decay , and which leads to a loss of the quantum speedup . additionally , we propose specific adiabatic evolution paths ( `` interpolations'')inspired by a procedure for minimization of time functionals for adiabatic evolution @xcite and use these to illustrate our results . we also examine the applicability of the traditional measure of the evolution time and contrast this to what we obtain from our analysis . this careful analysis enables us to show explicitly how one can reduce the adiabatic error as a function of the evolution time . the structure of this paper is as follows . we start with some general background in sec . [ ad - int ] by delineating the framework of adiabatic quantum computation , the definition of the adiabatic error , and the adiabatic theorem . in sec . [ qsham ] , we specialize to the specific problem of quantum search . there we introduce a general hamiltonian interpolation for the search problem , and analytically solve the corresponding schrdinger equation . section [ aderror ] contains the core of our results . we first derive an exact expression for the adiabatic error in subsection exactr . next , in subsection [ approx ] , we approximate this expression and show how the polynomial and exponential behaviors emerge . since estimation of the adiabatic error requires specific interpolation paths , we derive a general interpolation for the search hamiltonian in subsection ham , and investigate in detail three special cases . the adiabatic error for the general interpolation together with the special cases is estimated in subsection [ erpaths ] . in subsection [ strategy ] , we discuss a strategy for controllably reducing the adiabatic error . there we show how one can employ a freedom in the interpolation to manipulate the adiabatic error . this accounts for the performance - resource tradeoff . we conclude with a summary of the results and an outlook in sec . we begin by defining the adiabatic error as it arises in the context of aqc . we also provide a brief review of some pertinent facts concerning the traditional adiabatic theorem , also in the context of aqc . however , we note that in the sequel we shall not use the adiabatic theorem ; rather , we shall treat the dynamics directly by solving the schrdinger equation , and later enforce adiabaticity by means of a @xmath2 expansion . let us treat the total evolution time @xmath0 as a parameter and define the the scaled ( dimensionless ) time@xmath3.\]]assume that we have an @xmath4-dimensional quantum system that evolves for a total time @xmath0 under the hamiltonian @xmath5 with spectral decomposition @xmath6 \label{h}\]]where @xmath7 for @xmath8 , except that the ground state energy is separated from the rest of the spectrum by a nonvanishing gap @xmath9possible level crossing among excited eigenstates @xmath10 shall not concern us here because in the following we shall only focus on the ground state @xmath11 . we assume that the system is closed , i.e. , the evolution is governed by the schrdinger equation @xmath12where from now on dot denotes @xmath13 , and we set @xmath14 . the initial state is assumed to be the ground state , i.e. , @xmath15 _ remark_. we shall use the subscript @xmath0 to indicate the parametric dependence on @xmath0 . when symbols already have a subscript we shall avoid the additional @xmath0 subscript so as not to clutter the notation . a central quantity of interest to us is the adiabatic error @xmath16 , which quantifies the distance between the instantaneous ground state and the actual state : @xmath17note that @xmath16 is a distance and ( in)fidelity measure in their rigorous sense . indeed , the trace distance @xmath18 and fidelity @xmath19 between two arbitrary density operators @xmath20 and @xmath21 are defined as @xcite @xmath22where @xmath23 is the trace norm ( sum of singular values of @xmath24 ) . for pure states @xmath25 it is not hard to show that @xmath26 and @xmath27 . thus@xmath28 an equivalent and useful formulation of @xmath16 is the following . let @xmath29 denote the ground - state eigenprojection . the connection between the initial preparation @xmath30 and the time evolved state @xmath31 is given by @xmath32where the unitary operator @xmath33called the adiabatic intertwiner @xcite determines the evolution in the eigenspace @xmath31 and its orthogonal complement @xmath34 : @xmath35one can assign a ( dimensionless ) adiabatic hamiltonian , @xmath36 , to this evolution defined via @xmath37on the other hand , the evolution operator generated as @xmath38dictates the actual dynamics : @xmath39thus the error @xmath16 in fact measures the difference between @xmath40 and @xmath33 , or equivalently how far the unitary operator @xmath41is from @xmath42 ( identity ) . indeed , we have @xmath43which vanishes iff @xmath44 . the above formulations have presumed that the degeneracy of the ground - state eigensubspace does not change in time . nonetheless , there may be situations in which this degeneracy preservation assumption does not hold . in fact , as we shall see later , the quantum search problem we study in this paper falls into this category . let us assume that the initial state is pure , @xmath45 , but the instantaneous ground - state eigenprojection @xmath31 is not necessarily rank-@xmath46 , accounting for the possibility of degeneracy . intuitively , if the actual state @xmath47 only has components in the support of @xmath31 the algorithm has achieved its goal at the instant @xmath48 , whereas a less than full overlap denotes lack of success at this instant . the overlap of @xmath49 and the support of @xmath31 can be quantified , e.g. , with @xmath50 , whence we define the following performance error : @xmath51it is evident that @xmath52 , with @xmath53 iff the actual state has vanishing overlap with the support of the instantaneous ground - state eigenprojection , while @xmath54 iff the actual state resides anywhere in the support . we can write @xmath55 $ ] and @xmath56=\mathrm{tr}[p(0)\omega _ { t}(\tau ) p(0)\omega _ { t}^{\dag } ( \tau ) ] $ ] . using this , note the difference between @xmath57|^{2}}\]]and @xmath58}.\]]namely , in the nondegenerate case @xmath59 iff @xmath60 , indicating that the evolution was perfectly adiabatic . in the degenerate case , on the other hand , algorithmic success does not require the evolution to be perfectly adiabatic since only nonvanishing overlap with the ground - state eigensubspace is required , i.e. , @xmath54 iff @xmath61 $ ] , whereas complete failure requires the dynamics to remove any overlap with the ground - state eigensubspace , i.e. , @xmath53 iff @xmath62 $ ] . while the error as defined in eq . ( [ error - final - form ] ) is not necessarily a distance in the strict sense , it is adequate for quantifying the adiabatic error . note also that @xmath63 reduces to @xmath16 when the ground state is nondegenerate . from now on we shall be using these various expressions for the adiabatic error as appropriate in the rest of this paper . one variant of the traditionaladiabatic theorem @xcite states that given an @xmath64 and a time - dependent hamiltonian @xmath65 with a nondegenerate ground state , the adiabatic error satisfies @xmath66 , provided that @xmath67 in which @xmath68 is the standard operator norm , defined as the maximum singular value , i.e. , @xmath69 which reduces to @xmath70 for normal operators . as remarked in the introduction , this condition is hardly quantitative due to the intrinsic vagueness of @xmath1 , and has been the subject of critique ( consistent with its lack of rigor ) @xcite , justifications @xcite , and rigorous improvements avron:871,avron:872,nenciu:93,martinez , hagedornjoye:02,jrs , ohara:042319,boixo , lrh . nevertheless , it remains a useful rule of thumb , as long as it is applied with appropriate care . an immediate implication of the adiabatic theorem is that , assuming it is initialized in the ground state , the system remains close to the final ground state at @xmath71 . thus , by choosing the hamiltonian such that @xmath72 corresponds to a _ simple _ ground state @xmath73 ( simple in the sense that it is easily preparable ) , and @xmath74 represents a hamiltonian whose ground state @xmath75 identifies the solution to a computationally _ hard _ problem , one can devise an adiabatic version for the corresponding algorithmic or computational task . this is precisely the insight that led to the advent of aqc @xcite . a simple `` annealing schedule '' , or `` path '' , between @xmath72 and @xmath74 is the following linear interpolation in @xmath48 : @xmath76 in physical situations , however , one often realizes the dynamics by tuning some time - dependent control knobs or couplings @xmath77 of the hamiltonian . this suggests that a generalization of eq . ( [ aaa ] ) can be introduced by assuming access to a controllable set of non - commuting , linearly - independent primitive hamiltonians @xmath78 combined , e.g. , as @xmath79=\sum_{i=1}^{k}x_{i}(\tau ) h_{i}$ ] . further generalizations can be introduced as well @xcite . however , for our purposes in this paper we shall consider the parametrization rolandcerf , qab @xmath80with the boundary conditions @xmath81 two remarks are in order regarding aqc . ( i ) a primary goal in aqc is to make @xmath82 decrease more rapidly for a given @xmath0 and problem size , or alternatively , to make @xmath0 smaller for a given @xmath83 and problem size . often the problem size is given by @xmath4 , the dimension of the hilbert space . however , in the context of many - body quantum systems , where the hilbert space is a tensor product of subsystems ( e.g. , qubits ) , @xmath84 is the correct measure of problem size , coinciding with system size . ( ii ) the run time complexity of a quantum algorithm should be defined as @xmath85not @xmath0 @xcite . this _ regularization _ is required because of the energy - time tradeoff in quantum mechanics , in the sense that multiplication of @xmath65 in eq . ( [ cond ] ) by some positive factor @xmath86 manifests itself as dividing @xmath0 by the same factor , making it possible to decrease @xmath0 arbitrarily by choosing @xmath86 sufficiently large . this tradeoff can also be understood via the schrdinger equation ( [ eqscalt ] ) , in which the final state of a system evolving under hamiltonian @xmath65 for @xmath0 is the same as that of a system evolving under @xmath87 for @xmath88 . this ambiguity is fixed by the definition of @xmath89 as in eq . ( [ runt ] ) . scaling of @xmath90 with system size , for a given upper bound on the error @xmath91 , determines the run time complexity of the corresponding quantum algorithm . grover s quantum search algorithm @xcite performs a search for @xmath92 marked items among @xmath4 items of an unsorted database , presuming that there is an oracle for distinguishing the marked from the unmarked items . the algorithm in its original form ( @xmath93 ) comprises the following steps : ( i ) assign orthonormal quantum states ( i.e. , labels ) @xmath94 to the items , ( ii ) prepare the quantum system in the equal superposition state @xmath95 , and ( iii ) apply the grover operator encompassing the oracle repeatedly @xcite . the algorithm finds a marked item after @xmath96 calls of the oracle a quadratic speedup over the best classical algorithm and is provably optimal for any @xmath4 ( not necessarily very large ) @xcite . various generalization of the algorithm have been introduced ( e.g. , refs . @xcite ) , and it has also been implemented experimentally in a number of physical settings ( e.g. , refs . @xcite ) . an adiabatic hamiltonian version of the search algorithm was first introduced in ref . @xcite , but failed to display the expected quadratic speedup as it relied on the linear interpolation of eq . ( [ aaa ] ) . this was fixed in ref . @xcite by using a non - linear but one - dimensional interpolation with @xmath97 [ recall eq . ( 2-d - interpol ) ] , which moves fast when away from the minimum gap , but slows down near it . as shown in ref . @xcite , this result can be further improved , in the sense of a smaller adiabatic error , by adopting a two - dimensional interpolation as in eq . ( [ 2-d - interpol ] ) , @xmath98with the two projective hamiltonians @xmath99where @xmath100 is the equal superposition of all of the labelstates ( items ) , @xmath101 is the projection over the subspace @xmath102 of the marked items ( @xmath103 ) , and @xmath104 is a dimensional constant which sets the energy scale . in other words , the initial state @xmath105 is the ground state of the initial hamiltonian @xmath106 , while any state supported fully on @xmath102 is a ground state of the final ( oracle ) hamiltonian @xmath107 . note that unlike previous treatments of adiabatic quantum search @xcite , the hamiltonian @xmath107 has a degenerate ground eigenspace spanned by @xmath108 . we remark that the search hamiltonian ( [ search - ham ] ) is a member of the following class of projective hamiltonians qab , aqc - intrinsic , aharonovta - shma : @xmath109where @xmath110 , @xmath111 , with @xmath112 and @xmath113 fixed ( normalized ) vectors in the system hilbert space , for which @xmath114 is a given function of @xmath4 . in the case of the search problem , we have @xmath115 , @xmath116 , whence @xmath117 . the results of this paper can be generalized to other members of the class of projective hamiltonians . in the computational basis , in which @xmath118 , the schrdinger equation ( [ eqscalt ] ) becomes @xmath119 , \]]with the initial value @xmath120 . it can be seen from this expression that the marked components all behave similarly , as do the unmarked components . hence we can rewrite the state @xmath121 as @xmath122where the subscripts u and m denote unmarked and marked , respectively . the normalization condition now reads @xmath123by defining the ( unnormalized ) two - dimensional vector @xmath124and the ( non - hermitian ) reduced hamiltonian matrix @xmath125 in which @xmath126is the fraction of the marked items , the schrdinger equation reduces to @xmath127with the initial condition @xmath128 . therefore , not only is the parameter space of the problem two - dimensional , it is described by an effectively two - dimensional hamiltonian ( in the m - u representation ) . this reduction from the real hamiltonian @xmath129 [ eq . ( [ search - ham ] ) ] to the effective hamiltonian @xmath130 [ eq . ( [ pseudoham ] ) ] will prove useful in our analysis below . later in the paper we shall need the norm of the hamiltonian as well . this can be calculated easily from eqs . ( [ op - norm ] ) and ( [ pseudoham ] ) , and yields : @xmath131and similarly , @xmath132 the hamiltonian @xmath133 ( [ search - ham ] ) has three distinct dimensionless eigenvalues @xmath134 , where @xmath135where @xmath136is the dimensionless gap ( hence @xmath137 ) and @xmath138 is @xmath139-fold degenerate . let @xmath140 and @xmath141 denote the pauli matrices . let us define the similarity matrix@xmath142and the unitary@xmath143/2}. \label{formofa}\]]then the effective hamiltonian @xmath144 [ eq . ( [ pseudoham ] ) ] satisfies @xmath145where @xmath146is the hermitian core of @xmath147 , and we easily find that @xmath148\openone+\delta ( \tau ) \widehat{a}(\tau ) \sigma _ { z}\widehat{a}^{\dag } ( \tau ) \right\ } . \label{uni - interp}\]]this last result is remarkable : it states that , up to an overall ( time - dependent ) shift @xmath149\openone$ ] and a conformal factor @xmath150 , the reduced hamiltonian @xmath151 is a _ unitary interpolation _ @xmath152 @xcite . we will exploit this observation below . the non - hermitian reduced hamiltonian @xmath144 and its hermitian core @xmath151 have the same set of eigenvalues @xmath153 , and we have the spectral resolution@xmath154where @xmath155and @xmath156 are the eigenvectors of @xmath157 , corresponding to the eigenvalues @xmath158 . the unitary operator @xmath159 acts as a reduced adiabatic intertwiner [ eq . ( [ ad - int ] ) ] , in the sense that for the reduced projection @xmath160 we have @xmath161 _ remark . _ we emphasize that throughout the paper hat and tilde denote states or operators in the reduced representation ; the only exception is @xmath162 . in solving the schrdinger equation and calculating the adiabatic error @xmath82 , it is more convenient to work with the normalized state @xmath163equation ( [ sch-1 ] ) now becomes @xmath164with @xmath165 . solving this equation results in @xmath166 in which @xmath167\ ] ] is the time - ordered reduced evolution operator . having observed that the grover search problem can be cast as a conformal unitary interpolation , we outline a general , systematic approach for solving the corresponding class of schrdinger equations using the dyson series expansion @xcite . consider as a specialization of the general time - dependent hamiltonian of eq . ( [ h ] ) the conformal unitary interpolation hamiltonian @xmath168 in which @xmath169is the spectral decomposition of the ( traceless ) initial hamiltonian @xmath72 , the unitary operator @xmath33 satisfies the adiabatic schrdinger equation ( [ ad - schr ] ) generated by the adiabatic hamiltonian @xmath170 @xcite , and ( the dimensionless gap ) @xmath171 is a smooth function with the initial value @xmath172 . it is also useful to think of the time - dependent hamiltonian @xmath5 [ eq . ( [ h_t ] ) ] as the adiabatic interaction picture hamiltonian , though normally an interaction picture transformation does not involve a time - dependent prefactor such as @xmath173 . it is evident that the eigenvalues and eigenprojections of @xmath5 satisfy @xmath174equation ( [ spect ] ) implies that the spectrum of @xmath5 evolves conformally with @xmath150 , i.e. , all the eigenvalues are multiplied by the same factor , while eq . ( [ ap : p ] ) implies that the eigenprojections are unitarily connected , and the degeneracy is constant in time . comparison of eqs . ( [ ad - int ] ) and ( [ ap : p ] ) reveals that @xmath175 and here too , @xmath33 plays the role of the adiabatic intertwiner . rather than solving the schrdinger equation @xmath176we solve the equation of motion for the adiabatic interaction picture propagator [ cf . ( [ omega ] ) ] @xmath177this provides a more direct tool for the calculation of the adiabatic error [ eq . ( [ def : ad - err ] ) ] . let us define @xmath178note that @xmath179 has dimensions of energy while @xmath180 is dimensionless . in the adiabatic interaction picture the ( dimensionless ) perturbation is @xmath181 , i.e. , it follows from eqs . ( [ ad - schr ] ) and ( [ wdot ] ) that @xmath182\omega _ { t}(\tau ) . \label{omega - dott}\]]we also define the two unitaries @xmath183 and @xmath184 through the following equations : @xmath185it is easily seen that @xmath186 also satisfies eq . ( [ omega - dott ] ) , so that @xmath187 to simplify the analysis , we only consider hamiltonians for which @xmath188 in which @xmath189 is an integrable function and @xmath190 is a constant ( @xmath48-independent ) and traceless operator belonging to the space of linear operators acting on the system hilbert space . thus from eq . ( [ h_1 ] ) we obtain @xmath191 note that , from eqs . ( [ h_0 ] ) and ( [ v_0 ] ) , @xmath192where @xmath193 is shorthand for @xmath194we shall use the similar shorthand @xmath195wherever convenient . inserting @xmath196 into eq . ( [ v_1 ] ) yields @xmath197 in which the kernel @xmath198 is defined as @xmath199equation ( [ v_1-eq ] ) , or equivalently the volterra equation @xmath200can be solved iteratively , yielding the dyson series @xmath201 now we apply the method described above to the adiabatic quantum search problem . recall that we are working with the reduced states and operators ( hence the hat over all reduced quantities , except the gap @xmath162 [ eq . ( [ gap ] ) ] ) comparing eq . ( [ hhat ] ) with eq . ( [ h_t ] ) implies that @xmath202after excluding the trivial term @xmath203 from @xmath204 . similarly , eqs . ( [ formofa ] ) and ( [ h_a ] ) yield @xmath205\right ] \notag \\ & \overset{\text{(\ref{gap})}}{=}&\sqrt{r(1-r)}\frac{x_{1}\dot{x}_{2}-\dot{x}_{1}x_{2}}{\delta ^{2 } } , \label{h_a } \\ \widehat{\xi } & = & \sigma _ { y}.\end{aligned}\]]as a result , from eqs . ( [ v_0-form ] ) and ( [ kernel ] ) we obtain @xmath206where the phase factor @xmath207~\mathrm{d}\tau ^{\prime } \]]compensates for the removal of the trivial term from @xmath208 . some simple algebra then yields @xmath209where , for @xmath210 , the dyson series terms are @xmath211and @xmath212 . this completes the derivation of @xmath213 in the previous section , we worked out the solution to the schrdinger equation in the quantum search problem . having collected the pertinent ingredients , we now return to calculating our main object of interest , @xmath16 . after deriving an exact formula for the error , we proceed with approximating it in the large system - size limit , identified with @xmath214 . we start with the well - known polynomial expansion of @xmath82 in terms of @xmath0 , which works well for large times . in refining this result , we show that in fact two regimes are discernible in the behavior of @xmath82 vs @xmath0 : ( i ) the onset of _ exponential _ decrease , followed by ( ii ) a _ polynomial _ tail . this dichotomy will appear to be crucial in a correct characterization of the scaling of the run time of the algorithm with system size . the exact behavior of the error in the algorithm depends strongly on the form of the interpolation one chooses for the hamiltonian . inspired by our earlier study aiming at minimizing the adiabatic error in quantum algorithms @xcite , we shall suggest a general class of interpolations , which includes three specific cases already studied in the literature . next , we investigate the specific behavior of the adiabatic error for each interpolation , separately . finally , we shall suggest methods for suppressing the adiabatic error even further . recall that the adiabatic quantum search hamiltonian ( [ search - ham ] ) has a nondegenerate ground state @xmath105 at the initial time @xmath215 , whereas the ground - state eigenprojection at the final time @xmath216 is @xmath217 , which is @xmath92-fold degenerate . here , any full superposition of the form @xmath218 will work equally well , whereas if @xmath219 does not have complete support over @xmath217 then this indicates that the algorithm has partially failed . hence , following the discussion in sec . ad - int [ eq . ( [ error - final - form ] ) ] , the adiabatic error at the final time is determined by @xmath220 an equivalent formulation can be obtained for the two - dimensional reduction we discussed in sec . [ qsham ] . in this representation @xmath221 is replaced by @xmath222 [ eq . ( [ state - red ] ) ] ; similarly , the instantaneous ground state is represented by the nondegenerate state @xmath223 [ eq . ( [ phi - hat ] ) ] . thus we can employ the error formula appropriate for the non - degenerate case [ eq . ( [ def : ad - err2 ] ) ] , whereby @xmath224 in which @xmath225 is given by eq . ( [ o - hat ] ) , and in the last line we used the unitarity of @xmath225 . the equality of eqs . ( [ er ] ) and ( [ ad - error ] ) is immediately seen from @xmath226 inserting @xmath225noting eqs . ( [ v0 ] ) and ( [ v1])into eq . ( [ ad - error ] ) yields the following _ exact _ expression for the adiabatic error at the final time : @xmath227which is upperbounded by @xmath228from the above equations we can in principle calculate the adiabatic error or its bound given that we know all @xmath229 s . _ the value of @xmath230 will be important later . in this case , we have @xmath231 , thence eq . ( [ ad - error ] ) yields @xmath232 in which we used the boundary conditions ( [ bc-2 ] ) in @xmath233 . this relation is valid for any interpolation that satisfies the boundary conditions . an exact calculation of the adiabatic error from eq . ( [ error ] ) can be challenging because of the infinite number of terms in the summand and the fact that each term contains a multiple integral . to alleviate this difficulty , in this subsection we approximate the upper bound on @xmath82 from eq . ( [ error ] ) from the first few @xmath234 s , and argue that this suffices for most algorithmic purposes . we start from an expansion in powers of @xmath2 , based on integration by parts , and explain its limitations . we then provide more careful analyses , based on the residue theorem and on the stationary phase method , both of which lead to an exponential error estimate . the most common rigorous adiabatic approximation employs an expansion in powers of @xmath2 , presuming that @xmath0 is large @xcite . let us now show how one can systematically expand the adiabatic error as a polynomial in@xmath2 by extracting powers of @xmath2 from each term @xmath235 through integration by parts . from the identity @xmath236 [ valid for any differentiable function @xmath237 , we obtain the following relation by integration by parts : @xmath238 . \notag\end{aligned}\]]notice how this extracted a @xmath2 in front of the first term . in the second integral on the right hand side we can iterate the same trick of replacing the exponential with its derivative ; which generates @xmath239 and higher order terms . this provides a systematic way for generating @xmath240 expansions of exponential integrals , as we shall see more specifically below for the @xmath235 s . using eq . ( [ ibp - identity ] ) , we obtain @xmath241 . \label{i1}\end{aligned}\]]applying once more the exponential identity ( [ ibp - identity ] ) for the second integral above gives rise to @xmath242 terms ; whence , @xmath243+o\bigl(\frac{1}{t^{2}}\bigr ) , \notag \\ & & \end{aligned}\]]i.e . , @xmath244 . in analogous fashion , for @xmath245 we obtain @xmath246 , \notag \\ & & \end{aligned}\]]from which @xmath247-\frac{\widehat{h}_{a}(0)}{(jt)^{2}}\bigl[\frac{\widehat{h}_{a}(\tau ) e^{-ijt\int^{\tau } \delta } } { \delta ( \tau ) } \notag \\ & & -\widehat{h}_{a}(0)\bigr]+o\bigl(\frac{1}{t^{3}}\bigr ) , \label{i2}\end{aligned}\]]and @xmath248 . by induction , one can conclude from eq . ( [ inttran ] ) that @xcite @xmath249for @xmath250 . thus , from eq . ( [ error ] ) the adiabatic error bound becomes @xmath251+o\bigl(\frac{1}{t^{2}}\bigr ) . \label{addelta}\end{aligned}\]]this relation can be simplified further . from the boundary conditions ( bc-1 ) and ( [ bc-2 ] ) , we obtain @xmath252 [ eq . ( [ gap ] ) ] , @xmath253 and @xmath254 [ eq . ( [ h_a ] ) ] . thus eq . ( addelta ) reduces to @xmath255+o\bigl(\frac{1}{t^{2}}\bigr ) . \label{er - simp}\]]proceeding in a similar manner , one can in principle obtain the exact form of the coefficient of each @xmath256 term , for arbitrary @xmath250 . _ remarks . _ let us make some remarks regarding the polynomial expansion , and in particular eq . ( [ er - simp ] ) . \(i ) notice that often the @xmath240 series is _ truncated _ after the first or at most the second order term , on the basis of the assumption that for sufficiently large @xmath0 the first couple of terms should give a reliable and accurate upper bound . however , without correctly defining what large @xmath0 means , a truncation after the first few terms might be unjustifiable . in fact , in addition to @xmath0 , the system size @xmath84 ( introduced here through @xmath257 ) and the gap @xmath150 are also key players in the estimation of @xmath16 . the minimum ( system - size dependent ) gap @xmath258 works in general as a bottleneck for the performance of quantum algorithms ( e.g. , refs . farhi1,farhi2,latorre , qab , schaller ) . specifically , where the gap closes or becomes small , the adiabatic approximation may not hold , indicative of a quantum phase transition ( in the thermodynamic limit ) sachdev : book , latorre , schutzhold:06,schaller2,aminchoi . this implies that the coefficients of some high order @xmath256 terms might have a stronger gap dependence than those of lower order terms . with this caveat , neglecting those higher order terms is not always possible . in fact , it is not difficult to see that the coefficient of the @xmath239 has a @xmath259 dependence ( see also ref . @xcite ) , stronger than the @xmath260 dependence of the coefficient of the @xmath2 term in @xmath261 [ eq . ( [ h_a ] ) ] . \(ii ) for similar reasons , an estimate of @xmath0 arising from @xmath262 ( for a given @xmath83 ) along with a truncated @xmath240 expansion , is not always reliable . we shall see this explicitly later in this section . \(iii ) one might argue that @xmath263 results from an energy - time uncertainty relation such as @xmath264 [ presuming @xmath82 is directly related to the uncertainty in measurement of energy ] . however , this argument is not rigorous and should not be considered as a replacement for the analysis leading to the @xmath240 expansion ( unless justified carefully ) . a rigorous energy - time uncertainty relation is given , for example , by the mandelstam - tamm inequality @xmath265\times t_{\psi } [ k]\geq 1/2,\]]in which @xmath266\equiv \delta _ { \psi } [ k]/|\mathrm{d}\langle \psi @xmath267\equiv \sqrt{\langle \psi |x^{2}|\psi \rangle -\langle \psi |x|\psi \rangle ^{2}}$ ] @xcite . hence , as is well known the naive relation @xmath268\times t\approx 1 $ ] should be treated with care . strictly , a relation between @xmath269 $ ] and @xmath82 can be constructed as the following . note that we can write @xmath270where @xmath271 is orthogonal to @xmath272 [ eq . [ ad - error ] ] . hence , after some algebra we obtain @xmath273\bigl|_{\tau = 1}}+o(\delta _ { \text{ad}}^{2 } ) . \hskip-3 mm & & \end{aligned}\]]despite this relation , connecting @xmath0 and @xmath274 $ ] is not straightforward . although using different versions of the energy - time uncertainty relation @xcite may provide additional insights , we shall not further pursue this here . [ [ residue - theorem - analysis ] ] residue theorem analysis + + + + + + + + + + + + + + + + + + + + + + + + in the previous subsection we used integration by parts to arrive at a polynomial expansion . let us now show that an alternative , more careful analysis based on the residue theorem of complex analysis , reveals that the adiabatic error decays _ exponentially _ for sufficiently short times . in some sense , this exponential behavior is reminiscent of the well - known landau - zener formula for two - state quantum systems @xcite , which in its simplest form states that the _ tunneling probability _ @xmath275 from @xmath276 to @xmath277 , for the hamiltonian @xmath278 , is @xmath279we notice that @xmath280 is in fact intimately related to @xmath82 in this simple two - state case , as @xmath281the exponentiality of the adiabatic error _ vs _ time and ( some power of ) the gap has been previously established in generality in rigorous treatments of the adiabatic theorem nenciu:93,martinez , hagedornjoye:02,lrh ( see also ref . @xcite in a more restricted setting ) . in what follows we focus on @xmath282 ; higher order terms can be treated similarly . we start from eq . ( [ error ] ) , whence @xmath283here denotes the higher order terms @xmath284 ( @xmath210 ) , whose neglect we justify below in the specific examples we discuss . an exponential error term can be obtained , for example , by extending the integral to the complex time plane and using an appropriate closed contour for the application of the residue theorem @xcite . a precursor to this treatment of the adiabatic error can be found , e.g. , in ref . @xcite . if @xmath285 ( @xmath286 ) is not constant it has poles at points @xmath287 where the gap vanishes : @xmath288 [ eq . ( [ h_a ] ) ] . from eq . ( [ gap ] ) , we obtain @xmath289 , or ( assuming @xmath290 ) @xmath291where @xmath292 . the poles @xmath287 can in principle be obtained by inverting this relation for a given interpolation @xmath293 . note , however , that there might exist other singularities ( typically at infinity ) arising from the exponential @xmath294 in eq . ( h1-int ) . therefore , estimating the integral ( [ h1-int ] ) requires finding all contributing singularities in a suitably chosen contour @xmath295 in the complex @xmath48-plane . the value of the integral @xmath296 can now be obtained by calculating the residues of the integrand at the poles enclosed inside the contour , @xmath297}\widehat{h}_{a}(z)e^{ijt\int^{z}\delta } \mathrm{d}z\bigr| . \label{tmp - eq-1}\end{aligned}\]]for example , if the poles @xmath287 are _ simple _ and arise from @xmath298 , calculating the integral becomes straightforward . recall that @xmath299 [ eq . ( [ h_a ] ) ] and the residue of a rational function @xmath300 at a simple pole @xmath301 is given by @xmath302 @xcite . thus , in this case we obtain @xmath303= \label{residue } \\ \hskip-5 mm & & \frac{\sqrt{r(1-r)}\dot{x}_{21}(z_{0})e^{ijt\int^{z_{0}}\delta } } { \frac{\mathrm{d}}{\mathrm{d}z}\bigl[\bigl(1-x_{21}(z)\bigr)^{2}+4rx_{21}(z)\bigr]\bigl|_{z_{0}}}\approx \frac{1}{4i}e^{ijt\int^{z_{0}}\delta } \notag,\end{aligned}\]]where in the last line we assumed @xmath304 . note that this computation of the residues does not necessarily hold when @xmath305 , or when @xmath287 is at infinity . this yields that an exponentially decreasing contribution emerges from @xmath306 $ ] of the exponential within the integrand perhaps in addition to a generically non - exponentially - decreasing term . thus , from eq . ( [ residue ] ) we find @xmath307 } + r_{t}^{\prime } ( 1 ) , \label{xx}\]]where @xmath308 results from the non - exponential contribution of the residue at @xmath287 and @xmath309 encapsulates the collective nonvanishing contribution of other segments of the contour as well as other non - exponentially - decreasing contributions emerging from the terms in eq . ( [ h1-int ] ) . ) , and from eq . ( [ tmp - eq-1 ] ) , we obtain : @xmath310 . this is in good agreement with eq . ( [ t=0-eq ] ) when @xmath311 and @xmath312 . ] [ [ stationary - phase - analysis ] ] stationary phase analysis + + + + + + + + + + + + + + + + + + + + + + + + + an alternative and complementary approach for obtaining the exponential contribution to the adiabatic error is to use the stationary phase method . this method is appropriate for obtaining asymptotic value of complex integrals of the form @xmath313 , where @xmath314 is a large number and @xmath315 is a path for the integration . assuming @xmath316 is a slowly - varying function over @xmath315 and @xmath317 is an analytic function , one can argue that the major contribution to the integral comes from the point(s ) @xmath287 at which @xmath317 has a minimum , whence hayek : book @xmath318 to apply this result to eq . ( [ h1-int ] ) , we replace @xmath317 and @xmath319 with @xmath320 and @xmath285 , respectively . accordingly , @xmath287 is where @xmath321 . the minimum point @xmath287 is often a complex number . in fact , in quantum many - body situations , the gap @xmath162 is a nonnegative function , often with a nonvanishing minimum , which becomes zero in the thermodynamic limit @xcite . in such cases , @xmath317 does indeed have a minimum . it is also required that @xmath322 be finite [ this of course is not satisfied when @xmath323 has a pole or singularity at @xmath287 ] . if all these conditions are satisfied , eqs . ( [ h1-int ] ) and ( [ spm ] ) yield @xmath325 } , \label{st - ph}\]]as we wished . this relation complements eq . ( [ residue ] ) in that it may be applicable when eq . ( [ residue ] ) is not . [ [ discussion ] ] discussion + + + + + + + + + + the emergence of the exponential dependence of @xmath326 on @xmath0 [ eqs . ( [ xx ] ) and ( [ st - ph ] ) ] in contrast to the polynomial dependence [ eq . ( [ er - simp ] ) ] is remarkable , as it indicates a much faster decay of the adiabatic error than what is suggested by the standard @xmath2 expansion . in the exponential regime it suffices that @xmath0 be large compared to @xmath327 $ ] , or roughly @xcite:@xmath328this is a less stringent condition than the standard condition ( [ cond ] ) or its more rigorous counterparts avron:871,avron:872,nenciu:93,martinez , hagedornjoye:02,jrs , ohara:042319,boixo , lrh , involving higher powers of the gap . the crossover point @xmath329 between the exponential and the polynomial regimes can be estimated by solving @xmath330}\approx \frac{\sqrt{r}}{jt^{\star } } \bigl(|\dot{x}_{1}(1)|+|\dot{x}_{2}(0)|\bigr ) , \label{crossover}\]]in which @xmath331 is a non - exponential prefactor given by eq . ( [ xx ] ) or ( [ st - ph ] ) . if @xmath332 ( @xmath333 ) the exponential ( polynomial ) behavior prevails . later in this section , we show explicitly that the expected run time for the quantum search algorithm is often given by the exponential contribution ; the polynomial regime may _ overestimate _ the minimum run time required for reasonable accuracy . since the scaling of the run time depends on which interpolation we choose for the hamiltonian , in the following we shall obtain specific interpolations by employing some recently developed results for ( partial ) minimization of the adiabatic error @xcite . the set of available control knobs ( @xmath334 ) , as well as the way one varies them , determine the specific hamiltonian interpolation implemented in a laboratory setting . theoretically , though , there are various ways by which one can obtain families of hamiltonians for an adiabatic quantum algorithm . one natural choice is interpolations which minimize physical cost . for example , in the setting of refs . rolandcerf , qab , time functionals were constructed from a _ local _ version of the adiabatic condition ( [ cond ] ) , whose minimization resulted in a set of euler - lagrange equations for the underlying interpolations . a different method was suggested in ref . @xcite , where it was shown that in the standard @xmath240 expansion of the adiabatic error @xmath16 [ general counterpart of eq . ( [ addelta ] ) or ( [ er - simp ] ) ] , the coefficient of the @xmath2 term has a _ geometric _ part , in a differential geometric sense . specifically , this geometric coefficient , in terms of the ground - state eigenprojection , @xmath31 is@xmath335\bigr\vert~\mathrm{d}\tau ^{\prime } . \label{geometric}\]]minimizing this coefficient yields adiabatic geodesic interpolations , which partially decrease @xmath336 for a given @xmath0 . it follows from standard variational calculus @xcite that the geodesic interpolations satisfy the following equation @xmath337\bigr\vert\bigl|_{\mathbf{x}_{\text{geo}}(\tau ) } = \text{const . } , \label{geo - eq}\]]where the constant is chosen to satisfy boundary conditions . we adopt this method in the following and derive geodesic interpolations for the adiabatic quantum search . in the reduced two - dimensional representation , @xmath338 is replaced with @xmath339 . thus we have @xmath340noting with the help of eqs . ( [ formofa ] ) , ( [ phi - hat ] ) , and ( [ h_a ] ) that @xmath341we find @xmath342\bigr\vert & = & \sqrt{2\langle \dot{\widehat{\phi } } _ { -}|\dot{\widehat{\phi } } _ { -}\rangle } \notag \\ & \overset{\text{(\ref{phi - hat})}}{=}&\sqrt{2}|\widehat{h}_{a}|.\end{aligned}\]]hence , according to eq . ( [ geo - eq ] ) , a class of adiabatic geodesics can be obtained from @xmath343which implies that the adiabatic hamiltonian @xmath344 is in fact constant . in other words , from @xmath345 [ eq . ( [ formofa ] ) ] we can see that @xmath346this equation suggests that a more general choice for the intertwiner @xmath347 can be obtained by @xmath348 , for some arbitrary differentiable @xmath349 ; i.e. , @xmath350 . in terms of @xmath323 , this translates into choosing @xmath351 . in terms of the geometric factor ( geometric ) , this simply means that on the geodesic , the time is determined by @xmath352 rather than @xmath48 . to see this , note that @xmath353\bigr\vert~\mathrm{d}\tau ^{\prime } & = & \int_{0}^{\theta ( \tau ) } \bigl\vert\lbrack \partial _ { \theta } p(\theta ) , p(\theta ) ] \bigr\vert~\mathrm{d}\theta , \notag \\ & & \end{aligned}\]]which in turn , from variational calculus , gives the following equation for the geodesics : @xmath354\bigr\vert\bigl|_{\mathbf{x}_{\text{geo}}(\theta ( \tau ) ) } = \mathrm{const . } \label{geo - eq2}\]]hence , @xmath355\bigr\vert & = & \frac{1}{\dot{\theta}}\bigl\vert\lbrack \dot{\widehat{p}}_{-},\widehat{p}_{-}]\bigr\vert \notag \\ & = & \sqrt{2}\frac{|\widehat{h}_{a}|}{\dot{\theta } } \notag \\ & = & \mathrm{const.},\end{aligned}\]]where we assumed @xmath356 , @xmath357 , and @xmath358 . as a result , we choose the adiabatic geodesic satisfying @xmath359 we remark that introducing an arbitrary nondecreasing function @xmath360 will serve as more than a generalization . in fact , we show below that @xmath349 allows us to substantially enhance the suppression of the adiabatic error . for example , choosing a @xmath349 such that it is a differentiable function of @xmath48 ( up to some controllable order , say , @xmath361 ) , with vanishing derivatives ( up to the same order @xmath361 ) at the initial and final times , can suppress the adiabatic error [ up to @xmath362 . this property which comes at the relatively small price of sufficient control at the beginning and end of the dynamics may have immediate applications in experimental realizations of quantum annealing and aqc . the above geodesic equation can be solved analytically . recall that @xmath363 , with the boundary values @xmath364 and @xmath365 [ eqs . ( [ bc-1 ] ) and ( [ bc-2 ] ) ] . then from eq . ( h_a ) we find @xmath366^{2}+4rx_{21}(\tau ) } = \varphi \dot{\theta}(\tau ) , \label{weak}\]]the solution to this equation can be written as follows : @xmath367}{\sin [ 2(1-\theta ( \tau ) ) \varphi ] } , \label{relat}\]]where we must choose @xmath368.\]]observe that @xmath369 has the following symmetry : @xmath370which is satisfied , for example , by requiring @xmath371equations ( [ relat ] ) and ( [ symmetry ] ) identify a two - dimensional interpolation for the quantum search hamiltonian . notice that , given eq . ( [ relat ] ) , we can also add another relation ( satisfying the boundary conditions ) so as to find other classes of interpolation . for example , we can choose @xmath372 in which @xmath373 can be a smooth function with the boundary values @xmath374 ; e.g. , @xmath375 or @xmath376 . choosing a form for @xmath373 corresponds to assuming a given @xmath377 [ eq . ( [ hnorm ] ) ] , which implies a given control over the maximum amount of the available energy in the system . thus , eqs . ( [ relat ] ) and ( [ bc-3 ] ) yield @xmath378}{2\sin ( \varphi ) \cos [ ( 1 - 2\theta ( \tau ) ) \varphi ] } \notag \\ & = & \frac{a(\tau ) } { 2}\bigl[1+\cot ( \varphi ) \tan [ ( 1 - 2\theta ( \tau ) ) \varphi ] \bigr ] , \label{x1 } \\ \hskip-6 mm x_{2}\bigl(\theta ( \tau ) \bigr ) & = & \frac{a(\tau ) \sin [ 2\theta ( \tau ) \varphi ] } { 2\sin ( \varphi ) \cos [ ( 1 - 2\theta ( \tau ) ) \varphi ] } \notag \\ & = & \frac{a(\tau ) } { 2}\bigl[1-\cot ( \varphi ) \tan [ ( 1 - 2\theta ( \tau ) ) \varphi ] \bigr ] . \label{x2}\end{aligned}\]]note that this interpolation also satisfies the symmetry ( [ symmetry ] ) . it is clear that one can consider other auxiliary or control conditions over the hamiltonian different from eq . ( [ bc-3 ] ) . in the following , we address three special cases : ( i ) the hamiltonian interpolation is linear in time , namely , @xmath379 ; ( ii ) hamiltonians with constant norm ; specifically , @xmath380 ; and , ( iii ) hamiltonians with constant gap ; specifically , @xmath381 . if we choose @xmath382,\]]and assume @xmath383 , from eqs . ( [ x1 ] ) and ( [ x2 ] ) we obtain the simple linear interpolation @xmath384 let us assume @xmath385 . the constraint @xmath386 implies that @xmath387 [ eq . ( [ hnorm ] ) ] , or equivalently @xmath388 [ eq . ( [ bc-3 ] ) ] . hence , in this case from eqs . ( [ x1 ] ) and ( [ x2 ] ) we obtain @xmath389 , \label{rcpath1 } \\ x_2\bigl(\theta(\tau ) \bigr ) = \frac{1}{2}-\frac{\sqrt{r}}{2\sqrt{1-r}}\tan\bigl[\bigl(1 - 2\theta(\tau)\bigr)\varphi\bigr ] . \label{rcpath2}\end{aligned}\ ] ] this interpolation is a generalization of the interpolation obtained in refs . @xcite by using a local adiabatic condition . rather than assuming condition ( [ bc-3 ] ) , here we consider the case in which the gap is constant , e.g. , @xmath390 . hence combining the following relation [ eq . ( [ gap ] ) ] : @xmath391^{2}+4rx_{1}\bigl(\theta ( \tau ) \bigr)x_{2}\bigl(\theta ( \tau ) \bigr)=1,\]]and eq . ( [ relat ] ) yields @xmath392 , \label{constgapx1 } \\ x_{2}\bigl(\theta ( \tau ) \bigr ) & = & \frac{1}{2\sqrt{r(1-r)}}\sin \bigl[2\theta ( \tau ) \varphi \bigr ] . \label{constgapx2}\end{aligned}\ ] ] having given a general recipe for adiabatic interpolations and having obtained three particular interpolations , we proceed to compute the adiabatic error for each of these interpolations . our main interest here is to analyze how the run time scales with system size for each of these three interpolations . we shall also contrast the result for each case with the estimate obtained from the traditional adiabatic theorem . as a result , we will demonstrate that the traditional adiabatic condition is not always reliable for estimation of the minimum run time ( given an error threshold ) , whereas the time we obtain from the exponential regime of the adiabatic error is in fact accurate . a remarkable feature of this result is that the estimated times ( in an appropriate sense ) need not be very large . for the linear interpolation @xmath393 , obtained by numerically solving the corresponding schrdinger equation , for @xmath394 . the dashed lines represent the exponential fits @xmath395 [ eq . ( [ b2 ] ) ] and the polynomial fit @xmath396 [ eq . ( [ b1 ] ) ] , from left to right , respectively.,width=321 ] in this case , @xmath397 , so that we have @xmath398hence the traditional adiabatic condition ( [ cond ] ) implies that , for @xmath262 , we should have @xmath399or equivalently from eq . ( [ runt ] ) , @xmath400that is , that the adiabatic quantum search with a linear interpolation hamiltonian requires a run time @xmath401 [ recall @xmath402 , hence performing no better than a classical search algorithm @xcite . note , however , that if we truncate the @xmath240 expansion ( er - simp ) after the first term , in the @xmath214 limit we obtain @xmath403i.e . , @xmath404 , which is not the right estimate . this illustrates the caveat we discussed in sec . [ aderror ] . now we employ the results we developed in sec . [ aderror ] for estimating the exponential regime of @xmath82 [ eq . ( [ xx ] ) ] . it is obvious that for this case there exist no point at which both @xmath405 and @xmath406 vanish . thus , the pole @xmath287 ( where the gap vanishes ) is obtained simply by inverting eq . ( [ poles ] ) , i.e. , @xmath407the integral over the gap can be evaluated explicitly as @xmath408 = \frac{\pi r}{8\sqrt{1-r}}.\]]for the integral ( [ residue ] ) , we choose the contour @xmath295 to be a rectangle composed of : ( i ) @xmath409 the real line @xmath410 $ ] , ( ii ) @xmath411 , the line connecting @xmath412 to @xmath413 , ( iii ) @xmath414 , the line connecting @xmath413 to @xmath415 , and ( iv ) @xmath416 , the line connecting @xmath413 to @xmath417 . from the form of @xmath323 , we can easily see that @xmath418 [ because @xmath419 and @xmath420 [ because @xmath421 . this means that for @xmath214 and in the regime in which @xmath422 gives the dominant contribution to @xmath82 , the error exhibits an exponentially decreasing behavior as @xmath423 in this regime , for @xmath262 it is sufficient to have @xmath424whereby we can estimate the following run time : @xmath425this result agrees perfectly with the expected @xmath401 scaling , with a _ logarithmic _ ( rather than inverse ) dependence on the error @xmath83 . figure [ fig.linear ] depicts the adiabatic error vs time , calculated by solving the corresponding schrdinger equation numerically . it illustrates the exponential and polynomial regimes . _ here we could not use eq . ( [ st - ph ] ) because @xmath323 has poles @xmath426 . here for simplicity , we assume @xmath427 . from eqs . ( rcpath1 ) and ( [ rcpath2 ] ) , the gap ( [ gap ] ) is @xmath428 , \label{g - c - n}\]]with the minimum value @xmath429 at @xmath430 , and eqs . ( [ hnorm ] ) and ( [ hdotnorm ] ) yield @xmath431,\end{aligned}\]]the latter with the maximum value @xmath432 at @xmath433 . thus , according to the traditional adiabatic condition ( [ cond ] ) , in order to have @xmath262 , we should require @xmath434and in turn , @xmath435which is larger than the expected @xmath436 grover - like scaling rolandcerf , qab . for the constant - norm interpolation [ eqs . ( [ rcpath1 ] ) and ( rcpath2 ) ] , obtained by numerically solving the corresponding schrdinger equation , for @xmath437 . the dashed lines represent the exponential fit @xmath438 [ eq . ( [ dellin ] ) ] and the polynomial fit @xmath439 [ eq . ( [ c1])].,width=321 ] on the other hand , truncation of the corresponding @xmath240 expansion ( [ er - simp ] ) results in @xmath440 in which we used @xmath441 [ eqs . ( [ rcpath1 ] ) and ( [ rcpath2 ] ) ] and @xmath442 . hence , @xmath443which of course is incorrect . now we show that a careful treatment of @xmath282 , as in sec . [ aderror ] , results in an exponential adiabatic error , and gives the correct scaling for the run time . here , we note that @xmath444 [ eq . ( [ h_a ] ) ] , from which @xmath445 ^{-i\frac{jt\sqrt{r}}{2\varphi } } \mathrm{d}\tau ^{\prime } , \notag \\ & & \end{aligned}\]]where we used the identity @xmath446 . for @xmath214 [ @xmath447 , this gives rise to @xmath448^{-i\frac{jt\sqrt{r}}{2\varphi } } \mathrm{d}\tau ^{\prime } \right\vert \notag \label{integralbeta } \\ & \approx & \frac{\pi } { 2}\int_{0}^{1}[\tan ( \pi \tau ^{\prime } /2)]^{-i\frac{jt\sqrt{r}}{2\varphi } } \mathrm{d}\tau ^{\prime } + o(\sqrt{r } ) . \notag \\ & & \end{aligned}\]]further simplification can be obtained by using the identity @xmath449^{-i\alpha } \mathrm{d}\tau ^{\prime } = \mathrm{sech}(\pi \alpha /2)$ ] ( for @xmath450 ) and the inequality @xmath451 ( for @xmath452 ) ; hence , @xmath453this is the corresponding _ exponential _ behavior for the constant - norm hamiltonian interpolation . it implies that for @xmath66 it is sufficient to have @xmath454or equivalently , @xmath455which is the expected grover - like @xmath436 scaling rolandcerf , qab , but with a logarithmic dependence on the error . although eq . ( [ residue ] ) may not be applicable to the case of constant - norm interpolation ( for @xmath323 does not have any singularity ) , we can apply eq . ( [ st - ph ] ) instead . in fact , eq . ( g - c - n ) implies that @xmath456hence , @xmath408 = \frac{\pi \sqrt{r}}{4\varphi } \overset{r\ll 1}{\approx } \frac{\sqrt{r}}{2},\]]and in turn [ eq . ( [ st - ph ] ) ] @xmath457with some non - exponentially decreasing @xmath458 . this results in a scaling similar to eq . ( [ mmm ] ) . in this case [ @xmath381 ] , from eqs . ( [ hnorm ] ) , ( [ hdotnorm ] ) , ( [ constgapx1 ] ) , and ( [ constgapx2 ] ) we obtain @xmath460|/\sqrt{r } , \\ & \vert \dot{h}(\tau ) \vert /j=&\varphi |\sin [ ( 1 - 2\tau ) \varphi ] |/\sqrt{r}. \label{temp1}\end{aligned}\]]hence , according to the traditional adiabatic condition ( [ cond ] ) , in order to have @xmath262 , it is sufficient to have @xmath461whereby @xmath462which is quadratically larger than the grover - like @xmath436 scaling . for the constant - gap interpolation [ eqs . ( [ constgapx1 ] ) and ( constgapx2 ) ] . here , @xmath463 ( blue and green , respectively ) , and the dashed lines ( red ) show the @xmath464 envelopes . the points @xmath465 , where @xmath466 , are given by eq . ( [ t_k]).,width=321 ] on the other hand , noting that @xmath441 [ eqs . ( [ constgapx1 ] ) and ( [ constgapx2 ] ) ] , the truncation of the corresponding @xmath240 expansion after the first term yields @xmath467or equivalently @xmath468clearly , eqs . ( [ rr1 ] ) and ( [ rr2 ] ) are not in agreement . it is interesting to note that here one can in fact solve the schrdinger equation exactly . the operator @xmath469 [ eq . ( [ o - hat ] ) ] satisfies the following equation : @xmath470 in which @xmath471}{\sqrt{r}}\openone+\sigma _ { z}\right ) -\varphi \sigma _ { y}. \notag \\ & & \end{aligned}\]]since @xmath472 is time - independent , integration of eq . ( [ temp3 ] ) is straightforward : @xmath473 . \notag \\ & & \end{aligned}\]]thus , from eq . ( [ ad - error ] ) the adiabatic error is exactly @xmath474 figure [ fig.constg ] depicts @xmath82 for two different values of @xmath257 . note that for large evolution times @xmath475 , we obtain @xmath476this implies that , in the @xmath214 limit , in order for @xmath262 it is sufficient to have @xmath477which is the grover - like scaling @xmath436 . notice that in this limit the adiabatic error behaves inverse - linearly , @xmath478 , which is in perfect agreement with eq . ( [ temp4 ] ) . in addition , we observe that there exist @xmath0s less than the above limit in in which the adiabatic error can vanish ( hence instantaneous full adiabaticity ) . according to eq . ( [ c - g - t ] ) , we have @xmath479 where @xmath480for @xmath481 . the existence of such @xmath465s is in agreement with ref . figure [ fig.constg ] shows @xmath82 for two different values of @xmath257 . _ as is evident here the adiabatic error does not show any exponential behavior . in fact , neither of the methods we discussed in subsection [ expl ] is applicable . here , we discuss the behavior of the exponential @xmath482}$ ] for the general interpolation we derived in subsection [ path - g ] . our analysis is based on a formal power series expansion of @xmath483 and @xmath484 in terms of @xmath257recall that we are interested in the regime @xmath485 . we further assume that @xmath352 does not depend explicitly on @xmath257 . consider the following formal expansions : @xmath486 in which @xmath487 and @xmath488 are some nonnegative numbers ( to be determined later ) , @xmath489 , and @xmath490 . we notice that the linear and constant - gap interpolations ( subsections lin - sec and [ c - gap ] ) do not admit expansions as in eqs . ( [ e1 ] ) and ( [ e2 ] ) . equation ( [ gap ] ) hence yields @xmath491similarly , inserting eqs . ( [ e1 ] ) and ( [ e2 ] ) into @xmath363 gives @xmath492on the other hand , eq . ( [ relat ] ) yields @xmath493the symmetry @xmath494 [ eq . ( [ symmetry ] ) ] requires that @xmath495which in turn implies @xmath496 . comparing the terms with the same powers of @xmath257 in eqs . ( [ x21 - 1 ] ) and ( [ x21 - 2 ] ) , we conclude that @xmath497and @xmath498after inserting the above relations back into eq . ( [ delta - r ] ) and using eq . ( [ fg - eq ] ) , we obtain @xmath499now we assume that @xmath500 everywhere , or if there exist points at which @xmath501 vanishes , their contribution to the integral @xmath502 $ ] is not substantial . note that the previous condition is in fact a condition on the norm of the hamiltonian because from eq . ( [ hnorm ] ) @xmath503therefore , provided that for large times the adiabatic error asymptotically behaves as in eq . ( [ residue ] ) or ( [ st - ph ] ) , we obtain @xmath504\bigr|},\]]where @xmath458 is a non - exponential function of @xmath0 ( which may also depend weakly on @xmath257 ) . hence , in the @xmath214 limit for @xmath66 , it is sufficient to have @xmath505\bigr|.\]]this in turn implies the following scaling for the run time : @xmath506\bigr|\bigr ) . \notag \label{generalt } \\ & & \end{aligned}\]]the overall @xmath257-dependence here comes from @xmath507 and @xmath508 ; e.g. , we recover the grover - like @xmath436 scaling if @xmath287 does not depend on @xmath257 . this analysis then highlights in a fairly general way the interplay between @xmath257 , @xmath0 , and @xmath82 in the quantum search algorithm . for most applications it is desirable to make the adiabatic error as small as possible . we have seen that @xmath82 contains exponential terms , suppressed by the polynomially - decaying terms . therefore , it is useful to somehow extend the dominance of the exponential term by reducing the contribution of the polynomial term , e.g. , by prolonging the dominance of the exponential regime or by enforcing higher order polynomial behavior . in the following we shall discuss a control strategy for reducing the adiabatic error by manipulation of the boundary conditions ( see , e.g. , refs . @xcite ) . equation ( [ er - simp ] ) demonstrates explicitly how the adiabatic error depends on the boundary conditions , up to @xmath509 . interestingly , if we choose @xmath510 , the prefactor of the @xmath2 vanishes , whence @xmath511 . in a similar fashion , one can see that by enforcing suitable ( extra ) boundary conditions on the interpolation @xmath512 the prefactor of the @xmath239 or even higher order terms can be made zero . this implies that by manipulating the boundary conditions of the interpolation , one may achieve smaller adiabatic errors . this observation is a manifestation of the following general theorem : if the hamiltonian @xmath65 is sufficiently differentiable , forcing _ all _ time derivatives of the hamiltonian up to some order @xmath361 to vanish at the boundaries,@xmath513is sufficient for @xmath514 sancho , hagedornjoye:02,jrs , lrh . it is interesting to note that the very same condition together with the assumption of the _ analyticity _ of @xmath65 in a small strip around the real axis in the _ complex _ @xmath48-plane give rise to @xmath515 , where @xmath516 [ up to an @xmath517 prefactor ] @xcite . we obtained above are in fact weaker than requiring @xmath518 [ sufficient for @xmath519 . ] this is a remarkable result , in that it guarantees that with sufficient smoothness at two points one can substantially suppress the adiabatic error . this is a fairly low price to pay for higher accuracy . in particular , in experimental realizations , manipulating hamiltonian interpolations only at the beginning and the end ( as opposed to from beginning to end ) , may offer a less demanding control strategy than one seeking to control the dynamics instantaneously along the entire evolution . now we show that in the framework we developed earlier , enforcing the required smoothness properties can be achieved by choosing an appropriate @xmath352 function . we recall that this function was fairly arbitrary ; we required that it be a monotonically increasing differentiable function ( @xmath356 ) satisfying the boundary conditions @xmath357 and @xmath358 . we require further that @xmath520 ( for a given @xmath481 ) have the following property : @xmath521namely , the first @xmath361 derivatives should vanish at the boundaries . this property is sufficient for fulfilling eq . ( [ hdz ] ) because @xmath522 . an example of such @xmath523 is the regularized incomplete beta function , @xmath524 in which @xmath525 , with @xmath526 , and @xmath527 @xcite . to demonstrate explicitly how the conditions ( [ hdz ] ) affect the adiabatic error , we employ the method developed in refs . hagedornjoye:02,lrh for approximating the coefficients in the @xmath240 expansion of @xmath82 . one can construct an approximate ( unnormalized ) ansatz for the solution to the schrdinger equation ( [ eqscalt ] ) in the powers of @xmath2 as follows : @xmath528,\end{aligned}\]]with the error @xmath529here , @xmath530 and @xmath531 are given as follows @xcite : @xmath532it is evident that @xmath533 . provided that all @xmath534 derivatives of @xmath65 vanish at the boundaries , eq . ( [ hdz ] ) , all the terms except @xmath75 and @xmath535 will vanish @xcite ( see also ref . @xcite , where with a condition different than eq . ( [ fs ] ) all the terms except @xmath536 vanish ) . let us define @xmath537since @xmath538 ( for normalized @xmath539 and @xmath540 ) is a distance [ eq . ( [ delta - d ] ) ] , from the triangle inequality @xmath541 we have : @xmath542notice that @xmath543 can also be written as follows : @xmath544^{2}[1+o(1/t)]^{2 } } } \notag \\ & \approx & \sqrt{1-(1 + 2\mathrm{re}[\langle \widehat{\psi } _ { n}(1)|\widehat{r}_{n}(1)\rangle ] ) } \notag \\ & = & o\bigl(\frac{1}{t^{n+2}}\bigr ) , \label{delta1-t-2}\end{aligned}\]]where @xmath545 on the other hand , a straightforward calculation ( supplemented with induction ) shows that @xmath546this bound holds true for any interpolating paths @xmath512 for which @xmath547 . as can be seen , @xmath543 is negligible in comparison to @xmath548 , hence we obtain @xmath549this result is a generalization of eq . ( [ er - simp ] ) . for example , in the case of the constant - gap interpolation ( subsection [ c - gap ] ) , this error reduces to @xmath550 in comparison with eq . ( [ cg-1 ] ) . figure [ fig : errorbeta ] depicts @xmath82 for the constant - norm interpolation [ eqs . ( rcpath1 ) and ( [ rcpath2 ] ) ] . it can be seen that by increasing @xmath361 the exponential regime dominates longer , while the polynomial regime is pushed farther away to the region of large evolution times . however , this improvement comes at a price . the rate of exponentiality decreases with increasing @xmath361 ; that is , if @xmath551 and @xmath66 for an @xmath83 in the exponential regime for the larger @xmath361 , then @xmath552 [ see the inset of fig . [ fig : errorbeta ] ] . in other words , for some values of @xmath83 , increasing @xmath361 might give rise to an increased run time . of course , if @xmath83 is such that the polynomial regimes dominate for both values of @xmath553 and @xmath554 , the interpolation with the larger @xmath361 ( @xmath554 ) results in a smaller run time . obtained by numerically solving the schrdinger equation corresponding to the constant - norm interpolation with @xmath555 the regularized beta function [ eq . ( [ beta ] ) ] , for @xmath556 ( here @xmath557 ) . the dashed ( red ) lines represent @xmath558 and @xmath559 [ eq . ( [ er])].,width=321 ] nevertheless , @xmath561 helps remove the polynomial terms arbitrarily , hence extending the exponential regime farther . additionally , the exponent of the exponential term is controllable through varying the parameters @xmath86 and @xmath568 . in this case , asymptotic evaluation of the integral ( [ inttran ] ) with the stationary phase method results in a faster - than - polynomial convergence to zero . for example , with @xmath569 and @xmath570 for the constant - norm interpolation , we can approach an exponential convergence similarly to the case with @xmath571 . the achievement of an exponentially small @xmath82 for such functions may be understood as an instance of rigorously derived exponentially small errors for a general class of functions called the gevrey class @xcite . it is evident that manipulating the @xmath352 function may also affect the exponent @xmath572 $ ] . this can be observed in fig . [ fig : errorbeta ] ( the inset ) through the change of the slope of the exponential lines in the log plot . a side consequence of using a @xmath573 with the desired boundary conditions ( [ theta - cond ] ) is that increasing @xmath361 may adversely increase the value of @xmath0 for which @xmath262 ( in the exponential regime ) . to quantify how choosing a @xmath523 for the hamiltonian interpolation affects the performance of the algorithm , we propose the following measure : @xmath574namely , the average adiabatic error up to time @xmath0 , for a given @xmath361 . a larger average error may be interpreted as less efficient performance . the above problem with the effect of @xmath361 on exponentiality may be partially alleviated in some cases . recall that in subsection [ path - g ] we found a fairly general interpolation , which resulted in the parametrization @xmath575 [ eq . ( [ relat])]it was later that we added further conditions so as to find @xmath576 and @xmath577 separately . rather than assuming the condition ( [ bc-3 ] ) , let us impose @xmath578for some @xmath579 . we should be mindful of the fact that , from eq . ( [ hnorm - f ] ) , modifying @xmath501 leads to a modification of the norm of the hamiltonian , or equivalently the maximum energy of the system ; this is a cost , which should be taken care of in the correct estimation of @xmath89 [ eq . ( [ runt ] ) ] . additionally , from eq . ( [ gg-0 ] ) it is seen that since we assumed @xmath580(up to the leading order in @xmath257 ) the gap @xmath162 vanishes where @xmath581 ; i.e. , @xmath582 . in this case , the exponent @xmath583 $ ] [ eq . ( [ xx ] ) ] becomes @xmath584 = \nonumber\\ \mathrm{im}\left[\int_0^{{z_0}_+}\csc\bigl(\pi\theta(\tau')\bigr)~\mathrm{d}\tau'\right ] + \frac{\zeta}{2}. \label{mm}\end{aligned}\]]therefore , by appropriately choosing @xmath585subject to the condition @xmath586we can tune the exponent of the exponential term in @xmath82 . this in turn gives us control over the run time in the exponential regime . furthermore , as we argued earlier , replacing @xmath587 causes the polynomial terms of @xmath82 to be @xmath588 . thus , we are now in possession of two control parameters @xmath361 [ more precisely @xmath523 ] and @xmath585 with which we can manipulate how the adiabatic error behaves in either of the exponential and polynomial regimes . this type of control may have applications in experiments in which adiabaticity plays a role . as an example , let @xmath523 be the regularized beta function [ eq . ( [ beta ] ) ] . for @xmath590 and after employing stirling s approximation for the factorial function ( @xmath591 @xcite ) , we obtain @xmath592whereby @xmath593as a result , for example , in order to keep the maximum energy constant , while having the advantages of @xmath594 , we should choose @xmath595 . adiabatic evolution is characterized by a tradeoff between the total time taken and the error in the final state reached , relative to the desired adiabatic state . motivated by a desire to understand and optimize this tradeoff , in this work we performed a detailed analysis of the adiabatic error for the case of an adiabatic quantum search algorithm . rather than using the traditional adiabatic condition , with its associated pitfalls , we chose to calculate the adiabatic error directly by solving the schrdinger equation . this enabled us to derive an exact relation for the adiabatic error . building on this exact result , we employed a formal polynomial series expansion in @xmath2 for calculating the error term by term . this also allowed us to bound the adiabatic error . we showed that the polynomial expansion should be truncated carefully if we aim to obtain a reliable estimate for the run time of the algorithm . we demonstrated that employing a different technique based on complex analysis reveals , in fairly general situations , a regime of exponential time - dependent decay of the adiabatic error , preceding a polynomial regime . the latter has been shown to be a general feature of adiabatic hamiltonians , whereas the existence of an exponential precursor is not always guaranteed . we showed how , in case these two regimes both exist , one can estimate the crossover region i.e . , the time at which a transition between the two regimes takes place . equipped with this , we provided an estimate for the minimum time required for the algorithm to achieve a given accuracy threshold . discerning the exponential regime enabled us to give an improved total time estimate , circumventing the overestimate arising from the error bound using only the polynomial expansion . indeed , the total time estimated from the exponential regime always gave the correct scaling with system size ( the well - known quadratic speedup over classical search ) , while the estimate resulting from the polynomial regime resulted in unreliable and even erroneous results . we also obtained a specific class of hamiltonian interpolations for the search problem . to this end , we employed a recently developed theory , based on the geometry of adiabatic evolutions , for obtaining suitable adiabatic interpolations . this theory separates the adiabatic error into geometric and non - geometric parts , and minimizes the former . we discussed three special cases of the resulting class of interpolations in detail : ( i ) linear , ( ii ) constant - norm , and ( iii ) constant - gap interpolations . finding strategies for minimizing the required total time as a function of a given resource ( system size , for example ) is a desirable goal for many applications , and is also of fundamental importance for the control of quantum systems . we demonstrated explicitly how by imposing fairly general controllability assumptions , which should be experimentally straightforward to realize in certain scenarios , one can achieve a significant reduction of the adiabatic error , and hence improve algorithmic performance . the method we used relied on a polynomial expansion of the adiabatic error , and resulted in the suppression of polynomial terms in @xmath2 by requiring smoothness for interpolations at the initial and final times . it is evident that controlling the interpolation in this manner , at only two points , has a substantial advantage over instantaneous control of the hamiltonian along the entire evolution . however , we demonstrated that there is an extra price to pay for this error reduction : the exponential regime ( if exists ) is extended , but with a slower rate of decay . this , in turn , may result in an overestimation of the run time of the algorithm for some values of the error threshold . we proposed a measure for quantifying the performance of an adiabatic interpolation with various controllability properties . in some cases , we also suggested a remedy for the above problem . this fix necessitated further control over the hamiltonian interpolation , directly related to the amount of accessible energy in the system . the interplay between the degree of required control over hamiltonian interpolations and the run time needed for achieving a given accuracy was thus clearly exhibited . although we focused on the quantum search problem , our methods and most of our results are applicable ( perhaps with minor modifications ) to a wider class of problems as discussed in the text . since a principal goal in adiabatic quantum algorithms , adiabatic quantum transport , quantum annealing , and other applications of the adiabatic theorem , is the design of algorithms with favorable performance - resource tradeoff , we hope that our results will be of use in related physical applications .

active galactic nuclei are powerful sources of radiation in a wide spectral range : from @xmath1 rays to radio waves . they derive their extraordinary luminosities ( sometimes more than @xmath2 times higher than luminosities of `` ordinary '' galaxies ) from energy release by matter accreting towards , and falling into , a central supermassive black hole ( see e.g. * ? ? ? according to the unification model of agn @xcite , their central engine consists of a supermassive black hole ( smbh ) with mass ranging from @xmath3 to @xmath4 solar masses @xmath5 @xcite , which is surrounded by an accretion disk that radiates in the x - ray band @xcite . the vast majority of the x - ray sources in the universe are agn . the integral emission of agn reflects the history of accretion onto smbh over cosmic time . emission lines are usually seen in the x - ray spectra of agn . the broad emission fe k@xmath0 spectral line ( 6.46.9 kev , depending on ionization state ) with asymmetric profile ( narrow bright blue peak and a wide faint red wing ) has been observed in a number of type 1 agn ( see e.g. * ? ? ? early results from the asca ( _ advanced satellite for cosmology and astrophysics _ ) era suggested that broad relativistic lines might be common in type 1 agn . however , surprisingly , they have been detected in significant number and their features described in only a small fraction of those sources @xcite . the first and the best studied one is mcg-6 - 30 - 15 @xcite . in some cases the line width corresponds to one third of speed of light , indicating that its emitters rotate with relativistic velocities . therefore , the line is probably produced in a very compact region near the central black hole of agn and can provide us some essential information about the plasma conditions and the space - time geometry in vicinity of the black hole . consequently , if the line is emitted close enough to the smbh , it shows a broad relativistic profile affected by smbh spin @xcite and gravitational redshift ( as well as other general relativistic effects ) @xcite . several studies have been performed over samples of local agn ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * etc ) , as well as from distant quasars @xcite in order to characterize the fe k@xmath0 emission . @xcite performed a spectral analysis of a sample of 26 type 1 to 1.9 seyferts galaxies ( @xmath6 ) observed by _ xmm - newton_. they found that a relativistic line is significantly detected in a half of their sample ( 54@xmath710 percent ) with a mean equivalent width ( ew ) of @xmath8 80 ev , but around 30% of selected agn showed a relativistic broad line that can be explained by the emission of an accretion disk . accretion disks could have different forms , dimensions , and emission , depending on the type of central black hole , whether it is rotating ( kerr metric ) or non - rotating ( schwarzschild metric ) . they represent an efficient mechanism for extracting gravitational potential energy and converting it into radiation , giving us the most probable explanation for the main characteristics of agn ( high luminosity , compactness , jet formation , rapid time variations in radiation and the profile of the fe k@xmath0 spectral line ) . here we discuss the x - ray radiation from relativistic accretion disks around supermassive black holes , supposed to exist in the centers of agn . especially , we discuss the fe k@xmath0 line profile which originates from the accretion disk . moreover , we also present some effects which can disturb the fe k@xmath0 profile , such as : perturbations in the disk emissivity , absorbtion by warm absorbers and gravitational microlensing . the aim of this chapter is to present a short overview of results of some recent investigations in this field , and it is divided into the following six sections : _ active galactic nuclei as hosts of supermassive black holes _ - where the main features , classification and unified model of active galaxies are briefly presented , _ space - time geometry in vicinity of supermassive black holes _ - where the basic definitions of schwarzschild and kerr metrics are given , _ accretion disk around a supermassive black hole _ - where we explain the standard model of an accretion disk , including its emission , accretion rate , luminosity , structure and spectral distribution , _ supermassive black holes and x - ray emission _ - where the focus is on the modeling of the observed x - ray radiation from relativistic accretion disk around a supermassive black hole of agn in both the fe k@xmath0 spectral line and x - ray continuum , _ variability of x - ray emission around supermassive black hole _ - where we present the main causes of rapid and irregular variability of the x - ray emission which can be due to disk instability , reflecting in perturbations of its emissivity , or it could be caused by some external effects , such as gravitational microlensing and absorption by x - ray absorbers , and finally _ conclusion _ - where the most important results from previous sections are pointed out and their brief summary is given . finally , we should note that a huge number of papers devoted to investigation of the x - ray emission of agn was published in the last two decades , so it was not possible to mention all of them here , but more details and references can be found in review papers such as e.g. @xcite , etc . active galaxies differ from the so called `` normal '' galaxies in the amount of energy emitted from their nuclei . the term _ active galactic nuclei _ ( agn ) refers to the energetic phenomena in the central region of a galaxy which can not be solely or directly produced by stars . the nebular - like emission spectra of ngc 1068 , ngc 4051 and ngc 4151 , with h@xmath0 , [ o ii ] @xmath93727 , [ ne iii ] @xmath93869 and [ o iii ] @xmath10 4363,4959,5007 were observed by @xcite , but the first classification of these objects was made by @xcite . he recognized these objects as a class of galaxies with strong and high - excitation optical emission lines localized in the nuclei . active galaxies have been the subject of an intensive astrophysical investigations for the last 3 4 decades . as the most luminous objects with one of the most powerful energy release rates and with the most compact dimensions , agn are of the great interest in modern astrophysics . agn ( i.e. quasars ) are the brightest objects in the universe and thus , these objects are important for the studies of the early universe and cosmology in general . for instance , studies of their luminosity functions or quasar host galaxies are crucial for understanding the formation and evolution of galaxies in general . the class of active galaxies contains many different objects , such as quasars , gigantic elliptic radio galaxies , luminous spiral seyfert galaxies , blazars , etc . agn are the most luminous objects in the universe and they have luminosities in the range from @xmath11 to @xmath12 erg s@xmath13 , that can be @xmath14 times greater than in the case of a typical galaxy . apart from their great luminosities , agn are emitting in the broad band of electromagnetic spectrum . the observed emission of agn is in the continuum and in lines , from the @xmath1 and x domains to the far infrared and radio bands ( see e.g. * ? ? ? one of the characteristics of agn is violent and fast variability observed in different parts of electromagnetic spectra . fast changes in the agn brightness are observed sometimes just for a couple of days , e.g. in the case of galaxy ngc 4151 @xcite . this leads to the conclusion that the emission regions in the agn have small dimensions ( from only couple of light days up to couple of light months ) . also , the brightness of agn can be so high that the brightness of the host galaxy can be neglected in the total brightness ( for instance in the case of quasars ) . consequently , they are compact regions which could not be directly observed in the most bands of electromagnetic spectrum . therefore , agn and their physical / kinematical properties are indirectly studied through the analysis of their spectra . in contrast to the spectral energy distribution ( sed ) of an ordinary galaxy ( that represents a sum of stellar spectra , thus the most of its luminosity comes within no more than one decade of frequency ) , agn emit across the broad range of frequencies ( of order @xmath3 hz ) . the agn continuum in a number of high - luminosity sources peaks in the ultraviolet , i.e. in the so called big blue bump ( bbb ) @xcite . the bbb is attributed to the thermal emission from an optically thick region of the accretion disk @xcite , or from optically thin regions , i.e. free - free emission @xcite . the observed sed in agn is remarkably different from the thermal ( black - body ) spectrum of a star or a regular galaxy @xcite . this featureless continuum observed in agn characterizes the main source of energy input and suggests that the mechanisms that produce it are common to all types of agn . the sed can be approximated with the power law function @xmath15 where @xmath0 is the spectral index @xcite . the observed continuum of a typical seyfert 1 galaxy is very strong compared with seyfert 2 galaxies ; as a result seyfert 1 galaxies appear to be more luminous than seyfert 2-s . some agn have the sed almost flat from the infrared to x - ray part of the spectrum , so that the spectral index is @xmath16 , although it is usually steeper . in the radio band , the brightness of active galaxies is usually an order of magnitude higher than of normal galaxies . but still , even if it is so intensive , the radio luminosity is never higher more than 1% of the bolometric luminosity . another property of agn is strong x - ray flux , with the x - ray continuum produced by lower energy photons which are compton scattered to higher energies by relativistic electrons @xcite . the fraction of the power emitted in the x - ray emission is three or four times larger in agn than in normal galaxies @xcite , therefore the x - emission can indicate the presence of agn . today , it is widely accepted that the mechanism which powers agn is accretion onto supermassive black hole ( smbh ) , i.e. the greatest part of the continuum emission comes from an accretion disk and its corona . the mass of the supermassive black hole is estimated to be from @xmath3 to @xmath17 @xcite . thus , in the center of an agn there is a supermassive black hole with an accretion disk surrounded by gas and dust in the form of a torus . there are also other emitting regions present in the vicinity of the supermassive black hole , such as ionized gas clouds that produce intensive emission lines or jets of matter that are mostly visible in the radio band , but sometimes they can also be seen in the optical band . the kinematics of these regions is very complex as well as the physical processes that produce such specific spectra of these objects . in the central part , there are mainly three emitting line regions : fe k@xmath0 ( in the x - ray range ) , broad and narrow line emitting regions ( blr and nlr , respectively ; in optical / uv spectra ) . in the optical and uv band , the total flux of emission ( and occasionally absorbtion ) lines contributes from several percent to tens of percent of the continuum flux . the existence of the broad and narrow emission lines comes from the fact that an agn contains two separated emission line regions with different kinematics , density , ionization , optical thickness and radiation transfer . the basic characteristics of the blr and nlr can be described as follows : - blr is a compact region , with dimensions from only couple of light days up to couple of light weeks @xcite , that is located in the vicinity of the black hole ( @xmath181 pc ) . the structure of this region is very complex and most likely it consists from at least two separated subregions ( see e.g.@xcite , etc ) . the ionized gas in this region is of relatively high density ( @xmath19 ) and temperature @xmath20 k. emitters are moving with high velocities , up to 10000 km / s . in this region the broad emission lines ( bels ) from allowed transitions are formed . nlr extends even up to 1 kpc far from the central source @xcite . the density in this region is significantly smaller ( @xmath21 ) than in the blr , while the temperature is on the same order of magnitude as in the blr . from this region the narrow emission lines ( nels ) arise , indicating a random motion of the emitting gas smaller than 1000 km@xmath22 . nels are often coming from the forbidden transitions . the physical properties of this region are much closer to the properties of emission nebulae than to the blr . according to the different characteristics mentioned above , there are several classes of agn . the first group of stellar systems with active nuclei consists of seyfert galaxies . these are spiral galaxies with very bright nuclei which spectra have strong emission lines of neutral and multi - ionized emitters . the presence of highly ionized emission lines indicates the existence of a non - stellar ionization continuum . the presence or absence of bels ( i.e. blr ) has been historically used to separate seyfert galaxies into two classes . seyfert 1 galaxies have broad permitted ( h i , he i and he ii ) and narrow permitted and forbidden lines ( such as [ o iii ] @xmath95007 ) , while seyfert 2 galaxies have only narrow permitted and forbidden lines @xcite . this simple seyfert classification scheme can be further sub - divided according to specific spectral properties @xcite , as e.g. seyfert galaxies with intermediate balmer profiles are classified as seyfert 1.5-s with apparent narrow balmer line components superimposed on broad wings . many low luminosity galaxies have a nucleus that resembles the seyfert 2 nucleus , but with the spectrum that shows the forbidden lines from lower ionization states . therefore , these galaxies are called low - ionization nuclear emission - line regions , i.e liners . active galaxies which emit intensively in the radio band are so called radio galaxies ( rg ) . radio galaxies are dominantly elliptic galaxies that are divided into two morphologically different types : fr i and fr ii . the classification is done according to the relative position of the surface brightness maximum ( called the hot spot ) in the radio lobes . for example , in the fr i radio galaxy the distance between the hot spot in the radio lobes is less than half of the maximum diameter of the radio source , while in the fr ii type this distance is larger @xcite . like seyfert galaxies , radio galaxies can be divided into subgroups according to the widths of their emission lines , i.e. narrow line radio galaxies ( nlrg ) that emit narrow emission lines characteristic for seyfert 2 galaxies , and broad line radio galaxies ( blrg ) that emit broad lines as seyfert 1 galaxies . being the most powerful sources in the universe , quasars belong to the group of active galaxies with the absolute magnitude @xmath23 @xcite . these quasi stellar ( quasars ) objects emit intensively in the radio domain and their spectrum shows broad emission lines significantly redshifted . in this group of agn there are also quasi - stellar objects ( qsos ) which have the same observed properties as quasars , except they are weak radio sources . besides , there is a group of qsos with broad absorption lines , so called broad absorption line quasars ( bal qsos ) . finally , the separate class of agn are so called bl lacerte ( lacs ) objects . bl lacs are nuclei of elliptical galaxies that emit highly variable and polarized radiation , and have a non thermal optical continuum and a strong radio emission . with this group of galaxies , we often connect another type of agn , optically violently variable ( ovvs ) quasars . the ovvs , in contrast to bl lacs which have featureless continuum , emit broad optical emission lines characteristic for quasars . both these groups of objects are often referred to as one class called blazars . as it was mentioned above , all types of agn have some common properties . it is widely accepted that central engine of an agn is accretion onto supermassive black hole . first observational evidence in favor of a unified model of agn was spectropolarometric observations of seyfert 2 galaxy ngc 1068 @xcite . @xcite found broad balmer lines and fe ii emission , characteristic of a seyfert 1 spectrum , in the polarized spectrum of ngc 1068 . in the unified model of agn @xcite both types of seyfert galaxies are intrinsically the same . the difference in spectral characteristics appears due to different angle of the visibility of the central regions ( see fig . [ fig23_1 ] ) . a geometrically and optically thick dusty molecular torus - like structure surrounds the central source , as well as the blr ( see fig . therefore the visibility of the nuclear engine depends on viewing angle with respect to the torus . the broad - line polarization in seyfert 2-s is probably due to electron scattering . this picture is consistent with the differences observed between continuum fluxes of the seyfert 1 and seyfert 2 galaxies @xcite . following the unified model , the line - of - sights to seyfert 2 galaxies ( or narrow line objects ) are obstructed by optically thick material corresponding to hydrogen column densities of @xmath24 @xcite . for column densities @xmath25 , photons above a few kev can penetrate the torus creating an un - obstructed view of the nuclear source . in this case the source is called _ compton thin_. for column densities @xmath26 , only high energy x - ray emission ( tens of kev ) can pass through the obscuring material @xcite . but in the case @xmath27 even high energy x - rays , above a few tens of kev , are compton scattered and the nuclear source is completely hidden from our direct view . therefore , by observing near the edge - on torus , one can detect narrow line objects , with weak ( or without ) x - ray emission ; while by observing near the face - on torus , one can detect blazars or ovvs . the broad line objects can be detected if one observes an inclined torus . as it was shown in the previous section , according to unified model of agn , in their heart there is a supermassive black hole with mass up to @xmath17 , surrounded by an accretion disk . a black hole is a region of space - time around some collapsed mass which gravitational field became so powerful that nothing ( including electromagnetic radiation ) could escape from its attraction , after crossing the certain boundary called the event horizon @xcite . all black holes in nature are commonly classified according to their mass as : supermassive black holes ( with masses @xmath28 ) , intermediate - mass black holes ( @xmath29 ) , stellar - mass black holes ( @xmath30 ) , mini and micro black holes ( @xmath31 ) . in this chapter our focus will be on supermassive black holes located in the centers of agn , because they are surrounded by accretion disks which x - ray radiation is the main subject of this discussion . space - time geometry in vicinity of a supermassive black hole depends on its type ( more precisely , on its angular momentum ) , being either non - rotating , where the local space - time geometry is described by schwarzschild metric , or rotating , where the kerr metric determines the geometry of local space - time . the term black hole was introduced in 1969 by the american scientist john wheeler , but the predecessor of the modern idea of black holes emerged in the late eighteenth century . it was independently developed by the british geologist john michell in 1783 and the french mathematician and astronomer pierre - simon laplace in 1796 . this idea was based on newton s theory of gravity and particle theory of light , and is commonly referred to as a dark star . a dark star is , according to michell , a sufficiently massive and compact star which have such a strong gravitational field that the corresponding surface escape velocity equals or exceeds the speed of light . therefore , any light emitted from the surface of such star would be dragged back by the gravitational field and we would not be able to see it , but we could still feel its gravitational attraction @xcite . after the wave theory of light was founded , the idea about a dark star was forgotten since , according to the wave theory , it was not clear that light could be affected by gravity at all . modern theory of black holes was founded in the twentieth century , more precisely in 1915 , shortly after einstein introduced the theory of general relativity , when the german physicist and astronomer karl schwarzschild found a particular exact solution to the einstein field equations , for the limited case of a single spherical non - rotating mass . the implications of the general relativity to the stellar evolution were firstly understood by subrahmanyan chandrasekhar in 1928 @xcite . he calculated a maximum limit for the mass of a star , above which the star would not be able to support itself ( by repulsion of electron degeneracy pressure which arises from pauli exclusion principle ) against its own gravity , after it had used up all its fuel . this limit ( @xmath32 ) is nowadays known as the chandrasekhar limit . if a star s mass is less than this limit , it can eventually stop contracting at a final state known as a `` white dwarf '' @xcite . a similar discovery was made in 1932 by the russian scientist lev davidovich landau . he found that there is another limiting mass ( @xmath33 ) , for which the corresponding final state is much smaller than a white dwarf and which would be supported against its gravity by the exclusion principle repulsion between neutrons and protons , rather than between electrons . such stars are therefore called neutron stars @xcite . all these results led to a surprising conclusion that a star which mass is above the chandrasekhar limit , when it comes to the end of its fuel , could experience a catastrophic gravitational collapse to a point with an infinite density . american scientist robert oppenheimer was the first who explained in 1939 that , according to general relativity , stars above @xmath34 would at the end of their lives collapse into black holes , due to reasons given by chandrasekhar @xcite . with the discovery of qsos in the early 60s , the speculations about how their enormous luminosities could be produced within such a small region occurred . thermonuclear reactions were not enough efficient , and therefore , they were quickly eliminated . hoyle and fowler suggested in 1963 that such luminosities could be produced during the collapses and disintegrations of stelar - type objects with masses of @xmath35 @xmath36 @xcite . soon , it became clear that the energy source was gravitational . for example , wheeler suggested a mechanism in which gravitational singularity at the center of a galaxy converted the falling matter into energy @xcite . zeldovich , novikov and salpeter proposed in 1964 the release of energy through accretion due the growth of a massive object at the center of a galaxy , while lynden - bell in 1969 made an attempt to explain the phenomena observed in qsos and seyfert galaxies in terms of a black hole formalism @xcite . a crucial event for the acceptance of black holes was the discovery of pulsars by jocelyn bell in 1967 , because it was the clear evidence of the existence of neutron stars , and therefore , confirmation of chandrasekhar limit . the first detection of a solar mass black hole came in 1972 , when the mass of the rapidly variable x - ray source cygnus x-1 was proven to be above the maximum allowed for a neutron star @xcite . the first convincing evidence for the existence of supermassive black holes is the discovery of the fe @xmath37 spectral line in x - ray spectrum of seyfert galaxy mgc-6 - 30 - 15 by @xcite . nowadays , the black hole theory is very well established , mainly due to the contributions of numerous authors during the second half of the twentieth century , like roy kerr , werner israel , john wheeler , brandon carter , roger penrose , stephen hawking and many others . schwarzschild metric describes the space - time geometry in spherically symmetric gravitational field around a time - steady non - rotating black hole in vacuum . the solution of the einstein field equations under these conditions was found by karl schwarzschild in 1916 @xcite . if we denote the schwarzschild radius which corresponds to a mass @xmath38 by @xmath39 ( where @xmath40 is the gravitational constant and @xmath41 is the speed of light ) , then the squared element of space - time interval @xmath42 in the case of schwarzschild metric is given by : @xmath43 where @xmath44 is the time coordinate measured by a stationary clock at infinity ( which must be distinguished from the proper time @xmath45 , measured by a clock moving with the particle ) , @xmath46 is the radial coordinate , @xmath47 is the colatitude ( angle from north ) in radians and @xmath48 is the longitude in radians . it is obvious from eq . ( [ eq31_1 ] ) that schwarzschild metric depends only on the mass @xmath38 through the schwarzschild radius @xmath49 . the schwarzschild metric has an infinite space - time curvature ( i.e. a singularity ) at @xmath50 . in the case when @xmath51 this metric is reduced to the minkowski metric of flat space - time : @xmath52 . if the complete mass @xmath38 collapses below its schwarzschild radius @xmath49 , then it will form a spherically symmetric schwarzschild black hole . a schwarzschild black hole has no charge or angular momentum and it can be distinguished from any other schwarzschild black hole only by its mass . to a first approximation , the matter surrounding a black hole may be assumed to rotate in circular keplerian orbits with velocity ( e.g. * ? ? ? * ) : @xmath53 . therefore , taking into account that a maximum allowed rotation velocity is speed of light ( @xmath54 ) , one can obtain so called gravitational radius @xmath55 : @xmath56 which is usually used as a unit for distance around a black hole . roy kerr , a new zealand mathematician , found an exact solution of the einstein field equations of general relativity in the case of gravitational field outside an uncharged rotating black hole . the properties of kerr solution , known as kerr metric , are given in numerous papers and books ( see e.g. * ? ? ? * ; * ? ? ? * ) and therefore , here we will present only its definition and some basic features , necessary for the further discussion . contrary to the schwarzschild case , the kerr metric is no longer spherically symmetric . in the case of a rotating black hole with mass @xmath38 and angular momentum @xmath57 , this metric is given by ( see e.g. * ? ? ? * ) : @xmath58 where @xmath59 above definition is given in boyer - lindquist coordinates for @xmath60 . as it can be seen from expressions ( [ eq32_1 ] ) and ( [ eq32_2 ] ) , in addition to black hole mass @xmath38 , the kerr metric also depends on its specific angular momentum @xmath61 . here @xmath62 because any additional angular momentum would increase the energy of the black hole and , therefore , its mass @xcite . in the case of a non - rotating black hole , i.e. for @xmath63 , it reduces to the schwarzschild metric . by solving the equation @xmath64 and taking only a root with @xmath65 sign , we obtain the radius of event horizon of a black hole : @xmath66 thus , when @xmath63 then @xmath67 ( because we assumed that @xmath60 ) , whereas when @xmath68 ( i.e. for a maximally rotating black hole ) then @xmath69 . the minimum allowed radius @xmath70 of a stable circular equatorial orbit , or so called marginally stable orbit , is given by the roots of the equation @xcite : @xmath71 , where the upper sign refers to co - rotating orbits , while the lower one to counter - rotating orbits . in the case of schwarzschild metric ( @xmath72 ) @xmath73 , whereas for a maximally rotating black hole ( @xmath74 ) @xmath75 @xcite . rotating black holes are surrounded by a region called the ergosphere , in which there can be no static observers and the negative energy orbits are possible @xcite . also , there is a general relativistic effect known as frame - dragging , according to which a rotating black hole drags the space - time around itself , but such effects will not be discussed here in more details . some observed quasars have luminosities of up to @xmath77 erg / s @xcite . the corresponding total energy emitted during the lifetime of such quasars can be estimated to @xmath78 erg , assuming that their minimum age is about @xmath79 years and that their luminosities do not change significantly over the lifetime @xcite . only two known mechanisms could produce such enormous amount of energy : nuclear fusion and accretion . the maximum efficiency @xmath80 ( defined as the mass fraction of fuel that is converted into energy , according to @xmath81 ) of thermonuclear reactions is @xmath82 . @xcite showed that this value is too low to explain such high energies of @xmath83 erg . accretion is the only remaining mechanism which can yield larger @xmath80 . its maximum efficiency is about @xmath84 for a non - rotating black hole and @xmath85 for a maximally rotating one @xcite . therefore , agn release their enormous energy by matter accreting towards , and falling into , a central supermassive black hole . if the cold matter , which was initially at rest and without magnetic field , was subjected to free radial infall it would accrete to the central black hole without any energy release or observational effects @xcite . however , in the case of agn the accreting matter has a significant angular momentum which does not allow its free infall . at the marginally stable orbit of central black hole with mass of @xmath86 , the specific angular momentum is @xmath87 , which is much less even in comparison to a typical galaxy where the speciffic angular momentum of orbiting material is @xmath88 @xcite . it means that approach of accreting material toward a black hole requires a loss of the greatest fraction of its initial angular momentum . mechanisms which can contribute to such loss of angular momentum are viscosity , nonaxisymmetric gravitational forces , magnetic forces , etc . @xcite . since the orbit of minimum energy for fixed angular momentum in any spherically symmetric potential is a circle , the infall of accreting material due to loss of its angular momentum will be in form of successively smaller and smaller concentric circles . matter traveling along orbits inclined to each other will eventually collide in the plane of intersection , and as a result , the angular momenta of different gas steams will be mixed and equalized . consequently , all accreting matter will orbit in a single plane and will have the same specific angular momentum at any given radius , meaning that accretion is most likely performed through an accretion disk @xcite . the accreting material is able to approach the marginally stable orbit of central black hole only if there is an efficient mechanism for transporting angular momentum outward . magnetic field that exists in matter flowing into the disk , as well as its turbulent motions , enable angular momentum to be transferred outward . hence , the accretion disk is one of the best candidates for a such mechanism , since the particles in it lose their angular momentum due to friction between its adjacent layers and spiral towards black hole , releasing their gravitational energy . part of this energy increases the kinetic energy of rotation and the rest is turned into the thermal energy and irradiated from the disk surface @xcite . the efficiency of angular momentum transport in the disk is characterized by viscosity parameter @xmath89 , which can be assumed as constant in most cases , since the spectrum of disk radiation and its surface temperature do not strongly depend on it , except in the case of a supercritical accretion regime @xcite . the total energy release and the spectrum of emitted radiation are mainly determined by the rate of matter inflow into the disk on its outer boundary , i.e. by its accretion rate which is usually denoted by @xmath90 . if the accretion converts matter to radiation with fixed radiative efficiency @xmath91 ( in rest - mass units ) , then a characteristic scale for the accretion rate is the eddington accretion rate @xcite : @xmath92 for witch the total release of energy in the disk @xmath93 is equal to the eddington luminosity : @xmath94 eddington luminosity defines a critical luminosity for any given mass , beyond which the radiation force overpowers gravity . observed agn have luminosities from @xmath11 to @xmath95 erg / s , which means that their central black holes must have masses from @xmath3 to @xmath17 , respectively @xcite . if we assume that accretion in agn occurs with rate of @xmath96 and with radiative efficiency @xmath97 , the resulting luminosity would be @xmath98 erg / s , which is the middle of the observed luminosity distribution of low redshifted agn @xcite . actual accretion rates @xmath90 of matter inflow into the accretion disks of agn could be many times less or higher than the critical value of the eddington accretion rate . at essentially subcritical accretion rates ( @xmath99 ) , the maximal surface temperatures @xmath100 are on the order of @xmath101 k in the inner regions of the disk , from which the most of energy is released , mainly in uv and soft x - ray bands @xcite . when the accretion rate increases , the luminosity also raises , as well as the effective radiation temperature . at accretion rates comparable to @xmath102 , the accretion disk becomes a powerful source of x - ray radiation with the effective radiation temperature @xmath103 from @xmath79 to @xmath104 k. in a strongly supercritical regime of accretion ( @xmath105 ) , the luminosity becomes fixed at the eddington critical limit @xmath106 and the most of energy is irradiated from accretion disk in uv and optical spectral bands @xcite . by the term `` standard newtonian model of accretion disk '' , here we will assume the disk model given by @xcite . this model was originally developed to describe the accretion disks around stellar sized black holes in the binary systems , but with certain modifications it could be also applied on accretion disks around supermassive black holes in agn . the standard newtonian model of accretion disk is based on the supposition that the released gravitational energy is emitted as a multitemperature blackbody radiation , where the surface temperature profile is given by ( see * ? ? ? * ) : @xmath107^{1/4 } , \label{eq42_1}\ ] ] where @xmath108 is the stefan constant , @xmath40 is the gravitation constant , @xmath38 is the mass of the central black hole , @xmath90 is the accretion rate and @xmath109 is the inner radius of the accretion disk . in above equation and throughout this section we will use the dimensionless disk radius @xmath46 defined as @xcite : @xmath110 where @xmath111 is disk radius expressed in gravitational radii @xmath55 . in the standard model of accretion disk , accretion occurs via an optically thick and geometrically thin disk . the effective optical depth in the disk is very high and photons are close to thermal equilibrium with electrons @xcite . the spectrum of thermal radiation emitted from accretion disk surface depends on its structure and temperature ( and therefore on the distance to the black hole ) , and can have several distributions such as @xcite : 1 . planck distribution in the outer regions of the disk , where free - free and free - bound processes , as well as absorption in the lines of heavy elements broadened by the gas pressure , give the main contribution to the shape of emitted spectrum : @xmath112 where @xmath113 , @xmath114 is planck and @xmath115 is boltzman constant , 2 . more complex spectrum which passed through the homogeneous or exponentially varying medium in intermediate regions of the disk where thomson scattering dominates and 3 . wien distribution in the inner regions where the comptonization processes strongly affect the shape of emitted spectrum : @xmath116 the surface temperature has different radial distributions in different parts of accretion disk and results in the multicolor black body spectrum . in the inner regions of accretion disks of agn where their x - ray radiation is generated , the radial distribution of surface temperature has the following form @xcite : @xmath117 where the temperature constant @xmath118 is chosen so that the corresponding effective radiation temperature is @xmath119 k. this modification of standard disk model is made , because the surface temperature given by eq . ( [ eq42_1 ] ) could not be successfully applied in the case of accretion disks of agn . the distribution of the temperature as a function of the accretion disk radius , according to eq . ( [ eq42_4 ] ) , is presented in fig . [ fig42_1 ] for two different values of angular momentum @xmath61 of central black hole @xcite . ( in the direction normal to the rotation axis , as seen by distant observer ) , given for two different values of angular momentum @xmath61 of central black hole @xcite . negative values of @xmath111 correspond to the approaching and positive to the receding side of the disk . ] one can see from fig . [ fig42_1 ] that the temperature depends not only on disk radius , but also on its other parameters , such as e.g. angular momentum @xmath61 . emitters located at different radii in accretion disk have different temperatures and therefore make different contributions to the observed flux . also , it is noticeable that the temperatures at smaller radii of approaching side of the disk are significantly higher than those on its receding side . intensity of radiation emitted from the total disk surface can be obtained by integration of all local spectra emitted from the above regions : @xmath120rdr , \label{eq42_5}\ ] ] where @xmath109 and @xmath121 are inner and outer radii of accretion disk , respectively . ( upper left ) , 30@xmath122 ( upper right ) , 55@xmath122 ( lower left ) and 80@xmath122 ( lower right ) . the disk rotation is in counterclockwise direction . ] magnetohydrodynamic simulations of an accretion disk as seen by a distant observer at four different inclinations are presented in fig . [ fig42_2 ] @xcite . the most distinctive details in this figure are especially bright regions in form of arcs within the turbulent flow , which trace out the photon trajectories close to the radius of marginally stabile orbit . these arcs are much brighter on approaching side of the disk due to higher surface temperature , as shown in fig . [ fig42_1 ] , and also due to doppler boosting and relativistic beaming which both enhance the flux observed in the direction of motion and diminish the flux in the opposite direction . an accretion disk around a supermassive black holes in the center of agn extends from the radius of marginally stable orbit @xmath123 to the several thousands of gravitational radii . according to its radiation emitted in different spectral bands , it can be stratified in several parts @xcite : 1 . innermost part close to the central black hole which emits x - rays and which extends from the radius of marginally stable orbit @xmath123 to the several tens of gravitational radii ( see e.g. * ? ? ? * ) , 2 . central part ranging from @xmath124 to @xmath125 which emits uv radiation and 3 . outer part extending from several hundreds to several thousands @xmath55 , from which the optical emission is coming @xcite . emissivity of the disk has an important role in the line and continuum shapes . in the rest frame of the emitting material ( i.e. in disk frame ) , emissivity @xmath126 is defined as energy emitted per unit proper time per unit proper area @xcite . in the same frame , it is related to the emitted intensity @xmath127 and emitted flux @xmath128 by @xcite : @xmath129 under assumption that the line emitted at frequency @xmath130 can be approximated by a @xmath131 function . although the standard model of the accretion disk does not predict the power - law for the disk emissivity , such law is usually accepted in the case of the hard x - ray emission from the inner parts of agn accretion disks @xcite . in that case the surface emissivity is assumed to vary with radius as power law with emissivity index @xmath132 @xcite : @xmath133 where @xmath134 is the emissivity constant . @xcite found that emissivity index @xmath132 varies between 0 and 3 for the fe k@xmath0 line emitting region of cygnus x-1 , where the most probable value is @xmath135 . some authors found that models describing the disk emissivity as a broken power law , rather than a single power law , achieve significantly better statistical fits of the x - ray emission in the fe k@xmath0 line @xcite . one such broken power law between some inner radius @xmath109 and outer radius @xmath121 with break radius @xmath136 is given by @xcite : @xmath137 where @xmath138 and @xmath139 are some emissivity indices . using this law @xcite fitted the fe k@xmath0 line of mcg6 - 30 - 15 , assuming a maximally spinning ( @xmath140 ) black hole , and found that the line emissivity followed such dependence for @xmath141 and @xmath142 , where the break radius was @xmath143 . the x - ray continuum can be also modeled e.g. by the following time - independent intrinsic emissivity of the disk as a function of its radius @xmath46 and photon energy @xmath144 @xcite : @xmath145 where @xmath146 is photon index and @xmath0 is radial emissivity power law index of a continuum emitting region . according to some studies of observed x - ray spectra of agn , the values of @xmath146 and @xmath0 are estimated to be about 1.5 and 2.5 , respectively @xcite . in the case of the disk outer parts ( i.e. for optical emitting region ) , the black - body emissivity law is assumed and , according to the eq . ( [ eq42_2 ] ) , emitted intensity is given by planck function @xcite : @xmath147 however , in the inner parts of the accretion disk planck function can not be used properly . in these regions so called `` modified '' black - body emissivity law is applied . according to the eq . ( [ eq42_3 ] ) , the corresponding emitted intensity is given by @xcite : @xmath148 and @xmath149 , around a supermassive black hole with mass @xmath150 @xcite . the continuum shapes are obtained assuming the following three emissivity laws ( all normalized to their maximum values ) : the power law defined by eq . ( [ eq43_1 ] ) with emissivity index @xmath151 , the black - body and `` modified '' black - body emissivity laws defined by eqs . ( [ eq43_3 ] ) and ( [ eq43_4 ] ) , respectively , assuming the radial distribution of the surface temperature given by eq . ( [ eq42_4]).,scaledwidth=80.0% ] a comparison between the shapes of the x - ray continuum ( 1 10 kev ) emitted from an accretion disk around a supermassive black hole , obtained assuming three different emissivity laws , is presented in fig . [ fig43_1 ] @xcite . from fig . [ fig43_1 ] it is clear that x - ray continuum shape strongly depends on adopted emissivity law . a disk geometry for the x - ray emitting regions of agn could be assumed , since the unified model of agn includes a supermassive black hole fed by an accretion disk . @xcite calculated spectral line profiles for radiation emitted from inner parts of accretion disks and later on such features of fe @xmath37 lines were discovered by @xcite in japanese _ asca _ satellite data for seyfert galaxy mgc-6 - 30 - 15 . the assumption of a disk geometry for the distribution of the x - ray emitters in the central parts of agn is also supported by the spectral shape of the fe k@xmath0 line @xcite . moreover , a bump in the uv spectra of agn is present very often , indicating that the uv and optical continua also originate from an accretion disk @xcite . we should note here that probably most of the x - ray emission in the 110 kev energy range originates from inverse compton scattering of photons from the disk by electrons in a tenuous hot corona @xcite . proposed geometries of the hot corona of agn include a spherical corona sandwiching the disk and a patchy corona made of a few compact regions covering a small fraction of the disk @xcite . on the other hand , it is known that part of the accretion disk that emits in the 110 kev rest - frame band ( e.g. the region that emits the continuum compton reflection component and the fluorescent emission lines ) is very compact and may contribute to the x - ray variability in this energy range . a typical x - ray spectrum of agn ( see fig . [ fig50_1 ] ) is composed from the following components @xcite : - an underlying power - law continuum due to thermally comptonized soft photons , a soft excess at low energies below 1 kev due to thermal ( black - body ) emission from an optically - thick accretion disk , a fluorescent / recombination fe k@xmath0 line and a compton hump due to x - ray reflection from the disk . in the following text we will discuss in more details power law component of x - ray continuum and the fe @xmath37 spectral line . agn are powerful sources of x - ray radiation in the continuum from 0.1 to 100 kev , which contains two components ( see fig . [ fig51_1 ] ) : the soft x - rays with a steep spectrum and the hard component with spectrum in form of a power law @xcite . therefore , the observed continuum flux is very often fitted with one or two black - body components in the soft x - rays , in addition to a power law component in the hard x - rays . it is believed that both , soft and hard components mostly arise from the inner region of agn disk , close to the central supermassive black hole ( @xmath152 ) . the first component is probably formed in the accretion disk , which is also a strong source of the soft uv and optical photons , or at least in cold ( @xmath153 k ) accreting gas clouds . the second one is caused by high - energy ( most likely relativistic ) electrons in hot corona above the disk , when they multiply inverse - compton scatter some of the low - energy uv and optical photons from the disk to x - ray energies @xcite . the resulting hard x - ray power law component irradiates the accretion disk and produces a reflection component which causes the observed spectrum to flatten above 10 kev ( see fig . [ fig51_1 ] ) , as compton recoil reduces the backscattered flux @xcite . apart from its own thermal radiation , the cold gas around a black hole is also irradiated by these hard x - ray photons , producing different spectral features through photoelectric absorption , fluorescence ( responsible for occurrence of fluorescent iron lines ) and scattering @xcite . therefore , the observed x - ray spectra of agn consist of an iron line spectrum superposed on a power law continuum @xcite . fluctuations of the x - ray radiation on timescales from several parts of an hour until several days are a common property of all agn @xcite . such fast variations confirm assumption that x - ray radiation is emitted from a very compact region in the center of agn . @xcite showed that a two - component model , consisting of a power - law and a reflection component , could better describe the observed variability of x - ray spectra in the case of several bright seyferts than a simple power - law model . this two - component model is presented in fig . [ fig51_1 ] . contrary to the uv and optical spectra of agn , there are very few strong spectral lines in their x - ray spectra . the most important one is the fe k@xmath0 line . iron abundance in accretion disks of agn is sufficient to produce very strong emission in this line . the ccd detectors on japanese _ asca _ satellite were the first instruments with sufficient spectral resolution and sensitivity in the x - ray band , by which @xcite obtained the first convincing proof for the existence of the fe k@xmath0 line in agn spectra . this discovery was made after four - day observations of seyfert 1 galaxy mcg-6 - 30 - 15 ( see fig . [ fig52_1 ] ) . line from seyfert 1 galaxy mcg-6 - 30 - 15 observed by _ asca sis _ detector @xcite and the best fit ( blue solid line ) obtained by a model of the accretion disk in schwarzschild metric , extending between 3 and 10 schwarzschild radii @xcite . image credit : tanaka et al . , copyright : nature , 375 , 659 , 1995 . image generated by dr . paul nandra nasa / gsfc . ] the fluorescent / recombination iron k@xmath0 line is an important indicator of accreting flows around compact objects , because it is produced in inner parts of their accretion disks . at the same time , it is the strongest line of the x - ray radiation , and it can be found in the spectra of all types of accreting sources : binary black hole and neutron star systems , cataclysmic variable stars and agn . the fe k@xmath0 line is produced when plasma is subjected to the influence of the hard x - ray radiation so that one of the two @xmath154-shell ( @xmath155 , where @xmath156 is the principal quantum number ) electrons of an iron atom ( or ion ) is ejected following the photoelectric absorption of an x - ray @xcite . the threshold for the absorption by neutral iron is @xmath157 kev @xcite . the resulting excited state decays when an @xmath158-shell ( @xmath159 ) electron drops into the @xmath154-shell , releasing @xmath160 kev of energy . this energy is either emitted as an emission - line photon ( 34% probability ) or internally absorbed by another electron ( 66% probability ) which is consequently ejected from the iron ion ( auger effect ) . the fluorescent yield ( i.e. the probability that photoelectric absorption is followed by fluorescent line emission rather than the auger effect ) is a weak function of the ionization state from neutral iron ( fe i ) up to fe xxiii @xcite . for lithium - like iron ( fe xxiv ) through to hydrogen - like iron ( fe xxvi ) , the lack of at least two electrons in the @xmath158-shell means that the auger effect can not occur . for he and h - like iron ions , the line is produced by the capture of free electrons ( recombination ) and the equivalent fluorescent yield is high and it depends on the plasma conditions @xcite . for the neutral iron , the fe k@xmath0 line energy is 6.4 kev ( more precisely , there are two components of the line @xcite : fe k@xmath138 at 6.404 and fe k@xmath139 at 6.391 kev ) , while in the case of ionization , the energy of both the photoelectric threshold and the fe k@xmath0 line are slightly increased . even for such high ionization states of he and h - like iron ions , the fe k@xmath0 line energy is increased only to @xmath161 and @xmath162 kev , respectively @xcite . fe k@xmath0 line is pretty narrow in itself , but in case when it originates from a relativistically rotating accretion disk of agn it becomes wider due to kinematical effects , and also its shape ( or profile ) is changed due to doppler boosting and gravitational redshift . such broadening of the line is very often observed in spectra of seyfert galaxies and is one of the main evidences for the existence of a relativistic accretion disk which extends deeply in the gravitational field of the central black hole @xcite . if the line originated from an arbitrary radius of a nonrelativistic ( keplerian ) accretion disk it would have a symmetrical profile ( due to doppler effect ) with two peaks : a `` blue '' one which is produced by emitting material from the approaching side of the disk in respect to an observer , and a `` red '' one which corresponds to emitting material from the receding side of the disk . the widest parts of the fe k@xmath0 line arise from the innermost regions of the disk , where the rotation of emitting material is the fastest . using _ asca _ satellite observations , @xcite found that , in case of 14 seyfert 1 galaxies , full - widths at half - maximum ( fwhm ) of their fe k@xmath0 lines correspond to velocities of @xmath163 km / s . in some cases ( e.g. for seyfert 1 galaxy mcg-6 - 30 - 15 ) , fwhm velocity reaches 30% of speed of light ( see e.g. * ? ? ? it means that in the vicinity of the central black hole , orbital velocities of the emitting material are relativistic , causing the enhancement of the fe k@xmath0 line `` blue '' peak in regard to its `` red '' peak ( relativistic beaming ) . taking into account the integral emission in the line over all radii of accretion disk , one can obtain the line with asymmetrical and highly broadened profile @xcite . the `` blue '' peak is then very narrow and bright , while the `` red '' one is wider and much fainter ( see fig . [ fig52_1 ] ) . besides , the gravitational redshift causes further deformations of the fe k@xmath0 line profile by smearing the `` blue '' emission into `` red '' one . since the observed fe k@xmath0 line profiles are strongly affected by such relativistic effects , they represent a fundamental tool for investigating the plasma conditions and the space - time geometry in the vicinity of the supermassive black holes of agn . one of the important features of the fe k@xmath0 line is variability of both , its shape and intensity . observed variations of this line are surprisingly less than those of the high energetic continuum , which is assumed to give rise to the line emission @xcite . also , it seems that there is a lack of corresponding line response to the continuum variations on time scales from several minutes to several days , or that these line and continuum variations are uncorrelated @xcite . in 6 of this chapter we will pay attention to some possible causes of such behavior of the fe k@xmath0 line and x - ray continuum . the disk emission can be analyzed by numerical simulations , based on so called ray - tracing method in kerr metric @xcite , taking into account only photon trajectories reaching the observer s sky plane . in this method one divides the image of the disk on the observer s sky into a number of small elements ( pixels ) . for each pixel , the photon trajectory is traced backward from the observer by following the geodesics in a kerr space - time , until it crosses the plane of the disk ( see fig . [ fig53_1 ] ) . then , the flux density of the radiation emitted by the disk at that point , as well as the redshift factor of the photon are calculated . in that way , one can obtain the color images of the accretion disk which a distant observer would see by a high resolution telescope . the simulated line profiles can be calculated taking into account the intensities and received photon energies of all pixels of the corresponding disk image . all illustrations of the accretion disk and the line shape in this chapter are obtained using such numerical simulations . here we will briefly describe a pseudo - analytical approach of ray - tracing proposed by @xcite . , and observed at a pixel with coordinates ( impact parameters ) @xmath164 on the disk image in the observer s reference frame ( figure courtesy : vesna borka jovanovi ' c ) . ] this method is based on the pseudo - analytical integration of the geodesic equations which describe the photon trajectories in the general case of a rotating black hole having some angular momentum @xmath165 , which gravitational field is therefore described by the kerr metric ( see 3.2 for more details ) . the kerr metric depends on the angular momentum normalized to the mass @xmath38 of black hole : @xmath166 . a photon trajectory in the kerr metric can be described by three constants of motion ( the energy at infinity and two constants related to the angular momentum , respectively ) which have the following forms when natural units @xmath167 are assumed @xcite : @xmath168 , @xmath169 and @xmath170 . here , @xmath171 are the usual boyer - lindquist coordinates and @xmath132 is the 4-momentum . as the trajectory of a photon is independent on its energy , it may be expressed using the two dimensionless parameters @xmath172 and @xmath173 which are very simply related to the two impact parameters @xmath0 and @xmath174 describing the apparent position on the observer s celestial sphere : @xmath175 and @xmath176 , where the sign of @xmath174 is determined by @xmath177 . in order to find the photon trajectories ( null geodesics ) which originate in the accretion disk at some emission radius @xmath178 and reach the observer at infinity , one must solve the following integral equation @xcite : @xmath179 @xmath180 } , \\ \theta \left ( { \theta , \lambda , q } \right ) = q^{2 } + a^{2}\cos ^{2}\theta - \lambda ^{2}\cot ^{2}\theta . \end{array } \label{eq53_2}\ ] ] above integral equation ( [ eq53_1 ] ) can be solved in terms of jacobian elliptic functions , and therefore it is a pseudo - analytical integration . for the exact expressions of the solutions , see e.g. @xcite , or more recent paper from @xcite . due to relativistic effects , photons emitted at frequency @xmath181 will reach infinity at frequency @xmath182 , and their ratio determines the shift due to these effects : @xmath183 . the total observed flux at the observed energy @xmath184 is given by @xcite : @xmath185 where @xmath186 is the disk emissivity , @xmath187 is the solid angle subtended by the disk in the observer s sky and @xmath188 is the rest energy . a simulated accretion disk image is obtained in the following way : 1 . values of the following input parameters are specified : inner and outer radii ( @xmath189 and @xmath190 ) of the disk , angular momentum @xmath61 of the central black hole , observer s viewing angle ( disk inclination ) @xmath191 ( also , often denoted by @xmath192 ) and parameters defining the disk emissivity 2 . constants of motion @xmath9 and @xmath193 are calculated for each pixel on imaginary observer s photographic plate ( i.e. for each pair of impact parameters @xmath0 and @xmath174 ) 3 . geodesic equation ( [ eq53_1 ] ) is integrated for each pair of @xmath9 and @xmath193 4 . values of shift due to relativistic effects @xmath194 and observed flux @xmath195 are calculated 5 . pixels on imaginary observer s photographic plate are colored according to the value of shift @xmath194 and a simulated disk image is obtained . the simulated line profiles can be calculated from the corresponding disk images by binning the observed flux at all pixels over the bins of shift @xmath194 . the examples of simulated disk images obtained in such way are presented in left panels of figs . [ fig54_1 ] - [ fig54_3 ] , and the corresponding simulated line profiles are presented in the right panels of the same figures . in general , black holes have three measurable parameters ( not including the hawking temperature ) : charge , mass ( and hence gravitational field ) and angular momentum ( or spin ) . in the case of supermassive black holes of agn , only the latter two are of sufficient importance because they are responsible for several effects which can be detected in the observed fe k@xmath0 line shapes @xcite . in order to study the size of the fe k@xmath0 line emitting region , as well as its location in the disk , one can assume that the line is emitted from a region in form of a narrow ring . for example , @xcite assumed a line emitting region with width equals to @xmath196 , located between : a ) @xmath197 r@xmath198 and @xmath199 r@xmath198 and b ) @xmath200 r@xmath198 and @xmath201 r@xmath198 . these two cases are presented in fig . [ fig54_1 ] . line emitting region in form of narrow ring with width @xmath202 , extending from : @xmath197 r@xmath198 to @xmath199 r@xmath198 ( top ) and @xmath200 r@xmath198 to @xmath201 r@xmath198 ( bottom ) . _ right : _ the corresponding fe k@xmath0 line profiles @xcite.,title="fig:",scaledwidth=48.0% ] + line emitting region in form of narrow ring with width @xmath202 , extending from : @xmath197 r@xmath198 to @xmath199 r@xmath198 ( top ) and @xmath200 r@xmath198 to @xmath201 r@xmath198 ( bottom ) . _ right : _ the corresponding fe k@xmath0 line profiles @xcite.,title="fig:",scaledwidth=48.0% ] from fig . [ fig54_1 ] one can see how the fe k@xmath0 line profile is changing as the function of distance from central black hole . when the line emitters are located at the lower radii of the disk , i.e. closer to the central black hole , they rotate faster and the line is broader and more asymmetric ( see fig . [ fig54_1 ] top - right ) . if the line emission is originating at larger distances from the black hole , its emitting material is rotating slower and therefore the line becomes narrower and more symmetric ( see fig . [ fig54_1 ] bottom - right ) . in majority of agn , where the broad fe k@xmath0 line is observed , its profile is more similar to the modeled profile as obtained under assumption that the line emitters are located close to the central black hole @xcite . angular momentum or spin of the central supermassive black hole of agn is a property of the space - time metric . to demonstrate how it affects the observed line profiles we will now assume that the fe k@xmath0 line emitting region extends between the following inner and outer radii : @xmath203 and @xmath204 . we will analyze the following two cases for accretion disk inclination in both schwarzschild and kerr metrics : ( i ) @xmath205 and ( ii ) @xmath206 . in the case of schwarzschild metric we have a stationary black hole and hence , angular momentum ( normalized to the mass of black hole ) is @xmath72 . in kerr metric ( i.e. for a rotating black hole ) , it can take any value from the @xmath207 $ ] range , but in these two examples we will assume an almost maximally rotating black hole with @xmath140 . line profiles ( right ) in the case of schwarzschild ( top ) and kerr metric with angular momentum @xmath140 ( bottom ) . the disk inclination is @xmath205 and its inner and outer radii are @xmath203 and @xmath208 r@xmath198 , respectively @xcite.,title="fig : " ] + line profiles ( right ) in the case of schwarzschild ( top ) and kerr metric with angular momentum @xmath140 ( bottom ) . the disk inclination is @xmath205 and its inner and outer radii are @xmath203 and @xmath208 r@xmath198 , respectively @xcite.,title="fig : " ] , but for a highly inclined disk with @xmath206 @xcite.,title="fig : " ] + , but for a highly inclined disk with @xmath206 @xcite.,title="fig : " ] illustrations of an accretion disk and the corresponding fe k@xmath0 line shapes in the first case are presented in fig . [ fig54_2 ] . as one can see from fig . [ fig54_2 ] , the red peak of the fe k@xmath0 line is brighter in case of almost maximally rotating black hole , but at the same time it is also more embedded into the blue peak wing and therefore less separable from it . [ fig54_3 ] contains illustrations of the line emitting regions and the corresponding line shapes in the case of a highly inclined disk ( @xmath206 ) . here , the line profiles are broader than in the first case , mostly due to higher inclination . as it can be seen from fig . [ fig54_3 ] , in the case of the kerr metric , the red peak of the line is again more embedded into its blue peak wing ( as in the first case ) and it confirms that this effect can be most likely attributed to angular momentum @xcite . consequently , angular momentum of the central black hole has significant influence on the line shape which supports assumption that the line originates from the innermost part of accretion disk , close to the central black hole ( see e.g. * ? ? ? this fact can be used for estimation of angular momentum of the central black hole in observed agn . above simulations of the strong gravitational field influence on the fe k@xmath0 line show that such effects can be detected in the observed line shapes and therefore , comparisons between the observed and modeled fe k@xmath0 line profiles ( see e.g. fig . [ fig52_1 ] ) can bring us some essential information about the strong gravitational field in the vicinity of central supermassive black holes of agn @xcite . rapid and irregular variability of the observed x - ray emission in the line , as well as in the continuum , is a common property of all agn . this variability could be due to disk instability , reflecting in perturbations of the disk emissivity , or it could be caused by some external effects , such as gravitational microlensing and absorption by x - ray absorbers . we developed a model of perturbations of the disk emissivity , a model of absorption region and three models of gravitational microlensing . in the following text we will pay more attention to all three mentioned causes of the x - ray variability in agn . in some cases the observed x - ray variability of agn can not be explained by the standard model of accretion disk . for example , in addition to the stable 6.4 kev core of the fe k@xmath0 line , a variable `` red '' feature of the line at 6.1 kev is also detected in x - ray spectrum of seyfert galaxy ngc 3516 ( see fig . [ fig61_1 ] ) , observed by _ xmm - newton _ satellite @xcite . this feature varies systematically in the flux at intervals of 25 ks and in energy between 5.7 and 6.5 kev . @xcite found that the spectral evolution of the `` red '' feature agrees well with hypothesis of an orbiting spot in the accretion disk . line red feature ( middle ) and the 6.4 kev line core ( bottom).,scaledwidth=80.0% ] many processes in the accretion disk may lead to perturbations in its emissivity , such as self gravity , disk - star collisions and baroclinic vorticity @xcite . different models of emissivity perturbing regions can be used to describe the observed variability in the fe k@xmath0 line profiles and intensities , like stochastically perturbed one given by @xcite , but here we will present a bright spot model given by @xcite . in this model a modification of the power - law disk emissivity is proposed in order to explain the observed profiles . the following emissivity law of the disk is assumed : @xmath209 where @xmath210 is the modified disk emissivity at the given position @xmath211 of perturbing region ( in gravitational radii @xmath55 ) , @xmath212 is the power - law disk emissivity at the same position , @xmath213 is emissivity of perturbing region and @xmath214 are its widths ( also in @xmath55 ) . an example of the shape of the perturbed emissivity for an accretion disk in schwarzschild metric , as well as the corresponding perturbed and unperturbed fe k@xmath0 line profiles , are presented in the left and right panels of fig . [ fig61_2 ] , respectively . , @xmath215 and @xmath216 @xcite . _ right : _ the corresponding perturbed ( dashed line ) and unperturbed ( solid line ) fe k@xmath0 line profiles . ] as one can see from fig . [ fig61_2 ] , when perturbation moves along receding side of the disk ( positive direction of @xmath217-axis ) , the perturbing model affects only `` red '' part of the line flux , while the `` blue '' one stays nearly constant , as well as the line core . therefore , this bright spot model of perturbing region could satisfactorily explain the observed variations of the fe k@xmath0 line flux . under assumption that perturbation moves by speed of light @xmath41 , one can also calculate time @xmath218 $ ] that corresponds to the current position @xmath211 of perturbation , using the following expression @xcite : @xmath219 = \dfrac{{r(x_p , y_p ) \left [ { r_g } \right]}}{{c\left [ { m \cdot s^ { - 1 } } \right ] } } = \dfrac{{r_{x , y } \cdot gm_{bh } } } { { c^3 } } , \label{eq61_2}\ ] ] where @xmath220 @xmath40 is newton s gravitational constant and @xmath221 is the mass of central black hole . using the time @xmath218 $ ] we are now able to obtain simulated light curves , produced as perturbation moves along the accretion disk . an example of such light curves , corresponding to displacement of perturbation along the receding side of the disk , is given in fig . [ fig61_3 ] . from this figure it can be seen that displacement of perturbing region results in variations of only `` red '' light curve ( 0.1 6.1 kev ) , while the `` blue '' one ( 6.7 12.8 kev ) and the line core ( 6.1 6.7 kev ) stay nearly constant . ] these variations are then reflected in total line flux in 0.1 12.8 kev energy band @xcite . thus , this perturbing model could explain the variable `` red '' feature of the fe k@xmath0 line observed in ngc 3516 . besides , the realistic durations of disk emissivity perturbations are also obtained if a central supermassive black hole with mass @xmath222 is assumed @xcite . line of seyfert 1 galaxy ngc 3516 , observed by _ asca _ satellite @xcite . the dashed line shows best fit with model of an accretion disk around a rotating ( kerr ) black hole.,scaledwidth=80.0% ] the x - ray emission of agn could be also significantly absorbed by an outflowing wind , especially in case of so - called low ionization broad absorption line ( lobal ) quasars . recent observations of such quasars ( e.g. mrk 231 @xcite and h 1413 + 117 @xcite ) confirmed the presence of x - ray absorbers in these objects . @xcite detected an absorption line at 5.8 kev in nearby ( @xmath223 ) seyfert 1.5 galaxy ngc 4151 . a variable absorption line at the same energy has been discovered by @xcite in ngc 3516 ( fig . [ fig62_1 ] ) and was interpreted as a fe k resonant absorption line , redshifted either by infalling absorbing material or by strong gravity in the vicinity of the black hole . @xcite found an evidence for a p cygni profile of the fe k@xmath0 line ( fig . [ fig62_2 ] ) in narrow line seyfert 1 galaxies . according to these authors , complex x - ray spectra of these objects show strong `` soft excess '' below 2 kev and a sharp drop at @xmath224 kev which can be explained either by reflection or by absorption from relativistic , partially ionized material close to the black hole . they showed that a sharp feature at @xmath224 kev results from absorption / scattering / emission of the iron k@xmath0 line in the wind . in the case of 1h 0707 - 495 ( fig . [ fig62_2 ] ) , this absorption feature can be satisfactorily fitted by the p cygni profile @xcite . there are different models of the x - ray absorbing / obscuring regions , like absorbing medium comprised of cold absorbing cloudlets by @xcite , but here we will focus on the model given by @xcite , developed in order to study how much warm absorbers can change the fe k@xmath0 spectral line profile , emitted from a relativistic accretion disk . in this model , absorption region is considered to be composed of a number of individual spherical absorbing clouds with the same small radii ( see fig . [ fig62_3 ] left ) , scattered in space so that projections of their centers to the observer s sky plane @xmath225 have bivariate normal distribution @xmath226 . here , @xmath227^t$ ] and @xmath228 $ ] , where @xmath229 and @xmath230 are the means of @xmath231 and @xmath232 , @xmath233 and @xmath234 are their standard deviations and @xmath235 is the correlation between them . absorbing region presented in fig . [ fig62_3 ] ( left ) is obtained for the following parameters : @xmath236 , @xmath237 , @xmath238 and @xmath239 , where @xmath240 is the center and @xmath241 is the radius of projection of entire absorbing region . spectral line profile ( solid line ) and corresponding absorbed profile ( dotted line ) caused by the absorbing / obscuring region presented in the left panel @xcite . ] the absorption coefficient @xmath242 for every spherical cloud in the absorbing region is given by @xcite : @xmath243 where absorption intensity coefficient @xmath244 describes the distribution of absorption over the whole region , @xmath245 is the central energy of absorption and @xmath246 is the width of absorption band ( velocity dispersion ) . a comparison between the unabsorbed fe k@xmath0 spectral line profile and the corresponding absorbed profile obtained by this absorption model is given in fig . [ fig62_3 ] ( right ) . as it can be seen from fig . [ fig62_3 ] , when the x - ray radiation from approaching side of the disk is significantly absorbed / obscured by the absorbing region , there is a very strong absorption of the iron line . in such case the emission fe k@xmath0 line looks redshifted at @xmath247 kev and is followed by a strong absorption line at @xmath248 kev ( fig . [ fig62_3 ] right ) , which indicates the p cygni profile of the iron line . thus , this model can satisfactorily explain the p cygni profile of the fe k@xmath0 line in the case when approaching side of the accretion disk is partially blocked from our view by the x - ray absorbing / obscuring material , while the rest of the disk is less absorbed / obscured and therefore is visible @xcite . some recent observational and theoretical studies suggest that gravitational microlensing can also induce variability in the x - ray emission of agn , especially in the case of gravitationally lensed quasars . the phenomenology of gravitational lensing effects and introduction to this field has been given in several review papers and books . references to some of them can be found in e.g. @xcite . therefore , in this section we will not discuss all aspects of such effects in the universe , but instead , we will briefly present the basic concepts of gravitational microlensing theory , as well as some examples for microlensing influence on the x - ray emission from agn . gravitational lensing is an universal natural phenomenon where the gravitational force of lensing object induce either the amplification of some background source ( * microlensing * ) , or the appearance of its multiple images ( * macrolensing * ) , due to light bending in a gravitational field of the deflector . separation angle between the images of background source depends on the mass of gravitational lens and therefore , multiple images can be observed only in the case of a massive lensing object , such as galaxy . one of the most famous multiple image lens systems is quasar qso 2237 + 0305 , also known as einstein cross , which is located at redshift @xmath249 ( see fig . [ fig63_0 ] ) . its four images are due to lensing effect of galaxy zw2237 + 030 , located between us and the quasar at redshift @xmath250 . in the case of a small mass lens ( e.g. a star ) , the separation angle is also small and therefore , different images of background source can not be resolved . instead , its intensity is amplified , causing the changes in the observed light curve . common name for both , macrolensing ( or simply lensing ) and microlensing is * strong lensing * , contrary to * weak lensing * , which causes distortions in observed images of distant objects which can be then used for studying the mass distribution along the line of sight , but such phenomena will not be discussed here . @xcite found that cosmologically distributed microlenses could significantly contribute to the x - ray variability of high - redshifted ( @xmath251 ) qsos . indeed , microlensing of the fe k@xmath0 line has been reported at least in three macrolensed qsos : mg j0414 + 0534 @xcite , qso 2237 + 0305 @xcite , and h 1413 + 117 @xcite . @xcite detected an increase of the fe k@xmath0 equivalent width in the image b of mg j0414 + 0534 which was not followed by the continuum and explained this behavior by assumption that the thermal emission region of the disk and the compton up - scattered emission region of the hard x - ray source lie within smaller radii than the iron - line reprocessing region . analyzing the x - ray variability of qso 2237 + 0305a , @xcite also measured amplification of the fe k@xmath0 line in component a of qso 2237 + 0305 but not in the continuum and suggested that the larger size of the continuum emission region in comparison to the fe k@xmath0 emission region could explain this result . in the case of h 1413 + 117 , @xcite found that the continuum and the fe k@xmath0 line were enhanced by a different factor . for more detailed discussion about observational and theoretical investigations of the x - ray variability in gravitationally lensed qsos see e.g. @xcite and @xcite , and also references therein . the influence of gravitational microlensing on the x - ray emission from a compact accretion disk of agn can be studied using numerical simulations based on ray - tracing method described in 5.3 of this chapter . in the following text we will briefly review three models of gravitational microlenses ( point - like , straight - fold caustic and quadruple microlens ) , as well as their influence on the x - ray radiation of agn . if @xmath252 and @xmath253 are the impact parameters ( i.e. the coordinates ) which describe the apparent position of each point of the accretion disk image on the celestial sphere as seen by an observer at infinity , then the amplified brightness ( due to gravitational microlensing influence ) is given by @xcite : @xmath254 where @xmath255 ( @xmath130 and @xmath182 are the transition and observed frequencies , respectively ) , @xmath256 is the shift due to relativistic effects ( @xmath181 is the emitted frequency ) , @xmath257 is the disk emissivity and @xmath258 is the amplification caused by microlensing . usually , we do not know the exact form of @xmath258 , so we are forced to consider the different approximations in order to estimate this quantity . the first of these approximations is called * point - like microlens * , and it is applied when an isolated compact object ( e.g. a star ) plays a role of gravitational microlens . such microlens is characterized by its einstein ring radius in the lens plane @xcite : @xmath259 or by the corresponding projection to the source plane : @xmath260 where @xmath40 is the gravitational constant , @xmath41 is the speed of light , @xmath261 is the microlens mass and @xmath262 , @xmath263 and @xmath264 are the cosmological angular distances between observer - lens , observer - source and lens - source , respectively . in this case , the amplification is given by the following relation @xcite : @xmath265 where @xmath266 corresponds to the angular separation between the microlens and a source . @xmath267 are the coordinates of the microlens with respect to the disk center ( given in @xmath55 ) . the total observed flux is then given by @xcite : @xmath268 where @xmath269 is the solid angle subtended by the disk in the observer s sky and the integral extends over the whole disk image . an illustration of a point - like gravitational microlens crossing over an accretion disk in the kerr metric with angular momentum @xmath140 is presented in fig . [ fig63_1 ] @xcite . the simulated unamplified and amplified fe k@xmath0 line profiles due to different positions of the microlens are shown in fig . [ fig63_2 ] @xcite . einstein ring of the microlens is schematically presented by yellow euclidian circle.,scaledwidth=75.0% ] from fig . [ fig63_2 ] it can be seen that point - like microlens could induce strong changes of the fe k@xmath0 line shape and intensity , depending on the location of the microlens . in the first place , we have in mind the changes in number of peaks , their relative separation and the peak velocity . secondly , such transit of the point - like microlens could cause an asymmetrical enhancement of the line profile , and the maximum of the amplification would be obtained for negative values of @xmath217-coordinate , i.e. for the approaching side of the rotating accretion disk , and therefore it would affect mainly the blue part of the fe k@xmath0 line . line profiles ( solid line ) for different positions of the point - like microlens , which coordinates ( expressed in @xmath55 ) in respect to the accretion disk center are given at the top of each panel @xcite . the simulations are performed for an accretion disk in kerr metric with angular momentum @xmath140 and with the following disk parameters : r@xmath270 , r@xmath271 , @xmath205 , q=2.5 . the einstein ring radius of the point - like microlens is 10 r@xmath272 . the relative intensity ( @xmath273-axis ) is in the range from 0 to 3 , and energy shift @xmath194 due to relativistic effects ( @xmath217-axis ) is in the range from 0.4 to 1.2 . ] in most cases we can not simply consider that microlensing is caused by an isolated compact object , but we should take into account that such micro - deflector is located in an extended object ( typically , the lens galaxy ) . therefore , when the size of the microlens projected einstein ring radius is larger than the size of the accretion disk and when a number of microlenses form a caustic net , we describe the microlensing effects by a * straight - fold caustic * approximation @xcite , where the amplification at a point of an extended source ( accretion disk ) close to the caustic is given by @xcite : @xmath274 in above expression @xmath275 is the amplification outside the caustic and @xmath276 is the caustic amplification factor , where @xmath174 is constant of order of unity . @xmath277 is the distance perpendicular to the caustic and @xmath278 is the minimum distance from the disk center to the caustic ( both in @xmath55 ) . the `` caustic size '' @xmath279 is the distance @xmath277 for which the caustic amplification is equal to 1 , and therefore this parameter defines a typical linear scale for the caustic . @xmath280 is the heaviside function which equals 1 for @xmath281 and otherwise it is 0 . @xmath282 is @xmath283 depending on the direction of caustic motion : if the direction of the caustic motion is from approaching side of the disk @xmath284 , otherwise it is + 1 . also , in the special case of caustic crossing perpendicular to the rotating axis @xmath285 for direction of caustic motion from @xmath286 to @xmath287 , otherwise it is @xmath288 . a microlensing event where a caustic crosses over an accretion disk can be described in the following way @xcite : before the caustic reaches the disk , the amplification is equal to @xmath275 because the heaviside function of eq . ( [ eq63_4 ] ) is zero . just as the caustic begins to cross the disk , the amplification rises rapidly and then decays gradually towards @xmath275 as the source moves away from the caustic - fold . an illustration of the straight - fold caustic crossing over an accretion disk in schwarzschild metric is presented in fig . [ fig63_3 ] , and the corresponding effects on the shapes of the x - ray continuum and the fe k@xmath0 line are shown in fig . [ fig63_4 ] @xcite . as it can be seen from fig . [ fig63_4 ] , caustics could also induce significant amplifications of both , the x - ray continuum and fe k@xmath0 line . , scaledwidth=75.0% ] line ( right ) due to influence of straight - fold caustic from fig . [ fig63_3 ] . undeformed profiles are presented by solid and deformed by dotted lines @xcite.[fig63_4 ] ] the most complex approximation for gravitational microlensing amplification is so called * quadruple microlens*. this model is applied to obtain a spatial distribution of magnifications in the source plane ( where an accretion disk of agn is located ) , produced by a random star field placed in the lens plane . such spatial distribution of magnifications is called * microlensing map , microlensing pattern or caustic network*. if we consider a set of @xmath289 compact objects ( e.g. stars ) which are characterized by their positions @xmath290 and their masses @xmath291 , then the normalized lens equation is given by @xcite : @xmath292\vec x , \label{eq63_5}\ ] ] where @xmath293 and @xmath294 are the normalized image and source positions , respectively . the sum describes the light deflection by the stars and the last term is a quadruple contribution from the galaxy containing the stars , where @xmath295 is a smooth surface mass density and @xmath1 is an external shear . the total surface mass density or convergence can be written as @xmath296 , where @xmath297 represents the contribution from the compact microlenses . the corresponding microlensing map is then defined by two parameters : the convergence - @xmath282 , and the shear due to the external mass - @xmath1 . for some specific lensing event one can model the corresponding magnification map using numerical simulations based on ray - shooting techniques ( see e.g. @xcite and references therein ) , in which the rays are shot from the observer to the source , through the randomly generated star field in the lens plane . an example of a magnification map for a `` typical '' lens system ( i.e. for a lens system where the redshifts of microlens and source are @xmath298 and @xmath299 , respectively ) is presented in the left panel of fig . [ fig63_5 ] , and the corresponding x - ray , uv and optical continuum variations are given in the right panel of the same figure @xcite . the size of this magnification map is @xmath300 and it is generated using the following arbitrary values for convergence and shear : @xmath301 and @xmath302 . the light curves in the right panel of fig . [ fig63_5 ] are produced when an accretion disk crosses over the magnification pattern in the left panel , along the path denoted by the white solid line . as one can see from fig . [ fig63_5 ] ( right ) , the variations of the x - ray continuum due to microlensing are much stronger and faster in comparison to the variations in uv and optical spectral bands @xcite . usually , so called high amplification events , i.e. asymmetric peaks in the light curves are also analyzed @xcite . the results presented in fig . [ fig63_5 ] ( right ) show that the rise times of such events are the shortest and their frequency the highest in the x - rays , in comparison to the uv / optical spectral bands . in order to explain the observed x - ray variability in gravitationally lensed quasars , @xcite considered a microlensing magnification map for a specific case of the qso 2237 + 0305a image ( fig . [ fig63_6 ] , top ) . this map with @xmath303 on a side was generated using the following parameters : @xmath304 , @xmath305 and the mean mass of randomly distributed deflectors @xmath306 . the corresponding variations of the total fe k@xmath0 line and x - ray continuum fluxes are presented in fig . [ fig63_6 ] ( bottom ) . on a side and a scheme of an accretion disk with separated fe k@xmath0 line and x - ray continuum emitting regions which outer radii are 20 and 100 r@xmath272 , respectively @xcite . the straight horizontal line represents a path of the disk center . _ bottom : _ the corresponding amplifications of the fe k@xmath0 line and x - ray continuum total fluxes for different positions of the disk center along the path on the magnification map in the left panel.,title="fig:",scaledwidth=75.0% ] + on a side and a scheme of an accretion disk with separated fe k@xmath0 line and x - ray continuum emitting regions which outer radii are 20 and 100 r@xmath272 , respectively @xcite . the straight horizontal line represents a path of the disk center . _ bottom : _ the corresponding amplifications of the fe k@xmath0 line and x - ray continuum total fluxes for different positions of the disk center along the path on the magnification map in the left panel.,title="fig : " ] as one can see from fig . [ fig63_6 ] ( bottom ) , there is a global correlation between the total line and continuum fluxes during the complete simulated microlensing event , but while the continuum flux variations are smooth and slow , the line flux varies much stronger and faster @xcite . moreover , there is a certain period in the middle of the microlensing event in which the fe k@xmath0 line flux changes rapidly , while the continuum flux remains nearly constant . this indicates that different behavior of the line and x - ray continuum variability , observed in some lensed quasars @xcite , may be explained by microlensing hypothesis if the line originates from the innermost part of the accretion disk , while the x - ray continuum is emitted from a larger region in the disk @xcite . on the basis of all facts presented in this section , one can conclude that gravitational microlensing can produce significant variations and amplifications of the fe k@xmath0 line and x - ray continuum . during a microlensing event , even very small mass objects could produce noticeable changes in the x - ray radiation from accretion disks of agn . these changes are significantly larger than the corresponding effects in the optical and uv emission . also , microlensing hypothesis can satisfactorily explain the excess in the iron line emission , observed in some gravitational lens systems . in this chapter we discussed the x - ray radiation from relativistic accretion disks around supermassive black holes , supposed to exist in the centers of all active galactic nuclei . the x - ray radiation is created when the accreting material loses its angular momentum due to friction between the adjacent layers in the disk and spiral towards the central black hole , releasing the gravitational energy . part of this energy increases the kinetic energy of rotation and the rest is turned into the radiation , which is then emitted from the disk surface in the x - ray , uv and optical spectral bands . it is the only known mechanism of converting gravitational potential energy into radiation , which is sufficiently efficient to explain the high luminosities of some observed agn . the standard model of an accretion disk , including its emission , accretion rate , luminosity , structure and spectral distribution was described in details . a significant part of this chapter was devoted to modeling of the observed x - ray radiation in both the fe k@xmath0 spectral line and x - ray continuum . the modeled x - ray radiation is then used to analyze the several effects of strong gravity in the vicinity of black holes which have the influence on the shapes and intensities of the observed fe k@xmath0 line and x - ray continum . we showed that the x - ray emission strongly depends on the angular momentum and mass of the central black hole , as well as on several parameters describing the x - ray emitting region in the disk . this was demonstrated on several examples , showing various aspects of the x - ray emission from the accretion disk around schwarzschild and kerr black holes . since the x - ray radiation is probably produced in a very compact region near the black holes , such comparison between observed and modeled x - ray radiation can provide us some essential information about the plasma conditions and the space - time geometry in the vicinity of those black holes . we also studied some possible causes for the rapid and irregular variability of the fe k@xmath0 line and x - ray continuum , observed in the x - ray spectra of some agn . our results show that , at least , three mechanisms could explain such variability : perturbations of disk emissivity , absorption by warm x - ray absorbers and gravitational microlensing . taking all this into account , we can conclude that the x - ray emission from accretion disks of agn can be used as a powerful tool for revealing the physics and geometry in the vicinity of their central supermassive black holes . but , to use the fe k@xmath0 line profile for such investigations , one should also take into account the effects which can disturb its profile ( such as e.g. absorption , perturbations in the disk and gravitational lensing ) . braito , v. , della ceca , r. , piconcelli , e. , severgnini , p. , bassani , l. , cappi , m. , franceschini , a. , iwasawa , k. , malaguti , g. , marziani , p. , palumbo , g. g. c. , persic , m. , risaliti , g. , salvati , m. _ a&a _ , 2004 , _ 420 _ , 79

thermodynamics forms part of the bedrock of our current understanding of the physical world . it has remained unchanged despite huge revolutions in physics , such as relativity and quantum theory , and few believe it will ever fail . over time , it has been applied to situations well outside its original domain ; from black holes @xcite , to quantum engines comprised of only a few qubits @xcite . a fundamental question is that of the applicability of thermodynamics to quantum systems @xcite ; it is this question that we wish to address in this paper . one of the most fundamental questions one would like to ask is how much work can be extracted from a system , given access to a thermal bath at temperature @xmath0 . this question , for generic quantum systems and hamiltonians has been investigated intensively , e.g. in @xcite yet , until very recently only partial results have been obtained and it remained an open problem . very recently however strong general results were obtained @xcite , made possible by employing a new conceptual framework , namely that of resource theories " . the advent of resource theories is one of the main paradigm shifts which occurred within quantum information theory ; the main idea is to bring to the forefront the study of _ resources _ needed to implement various tasks . the archetypal resource theory in this context is entanglement theory @xcite . however , numerous other examples have now been developed , including reference frames @xcite , purity @xcite and asymmetry @xcite , which has culminated in a general formalised notion of a resource theory @xcite . in @xcite the authors introduced an appealing resource theory of thermodynamics , based on the class of allowed operations first studied in @xcite . here we revisit and modify the resource theory framework of thermodynamics as originally put forward in @xcite . in this new context we achieve two main results : first we prove the second law of thermodynamics . second , as described in more detail below , we show that free energy is a relevant quantity for individual quantum systems . the latter result is particularly relevant following the results of @xcite which call into question the role of free energy for individual quantum systems . classical thermodynamics tells us that the total amount of work we are able to extract from a system is given by its change in _ free energy _ , which was also supported by previous partial quantum results @xcite . yet in @xcite it was shown that work equal to free energy can be extracted only if we _ collectively _ process many copies of the same system . when acting on each copy individually , the amount of work that can be extracted is generally significantly less than the free energy . these results therefore suggest that the free energy is not the relevant quantity for individual systems . here we revisit the issue of work extraction and show that free energy is a significant quantity for individual systems . our paradigm is similar to that of @xcite but differs in an essential aspect . in @xcite they considered _ almost deterministic _ of failure ] work extraction , from the ` single - shot ' viewpoint which has received much attention lately @xcite . here , in contrast , we will consider _ average _ work extraction . we prove that there exist simple protocols which extract _ average _ work equal to the free energy change of the system . crucially , for systems in ` classical ' states ( mixtures of energy eigenstates ) our protocol works on a single copy of the system . this establishes the fact that free energy is a meaningful notion even for individual systems , when their states are classical . however , for non - classical states , where coherences between energy levels exist , we prove that collective actions are necessary , so long as no external sources of coherence are used . more precisely , we prove that when acting on an individual system the amount of average work that can be extracted is equal to the free energy of its decohered state the free energy due to coherences is ` locked ' and can not be used . as more copies are collectively processed , some of this locked free energy is released we can extract the free energy of the decohered state of @xmath1 copies , which is larger than @xmath1 times the free energy of a single decohered system . in the limit of infinitely many copies , the average work extracted per copy reaches its entire free energy . whether or not external coherent resources allow extraction of work equal to the free energy for individual systems , and how to account for such resources ( given that they may provide free energy themselves even without the system ) remains an open question . the paper is organised as follows . in sec . [ s : preliminaries ] we go through the preliminaries , including a brief description of the resource theory of thermodynamics in sec . [ ss : resource theory ] , work storage systems in sec . [ ss : work storage ] and a proof that the second law holds in sec . [ s : second_law ] . in sec . [ s : diagonal ] we study work extraction from diagonal systems , beginning with a motivating example in sec . [ s : motivating example ] , before moving onto a full qubit protocol in sec . [ ss : full protocol ] . we discuss a special case in sec . [ ss : maxwell demon ] , present a full protocol for arbitrary dimension systems in sec . [ ss : qudits ] and prove optimality in sec . [ s : optimality ] . in sec . [ s : coherence ] we show how work can be extracted from coherence by acting collectively on multiple copies of the system . we finish by briefly discussing how the results of this paper can be understood from the perspective of ` virtual temperatures ' in sec . [ s : virtual ] , before concluding in sec . [ s : conclusions ] . the first ingredient in a resource theory is a set of free resources . by free we mean that we have an unlimited supply of this resource , since it is deemed easy to acquire . in the present context of thermodynamics this free resource is _ a thermal bath at temperature @xmath0_. more precisely , we assume that we have an unlimited supply of finite - dimensional systems , with any desired hamiltonian @xmath2 , in their thermal state @xmath3 , @xmath4 where @xmath5 is the inverse temperature throughout for convenience ] , and @xmath6 is the partition function . the second ingredient is a set of allowed operations , those which one has the ability to perform with ease . in the original resource framework these allowed operations were _ any energy conserving unitary transformation_. in particular , any unitary operation which commutes with the total ( non - interacting ) hamiltonian of all the systems under consideration . this allows one to transform systems in a controlled way , but without admitting an external source of energy . however , we will also introduce an additional element into the resource framework , a _ work storage system _ , whose sole purpose is to provide and store work for thermodynamic transformations . in the next section , we will place some additional constraints on the allowed operations , which prevent us from ` cheating ' by using work storage system for other purposes ( e.g. by using it as a cold reservoir in a heat engine ) , and allow us to use the same work storage system for multiple thermodynamic protocols . given these ingredients , any additional system which is not in thermal equilibrium with a bath at temperature @xmath0 is a useful resource . the question that remains is to quantify how useful different resource states are . one way to quantify the usefulness of different resource states is to find a transformation from every resource state to some _ fiducial _ resource which allows for comparison in a well - defined and easy manner . in the resource theory of entanglement , for example , one can quantify the entanglement of a state by the optimal asymptotic number of maximally entangled states that can be created per resource state , or in other words the optimal rate at which states can be converted into maximally entangled states . different states will have different optimal rates , and thus the rate quantifies how useful they are . we seek here an analogous fiducial resource for thermodynamics . work is the desired output in this case , and different states can produce work at different rates . because we wish to study a closed quantum system , with conserved total energy , we desire a system where we can store the work extracted from resource states . similarly to entanglement theory , where the maximally entangled state is chosen for convenience , here we wish to find a work storage system which we deem to be most convenient . in previous work @xcite a candidate was put forward raising a qubit deterministically from its ground state to it s excited state . this qubit was termed a _ wit _ , short for _ work bit_. however , choosing the energy gap of the work bit requires advance knowledge of the work to be extracted , and this model does not translate well to non - deterministic work extraction , which we will be interested in here . furthermore , we would prefer to be able to use a single work storage system as a ` battery ' capable of gaining and expending work in multiple thermodynamic processes . the alternative work storage system we will consider here is a suspended _ weight _ , which is raised or lowered when work is done on or by it . in particular , we will consider a quantum system whose height is given by the position operator @xmath7 , with hamiltonian @xmath8 representing its gravitational potential energy we show that the weight can be replaced by a system with discrete equally - spaced energy levels . this introduces a small error in the results , but this can be made as small as desired by appropriately scaling the spacing of the weight . ] . for simplicity , we choose @xmath9 , such that the value of @xmath7 directly denotes the work stored by the mass . such a system has a long history of being used as a work storage system in classical thermodynamics @xcite . we will place two additional constraints on the allowed dynamics governing interactions with the weight : 1 . the average amount of work extracted in an allowed protocol must be independent of the initial state of the weight.[i : i ] 2 . all allowed unitaries should commute with translation operations on the weight . this reflects the translational symmetry of the weight system , and the fact that only displacements in its height are important.[i : ii ] these constraints are intended to prevent us from ` cheating ' by using the weight for purposes other than as a work storage system ( e.g. as a cold reservoir , or a source of coherence ) . for example if the weight starts in a pure state , it has zero entropy , so one needs to be careful to avoid unwittingly using this as a source of free energy , in addition to the system . constraint ( i ) encapsulates this requirement . constraint ( ii ) is a bit more subtle . on the one hand , it is very plausible that it is a consequence of constraint ( i ) . we however have not been able to show this . on the other hand , even if it does not follow from ( i ) , we would still like to assume it , as it seems wrong to use properties of the weight in the protocol , even if they do not affect the work extracted . furthermore , constraint ( i ) ensures that we can use the same work storage system for multiple thermodynamics protocols ( or on several copies of the same state ) without having to worry how the initial state has been modified by earlier procedures . as we are free to choose any initial state of the weight due to constraint ( i ) above , for simplicity we will usually take the initial state of the weight to be a narrow normalised wavepacket @xmath10 centred on the origin ( i.e. @xmath11 ) . when the same wavepacket is raised above the origin by a distance @xmath12 , we will denote the state by @xmath13 , where @xmath14 is the translation operator , where @xmath15 is the usual momentum operator satisfying @xmath16 = i \hbar$ ] . ] . we now show that the second law of thermodynamics holds in our framework , by proving a probabilistic version of the second law that there is no protocol which extracts a positive quantity of average work from a thermal bath ( i.e. that there is no way on average of turning heat into work ) @xcite . to show this we will use proof by contradiction . consider a thermal bath at temperature @xmath0 and the weight , i.e. there is no additional system out of thermal equilibrium . let us first consider the energy changes during the protocol . suppose that we are able to extract work from the bath and store it in the weight , @xmath17 . the average energy of the thermal bath must change by @xmath18 due to energy conservation . now consider the entropy changes during the same protocol ( in particular , the changes in von neumann entropy @xmath19 . as the bath and weight are initially uncorrelated , their initial entropy is simply the sum of their individual entropies . unitary transformations conserve the total entropy , @xmath20 . however , as correlations can form during the protocol , the sum of the final entropies of the bath and weight can be greater than the sum of their initial entropies ( as the entropy is subadditive ) . this means that @xmath21 given an initial thermal state for the bath ( with positive temperature ) , any change of the state which reduces its average energy must also reduce its entropy ] , @xmath22 . however , within our framework all allowed protocols are such that the average work extracted is independent of the initial state of the weight ; we are therefore free to choose any initial state of the weight we like . consider now that the initial state of the weight is a very wide wavepacket . we show in appendix [ a : second_law ] that the entropy change of the weight in this case can be made as small as desired , in particular we can make @xmath23 . this would result in violating . hence there is a contradiction , and thus there is no way to extract work from the bath . we will begin our exploration of work extraction by considering the special case of extracting work from systems whose states are diagonal in the energy eigenbasis . this represents systems which are essentially classical , since coherence plays no role , and states are simply mixtures of energy eigenstates . we will present a protocol which extracts work equal to the change in free energy of the system . in contrast to existing protocols , here our protocol will take one copy of the system , and extract average work equal to the free energy of the system . the protocol contains at its core a single elementary building block which is able to extract a vanishing amount of work blue , with maximal efficiency . the full protocol consists of many repetitions of this one step . we shall present first this building block in a motivating example and then show precisely how it is used to build the full protocol . in the main text the emphasis will be on the results of the protocol , with the more detailed analysis provided in the appendices . let us consider the task of performing the following transformation on a single qubit @xmath24 whose hamiltonian is @xmath25 and where @xmath26 . that is , we wish to consider the small transformation from a diagonal state to a nearby diagonal state . the initial and final state differ in free energy , and hence the goal is to extract an amount of work as close as possible to this difference , namely @xmath27 where @xmath28 is the free energy of @xmath29 with respect to a bath at inverse temperature @xmath30 , @xmath31 is the average energy , @xmath32 is the ( classical ) binary entropy , @xmath33 , and we have taken the first order expansion in @xmath34 to arrive at the final line . we shall now see that to first order we can extract exactly this amount of work from the system . to do so let us bring in a thermal state from the bath the free resource at our disposal and choose its hamiltonian such that the population in the excited state is @xmath35 . for thermal qubits the relation between excited state population @xmath36 and energy spacing @xmath37 is @xmath38 thus we will choose the hamiltonian of the bath qubit to be @xmath39 with @xmath40 let us take the weight to be prepared initially in the state @xmath41 of zero average energy , and consider applying an energy conserving unitary transformation @xmath42 which interchanges @xmath43 for any @xmath44 , whilst leaving all orthogonal states unchanged can be expressed in terms of the translation operator @xmath14 as @xmath45 where @xmath46 . note that this commutes with the free hamiltonian @xmath47 . ] . it is the change in average energy of the weight which is of interest . a straightforward calculation ( given for completeness in the appendix ) shows that this is given by @xmath48 which is equal to @xmath49 to first order in @xmath34 . thus we observe that if we wish to to make a change of state from @xmath50 to @xmath51 , a way to proceed is to bring in a qubit from the bath in the state @xmath52 . whenever @xmath50 is not the thermal state ( given it s hamiltonian @xmath53 ) , we find that @xmath54 , and therefore there is a difference in energy between the excited states of the two qubits . it is thus not possible to swap the state of the two qubits using an energy conserving unitary . however , by bringing in the weight we can perform the unitary which is energy conserving , and , at the level of the reduced states of the system and bath , indeed performs a swap . the system therefore ends up in the desired state @xmath51 , and we find that the change in average energy of the weight , when the state of the system and bath are close , approaches exactly the change in free energy of the system . the above insight can be used to build a full protocol which is able to extract an amount of work approaching the free energy change in transforming a system from a state @xmath50 to its thermal state @xmath55 . the basic idea is that , since in making a small change of state we have seen that the work extracted approaches the free energy change , we can divide a large change of state into many small changes and at each stage extract the desired amount of work . therefore , let us fix an integer @xmath56 , and consider @xmath57 thermal qubits @xmath58 , @xmath59 , @xmath60 where @xmath61 is the excited - state probability in the thermal state @xmath55 . this collection of bath qubits have excited - state probabilities which vary linearly from that of the initial state of the system @xmath62 to that of the thermal state of the system @xmath55 , in increments of @xmath63 . we will apply , in turn , the sequence of unitary transformations @xmath64 , where @xmath65 acts on bath qubit @xmath66 , system and weight and swaps the states @xmath67 whilst leaving all other orthogonal states unchanged . assuming once again that the weight starts in the state @xmath41 , it is the final state , and in particular the final average energy of the weight which is of interest , as the number of steps @xmath57 tends to infinity . in appendix [ a : weight ] we calculate explicitly this state in the asymptotic limit . the result is that the final reduced state in the infinite limit , denoted @xmath68 , is @xmath69 the average energy is thus @xmath70 where @xmath71 is the relative binary entropy . we observe that the average energy of the weight at the end of the protocol , and hence the average work extracted , approaches precisely the free energy difference in the asymptotic limit of infinitely many thermal qubits . although this is promising , it should be noted that the final distribution of the weight does not have the form one may intuitively have asked for ( i.e. a peak around a single value , at the free energy itself ) . instead we find that the solution has perfect correlation between the initial state of the system and the final energy of the weight . it is worth noting that this interesting distribution coincides with the results found in @xcite where , even though the details of the allowed transformations differ , it was shown that for classical systems , optimal work extraction protocols converge on this distribution in probability . note also that in the large @xmath57 limit , although many thermal qubits are required , each of them is only slightly perturbed by the process . finally , we consider the case of @xmath1 independent copies of the system i.e. the macroscopic limit . to do so we simply repeat the protocol above on each qubit separately , using a single weight to store the work . in each case we have seen that the weight is transformed from a single peak ( at @xmath41 above ) to two peaks . since the protocol is linear , it follows immediately the the final distribution of the weight will be the @xmath1-fold convolution of the above distribution for a single qubit . from the central limit theorem one concludes that the final average energy of the weight is @xmath72 with fluctuation of the order @xmath73 . for large @xmath1 these fluctuations will be insignificant compared to the average , and we obtain an almost deterministic quantity of work , with the amount of work extracted per copy of the system equal to the free energy difference . the results presented here are completely consistent with those presented in @xcite . it should be noted however the stark difference in which the two protocols achieve the same goal . in @xcite the analysis first considers all @xmath1 systems together and uses ideas from information theory to compress information within large typical subspaces . here , on the other hand , each qubit is manipulated separately and nowhere is the idea of a typical subspace ever introduced , nevertheless they both achieve the same end result . the protocol outlined in the previous subsection is the general protocol which allows for the extraction of work from an arbitrary diagonal state of a qubit , for any given hamiltonian . there is one particular choice of state that will be particularly important in later sections . as such , we shall briefly look at it in more detail . consider a qubit for which the hamiltonian vanishes ( @xmath74 ) and for which all the probability is initially in the state @xmath75 ( @xmath76 ) . since the hamiltonian vanishes , the system lives in a degenerate space , and since it is pure we have maximal information about the state of the system . this information can be used to extract work from the heat bath . since all states of the system have equal energy , it is clear that the system provides only information , whilst it is the bath which provides the energy . for a degenerate system the equilibrium state @xmath77 , thus after applying the protocol from the previous section , in the asymptotic limit of infinitely many bath qubits , the final state of the weight is given by @xmath78 corresponding to the extraction of an amount of work @xmath79 from the system . this is an analogous process to the isothermal expansion of a single - molecule gas from one side of a box into the full volume , as discussed in szilard s well - known treatment of maxwell s demon @xcite . it should be noted that contrary to the previous subsection , where the final state of the weight was found to consist of two peaks , in this special case of a pure initial state it consists of only a single peak . thus it is not necessary to go to the many - copy regime to extract a deterministic amount of work in this case . to conclude the analysis of diagonal systems it is now necessary to move beyond qubits , to quantum systems of arbitrary finite dimension @xmath80 , and arbitrary hamiltonians . our strategy is to connect a pair of energy levels in the system to a sequence of thermal qubits in the bath , shifting the occupation probability between these levels as in the qubit case ( sec . [ ss : full protocol ] ) . we then repeat this process for different pairs of levels until we have the desired final state . as the amount of work extracted in each step does not depend on the initial state of the weight ( as proven in appendix [ a : work_independence ] ) , we can consider each step separately and simply sum the total work extracted . an important difference from the qubit case is that the occupation probability of the two levels in the system that are coupled to the bath at any given stage need not sum to unity . for example consider a qutrit ( or larger system ) in the state @xmath81 where we wish to shift a small amount @xmath34 of probability from @xmath75 to @xmath82 . to do this , we take a thermal qubit with excitation probability @xmath83 , so that the _ ratio _ of occupation probabilities in the bath qubit is the same as in that in the desired final state of the system . we then apply the unitary @xmath84 where @xmath85 is the energy of the state @xmath82 , and all orthogonal states are left unchanged . we show in appendix [ a : qudits ] that the change in the average energy of the weight during this process is equal to the free energy change of the system , to first order in @xmath34 . from this it is straightforward to extend the qubit protocol to arbitrary diagonal states . for sufficiently large @xmath57 we can choose a sequence of @xmath86 states for the system where two adjacent states differ in occupation probability between a pair of levels by @xmath87 , with the first and last states equal to the initial state of the system and its thermal state respectively than in @xmath55 to the @xmath75 state , then move probability from @xmath75 to the remaining levels , using @xmath88 steps for each pair of levels . ] . we then pick the hamiltonian for @xmath57 thermal qubits such that unitaries of the form take the reduced state of the system from from one state in the sequence to the next . this is done by taking the thermal qubits to have populations equal to those of the pair of levels of interest of the next state in the sequence , after renormalising . in the limit @xmath89 the average amount of work put into the weight equals the free energy change of the system , regardless of the precise choice of path . interestingly , an equivalent protocol would work for transitions between any two diagonal states , @xmath50 and @xmath29 , extracting an average amount of work equal to @xmath90 . this implies that creating a diagonal state @xmath50 from a thermal state costs the same average amount of work as can be extracted from @xmath50 when it is thermalised . in this sense thermodynamic transitions between diagonal states are reversible . this differs from the results of @xcite , who show that such transitions are irreversible when considering ( almost ) deterministic work extraction , rather than average work . we now show that the above protocol is optimal . in particular , we will argue that the maximum free energy that can be extracted from a transition between diagonal states @xmath50 and @xmath29 is @xmath90 , using a modification of our proof of the second law ( in section [ s : second_law ] ) . suppose that we have a protocol which extracts average work @xmath91 ( where @xmath92 ) when the system is transformed from @xmath50 and @xmath29 . we can then use the above protocol to return the state from @xmath29 to @xmath50 in a finite number of steps , extracting work @xmath93 , where we choose the number of thermal qubits such that @xmath94 is in the range @xmath95 . the net effect is that a positive average work @xmath96 is extracted into the weight , and the system begins and ends the combined procedure in the same state @xmath50 . considering total energy conservation for the combined protocol , we note that the average energy of the bath must have decreased by at least @xmath97 , which means that the bath s entropy must also have decreased by a finite amount . however , the system does not change in entropy , and the change in entropy of the weight can be made as small as desired by choosing its initial state to be a sufficiently broad pure state ( see appendix [ a : work_independence ] ) . hence entropy conservation can not be satisfied by our hypothetical protocol , and thus no protocol exists which extracts more average work than the free energy change . so far all of our discussions have involved only states which are diagonal in the energy eigenbasis and the protocol presented involved shifting probability only between these energy eigenstates . we can of course consider more general states , and ask the question of how to extract the maximum amount of work from such states . as a motivating example , let us consider a system with hamiltonian @xmath98 prepared initially in the pure state @xmath99 this state has the same average energy as the thermal state @xmath55 , however , since this state is pure , it has zero entropy and hence its free energy differs from that of the thermal state . thus one would like to extract work from this state , with the limit on extractable work given by @xmath100 previously we were able to extract work from single copies of diagonal states . however , we show in the next section that it is impossible to extract work from a single copy of @xmath101 given the assumptions of the framework . this is similar to the analysis of @xcite where they also found that the state is useless for ( almost ) deterministic work extraction . however , it is only a single copy which is a useless resource ; two copies can be used to extract work , although the amount is less than that given by ( [ e : opt work ] ) . the important point is the way in which work is extracted , which will suggest a different way in which one can understand an asymptotic protocol ( in the number of system qubits ) able to extract the optimal amount of work . consider two copies of the state @xmath101 , which written out in full is given by @xmath102 let us consider that we now dephase this state in the energy eigenbasis , so called collective - dephasing . we can achieve this in two ways , either by allowing the state to evolve under its hamiltonian freely for some unknown duration of time , or more systematically by applying controlled-@xmath103 operations between maximally mixed ( therefore degenerate ) bath qubits and the system . note that both of these are energy conserving unitaries . in either case , at the end of the dephasing the state @xmath50 is @xmath104 where @xmath105 . we note that since @xmath106 and @xmath107 have the same energy , the definite phase between them is preserved . the important point is that when considering the combined system we find that we have a two - dimensional degenerate subspace and within this subspace we are in the pure state @xmath108 , and _ not _ the orthogonal state @xmath109 . thus we have knowledge that we are with certainty in one of two orthogonal states and we can use this information to extract work from a thermal bath by allowing it to ` isothermally expand ' into the full subspace . randomising these two states brings the pair of qubits to their thermal state . we can thus apply the analogue of the protocol in sec . [ ss : maxwell demon ] , coupling only the two states inside the degenerate subspace to the thermal bath , and it follows immediately that the final average energy of the weight , in the asymptotic limit , is @xmath110 except for the limiting case @xmath111 , corresponding to the unphysical condition @xmath112 , this value is strictly less than the change in free energy of two system qubits , @xmath113 this inequality is due to the fact that the collective - dephasing changed the entropy of the system but extracted no work in the process . one may wondering therefore why this operation is considered in the first place . in appendix [ a : coherence ] we show that we can apply the same analysis , including the collective - dephasing , to @xmath1 copies of @xmath101 and extract , in the asymptotic limit @xmath114 , work equal to the free energy change @xmath115 . this is possible since , as we also show , collective - dephasing increases the entropy of @xmath1 systems by an amount which grows as @xmath116 , which becomes negligible in the asymptotic limit . we thus see that coherence between states of different energy in fact plays no thermodynamic role asymptotically in work extraction . altogether this highlights an essential difference between diagonal and non - diagonal states ; when @xmath117 , the state of multiple qubits in a diagonal state will be maximally mixed inside each energy eigenspace . however , for non - diagonal states , these subspaces take on non - trivial states , and work can be extracted from them by allowing them to ` isothermally expand ' into the full subspace . we now present a general strategy to extract work from @xmath1 copies of an arbitrary state @xmath50 , of arbitrary dimension @xmath80 . it is helpful to divide the protocol into three steps : 1 . collectively dephase the state @xmath118 in the joint energy eigenbasis . [ i : dephase ] 2 . allow the resulting state to ` isothermally expand ' inside each energy eigenspace , until it equals @xmath119 , where @xmath120 is the state obtained by dephasing @xmath50 in its energy eigenbasis . [ i : expand ] 3 . individually transform each copy of @xmath120 into @xmath55 . [ i : transform ] it follows from the results of sec . [ ss : qudits ] that steps ( [ i : expand ] ) and ( [ i : transform ] ) can be performed whilst extracting an average amount of work equal to the free energy change . the free energy change in the first step is not captured as work , but is simply lost . however , we show in the appendix [ a : coherence ] that the amount of work lost in this step is at most @xmath121 . the average amount of work extracted per copy of the original system is therefore @xmath122 in the limit as @xmath123 , we therefore extract an average work per copy of @xmath50 equal to its free energy change , as desired . if we only have a single copy of the system , the amount of work we can extract from @xmath50 is given by the free energy change of the decohered state , @xmath124 . this can be obtained by first dephasing @xmath50 into @xmath120 and then using the protocol in sec . [ ss : qudits ] . we now show that this is optimal , following the approach given in @xcite . firstly , note that as we require the average work extracted by a protocol to be independent of the initial state of the weight , we are free to choose that state however we like here we choose it to be a very narrow wavepacket centred on zero . now consider a decomposition of the total state space into subspaces , each of which has total energy ( of the system , bath and weight ) close to one of the energy eigenvalues of the system and bath subspace corresponds to the total energy @xmath125 lying in the range @xmath126 , where @xmath127 are the energy eigenvalues of the system and bath . furthermore we choose the width of the weight s initial wavepacket to be narrower than the smallest subspace . ] . note that any work extraction protocol can be followed by a transformation which decoheres the state with respect to these total energy subspaces , without affecting the average work extracted . however , this decohering operation commutes with the unitaries used in the protocol , so we can move it to the beginning of the protocol without changing the work extracted . at the beginning of the protocol , this operation has the sole effect of decohering the system in its energy eigenbasis ( changing @xmath50 to @xmath120 ) . hence the protocol extracts the same amount of work as it would have if it had operated on @xmath120 . consider the model of the smallest possible heat engines @xcite , consisting of two qubits , one with energy spacing @xmath128 in thermal contact with a bath at temperature @xmath129 , and a second with energy spacing @xmath85 in contact with a bath at @xmath130 . as a composite system this pair of qubits is not in a global thermal state . nevertheless , let us assign a ` virtual temperature ' to each transition via the identification @xcite @xmath131 where @xmath132 and @xmath133 are the probability of being in the ` excited ' state ( the state of higher energy ) and ground state ( the state of lower energy ) respectively , with @xmath134 and @xmath135 being their energy eigenvalues , and @xmath136 is the ` inverse virtual temperature ' . for the case in hand , the virtual inverse temperature for the states @xmath137 and @xmath138 is @xmath139 where we have denoted the probability for qubit @xmath140 to be in the excited state by @xmath141 , and we have used the relation @xmath142 for a thermal state . the motivation for introducing the concept of virtual temperatures is that for the smallest machines it gives a concise characterisation of their behaviour ; the function of the machine is determined by the virtual temperature refrigerator , heat pump or heat engine , if colder , hotter , or negative temperature respectively ; carnot efficiency is approached when the virtual temperature approaches the external temperature for the refrigerator and pump , and when approaching minus infinite temperature for the heat engine . it is the final machine mentioned , the heat engine , which is relevant here , since this is the machine which produces work . we can think of the system qubit and bath qubit together to be a two - qubit heat engine and that the operating point of the protocol is the carnot point of the engine . let us return to the motivating example of section [ s : motivating example ] . our system qubit had spacing @xmath143 with excited state probability @xmath144 , whilst the bath qubit had spacing @xmath37 with excited state probability @xmath145 . let us rename the state @xmath146 and @xmath147 . with this renaming , the interaction becomes @xmath148 where @xmath149 is the spacing of the virtual qubit . thus from this perspective it is explicit that the unitary is the one which exchanges energy from the virtual qubit into the weight , analogous to the hamiltonian of the smallest heat engine @xcite . the operating point of the protocol was to choose @xmath37 such that @xmath150 . in this regime , to first order , the inverse virtual temperature becomes @xmath151 where @xmath152 . thus the inverse virtual temperature becomes vanishingly small and negative as @xmath153 , which is exactly the carnot limit of the small heat engine . the picture that one obtains from this perspective is that the reason why any state which is not a thermal state at temperature @xmath0 is a resource is because the _ composite system of resource state + thermal state _ contain non - trivial virtual temperatures . one can extract work from a system approaching its free energy when the only exchanges between the composite system and weight involve virtual qubits whose inverse temperatures become negatively infinite . finally , we conclude by noting the protocol for systems with many energy levels becomes easier to explain using this perspective . a non - thermal state will have transitions ( i.e. pairs of levels ) at many different temperatures , and each pair needs to be bought to the same temperature in the thermalisation process . in the protocol the bath qubits interact only with a single pair of levels at a time . to approach reversibility it must be the case that any systems which interact only differ in temperature by an infinitesimal amount , but this is exactly the condition enforced by choosing the renormalised probability of the pair to be close to the bath qubit populations . finally , the unitary which is applied is the basic thermalising unitary and the energy exchanged in the limit is precisely the free energy . to summarise , in this paper we have analysed the fundamental laws of thermodynamics using the recently proposed concept of resource theories . we proposed an alternative resource theory which differs in certain aspects from the original one . in this conceptual framework we have proved a number of results . first we proved the second law of thermodynamics . second we proved that for systems that are diagonal in the energy eigenbasis , but not at equilibrium with the thermal bath , we can extract work equal to the free energy by processing each system _ individually _ , hence showing the free energy is a relevant quantity for individual systems . third , we proved that when systems are not diagonal in the energy eigenbasis , i.e. they have coherences between different energy eigenstates , it is not possible to extract work equal to the free energy by processing them individually . work equal to the free energy can however be extracted if we allow for collective processing of such systems ( in the asymptotic limit of many copies ) . a number of important questions open immediately . first of all , whether the assumptions which underlie our resource theory framework are too restrictive . in particular , this may be the reason why we can not extract the entire free energy by processing single systems with coherences between different energy levels . indeed , given our assumptions we are very limited in how we can access and manipulate the relative phase between two different energy levels , hence we can not interfere them in a controlled way . as such , for all practical purposes the different energy levels behave as if they were decohered ; it is precisely the free energy of this decohered state that is the maximum we can extract in our framework . to be able to interfere different energy levels , i.e. to control their relative phase , we need a frame of reference " for it . one way to look at this is that we need some other resource with coherent superpositions between energy levels . another way to express this is that we essentially need a clock : the phase between two energy levels varies in time , so a frame of reference " for it is essentially a clock . when collectively processing many copies the different systems act effectively as reference systems for each other . for individual processing however we essentially need to add an external clock . in our present framework we did not include such a clock since we were not able to guarantee that one can not unwittingly cheat by using the clock itself as a source of free energy . it is conceivable however that a more careful formulation of the resource theory would permit integrating a clock in a consistent and controlled manner . in this case , it may be possible that work equal to the free energy can be extracted from all quantum states by only processing individual copies . second , while in the above we asked if we can relax some of the constraints imposed in the resource theory , for example by including a clock , we must also enquire whether we may indeed need to strengthen some of the constraints . in particular , one may be concerned that we have used unitary transformations as opposed to hamiltonian evolution . unitaries imply controlled evolution , i.e. turning on and off interactions at precise time intervals . one may wonder whether this does not require some free energy to achieve . if this is the case , then it is only an illusion that we have extracted work equal to the free energy from the system , since we had to pay to implement the process . although such a detailed analysis is missing , the fact that all of the results obtained within the resource theory framework so far are consistent with our expectations is an implicit indication that the assumptions are indeed justified . to conclude , the resource theory framework seems to be a natural and powerful way to approach thermodynamics . it has already delivered significant results and we believe that further investigation along these lines will lead to a much deeper understanding of the foundations of thermodynamics . sp and ps acknowledge support from the european project ( integrated project q - essence " ) . sp acknowledges support from the erc ( advanced grant `` nlst '' ) , and the templeton foundation . ajs acknowledges support from the royal society . ps is grateful to lidia del rio , johan berg , joe renes , philippe faist , renato renner , oscar dahlsten and luis masannes for interesting discussions . in each step of our protocols in which work is extracted , the state of a thermal qubit is swapped with a pair of energy levels in the system , with a corresponding translation of the weight in order to conserve total energy . consequently , each work - extraction step can be represented by a unitary transformation of the form @xmath154 where the states @xmath155 are an orthonormal basis of energy eigenstates for the system and the relevant portion of the bath , such that @xmath156 , @xmath14 is the translation operator on the weight , and @xmath157 is a permutation matrix . furthermore , apart from the initial decohering of non - classical states ( discussed in section [ s : coherence ] ) , which is entirely independent from the weight , the entire protocol can be cast in this form . it is easy to see that all such unitaries commute with translations on the weight . we now show that the average work extracted by a unitary of this form does not depend on the initial state of the weight ( even if it is initially correlated with the state of the system ) . let us denote the initial state of the system , bath and weight by the density operator @xmath50 . the average work extracted is given by @xmath158 where @xmath159 is the hamiltonian of the weight ( defined for convenience as an operator on the entire system ) . now note that @xmath160 where in the third line we have used the fact that for permutation matrices @xmath161 ( where @xmath162 is the kronecker delta function ) , and in the fourth line we have used the fact that @xmath163 . inserting this expression in ( [ eq : workindependence ] ) , we obtain @xmath164 where @xmath165 is the reduced density matrix of the system and bath . hence the average amount of work extracted is independent of the initial state of the weight as desired . note that the most general energy - conserving unitary transformations which commute with all translations on the weight have the form @xmath166 where @xmath167 is an arbitrary unitary matrix . however , for such unitaries the work extracted would generally depend on the initial state of the weight . in this appendix , we provide further details for our proof of the second law in section [ a : second_law ] . we first argue that any reduction in the average energy of an initially thermal state ( with positive temperature ) must also yield a reduction in its entropy . we will only need to use the fact that a thermal state is the maximal entropy state with a given average energy . if a system starts in a thermal state and is transformed to a final state with fixed average energy , the entropy change @xmath168 will be maximised when the final state is also thermal . in the case where the average energy decreases , and the initial state has positive temperature , the final state will be a thermal state of lower temperature , and thus lower entropy . we now show that the entropy change of the weight in any particular protocol may be made as small as desired by choosing its initial state to be a sufficiently broad pure state ( e.g. a state @xmath169 with wavefunction @xmath170 equal to @xmath171 for @xmath172 , and 0 otherwise , where @xmath173 ) . first note that the requirement that transformations are translationally invariant on the weight means that they have the form given in ( [ e : general_unitary ] ) . when applied to a thermal state of the bath , the effect of the protocol on the weight is to transform it into a finite mixture of translations of the initial state @xmath174 , where @xmath175 are probabilities and @xmath140 and @xmath176 label energy eigenstates ( of energies @xmath127 and @xmath177 ) of the bath . if the thermal systems used in the protocol have total dimension @xmath80 , and the initial state of the weight is pure , then the final state of the weight lies in a @xmath178-dimensional subspace spanned by the states @xmath179 for @xmath180 . as the final state of the weight lives in a finite dimensional subspace , its entropy can be shown to be continuous due to fannes inequality @xcite . @xmath181 where @xmath182 is the trace distance between @xmath183 and @xmath184 . we can make @xmath182 as small as we like by taking sufficiently large @xmath185 and therefore make the entropy change as small as we like . for the case in hand , a straightforward calculation shows that @xmath186 , where @xmath187 is the maximum energy gap of the system - bath hamiltonian . note that all of the above arguments apply equally well if we consider a system in addition to the bath which is initially in a mixture of energy eigenstates , as considered in section [ s : optimality ] . in this appendix we shall provide more details into the example of section [ s : motivating example ] , where a state undergoes a transformation and in the process supplies an amount of work @xmath188 . consider the state @xmath189 with hamiltonian @xmath25 , along with a thermal qubit @xmath190 with hamiltonian @xmath191 , where @xmath192 . let us bring in a weight in the state @xmath41 , and apply the unitary transformation @xmath42 , which interchanges @xmath193 whilst leaving all other orthogonal states unchanged . applying this unitary to the combined system we find that @xmath194 from this one can finds that the final reduced state of the system and bath qubit are @xmath195 and @xmath196 respectively . thus we see that at the level of reduced states these two systems have swapped states . the change in free energy of the system @xmath197 is @xmath198 moving on to the weight its final reduced state @xmath199 is @xmath200 the change in average energy of the weight @xmath201 is thus @xmath202 now , let us take @xmath203 and expand to second order in @xmath34 both @xmath197 and @xmath201 , @xmath204 the change in free energy and the change in average energy of the weight thus coincide to first order in @xmath34 and differ only by a factor of 2 in the second order term . since @xmath205 , we see immediately that @xmath206 , so that the change in free energy is always larger than the change in average energy of the weight , as we would expect . in this appendix we will show how to arrive at equation for the asymptotic state of the ladder . to do so let us consider first those instances where the system qubits starts in state @xmath75 . given this we shall then calculate the expected energy of the weight and its variance in the limit of large @xmath57 to first order , from which we shall be able to conclude that it approaches a definitive value in the asymptotic limit at the desired energy . a analogous method can also be employed assuming the system starts instead in the state @xmath207 , with the desired result also obtained . by calculating the first few stages of the protocol and then by induction it is easy to show that the expected energy of the weight at the end of the protocol is @xmath208 } = -p_{eq } e_s + p e_b^{(1 ) } + \sum_{k=1}^n ( q^{(k ) } - q^{(k-1)})e_b^{(k)}\ ] ] where the superscript @xmath209 $ ] is a reminder that this is the case for the system starting in the state @xmath75 , and as defined in the main text we have @xmath210 by expanding the logarithms in powers of @xmath211 and denoting @xmath212 , after some basic manipulations this can be re - written as @xmath213 } = -p_{eq } e_s + p e_b^{(1 ) } + \tfrac{a}{\beta}\log\left(\tfrac{1-p}{p}\right ) \\ + \frac{a}{\beta n}\sum_{k=1}^n \sum_{n=1}^{\infty } \left(-\frac{1}{n}\left(\frac{ak}{(1-p)n}\right)^n + \frac{1}{n}\left(\frac{-ak}{pn}\right)^n \right)\end{gathered}\ ] ] the advantage of performing this expansion is that we can now interchange the order of summation and note that the only term which depends upon @xmath66 is the term @xmath214 which is common to both terms in the sum . we can thus use the expansion @xmath215 to write @xmath216 } = -p_{eq } e_s + p e_b^{(1 ) } + \tfrac{a}{\beta}\log\left(\tfrac{1-p}{p}\right ) \\ + \frac{1}{\beta } \sum_{n=1}^{\infty } \left(\frac{a^{n+1}}{n}\left(\frac{1}{(-p)^n}-\frac{1}{(1-p)^n } \right)\left(\frac{1}{n+1 } + \frac{1}{2n}\right)\right ) \\+ \mathcal{o}\left(\tfrac{1}{n^2}\right)\end{gathered}\ ] ] this sum can now be evaluated explicitly , and after re - expressing @xmath143 in terms of @xmath217 , expanding @xmath218 to first order in powers of @xmath219 , and an amount of manipulation we arrive at @xmath213 } = \tfrac{1}{\beta}\log\left(\tfrac{1-p}{1-p_{eq}}\right ) + \tfrac{1}{\beta}(p_{eq}-p)\big(\tfrac{1}{2}\big(s_c'(p_{eq})\\-s_c'(p)\big ) -\tfrac{1}{1-p}\big)\tfrac{1}{n } + \mathcal{o}\left(\tfrac{1}{n^2}\right)\end{gathered}\ ] ] thus demonstrating that the weight is peaked around the point expected when the system starts off in the state @xmath75 with an offset which vanishes in the asymptotic limit . considering the case now when the system starts off in the state @xmath207 and again calculating the first few stages of the protocol and then by induction it can be shown that @xmath220 } = { \langlee\rangle}_w^{[0 ] } + e_s - e_b^{(1)}\ ] ] thus by re - expressing @xmath143 in terms of @xmath217 and expanding @xmath218 to first order it is straightforward given the previous results to show that @xmath221 } = \tfrac{1}{\beta}\log\left(\tfrac{p}{p_{eq}}\right ) + \tfrac{1}{\beta}(p_{eq}-p)\big(\tfrac{1}{2}\big(s_c'(p_{eq})\\-s_c'(p)\big ) + \tfrac{1}{p}\big)\tfrac{1}{n } + \mathcal{o}\left(\tfrac{1}{n^2}\right)\end{gathered}\ ] ] which again gives the desired result . let us move on now to a calculation of the variance @xmath222 } = { \langlee^2\rangle}_w^{[i ] } - { \langlee\rangle}_w^{2[i]}$ ] of each of the two packets , where for simplicity we will neglect the finite width of the initial wavepacket . again by induction it can be shown that the spread of both packets is identical ( and hence we shall drop the superscript @xmath223 $ ] ) and equal to @xmath224 let us consider the @xmath225 term in this sum , multiplied by @xmath57 . defining @xmath226 and expanding this summand in terms of @xmath219 we find @xmath227 thus for large @xmath57 the leading order contribution to this sum becomes equal to @xmath228 evaluating this integral explicitly we find finally that @xmath229 thus we see that each peak has a spread of the order of @xmath230 which vanishes in the asymptotic limit @xmath231 . therefore in this limit we can conclude that the distribution of the ladder approaches a two - peaked distribution , as given by equation . in this appendix we wish to show that the average work extracted when moving the occupation probability between energy levels is still equal to the free energy in a @xmath80-dimensional system . it will suffice to consider a qutrit , and moving probability @xmath34 from the ground state @xmath75 to the second excited state @xmath82 . let us take our system to be initially in the state @xmath81 and thus to end in the state @xmath232 where the system hamiltonian is @xmath233 to achieve this , we append a bath qubit with energy level spacing @xmath37 and ground state population @xmath36 and apply the unitary transformation which interchanges only the states @xmath234 this induces on the system and weight the transformation @xmath235 where we now carry out the final trace over the bath qubit for brevity of presentation . thus in order that the final reduced state of the system is @xmath236 we find that the ground state population of the bath must be chosen to be @xmath237 the denominator is the total probability to be in either of the states @xmath75 or @xmath207 , and hence this expression can be seen as a renormalised version of the qubit case , where the total probability to be in either state always sums to unity . we see that it is the ratio of probabilities which must be close to the ratio of populations of the thermal qubits , which is equivalent to saying that they must be close in temperature ( see sec . [ s : virtual ] ) . the change in energy of the weight is given by @xmath238 thus using the fact that @xmath239 and expanding to first order in @xmath34 we find that the work extracted from the system is given by @xmath240 which a straightforward calculation verifies is equal to the change in free energy @xmath241 to first order in @xmath34 . if we do a sequence of @xmath57 such transitions between the states @xmath50 and @xmath29 , in each of which @xmath242 , then we will obtain @xmath243 . in the asymptotic limit in which the number of thermal qubits @xmath57 goes to infinity , we will therefore extract average free energy equal to the free energy change . in this appendix we will present the full details for how work can be extracted from coherence . we shall first consider the special case presented in the main text for @xmath1 systems , and then move on to the general case . consider @xmath1 copies of @xmath101 , where @xmath244 there will be @xmath245 energy eigenspaces with energies @xmath246 , @xmath247 . the dimension of the eigenspace with energy @xmath246 will be @xmath248 . after collective - dephasing in each eigenspace the system will be a pure superposition over all states , and the probability to be in each eigenspace will be @xmath249 . we then cause the state to ` isothermally expand ' in each degenerate subspace , taking the state from the initial pure state to the final maximally mixed state and in the process extracting @xmath250 energy from the bath , where @xmath80 is the dimension of the subspace . in total , the final average energy of the ladder will be @xmath251 in the asymptotic limit , as @xmath252 it is straightforward to show that @xmath253 since in the limit the only contribution to the sum comes from @xmath254 . we thus see that asymptotically the expected amount of work is extracted . there are a couple of comments which are worth mentioning . first we note that due to concentration effects , we know that almost all weight lies in those subspaces with energy @xmath255 . we could therefore ignore those subspaces which lie outside this range and obtain , to leading order , the same amount of work . the results obtain however are independent of the concentration effects . second we recall that the protocol starts with collective - dephasing in the energy eigenbasis . it might appear that this operation , which changes the entropy of the system , throws away work . however , asymptotically we obtain to first order the optimal amount of work , and so we conclude that the amount thrown away must scale sub - linearly in @xmath1 . in fact the amount of entropy thrown away is only @xmath256 , the entropy of the binomial distribution , since the dephasing leaves the system binomially distributed amongst energy eigenspaces which have orthogonal support . this shows that asymptotically the coherence between states of different energy are thermodynamically insignificant . finally , for the case of arbitrary states of arbitrary dimension @xmath80 , we note that the same conclusion can also be drawn . first we need to show that the operation of collective - dephasing never increases the entropy of a collection of systems too much in the asymptotic limit . let us denote the projectors onto the energy eigenspaces by @xmath257 , and the total number of different energies by @xmath258 . after the operation of collective - dephasing the state becomes block diagonal in the energy eigenbasis , namely @xmath259 we are interested in the difference in entropy @xmath260 , i.e. the entropy generated by this operation . to do so , let us first consider the coherent version of collective - dephasing . let us introduce an ancillary system @xmath261 with @xmath258 states @xmath262 prepared initially in the state @xmath263 . let us also consider the unitary operation @xmath264 @xmath265 applied to the state @xmath266 this performs the transformation @xmath267 it is straightforward to see that @xmath268 whilst @xmath269 considering now the entropies , we have @xmath270 , since the two states differ by the appended pure ancilla and the joint unitary transformation . thus , from the araki - lieb inequality @xmath271 we have @xmath272 next , note that we can rewrite @xmath51 in the form @xmath273 where the sum is restricted to those @xmath66 for which @xmath274 . comparing this to ( [ e : rhoa ] ) , we note that @xmath275 , as a mixture of orthogonal mixed states always has higher entropy than the same mixture of orthogonal pure states . hence the absolute value can be omitted on the right - hand - side of equation and after rearranging we get @xmath276 next , note that @xmath277 , since in the worst case @xmath278 is maximally mixed over @xmath258 states . hence we find @xmath279 the number of different energy levels @xmath258 is upper bounded by the number of ways of distributing @xmath1 indistinguishable particles between @xmath80 distinguishable energy levels , which is given by @xmath280 . thus @xmath281 this demonstrates that the operation of collective - dephasing on @xmath1 qubits can only increase the entropy by a logarithmic amount in @xmath1 . the increase _ per qubit _ is thus of the order @xmath282 , which vanishes in the asymptotic limit . this result does not follow from any concentration of measure , but simply by the fact that the the number of energy eigenvalues of @xmath1 non - interacting systems is logarithmic in @xmath1 . in this appendix we will show that the energy level spacing of the weight need only become continuous in the asymptotic limit and that for any finite number of bath qubits @xmath57 it is sufficient to consider the levels to have a minimum spacing @xmath283 which scales as @xmath284 . let us begin by recalling how the protocol works . we pick a collection of @xmath57 bath qubits which have excited state probabilities @xmath285 given by @xmath286 and therefore excited state energies @xmath287 . at each stage of the protocol the weight either gains or loses energy @xmath288 . let us now assume instead that the of being a continuous degree of freedom that the weight has a discrete ( but infinite ) set of states @xmath289 and hamiltonian @xmath290 accordingly , we shall adjust the energy levels of the bath qubits to @xmath291 such that @xmath292 where the @xmath293 are integers , and we take the smallest possible @xmath294 possible such that this condition is met , meaning that @xmath295 . to proceed , we first note that can alternatively be re - written as @xmath296 } & = -p_{eq } e_s + p e_b^{(1 ) } + \sum_{k=1}^n ( q^{(k ) } - q^{(k-1)})e_b^{(k ) } \nonumber \\ & = \sum_{k=0}^{n-1 } q^{(k ) } ( e_b^{(k ) } - e_b^{(k+1)})\end{aligned}\ ] ] thus upon making the replacement @xmath297 for the case of a discrete weight , the average energy , conditional on the qubit starting in the ground state becomes @xmath298 } & = \sum_{k=0}^{n-1 } q^{(k ) } ( \tilde{e}_b^{(k ) } - \tilde{e}_b^{(k+1 ) } ) \nonumber \\ & = \sum_{k=0}^{n-1 } q^{(k ) } ( e_b^{(k)}\ ! -\ ! e_b^{(k+1 ) } ) + \sum_{k=0}^{n-1 } q^{(k ) } ( \varepsilon^{(k)}\ ! -\ ! \varepsilon^{(k+1 ) } ) \nonumber \\ & = { \langlee\rangle}_w^{[0 ] } + \sum_{k=0}^{n-1 } q^{(k ) } ( \varepsilon^{(k ) } - \varepsilon^{(k+1)})\end{aligned}\ ] ] it is the final term in the last line which is the error introduced by considering a discrete weight , and thus it is the magnitude of this term which is of interest . this can easily be bounded as follows @xmath299 therefore as long as the product @xmath300 as @xmath301 , then this term will asymptotically vanish . choosing @xmath302 would ensure that this correction is @xmath87 and thus the same order as the leading order correction to @xmath303}$ ] for finite @xmath57 . an identical analysis holds also for @xmath304}$ ] , the average energy of the weight conditional on the system starting in the state @xmath207 , and therefore we conclude that the results more generally do not rely crucially on the continuous nature of the weight .

we are concerned with underlying connections between fluids , elasticity , isometric embedding of riemannian manifolds , and the existence of wrinkled solutions of the underlying nonlinear partial differential equations . one of the main purposes of this paper is to develop such connections for the case of two spatial dimensions and examine whether the continuum mechanical equations can be mapped into a corresponding geometric framework and the inherent direct application of the theory of isometric embeddings and the gauss - codazzi equations . another motivation for such study is to explore the possibility whether the geometric theory can serve an avenue for addressing the admissibility criteria for nonlinear conservation laws in continuum mechanics . in recent years , a theory of _ wild _ solutions to both the incompressible and compressible euler equations in two and higher spatial dimensions has been developed by de lellis , szkelyhidi , and others in @xcite and the references cited therein . the approach is based on the analogy with the highly irregular ( static ) solutions of the isometric embedding of a two - dimensional riemannian manifold into three - dimensional euclidean space given by the nash - kuiper theorem @xcite . specifically , the analogy arises because of the applicability of gromov s h - principle and convex integration to both problems ( _ cf_. @xcite ) . this suggests that perhaps the initial value problem in fluid dynamics and the embedding problem in differential geometry would be more than just two analogous issues , and could in fact be mapped one to the other . a first thought on this issue would suggest that the question is not even meaningful : the fluid problem is dynamic and the embedding problem is static . thus , if the linkage is to make any sense at all , we must think of the embedding problem as dynamic and derive the equations of a time evolving two - dimensional riemannian manifold . of course , conceptually it is easy to visualize this time evolving two - dimensional riemannian manifold as a two - dimensional surface moving in three - dimensional space that can be seen in such an everyday phenomenon as the vibration of the surface of a drum . thus , to make the link , we must derive the equations of a new type of geometric flow and then interpret the consequences of this fluid - geometric duality . in fact , once we have mapped a solution of the euler equations onto the evolving surface , it is rather easy to see that the dynamic metric @xmath0 is a short metric in the sense of the nash - kuiper theorem @xcite with respect to a metric associated with a developable surface . hence , an immediate consequence of our theory is that the evolving manifold can be approximated by wrinkled manifolds ( _ i.e. _ , manifolds with discontinuous second derivatives ) and thus gives some indication of the existence of _ wild _ solutions of the dual fluid problem . a simple mental picture of the geometric image of the fluid problem would be the motion of @xmath1 wrinkles on a piece of paper . moreover , the _ wild _ solutions of the geometric problem are completely time - reversible and are in reality a sequence of the nash - kuiper solutions of the embedding problem with the metric given by @xmath2 , where @xmath3 corresponds to a developable surface that is time - independent and has non - vanishing mean curvature . the implication of this fact is immediate : the geometric nash - kuiper _ wild _ solutions are time - reversible . thus , the question raised in @xcite as to which is the correct admissibility criterion to choose the relevant solution from the infinite number of non - unique solutions to the euler equations , has at least a partial answer . namely , no dynamic admissibility criterion such as the energy inequality , entropy inequality , entropy rate can serve the purpose , since all the inequalities become the identities for such solutions . the only possible useful criteria must be meaningful for time reversible fluid flow , such as energy minimization , artificial viscosity @xcite , or viscosity - capillarity @xcite . the equations for geometric flow are abundant : the einstein equations of general relativity and the ricci flow equations are two of the better known examples . in both of these problems , the metric for a riemannian manifold becomes the dynamic unknown . in the theory developed in this paper , the same situation arises : a dynamic metric @xmath0 is our unknown along with the second fundamental form for the evolving manifold . furthermore , just as in the case of the einstein equations ( _ cf . _ @xcite ) , the initial data must be consistent with the problem , that is , initially an embedded manifold does exist and the map is also consistent with the divergence free condition on the velocity for the incompressible fluid case . for the einstein equations , this consistency of the initial data yields the einstein constraint equations , while in our case a system of constraint equations is also required . in this paper we show that our constraint equations have local analytic solutions ; moreover , there is a velocity @xmath4 that defines both a solution of the incompressible euler equations and an evolving two - dimensional riemannian manifold isometrically immersed in @xmath5 . a similar result is given for the compressible case , as well as for neo - hookean elasticity . the most important issue is the physical meaning of the results mentioned above . in short , we emphasize the comments which have appeared in szkelyhidi @xcite , bardos - titi - wiedermann @xcite , bardos - titi @xcite , and bardos - lopes filho - niu - nussenzveig lopes - titi @xcite . the appearance of the _ wild _ solutions is due essentially to the application of the euler equations on the initial data that form a discontinuous shear , _ i.e. _ , a vortex sheet . hence , the euler equations , which have no small scales built into their theory as opposed to the compressible or incompressible navier - stokes equations , have been used in a case where they should not be expected to apply . to this , we must add the proviso that as again mentioned in de lellis - szkelyhidi @xcite ; also see elling @xcite : these _ wild _ solutions could be a demonstration of fluid turbulence . the results given in this paper show that , if this is the case , then this _ wild _ fluid turbulent behavior is mirrored by the solutions produced by the non - smooth moving wrinkled surfaces produced by the nash - kuiper theorem @xcite . this paper consists of nine sections after this brief introduction . in 2 , we recall the euler equations for an inviscid incompressible fluid and the gauss - codazzi equations for the isometric embedding of a two - dimensional riemannian manifold into three - dimensional euclidean space . in 3 , as discussed above , we derive the equations of the geometric flow , as well as the constraint conditions on the initial data . in 4 , we prove the solvability of the constraint equations for the initial data . we emphasize that the constraint conditions require the initial fluid velocity _ not _ to be a shear flow . this assumption pairs nicely with a theorem in de lellis - szeklyhidi @xcite that non - smooth shear flow initial data for both the compressible and incompressible euler equations are _ wild _ data and yield non - unique solutions to the cauchy problem for both the compressible and incompressible euler equations . in fact our computation suggests that this is the only _ wild _ initial data . on the other hand , we note that , for the degenerate case , when the fluid motion is a shear flow , we still have a metric that provides the desired map , namely , metric @xmath3 . in 5 , we state and prove our main result : evolving from the initial data , there exists a metric @xmath0 of geometric flow equations which yields a solution of both the euler equations and the equations describing the evolving isometrically immersed riemannian manifold @xmath6 . in 6 , we continue the discussion of the initial data issue and show that , for the shear flow initial data in the hypotheses of lemma [ l41 ] , the symbol of the underlying system of second order partial differential equations vanishes so that the equations for @xmath0 are degenerate ; however , as we just commented , metric @xmath3 suffices in this singular case . in 7 , we give our principal result : the evolving manifold arising from the euler equations can be approximated by wrinkled @xmath7 manifolds @xmath8 , @xmath9 . in fluid variables , these wrinkled manifolds correspond to discontinuous shear flows , which thus show conceptually that the discontinuous shear flows are dense in the set of solutions of the incompressible euler equations . furthermore , these wrinkled solutions can be arranged as a time - sequence of solutions which render the initial value problem for the evolving manifold to have an infinite number of constant energy solutions . in 8 , a short discussion is provided to show that many of our earlier results for the incompressible euler equations carry over to the compressible case . based on the knowledge that has been obtained from the fluid equations , we show in 9 how the case of general continuum mechanics can be placed in our mechanics - geometry framework . as an illustrative example , we demonstrate results for elastodynamic motion of a neo - hookean solid . finally , our last section , 10 , provides a discussion of admissibility criteria for the ( incompressible and compressible ) euler equations which asserts that , for the multidimensional euler equations , no dynamic admissibility condition would eliminate _ wild _ solutions , and hence the only meaningful one must be the one which eliminates or at least reduces the number of wrinkles of our dual geometric problem . furthermore , these results suggest that more refined continuum mechanical theories relying on micro - structure could play a key role in choosing admissible solutions . we start with some basic geometric equations and notations for subsequent developments . for more details , see han - hong @xcite and the references cited therein . let @xmath6 be a two - dimensional riemannian manifold with @xmath10 denoting a point on the manifold , @xmath11 . the unit normal vector @xmath12 to the manifold is given by @xmath13 and the second fundamental form is @xmath14 where @xmath15 , @xmath16 , and @xmath17 with @xmath18 for @xmath19 . we will use the alternative version of the second fundamental form @xmath20 and recall that @xmath21 where @xmath22 is the gauss curvature , and @xmath23 is the riemann curvature tensor . furthermore , for notational simplicity , we henceforth drop the ' " superscript in the alternative version of the second fundamental form . then the codazzi equations are @xmath24,\\[1 mm ] \gamma_{12}^{1 } = \frac1{2\det { { \bf g}}}\left[g_{22}(\partial_{2}g_{11})-g_{12}(\partial_{1}g_{22})\right],\\[1 mm ] \gamma_{22}^{1 } = \frac1{2\det { { \bf g}}}\left[g_{22}(2\partial_{2}g_{21}-\partial_{1}g_{22})-g_{12}(\partial_{2}g_{22})\right],\\[1 mm ] \gamma_{11}^{2 } = \frac1{2\det { { \bf g}}}\left[-g_{12}(\partial_{1}g_{11})+g_{11}(2\partial_{1}g_{12}-\partial_{2}g_{11})\right],\\[1 mm ] \gamma_{12}^{2 } = \frac1{2\det { { \bf g}}}\left[-g_{12}(\partial_{2}g_{11})+g_{11}(\partial_{1}g_{22})\right],\\[1 mm ] \gamma_{22}^{2 } = \frac1{2\det { { \bf g}}}\left[-g_{12}(2\partial_{2}g_{21}-\partial_{1}g_{22})+g_{11}(\partial_{2}g_{22})\right ] , \end{cases } \ee and the riemann curvature tensor is given by \begin{equation}\label{2.4-a } r_{iljk}=g_{lm}\big(\partial_{k}\gamma_{ij}^{m } -\partial_{j}\gamma_{ik}^{m}+\gamma_{ij}^{n}\gamma_{nk}^{m } -\gamma_{ik}^{n}\gamma_{nj}^{m}\big),\ ] ] where we have used the einstein summation convention that the repeated indices are implicitly summed over in the terms , which will also be used from now on . in particular , we have @xmath25 with , we have gauss s theorem egregium for the gauss curvature . a convenient form is given by brioschi s formula : [ e24 ] = & 12(*g*)^2 - _ 22g_11 + 2_12g_12- _ 11g_22 & _ 1g_11 & 2_1g_12- _ 2g_11 + 2_2g_12-_1g_22 & 2g_11 & 2g_12 + _ 2g_22 & 2g_12 & 2g_22 + & -12(*g*)^2 0 & _ 2g_11 & _ 1g_22 + _ 2g_11 & 2g_11 & 2g_12 + _ 1g_22 & 2g_12 & 2g_22 . we recall the fundamental theorem of surface theory states that the solvability of the gauss - codazzi equations is a necessary and sufficient condition for the existence of an isometric embedding , _ i.e. _ , a simply connected surface @xmath26 which satisfies @xmath11 . a convenient reference for the smooth version of the fundamental theorem is do carmo @xcite , while a non - smooth version can be found in mardare @xcite . the equations for the balance of linear momentum are [ e25 ] _ 1(u^2+p)+_2(uv)=-_tu , + _ 1(uv)+_2(v^2+p)=-_tv , where the constant density @xmath27 is taken . the condition of incompressibility for constant density is then given by the equation : [ e26 ] _ 1u+_2v=0 . in addition , taking the divergence of equations and using the incompressibility condition , we have [ e27 ] _ 11(u^2 ) + 2_12(uv)+_22(v^2 ) = -p . just as in chen - slemrod - wang @xcite , it is convenient to write the geometric equations in fluid variables . set [ e28 ] l = v^2+p , m =- uv , n = u^2+p . then the gauss equation becomes @xmath28 that is , @xmath29 where @xmath30 . this quadratic equation then gives [ e29 ] p=-12q^212 . this means that @xmath31 must be required . as it will be seen in the analysis below , this condition is always satisfied . we have just shown that @xmath32 can be written in fluid variables . now we can write the fluid variables in terms of the geometric variables @xmath32 . to do this , simply substitute formula into to find @xmath33 write @xmath34 to see that @xmath35 , which yields that @xmath36 . then we see @xmath37 and , using @xmath38 , @xmath39 this shows that the `` + '' sign must be chosen in the above formulas so that [ e210 ] u^2=12(-(l - n)+ ) , + v^2=12((l - n)+ ) . note that the condition : @xmath40 , with @xmath41 , is equivalent to @xmath42 that is , @xmath43 which is always satisfied . thus , we have shown that @xmath44 are determined by @xmath32 , since @xmath45 we now construct a _ dual _ solution , which simultaneously satisfies the incompressible euler equations and the gauss - codazzi equations of isometric embeddings . as before , the gauss - codazzi equations are [ e31 ] _ 1n-_2 m = -_22 ^ 1l+2_12 ^ 1m-_11 ^ 1n , + _ 1m-_2l=_22 ^ 2l-2_12 ^ 2m+_11 ^ 2n , and [ e32 ] ln - m^2=. hence , for any @xmath4 to be a solution of the euler equations , we must have [ e33 ] _ tu=_22 ^ 1l-2_12 ^ 1m+_11 ^ 1n , + _ ^ 2l-2_12 ^ 2m+_11 ^ 2n , and [ e34 ] _ 1u+_2v=0 . taking the divergence of and using , we have [ e35 ] _ 1(_22 ^ 1l-2_12 ^ 1m+_11 ^ 1n ) + _ 2(_22 ^ 2l-2_12 ^ 2m+_11 ^ 2n)=0 . for convenience , define [ e36 ] & u(l , m , n):=(12)^12 , + & v(l , m , n):=(12)^12 , so that [ e37 ] ( u , v)=(u(l , m , n),v(l , m , n ) ) . the other choice of @xmath46 can be handled similarly . in summary , we have the two evolution equations and four closure relations and . since @xmath47 and @xmath22 are the functions of @xmath48 through , we obtain the six equations and for the six unknowns @xmath48 . in this section , we exposit the constraint equations on the initial data and the consequences of their solvability . our first result is [ theorem41 ] assume that the initial data @xmath48 at @xmath49 satisfy the five constraint equations : @xmath50 which mean that the initial data are consistent with both the incompressibility of the fluid and that the evolving manifold is initially indeed a riemannian manifold . then , if the system of six equations and in the six unknowns @xmath48 is satisfied , it produces simultaneously a solution of both the gauss - codazzi equations and the incompressible euler equations . since @xmath51 , we see from the first two evolution equations and that @xmath52 since the initial data satisfy the constraint equations , we conclude that , for @xmath53 , @xmath54 next , since the gauss - codazzi equations are satisfied for all @xmath53 , the fluid representation for @xmath55 allows to write the codazzi equations as @xmath56 and the gauss equation as @xmath57 by , we see that the balance of linear momentum equations is also satisfied . the proof is complete . we next examine the solvability of the initial data system . note that imply [ e46 ] _ 1(_1n-_2 m ) + _ 2(-_1m+_2l)=0 , and the initial data system can be written as @xmath58 we can reverse the above computation . if are satisfied , we take the divergence of the left - hand sides of and employ to yield . from , we find that and becomes an undetermined system of five equations in the six unknowns @xmath48 . our existence result for the initial data satisfying the constraint equations reads as follows : [ l41 ] let an analytic divergence free velocity @xmath59 be prescribed in a neighborhood of a point @xmath60 such that @xmath61 at this point . set @xmath62 . on @xmath63 ( respectively @xmath64 , prescribe the analytic initial data : @xmath65 , and @xmath66 satisfying the ordinary differential equation : @xmath67 with initial condition @xmath68 at @xmath69 so that @xmath70 at @xmath71 ( respectively , @xmath72 and a similar ordinary differential equation and initial data ) . then the initial data system has a local analytic solution @xmath73 . we divide the proof into eight steps . in fluid variables , three of our equations are @xmath74 next , prescribe the velocity to make this a determined system . if a divergence free velocity @xmath59 is prescribed , then is immediately solvable for @xmath75 under the standard regularity assumptions on @xmath59 , and hence defines the gauss curvature @xmath22 . thus , @xmath55 are known with no dependence on metric @xmath0 . now , determine @xmath0 by solving the three equations : @xmath76 we use to solve for @xmath77 and @xmath78 . simply use to write these two equations as [ e421 ] & 1*g * + & = _ 1n-_2 m + 12*g*l + & -m + 12*g * n , [ e422 ] & 1*g * + & = -_1m+_2l + 12*g*l + & - m + 12*g*n . in the matrix form , we have 1*g * g_12n & -g_22l + -g_11n & g_12l _ 1g_12 + _ 2g_21 = _ 1n-_2 m + -_1m+_2l + 1*g * g_1 + g_2 , where g_1=&l -m + & + n , + g_2=&l -m + & + n . the inverse of the coefficient matrix is @xmath79 \frac{g_{11}}{l } & \frac{g_{12}}{l } \end{bmatrix},\ ] ] which gives [ e423 ] _ 1g_12 + _ 2g_21 = & + & _ 1n-_2m+1*g*g_1 + -_1m+_2l+1*g*g_2 . from , the equality of cross partials gives us the additional consistency equation : [ e424 ] & _ 2((_1n-_2m+1*g*g_1 ) + ( -_1m+_2l+1*g*g_2 ) ) + & = _ 1((_1n-_2m+1*g*g_1 ) + ( -_1m+_2l+1*g*g_2 ) ) . the gauss curvature equation has the form : [ e425 ] = 1*g*+l.o.t . substitution of into yields the two second - order equations for @xmath80 . now compute the highest order terms of the differential operators associated with ( [ e423 ] ) at the origin with @xmath70 : @xmath81 \partial_{2}g_{21}\end{bmatrix}&= \begin{bmatrix } 0 & \frac{1}{n } \\[1.5 mm ] \frac{1}{l } & 0\end{bmatrix}\begin{bmatrix } \mathbf{\ } \partial _ { 1}n-\partial _ { 2}m-{\frac12}\partial_{1}g_{22}l-\partial_{2}g_{11}m+{\frac12}\partial _ { 1}g_{11}n \\[1.5 mm ] \mathbf{-}\partial _ { 1}m+\partial _ { 2}l+{\frac12}\partial_{2}g_{22}l-\partial _ { 1}g_{22}m+-{\frac12}\partial _ { 2}g_{11}n\end{bmatrix}\\[2 mm ] & = \begin{bmatrix } \frac{1}{n}(\mathbf{-}\partial _ { 1}m+\partial _ { 2}l+{\frac12}\partial_{2}g_{22}l-\partial_{1}g_{22}m-{\frac12}\partial_{2}g_{11}n ) \\[1.5 mm ] \frac{1}{l}\mathbf{\ ( } \partial_{1}n-\partial_{2}m-{\frac12}\partial _ { 1}g_{22}l-\partial _ { 2}g_{11}m+{\frac12}\partial _ { 1}g_{11}n)\end{bmatrix}. \end{split}\ ] ] then becomes [ e426 ] = & + + & + + l.o.t . and equation becomes @xmath82 in the matrix form , are written as [ e427 ] 14 & -14 + -12 & 12 _ 11 g_11 + g_22 + - & - + & - _ 12 g_11 + g_22 + -14 & 14 + -12 & 12 _ 22 g_11 + g_22 = l.o.t . make the change of independent variables : @xmath83 so that , for any @xmath84 , @xmath85 then we have [ e427-a ] a g_11 + g_22 + a g_11 + g_22 + ( + 1 ) & - + -1 & 1 g_11 + g_22 = l.o.t . where the coefficient matrix @xmath86 is @xmath87 -{\frac12}\frac{n}{l } & { \frac12}\end{bmatrix}+\begin{bmatrix } -\frac{m}{2l } & -\frac{m}{2n } \\[1.5 mm ] \frac{m}{l } & -\frac{m}{n}\end{bmatrix}+\begin{bmatrix } -{\frac14 } & { \frac14}\frac{l}{n } \\[1.5 mm ] -{\frac12 } & { \frac12}\frac{l}{n}\end{bmatrix}= \begin{bmatrix } { \frac14}\frac{n}{l}-\frac{m}{2l}-{\frac14 } & -{\frac14}-\frac{m}{2n}\text { } + { \frac14}\frac{l}{n } \\[1.5 mm ] -{\frac12}\frac{n}{l}+\frac{m}{l}-{\frac12 } & { \frac12}-\frac{m}{n}+{\frac12}\frac{l}{n}\end{bmatrix}\ ] ] with determinant equal to @xmath88 if the term : @xmath89 , then @xmath90 . thus , the conditions that @xmath91 and @xmath92 make this coefficient matrix non - singular . since @xmath93 in fluid variables , we need @xmath61 and curvature @xmath22 determined by the poisson equation with @xmath94 . of course , when @xmath95 , @xmath96 , and @xmath97 , then @xmath98 . clearly , if pressure @xmath75 determined by the poisson equation is positive , then the condition : @xmath91 , is automatically satisfied . then line @xmath99 is non - characteristic for the equation in . since @xmath100 , gives a partial differential equation for @xmath66 in the direction normal to the initial data line @xmath101 : @xmath102 we now determine the initial data for system and on line @xmath63 that is non - characteristic for the system . since @xmath103 the initial conditions : @xmath104 also give us @xmath105 this yields @xmath106 then , from the formula : @xmath107 , , and @xmath108 \partial_{2}g_{21}\end{bmatrix}= \begin{bmatrix } \frac{g_{12}}{n } & \frac{1}{n } \\[1.5 mm ] \frac{1}{l } & \frac{g_{12}}{l } \end{bmatrix}\begin{bmatrix } \mathbf{\ } \partial_{1}n-\partial_{2}m \\[1.5 mm ] \mathbf{-}\partial_{1}m+\partial_{2}l \end{bmatrix},\ ] ] we obtain the ordinary differential equation on line @xmath63 . this differential equation and the initial condition : @xmath68 at @xmath109 determine the initial data for @xmath66 on line @xmath63 . therefore , we have specified analytic initial data on the non - characteristic line @xmath63 for system and . finally , we see that the cauchy - kowalewski theorem delivers a local analytic solution @xmath73 of system and with the analytic data on the non - characteristic line @xmath63 , where we have used the analyticity of @xmath75 as the solution to our poisson equation and the analyticity of @xmath22 from the formula : @xmath110 . the positivity condition on @xmath75 is irrelevant and can always be satisfied by simply adding to any @xmath75 solving the poisson equation a sufficiently large positive constant . this completes the proof . notice that the shear flow data @xmath111 are divergence free , no matter how smooth or in fact how irregular @xmath112 is . in this case , @xmath113 , so that lemma [ l41 ] does not apply to yield a metric @xmath0 . however , in this case , we have the following trivial result . [ l42 ] if @xmath113 in any open set of @xmath114 , then there exists a developable surface @xmath115 with corresponding metric @xmath3 given by @xmath116 with @xmath117 similarly , the corresponding developable surface with corresponding metric @xmath3 can be obtained for @xmath118 as another shear flow . in particular , @xmath119 and @xmath120 for @xmath121 . this metric @xmath3 yields a solution of our initial data system . furthermore , this solution must be a shear flow @xmath122 , or @xmath123 , owing to the incompressibility . consider the surface : @xmath124 then @xmath125 so that @xmath126 and @xmath127 furthermore , we have @xmath128 so that @xmath129 at @xmath130 , @xmath131 set @xmath132 so that @xmath133 . then , at @xmath134 , @xmath135 which implies that @xmath136 with @xmath137 prescribed by the solution of @xmath138 we have constructed the developable surface , while the solution of is determined by @xmath139 similarly , we can obtain the corresponding results for @xmath118 as another shear flow . [ r41 ] notice the very important fact that the shear flow initial data given by lemma [ l42 ] require only the square integrability of the shear functions @xmath140 and @xmath141 . [ r42 ] the assumption of analyticity is made only for convenience . in fact , if one follows the presentation of deturck - yang @xcite for a similar problem where the theory of systems of real principal type is exploited , it is quite evident that the analyticity could be replaced by @xmath142 . now we prove our main existence theorem . [ t51 ] for the initial data constructed in lemma [ l41 ] , there exists a solution of system and , locally in space - time , which satisfies both the incompressible euler equations and a propagating two - dimensional riemannian manifold isometrically immersed in @xmath5 . for the given fluid data , by the cauchy - kowalewski theorem , there exists a local space - time solution @xmath143 of the incompressible euler equations with @xmath144 , and . define the gauss curvature via the relation : @xmath145 so that @xmath146 . thus system and is satisfied if the following system can be solved : _ tu=_22 ^ 1l-2_12 ^ 1m+_11 ^ 1n , + _ tv=_22 ^ 2l-2_12 ^ 2m+_11 ^ 2n , + * g*=g_2m(_2_11^m-_1_12^m + _ 11^n_n2^m-_12^n_n1^m ) for the time evolving metric @xmath0 , where u^2=12 , + v^2=12 , + q^2=. this is the same system for @xmath0 that has been solved in lemma [ l41 ] . this completes the proof . [ c51 ] let @xmath4 be a solution , local analytic in space - time , of the incompressible euler equations , satisfying the initial condition @xmath61 at a point @xmath147 . then there is an evolving metric @xmath0 so that solution @xmath4 of the euler equations also defines a time evolving two - dimensional riemannian manifold @xmath6 immersed in @xmath5 . [ c52 ] conversely , let @xmath0 be an isometrically embeddable manifold in @xmath5 with second fundamental form @xmath148,m[{{\bf g}}],n[{{\bf g}}])$ ] that is a solution of the gauss - codazzi equations for an as yet undetermined metric @xmath0 ) and @xmath149 . identify the fluid variables @xmath44 by our relations : [ 5.1-a ] u^2= , + v^2= , + p^2+pq^2= , q^2=u^2+v^2 , so that @xmath150 , v[{{\bf g } } ] , p[{{\bf g}}])$ ] are functionals of the unknown metric field @xmath0 . then the equations : [ 5.2-a ] _ tu = _ 22 ^ 1l-2_12 ^ 1m+_11 ^ 1n , + _ tv = _ 22 ^ 2l-2_12 ^ 2 m + _ 11 ^ 2n , + _ 11(u^2)+2_12(uv)+_22(v^2 ) = -p define a system of three equations in the three unknowns @xmath73 . if this implicit system for @xmath0 has a solution , then the three functions @xmath44 satisfy the incompressible euler equations . substitute the first two equations in into the codazzi equations . we see that the balance of linear momentum is satisfied , and the third equation in implies , via the balance of linear momentum , the incompressibility condition : @xmath151 the difficulty in employing corollary [ c52 ] would be that the equations for @xmath0 are only known implicitly . notice that equations satisfied by @xmath0 can be written as [ e51 ] _ t u=_22 ^ 1(v^2+p)+2_12 ^ 1(uv)+_11 ^ 1(u^2+p ) , + _ t v=_22 ^ 2(v^2+p)+2_12 ^ 2(uv)+_11 ^ 2(u^2+p ) , + _ 1(_22 ^ 1(v^2+p)+2_12 ^ 1(uv)+_11 ^ 1(u^2+p ) ) + + _ 2(_22 ^ 2(v^2+p)+2_12 ^ 2(uv)+_11 ^ 2(u^2+p))=0 . here @xmath4 are defined by in terms of @xmath55 which in turn depend on @xmath0 , and @xmath75 is given by @xmath152 from . thus , this is a system of three equations for the three components of @xmath0 . the difficulty here is that the dependence of @xmath59 on @xmath0 is implicit . it would be interesting to identify some suitable ways to check their solvability . as we noted in remark [ r41 ] , the shear flow data do not satisfy the hypothesis : @xmath61 in lemma [ l41 ] . hence , this is a particularly interesting case to study . namely , if the fluid data of @xmath4 are @xmath153 , the question is whether a riemannian manifold @xmath6 locally immersed in @xmath5 can be determined . for this data , we find immediately that @xmath154 so that @xmath75 is harmonic . furthermore , @xmath155 , and our constraint equations reduce to the equations : [ e61 ] & _ 1p=-_22 ^ 1p-_11 ^ 1(u^2(x_2)+p ) , + & _ 2p=-_22 ^ 2p-_11 ^ 2(u^2(x_2)+p ) , and [ e62 ] pu^2(x_2)+p^2=&12(*g*)^2 -_22g_11 + 2_12g_12-_11g_22 & _ 1g_11 & 2_1g_12-_2g_11 + 2_2g_12-_1g_22 & 2g_11 & 2g_12 + _ 2g_22 & 2g_12 & 2g_22 + & -12(*g*)^2 0 & _ 2g_11 & _ 1g_22 + _ 2g_11 & 2g_11 & 2g_12 + _ 1g_22 & 2g_12 & 2g_22 . as in [ s4 ] , give equation that is now [ e63 ] & -14 + - & 12 _ 11 g_11 + g_22 + -14 & + -12 & _ 22 g_11 + g_22 = l.o.t . the symbol of the above differential operator is given by & ( & -14 + - & 12 _ 1 ^ 2 + -14 & + -12 & _ 2 ^ 2 ) + & = _ 1 ^ 2 -14_2 ^ 2 & -14_1 ^ 2+_2 ^ 2 + -_1 ^ 2 - 12_2 ^ 2 & 12_1 ^ 2 + _ 2 ^ 2 + & = 0 . thus , for the shear flow initial data , our system of partial differential equations for @xmath0 is degenerate . on the other hand , we know from lemma [ l42 ] that , if a shear flow is given in an open neighborhood of space - time , then metric @xmath3 provides the desired map . now that the existence of a locally analytic riemannian manifold with metric @xmath0 has been established in corollary [ c51 ] , we wish to know the local form of the metric . to do this , we adapt a new local system of normal coordinates @xmath156 , @xmath157 ( _ cf . _ @xcite and @xcite ) so that the origin of the old coordinate system remains the origin of our new coordinate system and @xmath158 . we can easily compute @xmath0 to the leading order in the normal coordinates by the formula : @xmath159 that is , & g_11(x_1^,x_2^,t ) = 1 + 13r_1221x_2^x_2^+h.o.t . = 1 - 13 ( x_2^)^2+h.o.t . , + & g_22(x_1^,x_2^,t ) = 1 + 13r_2112x_1^x_1^+h.o.t . = 1 - 13 ( x_1^)^2+h.o.t . , + & g_12(x_1^,x_2^,t ) = 13r_1212x_1^x_2^+h.o.t . = 13 x_1^x_2^+h.o.t .. hence , as a quadratic form , @xmath160 however , the matrix @xmath161 -{\frac12}x_{1}^{\prime \prime } x_{2}^{\prime \prime } & ( x_{1}^{\prime \prime})^2\end{bmatrix}\ ] ] has determinant @xmath162 and hence is positive definite away from @xmath163 furthermore , on line @xmath164 , the quadratic form is @xmath165 thus , we have proved that map @xmath166 locally induced by metric @xmath0 is shorter than that induced by @xmath3 or , in language of the nash - kuiper theorem @xcite , embedding @xmath167 induced by @xmath0 is a _ short embedding_. we state this in the following lemma : [ l71 ] the metric , @xmath168 , given in theorem [ t51 ] and corollary [ c51 ] is locally a short embedding that satisfies @xmath169 in the sense of quadratic forms . we then recall the nash - kuiper theorem as given in @xcite . [ t71 ] let @xmath170 be a smooth , compact riemannian manifold of dimension @xmath171 . assume @xmath172 is in @xmath142 . then if @xmath173 , any short embedding can be approximated by isometric embeddings of class @xmath142 ( cf . nash @xcite and gromov @xcite ) ; if @xmath174 then any short embedding can be approximated in @xmath175 by isometric embeddings of class @xmath176 ( cf . nash @xcite , kuiper @xcite ) . we note that part ( ii ) of theorem [ t71 ] has been extended for an analytic metric @xmath172 by borisov @xcite to obtain the greater regularity @xmath177 , and the condition of analyticity has been weakened by conti - delellis - szelelyhidi @xcite . we are now interested in case ( ii ) with @xmath178 , and @xmath179 so that the analyticity of the metric is immediate and @xmath9 . thus , an immediate consequence of the results by nash - kuiper @xcite , borisov @xcite , and conti - delellis - szelelyhidi @xcite is the following theorem . [ t72 ] the short embedding @xmath180 established by theorem [ t51 ] and corollary [ c51 ] can be approximated in @xmath175 , by wrinkled embeddings @xmath181 in @xmath7 , @xmath9 , locally in space - time . in elementary terms , the geometric image of our locally analytic solution to the incompressible euler equations , i.e. , our surface propagating in time can be approximated in @xmath1 , @xmath9 , by @xmath7 wrinkled manifolds . [ t74 ] there are an infinite number of ( non @xmath182 ) evolving manifolds for metric @xmath179 , all with the same initial wrinkled data at @xmath49 . furthermore , for all these solutions , the energy remains constant : @xmath183 where the manifold is defined by the @xmath1solution of @xmath184 which trivially satisfies @xmath185 so that @xmath186 is in @xmath187 in time . on any sequence of time interval , just switch back and forth from any of the infinite choices of wrinkled solutions . specifically , let @xmath188 $ ] be a time interval in which there exists a unique analytic solution to the cauchy initial value problem for the euler equations . by theorem [ t72 ] , this solution is the pre - image of a short embedding @xmath180 and can be approximated by wrinkled embedding @xmath181 . divide the interval @xmath188 $ ] into sub - intervals @xmath189 , [ t_1 , t_2 ] , \dots , [ t_{n-1 } , t_n]$ ] with @xmath190 and @xmath191 . take any sequence @xmath192 . then , by theorem [ t72 ] , we have a sequence of wrinkled solution @xmath193 such that @xmath194 now define the wrinkled solution @xmath195 on @xmath188 $ ] by @xmath196 fix @xmath197 as the cauchy data , but allow @xmath198 and @xmath199 to vary for @xmath200 . in this way , we have produced an infinite number of wrinkled solutions satisfying the same cauchy data . then the solutions are in @xmath1 in space on each sub - interval @xmath188 $ ] for time , so that they are in @xmath201 ; c(\omega))$ ] , but not necessarily in @xmath202 ; c(\omega))$ ] . to compute the energy , use @xmath203 in the results given in theorems [ t71][t74 ] , by construction , embedding @xmath204 is the geometric image of a smooth solution of the euler equations . on the other hand , the nash - kuiper @xmath1 embeddings have not been shown to be the image of solutions in any sense of the euler equations . we now give a definition of weak solutions to the euler equations , which provides this link . let @xmath181 be a nash - kuiper solution , @xmath205 , @xmath206 , of the isometric embedding equations : @xmath207 in an open domain . since @xmath208 , it is in @xmath209 . by equality of cross partials , we have @xmath210 in the sense of distributions , where the second partials of @xmath181 lie in @xmath211 . the motivation behind this definition is as follows : for nash - kuiper solutions , we know that the normal to the second fundamental form is not well defined so that the gauss - codazzi equations that are the geometric image of the euler equations for smooth flows are also not well defined . therefore , we retreat this one step back and use the equality of cross partials which is the starting point for the derivation of the gauss - codazzi equations . [ weak_shear_embedding ] we define an embedding @xmath212 , @xmath213 , as a _ weak euler shear solution inducing embedding _ if ( [ cross_partials ] ) is satisfied in the sense of distributions . certainly , the nash - kuiper embeddings satisfy this definition . the arguments we have used for the incompressible case also carry over to the compressible euler equations . then the equations for the balance of linear momentum are [ e81 ] _ 1(u^2+p)+_2(uv)=-_t(u ) , + _ 1(uv)+_2(v^2+p)=-_t(v ) , the equation for the conservation of mass is given by the equation : [ e82 ] _ 1(u)+_2(v)=-_t . set [ e83 ] l = v^2+p , m =- uv , n = u^2+p , and the gauss equation becomes @xmath214 _ i.e. _ , [ e84 ] p^2+pq^2= , where @xmath30 . for the compressible case , we take [ e85 ] p= , 1 . substitute into to obtain [ e86 ] ( ) ^2 + q^2=. for simplicity , let us first take the isothermal case @xmath215 so that [ e87 ] ^2(1 + q^2)= , that is , [ e88 ] = . from , @xmath216 as the density is determined explicitly as a function of @xmath22 and @xmath217 . on the other hand , @xmath216 must satisfy the mass balance equation so that [ e89 ] _ 1(u)+_2(v ) = -_t ( ) . we see that , for given @xmath59 , is a scalar conservation law for the gauss curvature @xmath22 . this leads us to the following elementary lemma : [ l81 ] for a given smooth solution @xmath218 of the compressible euler equations , the initial value problem for the gauss curvature @xmath22 given by with initial data @xmath219 at @xmath49 has a global smooth solution @xmath22 in space - time , which satisfies . notice that equation is a linear transport equation for @xmath220 of conservation form . then the result holds simply by using the standard local existence - uniqueness theorem for the conservation law and the direct computation @xmath221 this equation has the unique solution @xmath222 the general @xmath223 case follows from noting that the left - hand side of is a monotone increasing function of @xmath216 so that , for any @xmath224 , there exists a function @xmath225 that solves . the regularity of this solution is seen by differentiation of with respect to @xmath226 . we can thus state the general version of lemma [ l81 ] as [ l82 ] for a given smooth solution @xmath218 of the compressible euler equations , the initial value problem for the gauss curvature for with initial data at @xmath49 has a global solution in space - time satisfying . again , we can also write the fluid variables in terms of the geometric variables @xmath55 . to do this , simply substitute with @xmath216 given by the solution to into @xmath227 , and @xmath228 to find @xmath229 write @xmath230 to see @xmath231 , which yields @xmath232 then @xmath233,\ ] ] and , with @xmath38 , @xmath234.\ ] ] this tells us to choose sign @xmath235 in the above formulas so that [ e810 ] u^2= , + v^2= , which are the desired relations . now the compressible analogs of theorem [ t51 ] and corollaries [ c51][c52 ] follow by the same arguments we have employed in [ s5 ] . [ t81 ] for locally analytic initial data @xmath218 satisfying and @xmath236 , there exists a local solution in space - time of system which satisfies both the compressible euler equations and a propagating two - dimensional riemannian manifold isometrically immersed in @xmath5 . [ c81 ] let @xmath218 be a local analytic solution in space - time of the compressible euler equations satisfying the initial condition : @xmath236 , at a point @xmath237 . then there is an evolving metric @xmath0 so that solution @xmath218 of the compressible euler equations also defines a time evolving two - dimensional riemannian manifold @xmath238 immersed in @xmath5 with the gauss curvature @xmath22 determined . theorems [ t72][t74 ] remain unchanged , except for the fact that metric @xmath3 for the wrinkled manifold now corresponds to the special incompressible solution of the compressible euler equations with @xmath239 . in particular , theorem [ t72 ] now gives the smooth solutions of the geometric image of the compressible euler equations being approximated by the wrinkled solutions that correspond weak shear solutions of the _ incompressible _ euler equations . the construction of the wrinkled solutions in this case is the same as done in [ s7 ] and produces the solutions of the incompressible euler equations . a wrinkled solution of the compressible euler equations is impossible since it would correspond to a vacuum via ( [ e84 ] ) . we note that the results given in 7 and this section on the existence of _ wild _ weak shear solutions have been given in terms of a local system of normal coordinates , rather than the original cartesian coordinates @xmath240 . this means for example that the _ wild _ weak shear solutions could in fact correspond to different local coordinates , _ e.g. _ , polar coordinates . the choice of local cartesian coordinates used earlier is only a convenience and the euler equations written in polar coordinates would suffice . motivated by our results for fluid dynamics , we now consider solutions in two - dimensional general continuum mechanics . denote by @xmath241 the ( symmetric ) cauchy stress tensor , and assume that fields @xmath4 , @xmath242 , and @xmath216 are consistent with some specific constitutive equation for a body and satisfy the balances of mass and the linear momentum ( the satisfaction of the balance of angular momentum is automatic ) . the equations for the balance of linear momentum in the spatial representation are [ e101 ] _ 1(u^2-t_11)+_2(uv - t_12 ) = -_t(u ) , + _ 1(uv - t_12)+_2(v^2-t_22 ) = -_t(v ) , and the balance of mass is @xmath243 or @xmath244 where @xmath245 is the density of the body in the reference configuration , and @xmath246 is the deformation gradient of the current configuration with respect to this reference . denote the geometric dependent variables as [ lmn ] l = v^2-t_22,n = u^2- t_11,m= -uv + t_12 , so that the gauss equation becomes @xmath247 under the assumption that @xmath248 at least on some initial time interval , such a solution of continuum mechanics corresponds to a positive gauss curvature for the corresponding isometric embedding problem . using @xmath55 as time - dependent data in ( [ e31])([e32 ] ) and expressing these three equations in terms of the metric components @xmath73 and their derivatives by using ( [ e23])([e24 ] ) , we have a system of three partial differential equations for the components of the metric . using the required analog of corollary [ c81 ] , then we have let @xmath249 be a local analytic solution in space - time of system at a point satisfying the condition : @xmath250 then there is an evolving metric @xmath0 so that the continuum mechanical solution defines a time evolving two - dimensional riemannian manifold @xmath6 immersed in @xmath251 . we note the ability to solve for the evolving metric @xmath0 in the fluid case relied on the initial data for the off - diagonal term @xmath252 in the second fundamental form being non - vanishing . in the cases of the incompressible and compressible fluids , this term was simply @xmath93 . however , for general continuum mechanics , @xmath253 that is , the expression for @xmath252 has an additional contribution . theorems [ t72][t74 ] remain unchanged , except for the fact that metric @xmath3 for the wrinkled manifold now corresponds to special _ steady _ smooth solutions of the equations of two - dimensional general continuum mechanics . we now identify these solutions . we want to define solutions to the mechanical equations in continuum mechanics from a smooth , degenerate isometric embedding problem for which @xmath254 , are satisfied . then , making the association analogous to that used in lemma 4.2 , we have [ 10.8-a ] uv - t_12 = 0 , + u^2 - t_11 = z(x_2,t ) , + v^2 - t_22 = 0 , where function @xmath255 is defined from the embedding , and _ 1(uv - t_12 ) + _ 2 ( v^2 - t_22 ) = 0 , + _ 1 ( u^2 - t_11 ) + _ 2 ( uv - t_12 ) = 0 are satisfied by the mechanical fields that are being defined . these would be a consistent set of the fields satisfying the balances of linear momentum and mass if the following constraints : _ t ( u ) = _ t ( v ) = 0 , + _ 1 ( u ) + _ 2 ( v ) = -_t are satisfied , _ i.e. _ , these geometric solutions define continuum mechanical solutions with steady momenta . this is easily done by noting that the conditions imply @xmath256 with the solution @xmath257 where @xmath258 and @xmath259 are arbitrary time - independent functions of the spatial variables , which are required to be so chosen that @xmath260 . we then define @xmath59 by integrating the pointwise ordinary differential equations : u_t = - _ 1 u , + v_t = - _ 1 v. with @xmath261 in hand , we define the stress components from to obtain a class of mechanical solutions in general continuum mechanics from the smooth degenerate isometric embedding problem . it is to be noted that , for such continuum mechanical solutions to be realizable for a specific material , the mechanical fields as defined have to be shown to be consistent with a constitutive equation for the stress for that material . alternatively , by seeking solutions to the system [ steady ] _ 1(uv - t_12 ) + _ 2 ( v^2 - t_22 ) = 0 , + _ 1 ( u^2 - t_11 ) + _ 2 ( uv - t_12 ) = 0 , + _ 1 ( u ) + _ 2(v ) = -_t , we can define solutions to the balance of mass and the _ steady equations of balance of linear momentum_. the solutions to system do not necessarily constitute exact , steady solutions of the balance of linear momentum , i.e. , @xmath262 and @xmath263 may not evaluate to @xmath264 from such motions , much in the spirit of quasi - static evolutions in solid mechanics and stokes flow in fluid mechanics . as is understood , such solutions are typically interpreted in an asymptotic sense when the velocities are assumed to `` equilibrate '' on a much faster time - scale than the evolution of driving boundary conditions or forcing , i.e. , @xmath265 is assumed to be the `` slow '' time scale and the right - hand sides of the first two equations actually have the terms : @xmath266 and @xmath267 , with @xmath268 , so that we deal with the singular perturbation with a first - order approximation . the goal in this section is to display the image of the isometric embedding problem in the smooth degenerate case in nonlinear elastodynamics for a neo - hookean material . we choose the constitutive equation for the cauchy stress of the compressible material as @xmath269 where @xmath246 is the deformation gradient from a fixed reference configuration of the body . the generic point on the fixed reference is denoted by @xmath270 , the motion as @xmath271 , and @xmath272 . for the sake of simplicity , we ignore a multiplicative scalar function of the invariants of @xmath273 that would ensure that the stress at the reference configuration would vanish , even in this simplest frame - indifferent elastic constitutive assumption . _ we now seek special solutions to the steady equations . after obtaining any such solution consistent with the posed steady problem , we will further check which of them also constitute ( steady ) solutions to the equations of balance of linear momentum_. thus , it needs to be demonstrated that the conditions : [ nh_gov_eq ] uv - b_12 = 0 , + v^2 - b_22 = 0 , + u^2 - b_11 = z(x_2,t ) are satisfied , and the balance of mass is satisfied by _ a - posteriori _ setting @xmath274 with @xmath275 calculated from the obtained solution to ( [ nh_gov_eq ] ) , where @xmath245 is the mass density distribution in the reference configuration . we always require that @xmath276 . the components of @xmath273 are given as @xmath277 thus , the equations required to be satisfied by a motion consistent with these constraints would be [ elastic_syst ] = + , + ( ) ^2 = ( ) ^2 + ( ) ^2 , + ( ) ^2 = ( ) ^2 + ( ) ^2 + z ( x_2(x , t),t ) . we define @xmath278 , and @xmath279 for subsequent use . if solutions exist to the above system , then they also satisfy the following conditions : the satisfaction of the first equation of ( [ elastic_syst ] ) by a solution satisfying the second and third is equivalent to @xmath280 which implies that @xmath281 , or else @xmath282 which is not acceptable by hypothesis . thus , the steady , neo - hookean image of the degenerate isometric embedding problem may be stated as the system for functions @xmath283 : [ es2 ] ( ) ^2 = |x_2|^2 , + ( ) ^2 = |x_1|^2 - ( ) ^2 , + z(x_2 , t ) |x_2|^2 = - ( j(x_1 , x_2))^2 , with the caveat that solutions to satisfy @xmath284 in order to generate steady solutions consistent with , we need to choose only the @xmath285 sign . when restricting attention to the class of solutions to the whole system that are scale - invariant in @xmath283 , _ i.e. _ , those solutions that remain solutions if the dependent and independent variables are scaled by the same constant , we note that the first two equations in ( [ es2 ] ) remain invariant under rescaling the independent and dependent variables by @xmath286 . similarly , @xmath287 is independent of @xmath286 . we now assume that the given function @xmath288_. [ elast_example ] consider a shearing motion of the form : [ ex_ansatz ] & x_1 = x_1 + f(x_2,t ) , + & x_2 = x_2 t. then @xmath289 , @xmath290 , @xmath291 , and @xmath292 . thus , @xmath293 , @xmath294 , and @xmath295 . then the first equation of ( [ es2 ] ) is identically satisfied , the second requires @xmath296 and the third is identically satisfied by our choice of @xmath297 . thus , the traveling waves of the form : @xmath298 define solutions to the steady , neo - hookean , degenerate isometric embedding problem , for each sign of @xmath299 in the second equation of ( [ ex_ansatz ] ) . in summary , we have shown that the elasticity analog of theorem [ t72 ] for fluids holds , with the identification that @xmath166 is one of the possible smooth degenerate ( _ i.e. , _ @xmath3 ) isometric embedding images of the neo - hookean motion and the corresponding _ @xmath181 are the weak euler shear solution inducing embeddings in the sense of definition [ weak_shear_embedding ] _ , where the shear would be with respect to the reference @xmath300configuration . for a spatially uniform density field on the reference configuration , it is easy to check that the equations of balance of linear momentum for the neo - hookean constitutive assumption we are considering ( the first piola - kirchhoff stress is given by @xmath301 ) reduces to the linear , second order system : @xmath302 thus , for the assumed ansatz ( [ ex_ansatz ] ) , the image of the degenerate isometric embedding problem has produced special _ exact _ non - steady , solutions to the balance of linear momentum . clearly , these are not steady solutions on the reference configuration in general . thus , it is an interesting question to check whether any of these are _ exact _ steady solutions on the current configuration . 1 . the solutions generated in example [ elast_example ] correspond to @xmath303 . let us consider each in turn . to check steadiness , we consider an arbitrarily fixed point @xmath304 in space . we have to show that the velocities of material points that occupy it at different times is the same , since the density is constant , @xmath305 , everywhere for this example . let the image in the reference configuration of point @xmath306 at time @xmath265 be @xmath307 for @xmath308 , @xmath309 which implies @xmath310 + now @xmath311 then @xmath312 thus , we indeed have a steady spatial velocity and momenta field . the way to physically understand it is the following : let @xmath313 be specified . the points on line @xmath314 have material points sitting on them at time @xmath265 with referential @xmath315coordinate given by @xmath316 . however , for such referential points , the horizontal velocity @xmath317 corresponds to @xmath318 hence , this is a very curious situation where the picture is completely steady on the current configuration ; however , there is unsteady wave propagation with piola - kirchhoff shear stress waves propagating on the reference configuration . now we consider the other solution @xmath319 . @xmath320 so that @xmath321 then we have unsteadiness in the current configuration as well as the reference . note that this case corresponds to choosing the wrong sign in solutions to to correspond to solutions of . however , if we now choose the ansatz for the motion to be @xmath322 then it can be checked that again @xmath323 are both solutions to system , but now only solution @xmath324 works ( for obvious reasons by following the previous argument ) : @xmath325 then again @xmath326 thus , we have a steady field on the current configuration , while , on the reference , stress / velocity / position waves are moving from the bottom to the top . while simple , it is important to appreciate that the generated exact solution in this extremely simple example can produce smooth analogs ( with continuous deformation gradient ) of the traveling wave profile shown in fig . [ f1 ] with arbitrarily small positive values of @xmath327 and @xmath328 . when @xmath327 is comparable to @xmath328 , this is a sequence of phase boundaries separating domains ; when @xmath329 , then the regions spanned within width @xmath328 are slip zones or shear bands separating undeformed blocks . the greater freedom that elastodynamics provides over elastostatics in producing microstructure can now be appreciated . while such deformations in elastostatics can only be produced with a multi - well energy function @xcite , such microstructures ( for @xmath330 ) in elastodynamics occur in this case of the simplified neo - hookean material ( with the energy density @xmath331 up to a multiplying constant ) , and persist for all times . indeed , multiple low - energy states corresponding to large total strains and static microstructures . ] are observed facts , and their representation consists of an important physical ingredient of solid mechanics . our intent here is to simply point out the greater freedom of displaying microstructure without any length scale in elastodynamics . by considering the case for ever smaller @xmath332 , it is also now perhaps easy to intuitively see why the wrinkled deformations of the isometric embedding problem may have a limiting status in elastodynamics , as the parameters @xmath327 and/or @xmath328 tend to zero . furthermore , it is an interesting question whether our elastodynamic model ( [ es2 ] ) corresponding to the degenerate isometric embedding also displays solutions that represent static and dynamic terminating lines of phase boundaries ( generalized disclinations ) and shear bands ( dislocations ) , ingredients of nature that may help to further understand physically the nature of the wrinkled embeddings from geometry . finally , it is important to recognize and accept that microstructure in solids is not about infinite refinement and , as is well - known , such scale - free deformations , no matter how exciting from the analytical point of view , are notoriously difficult to deal with practically in modeling endeavors ( following nature s guide , in some sense ) . thus , our considerations seem to point to the need for models that can represent both static , quasi - static ( _ i.e. _ , evolving microstructure in the absence of material inertia ) , as well as fully dynamic microstructures with in - built length scales and accounting for the dissipation due to microstructure evolution . suppose that we have a smooth , time - dependent two - dimensional manifold immersed in @xmath251 . then @xmath73 and @xmath32 can be defined from the manifold as in [ s2 ] . consider the initial data to be available on @xmath218 denoted by @xmath333 pointwise . then define the following quantities @xmath334 through the ordinary differential equations : [ momenta ] _ t m_u = _ 22 ^ 1l-2_12 ^ 1m+_11 ^ 1n , + _ t m_v = _ 22 ^ 2l-2_12 ^ 2m+_11 ^ 2n , with initial condition on @xmath335 specified as @xmath336 and on @xmath337 as @xmath338 . with fields @xmath334 available , solve @xmath339 with initial condition on @xmath216 to be @xmath245 . then , using this @xmath216 field along with @xmath340 and @xmath55 as time - dependent data , solve the pointwise ordinary differential equations for fields @xmath59 given by ( [ momenta ] ) with @xmath334 replaced by @xmath341 , with the initial condition on @xmath59 being @xmath342 , respectively . clearly , then we have @xmath343 however , then ( [ bal_mass ] ) implies that the constructed fields @xmath344 satisfy the balance of mass . finally , defining the stress components from ( [ lmn ] ) by using @xmath55 and the constructed fields @xmath218 as the data , and noting ( [ momenta ] ) and ( [ e31 ] ) , we find that every smooth time evolving two - dimensional manifold immersed in @xmath251 defines a solution of the balance laws of two - dimensional general continuum mechanics . it requires the imposition of further constraints to obtain the motions within this class that are consistent with the constitutive response of any specific material . for a given constitutive relation , the cauchy stress is determined from the velocity field @xmath4 and density @xmath216 . the case for ( incompressible and compressible ) inviscid fluids has been exposited , and the case of elasticity requires the deformation gradients to be calculated from the deformation field and to be inferred from velocity @xmath4 via time integration . hence , since @xmath345 have been determined by our argument above , the ability to satisfy the constitutive relations requires that metric @xmath0 be consistent with these extra constraints . let @xmath246 be the deformation gradient field of the mechanical body from a fixed stress - free elastic reference configuration . let @xmath346 denote the velocity gradient field . for simplicity , assume a local constitutive equation for the stress of the form : @xmath347 we also use notation @xmath348 and @xmath349 for the known functions for the christoffel symbols and the gauss curvature . a time - dependent set of mechanical fields @xmath350 and a time - dependent metric field @xmath0 related by the constitutive assumption @xmath351 are consistent in the sense that the balance of linear momentum and balance of mass are satisfied , and the metric is isometrically embedded in @xmath251 at each instant of time , provided that the following system of constrained partial differential equations of evolution are satisfied : @xmath352 \partial_k f_{ij } v_k - ( { { \bf l}}_v)_{ik}f_{kj } = -\partial_t f_{ij},\\[1 mm ] \partial_{1}\big(\rho u^{2}-\hat t_{11}({{\bf l}}_v,\rho , { { \bf f } } , { { \bf g}})\big ) + \partial_{2}\big(\rho uv- \hat t_{12}({{\bf l}}_v,\rho , { { \bf f } } , { { \bf g}})\big ) = -\partial_{t}(\rho u),\\[1 mm ] \partial_{1}\big(\rho uv - \hat t_{12}({{\bf l}}_v,\rho , { { \bf f } } , { { \bf g}})\big ) + \partial_{2}\big(\rho v^{2 } - \hat t_{22}({{\bf l}}_v,\rho , { { \bf f } } , { { \bf g}})\big)=-\partial _ { t}(\rho v ) , \end{cases}\ ] ] and @xmath353 \partial_{1 } m -\partial_{2}l = \gamma_{22}^{2}({{\bf g}},\nabla { { \bf g } } ) l -2\gamma_{12}^{2}({{\bf g}},\nabla { { \bf g } } ) m + \gamma_{11}^{2}({{\bf g}},\nabla { { \bf g } } ) n,\\[1 mm ] ln - m^2 = \kappa({{\bf g}},\nabla { { \bf g } } , \nabla^2 { { \bf g}}),\\[1 mm ] l \mathbf{e}^1 \otimes \mathbf{e}^1 + m \left(\mathbf{e}^1 \otimes \mathbf{e}^2 + \mathbf{e}^2 \otimes \mathbf{e}^1 \right ) + n \mathbf{e}^2 \otimes \mathbf{e}^2 = \mathbf{f } \left ( \rho \mathbf{u } \otimes \mathbf{u } - \mathbf{t}\left(\mathbf{l}_v , \rho , \mathbf{f } , \mathbf{g } \right ) \right ) , \end{cases}\ ] ] where @xmath354 is a tensor - valued function of its tensorial argument , and @xmath355 is the dual basis corresponding to the natural basis on the surface given by @xmath356 . the last tensorial equation consists of three independent equations . the equations in are to be considered as the constraints that determine the family of pairs of first and second fundamental forms of embedded manifolds consistent with a given mechanical state at any give time . conversely _ , the evolution of the mechanical fields following the first four equations must be constrained to the _ manifold _ defined by in the state - space of spatial @xmath357 fields . one may consider eliminating variables @xmath55 from the equations in to obtain three constraint equations for the three components of the metric field . abstractly , one may think of eliminating all of the equations in and replacing @xmath0 in the mechanical set of the first four equations as a spatially non - local term in the mechanical fields representing a solution of . notice that this remark applies as well to the steady problem of continuum mechanics ( where the right - hand sides of the third and fourth equations of are assumed to be @xmath264 ) and the time - dependence of boundary conditions ( or body forces that have been assumed to vanish here for simplicity ) drive the evolution of the mechanical problem . we note that the earlier sections of this paper dealing with the equations of incompressible and compressible fluid dynamics , and neo - hookean elastodynamics are special cases of the above system where the constitutive equation for the cauchy stress are independent of metric @xmath0 . we close by discussing some broad implications and possible extensions of the presented work . since the dynamics of the wrinkled solutions shadowing the incompressible or compressible euler equations are completely reversible nash - kuiper solutions corresponding to the metric for the developable surface , the usual irreversible entropy admissibility criteria become useless as selection criteria . this tells us that the only hope of selecting a unique physically meaningful solution would be chosen without recourse to time evolution , _ e.g. _ , artificial viscosity or energy minimization . for example , an energy minimization which penalizes second derivatives of @xmath166 would prefer the affine initial data over data with folds , so that the _ wild _ initial data would be ruled out . similarly , szkelyhidi @xcite , bardos - titi - wiedermann @xcite , bardos - titi @xcite , and bardos - lopes filho - niu - nussenzveig lopes - titi @xcite have shown that the viscosity criterion also eliminates the _ wild _ solutions for the euler equations . the condition of incompressibility and the poisson equation for the pressure remain as @xmath359 & \partial_{11}^{2}(u^{2})+2\partial_{12}(uv)+\partial_{22}(v^{2 } ) = -\delta p. \label{e93}\end{aligned}\ ] ] if we review all the previous arguments made for the inviscid euler equations leading up to and including [ s8 ] , we see all the conclusions we have made regarding the euler equations hold true for the navier - stokes equations modulo one crucial point . for the euler equations , metric @xmath3 provides the map from the steady shear , @xmath360 , a solution of the euler equations to the gauss - codazzi equations . however , for a shear solution of the navier - stokes equations , the right - hand sides of must vanish . therefore , instead of the fluid pre - image being a steady shear , the fluid pre - image must satisfy the diffusion equation : @xmath361 or @xmath362 thus , the pre - image is smooth so that , by , we have a smooth second fundamental form and a smooth embedding @xmath166 . however , this contradicts the non-@xmath182 property of our nash - kuiper wrinkled solution . thus , the only possibility is that , for the navier - stokes equations , there is no fluid pre - image of the nash - kuiper wrinkled solutions . one may be tempted to discount the physical relevance of the nash - kuiper theorem outlined here and in the recent work in @xcite . on the other hand , as noted in chen - glimm @xcite and others ( _ cf_. @xcite ) , viscous fluid turbulence arises with the imposition of an external force , which overcomes the viscous dissipation . hence , it may be that the addition of an external force would indeed bring us back to the nash - kuiper - gromov turbulence scenario . furthermore , the dynamics of defect microstructure like dislocations , phase and grain boundaries , triple junctions , and point defects in crystalline solids furnish a compelling physical argument for accepting / developing physically rigorous and practically computable models that account for the representation of microstructure , necessarily then not of infinite refinement and with a modicum of uniqueness in the predicted evolution of their fields . in analogy with a crumpled piece of paper that does not produce infinitely fine terminated folds and neither unfolds itself back to its original flat state , perhaps the correct notion of admissibility is to move to augmented physical models of continuum defect dynamics involving extra kinematics representing the microscopic , and hence smoothed , dynamics of discontinuity surfaces , their terminating lines , and point singularities of the fields of the original macroscopic model ( like nonlinear elasticity and navier - stokes ) , while accounting for the energetics of these defects and the dissipation produced owing to their motion . such partial differential equation - based augmentations of nonlinear elasticity theory have begun to emerge , _ e.g. _ , acharya - fressengeas @xcite , along with their interesting predictions of soliton - like dynamical behavior of nonsingular defects ( zhang et al . @xcite ) and their collective behavior . [ [ a - plausible - minimal - model - for - internally - stressed - elastic - materials ] ] a plausible minimal model for internally stressed elastic materials ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ the constrained evolution system ( [ constrained_system ] ) , written on the deforming configuration of an elastic body being tracked in lagrangian fashion , appears to pose an interesting model for internally stressed elastic bodies whose constitutive response in terms of the deformation gradient and a metric representing a stress - free state is known : @xmath363 \partial_{1}\hat t_{11}({{\bf f } } , { { \bf g}})+\partial_{2}\hat t_{12}({{\bf f } } , { { \bf g}})=\rho\ , d_t u,\\[1 mm ] \partial_{1}\hat t_{12}({{\bf f } } , { { \bf g}})+\partial_{2}\hat t_{22}({{\bf f } } , { { \bf g}})= \rho \,d_t v , \end{cases}\ ] ] and @xmath364 \partial_{1}m -\partial_{2}l = \gamma _ { 22}^{2}({{\bf g}},\nabla { { \bf g } } ) l -2\gamma_{12}^{2}({{\bf g}},\nabla { { \bf g } } ) m + \gamma_{11}^{2}({{\bf g}},\nabla { { \bf g } } ) n,\\[1 mm ] ln - m^2 = \kappa({{\bf g}},\nabla { { \bf g } } , \nabla^2 { { \bf g}}),\\[1 mm ] l \mathbf{e}^1 \otimes \mathbf{e}^1 + m \left(\mathbf{e}^1 \otimes \mathbf{e}^2 + \mathbf{e}^2 \otimes \mathbf{e}^1 \right ) + n \mathbf{e}^2 \otimes \mathbf{e}^2 = \mathbf{\hat{f } } \left ( \mathbf{t}\left(\mathbf{f } , \mathbf{g } \right ) \right ) , \end{cases}\ ] ] where @xmath365 , represent the spatial derivatives on the current configuration , @xmath366 represents the material time derivative operator , @xmath367 is a tensor - valued function of its tensorial argument , and @xmath355 is the dual basis corresponding to the natural basis on the surface given by @xmath356 . the last tensorial equation consists of three independent equations . first of all , we note that , on dimensional grounds , the identification of the second fundamental form with mechanical objects with physical dimensions of stress implies from the codazzi equations that the metric is physically dimensionless and the christoffel symbols have dimensions of reciprocal lengths . thus , the metric may be interpreted as describing strain . of course , this identification also implies that a material parameter with physical units of ( @xmath368 is required to make the gauss curvature equation dimensionally consistent , while introducing a length - scale into the traditional elastic problem of internal stress . next , we consider the equations in ( [ elast_constrained_system - a ] ) for given @xmath246 . considering , for the moment , the situation when the constitutive equation is independent of @xmath0 , this becomes a question of determining the metric , given the second fundamental form , such that an embedding exists in @xmath251 , which is the opposite of the isometric embedding problem that may be interpreted as the question of determining the second fundamental form , given a metric @xmath0 . at any rate , as some of our results show , this problem has a solution in many instances . it is also perhaps reasonable to expect that the situation does not change drastically even when the constitutive equation depends on @xmath0 , and is not degenerate in the sense that , for almost all @xmath246 , there are many solutions for @xmath0 ( this would necessitate the evolution equations for @xmath0 ) . based on this premise , the requirement of an embedding of @xmath0 in @xmath251 assumes a physical status replacing a separate constitutive equation for the evolution of @xmath0 that would be required for the mechanical problem otherwise . moreover , much like in systems displaying relaxation oscillations , the geometry and stability of states on the constraint manifold can , on occasion , lead to interesting dynamical behavior of the mechanical fields . the above model appears to have the possibility of being considered as a minimal model for the statics and dynamics of soft and biological materials ( _ e.g. _ , efrati et al @xcite , jin et al @xcite , and ambrosi et al . @xcite ) depending only on the knowledge of the elastic response of the material and the material constant needed to define the gauss curvature equation . * acknowledgments*. a. acharya acknowledges the support of the rosi and max varon visiting professorship at the weizmann institute of science , rehovot , israel , and was also supported in part by grants nsf - cmmi-1435624 , nsf - dms-1434734 , and aro w911nf-15 - 1 - 0239 . chen s research was supported in part by the uk epsrc science and innovation award to the oxford centre for nonlinear pde ( ep / e035027/1 ) , the uk epsrc award to the epsrc centre for doctoral training in pdes ( ep / l015811/1 ) , and the royal society wolfson research merit award ( uk ) . m. slemrod was supported in part by simons collaborative research grant 232531 . d. wang was supported in part by nsf grant dms-1312800 . a. acharya and c. fressengeas , continuum mechanics of the interaction of phase boundaries and dislocations in solids . in : _ differential geometry and continuum mechanics _ , pp . 123165 , eds . chen et al . ( eds . ) , springer proceedings in mathematics and statistics , * 137 * , 2015 . d. ambrosi , g. a. ateshian , e. m. arruda , s. c. cowin , j. dumais , a. goriely , g. a. holzapfel , j. d. humphrey , r. kemkemer , e. kuhl , j. e. olberding , l. a. taber , and k. garikipati , perspectives on biological growth and remodeling . _ j. mech . * 59 * ( 2011 ) , 863883 . c. w. bardos , m. c. lopes filho , d. niu , h. j. nussenzveig lopes , and e. s. titi , stability of two - dimensional viscous incompressible flows under three - dimensional perturbations and inviscid symmetry breaking . _ siam j. math . _ * 45 * ( 2013 ) , 18711885 . j. f. borisov , on the question of parallel displacement on a smooth surface and the connection of space forms of smooth surfaces with their intrinsic geometries . _ vestnik leningrad . _ * 15 * , 19 ( 1960 ) , 127129 . s. conti , c. de lellis , and l. szkelyhidi , the @xmath371-principle and rigidity for @xmath7 isometric embeddings . in : _ nonlinear partial differential equations _ , pp . 83116 , abel symp . * 7 * , springer : heidelberg , 2012 . d. deturck and d. yang , local existence of smooth metrics with prescribed curvature . in : _ nonlinear problems in geometry ( mobile , ala . , 1985 ) _ , pp . 3743 , _ contemp . _ * 51 * , ams : providence , ri , 1986 . x. zhang , a. acharya , n. j. walkington , and j. bielak , a single theory for some quasi - static , supersonic , atomic , and tectonic scale applications of dislocations . * 84 * ( 2015 ) , 145195 .

in multi - user mimo systems , one main challenge is the increased complexity and energy consumption of the signal processing to mitigate the interferences between multiple co - channel users . to achieve energy efficient transmission , the multi - user mimo system with very large antenna arrays at each base station ( known as `` massive mimo '' system ) has been advocated recently @xcite . the key result is that , with very large antenna arrays at each base station , both the intracell and intercell interferences can be substantially reduced with simple linear beamforming ( bf ) processing @xcite . on the other hand , full - duplex relaying ( fdr ) is a promising approach to improve the spectral efficiency ( se ) of relaying network while retains the merits of half - duplex relaying ( hdr ) ( e.g. , path loss reduction ) . in fdr , the relay transmits and receives simultaneously at the same frequency and time , but at the cost of a strong echo interference ( ei ) due to signal leakage between the relay output and input @xcite . to mitigate ei , three approaches have been investigated , i.e. , 1 ) passive cancellation , 2 ) time - domain cancellation @xcite and 3 ) spatial suppression @xcite . the passive cancellation relies on a combination of path loss , cross - polarization and antenna directionality @xcite . the time - domain cancellation is based on the fact that ei signal is known at full - duplex node . thus , it can be regenerated and removed in time - domain @xcite . in spatial suppression , ei is mitigated with the multiple transmit / receive antennas by approaches such as null - space projection @xcite . inspired by these works , a number of works have dedicated to the study of fdr protocol on both theory and testbed ( see @xcite and the references therein ) . to achieve spectral and energy efficient transmissions of multiple source - destination pairs , recent works sought to incorporate both hdr @xcite and fdr @xcite-@xcite with massive mimo . however , the aforementioned works on massive mimo are actually based on the assumption that the base stations or relays are equipped with a large number of high - quality transmit / receive radio frequency ( rf ) chains ( which are expensive and power - hungry ) . in contrast to conventional mimo system ( e.g. , at most 8 antennas in lte system ) , massive mimo must be built with low - cost components @xcite since the deploy cost and energy consumption of circuits will increase dramatically as the number of antennas grows very large . such low - cost components are prone to hardware imperfections ( e.g. , phase noise , nonlinear power amplifier , i / q imbalance , nonlinear low - noise amplifier and adc impairments ) , which must be considered in the design of practical massive mimo system . this paper focuses on the transceiver design for massive mimo full - duplex relaying ( mm - fdr ) with hardware impairments . the effect of hardware impairments is modeled using transmit / receive distortion noises @xcite . there are several challenges in the design of practical transceiver scheme in the considered system . the first is : _ how to deal with the ei cancellation without instantaneous ei channel ? _ ei cancellation is a critical problem in mm - fdr transceiver design which is not only important to reduce ei power , but also useful to reduce distortion noises caused by hardware impairments at the relay ( as will be shown in section iv ) . different from fdr with small - scale relay antenna arrays @xcite-@xcite , the instantaneous ei channel is usually not easy to obtain in mm - fdr . this is because the learning of ei channel requires training sequence with length not less than the number of relay antennas @xcite , which is prohibitive in mm - fdr as the channel coherent time is limited . the lack of ei channel makes traditional ei cancellation techniques ( e.g. , time - domain cancellation and spatial suppression ) difficult to apply . although passive cancellation does not relies on ei channel estimation , it usually can not provide satisfactory performance when used alone @xcite . another problem is : _ how to suppress distortion noises caused by hardware impairments at sources and destinations . _ different from multi - user interference ( mui ) , the distortion noise caused by transmit imperfection of source can be viewed as an interference signal with the same channel as transmit data . thus , it can not be suppressed by relay antenna arrays during coherent combining ( similar problem appears in reception at destinations ) . this causes performance ceiling on achievable rate as the number of relay antennas grows large , which degrades the gain of massive mimo significantly . in this paper , we propose practical transceiver scheme for mm - fdr with hardware impairments considering the above problems . different from the traditional mm - fdr protocol @xcite ( where sources and destinations are equipped with single antenna ) , we consider a general model where sources and destinations are allowed to equip with multiple antennas . the contributions are summarized as follows : * we first examine the limitation of traditional mm - fdr protocol under hardware impairments . in particular , we derive the upper bound on end - to - end achievable rate for traditional mm - fdr protocol with linear processing at the relay . the bound reveals that the achievable rate is limited by the hardware impairments at the sources and destinations , and performance ceiling appears as the number of relay antenna grows large . the result also implies that it is impossible to cancel the `` ceiling effect '' with linear processing , if sources and destinations are only equipped with single antenna . * based on the upper bound analysis , we propose a hardware impairments aware transceiver scheme ( hia scheme ) to mitigate the distortion noises by exploiting the statistical channel knowledge and antennas arrays of sources / destinations . the scheme needs no instantaneous knowledge of ei channel . the asymptotic end - to - end achievable rate of mm - fdr with hia scheme ( hia - mm - fdr ) is derived and the scaling behaviors of mui , ei and distortion noises are determined . * a joint degree of freedom and power optimization ( jdpo ) algorithm is presented to further improve the se of hia - mm - fdr . the paper is organized as follows . in section ii , we review the related work . the system model is described in section iii . the upper bound of achievable rate is analyzed in section iv . the hia scheme is proposed in section v and the jdpo algorithm is presented in section vi . simulation results are presented in section vii and some conclusions will be drawn at last . : @xmath0 and @xmath1 denote the expectation and variance . @xmath2 is @xmath3 identity matrix . @xmath4 , @xmath5 and @xmath6 denote conjugate , transpose and conjugate - transpose , respectively . @xmath7 , @xmath8 , @xmath9_{ij}$ ] , @xmath10 and @xmath11 denote the spectral radius , trace , @xmath12th element of matrix , @xmath13th largest eigenvalue and eigenvector with respect to @xmath13th largest eigenvalue of matrix @xmath14 , respectively . @xmath15 scales with @xmath16 means @xmath17 . the design and performance of using unlimited number of antennas at the base station in cellular system were first considered for independent antennas @xcite , and soon extended to the scenarios with spatial correlated antennas @xcite-@xcite . the asymptotic sinr for single cell cellular system was analyzed in @xcite . it has been shown that , with very large antenna arrays at the base station , a deterministic sinr ( also called the `` deterministic equivalent '' of sinr ) which depends only on the large - scale fading of channels can be achieved , if the transmit power is scaled by @xmath18 with perfect csi and @xmath19 with imperfect csi ( @xmath18 denotes the number of base station antennas ) . the authors in @xcite-@xcite have done considerable work to derive the `` deterministic equivalent '' of sinr for massive mimo system with spatial correlated antennas . in the field of massive mimo relaying , the se and ee of hdr with very large relay antenna arrays were investigated in @xcite . the mm - fdr with decode - and - forward ( df ) relay was first introduced in @xcite , and analyzed in end - to - end achievable rate as linear processing is employed . the asymptotic performance of amplify - and - forward based mm - fdr was considered in @xcite and the scaling behavior of the infinitely repeating echo interference was determined . only a few works considered the effect of hardware impairments on massive mimo system . a constant envelope signal design has been proposed in @xcite to facilitate the use of power - efficient rf power amplifiers . the authors in @xcite presented a low peak - to - average - power ratio ( par ) precoding solution to enable efficient implementation using non - linear rf components in massive mimo system . in fdr with small - scale antennas , the optimal precoding under limited adc dynamic range was studied in @xcite . however , these works can not provide much insight for the effect of hardware impairments on mm - fdr . consider the network with @xmath20 source - destination ( s - d ) pairs and a single full - duplex relay @xmath21 , where source @xmath22 wishes to communicate with destination @xmath23 ( @xmath24 ) with the help of @xmath21 . the relay adopts the df policy . it is assumed that the sources and destinations are equipped with @xmath25 and @xmath26 antennas respectively , while the relay is equipped with @xmath27 antennas ( @xmath28 for reception and @xmath29 for transmission ) . we are interested in the large-@xmath30 regime , i.e. , @xmath31 . @xmath25 , @xmath26 and @xmath20 can be either fixed or scale with @xmath32 . let @xmath33 be the channel matrix from @xmath22 to receive antenna array of relay and let @xmath34 be the channel matrix from @xmath23 to transmit array of relay . let @xmath35 denote the ei channel matrix between transmit and receive arrays of relay . the spatial correlation of each mimo channel is characterized by the kronecker model @xcite . thus , @xmath36 can be expressed as @xmath37 , where @xmath38 represents the large - scale fading . @xmath39 and @xmath40 characterize the spatial correlation of received signals across receive array of relay and that of transmitted signals across transmit array of @xmath22 . @xmath41 consists of the random components of channels whose elements are i.i.d with distribution @xmath42 . based on the same model , the channel matrices @xmath43 and @xmath44 can be expressed as @xmath45 and @xmath46 . to facilitate the analysis , we assume the following conditions on correlation matrices , i.e. , @xmath47 , where @xmath48 , * * a1 * : the spectral radius of @xmath49 is bounded by a constant , i.e. , @xmath50 . * * a2 * : @xmath49 is a hermitian and teoplitz matrix and has unit diagonal elements . the former is a common assumption in the studies of massive mimo which follows from energy conservation @xcite and * a2 * corresponds to the case of uniform linear array ( ula ) @xcite . to characterize the effect of hardware imperfections , we adopt the new signal model from @xcite . we model the effect of imperfect transmit rf chain by adding , per transmit antenna , an independent zero - mean gaussian `` transmit distortion noise '' , whose power is proportional to the signal power transmitted at that antenna . the experimental results in @xcite have shown that the independent gaussian distortion noise model closely captures the joint effect of imperfect components in transmit rf chain . let @xmath51 \in \mathbb c^{n_s\times 1}$ ] and @xmath52\in \mathbb c^{n_t\times 1}$ ] denote the transmit vectors of source @xmath22 and relay at time instant @xmath53 . based on the above model , the distorted transmit signals can be expressed as @xmath54 = { { \mathbf{x}}_{s , k}}\left [ u \right ] + { { \mathbf{t}}_{s , k}}\left [ u \right ] \hspace{.5em } { \rm and } \hspace{.5em } { { \mathbf{\tilde x}}_r}\left [ u \right ] = { { \mathbf{x}}_r}\left [ u \right ] + { { \mathbf{t}}_r}\left [ u \right]\ ] ] where the distortion noises @xmath55\sim { \cal cn}\left ( { 0,{\nu _ { s , k}}{\text{diag}}\left ( { \mathbb{e}\left [ { { { \mathbf{x}}_{s , k}}\left [ u \right]{\mathbf{x}}_{s , k}^h\left [ u \right ] } \right ] } \right ) } \right)$ ] and @xmath56\sim{\cal c}{\cal n}\left ( 0,\nu _ r \times \right.$ ] @xmath57{{\bf{x}}_r^h}\left [ u \right ] } \right ] } \right ) \right)$ ] . note that @xmath58 ( @xmath59 ) characterizes the level of transmit imperfection . for example , @xmath60 ( @xmath61 ) corresponds to the conventional assumption of perfect transmit rf chains . the quality of transmit rf chains degrades as @xmath62 ( @xmath63 ) increases . we model the effect of imperfect receive rf chain by adding , per receive antenna , an independent zero - mean gaussian `` receive distortion noise '' , whose variance is proportional to the signal power received at that antenna . more precisely , let @xmath64\in \mathbb c^{n_r\times 1}$ ] and @xmath65 \in \mathbb c^{n_d\times 1}$ ] be the undistorted received signals of the relay and destination @xmath23 at time instant @xmath53 , the distorted received signals can be expressed as @xmath66 = { { \bf{y}}_r}\left [ u \right ] + { { \bf{r}}_r}\left [ u \right ] \hspace{.5em } { \rm and } \hspace{.5em } { { { \bf{\tilde y}}}_{d , k}}\left [ u \right ] = { { \bf{y}}_{d , k}}\left [ u \right ] + { { \bf{r}}_{d , k}}\left [ u \right ] \end{aligned}\ ] ] where the distortion noises @xmath67 \sim { \cal c}{\cal n } ( { 0,{\mu _ r}{\rm{diag } } ( { { \mathbb{e } } [ { { { \bf{y}}_r } [ u ] { { \bf{y}}_r^h } [ u ] } ] } ) } ) $ ] and @xmath68 \sim{\cal cn } ( 0,{\mu _ { d , k } } { \text{diag } } ( { \mathbb{e } [ { { { \mathbf{y}}_{d , k } } [ u ] { \mathbf{y}}_{d , k}^h [ u ] } ] } ) ) $ ] . @xmath69 ( @xmath70 ) characterizes the level of receive imperfection . the experimental studies in @xcite have shown that the independent gaussian noise model is a good approximation to the joint effect of imperfect components in receive rf chain . at time instant @xmath53 , all sources transmit signals @xmath71 $ ] ( @xmath72 ) to the relay simultaneously . meanwhile , the relay broadcasts the decoded signals to the destinations . based on the models in ( [ 2 - 3 ] ) and ( [ 2 - 4 ] ) , the received signals at the relay and @xmath23 can be expressed as @xmath73 = \sum\limits_{k = 1}^k { { \mathbf{h}}_{sr , k}}\left ( { { { \mathbf{x}}_{s , k}}\left [ u \right ] + { { \mathbf{t}}_{s , k}}\left [ u \right ] } \right ) \\ & + { { \mathbf{h}}_{ei}}\left ( { { { \mathbf{x}}_r}\left [ u \right ] + { { \mathbf{t}}_r}\left [ u \right ] } \right ) + { { \mathbf{r}}_r}\left [ u \right ] + { { \mathbf{n}}_r}\left [ u \right ] \\ & { { \mathbf{\tilde y}}_{d , k}}\left [ u \right ] = { \mathbf{h}}_{rd , k}^h\left ( { { { \mathbf{x}}_r}\left [ u \right ] + { { \mathbf{t}}_r}\left [ u \right ] } \right ) + { { \mathbf{r}}_{d , k}}\left [ u \right ] + { { \mathbf{n}}_{d , k}}\left [ u \right ] \end{aligned}\ ] ] to keep the complexity low , it is assumed that a singe data steam is transmitted at each source and each node employs only linear processing . in particular , @xmath22 transmits the unit - power symbol @xmath74 $ ] using the unitary bf vector @xmath75 . therefore , the transmit vector of @xmath22 can be expressed as @xmath71 = \sqrt{e_{s , k}}{\bf p}_{s , k } s_k[u]$ ] , where @xmath76 denotes the transmit power . relay combines the received signal by multiplying the receive bf matrix @xmath77 $ ] , i.e. , @xmath78 = { \bf{w}}_r^h{{{\bf{\tilde y}}}_r}\left [ u \right]$ ] . the @xmath79th element of @xmath78 $ ] @xmath80 = \sqrt{e_{s , k}}{\mathbf{w}}_{r , k}^h{{\mathbf{h}}_{sr , k}}{{\mathbf{p}}_{s , k}}{s_k}\left [ u \right ] \\ & + \sqrt{e_{s , j } } \sum\limits_{j = 1,j \ne k}^k { { \mathbf{w}}_{r , k}^h{{\mathbf{h}}_{sr , j}}{{\mathbf{p}}_{s , j}}{s_j}\left [ u \right ] } + { \mathbf{w}}_{r , k}^h\sum\limits_{j = 1}^k { { { \mathbf{h}}_{sr , j}}{{\mathbf{t}}_{s , j}}\left [ u \right ] } \\ & + { \bf{w}}_{r , k}^h{{\bf{h}}_{ei}}\left ( { { { \bf{x}}_r}\left [ u \right ] + { { \bf{t}}_r}\left [ u \right ] } \right ) + { \bf{w}}_{r , k}^h{{\bf{r}}_r}\left [ u \right ] + { \bf{w}}_{r , k}^h{{\bf{n}}_r}\left [ u \right ] \end{aligned}\ ] ] is used to decode the symbol of @xmath22 . the relay forwards the decoded symbol using transmit bf matrix @xmath81 $ ] . the transmit vector of the relay can be expressed as @xmath82\ ! = \ ! { { \mathbf{w}}_t}{\mathbf{\lambda } _ r^{1/2}}{\mathbf{s}}\left [ { u - d } \right]$ ] , where @xmath83\!=\![{{s_1}}\left [ { u - d } \right ] , \cdots,{{s_k}}\left [ { u - d } \right]]^t$ ] and @xmath84 is the power allocation matrix at the relay . @xmath85 denotes the processing delay of the relay . to meet the relay s power constraint , @xmath82 $ ] must satisfy @xmath86{\bf{x}}_r^h\left [ u \right ] } \right ] } \right ) \!\ ! = \!\ ! \sum_{l = 1}^k { { e_{r , l } } } \!\!\le\ ! \!e_r^{\rm max}$ ] . @xmath23 uses the unitary bf vector @xmath87 to combine the received signal @xmath88 $ ] . the combined signal is expressed as @xmath89 = \sqrt{e_{r , k}}{\mathbf{p}}_{d , k}^h{\mathbf{h}}_{rd , k}^h{{\mathbf{w}}_{t , k}}{s_k}\left [ { u - d } \right ] \\ & + { \mathbf{p}}_{d , k}^h\sum\limits_{j = 1,j \ne k}^k \sqrt{e_{r , j } } { { \mathbf{h}}_{rd , k}^h{{\mathbf{w}}_{t , j}}{s_j}\left [ { u - d } \right ] } \\ & + { \mathbf{p}}_{d , k}^h{\mathbf{h}}_{rd , k}^h{{\mathbf{t}}_r}\left [ u \right ] + { \mathbf{p}}_{d , k}^h{{\mathbf{r}}_{d , k}}\left [ u \right ] + { \mathbf{p}}_{d , k}^h{{\mathbf{n}}_{d , k}}\left [ u \right ] \end{aligned}\ ] ] this section analyzes the upper bound on achievable rate of mm - fdr with single antenna sources and destinations . the goal is to reveal the fundamental limitation of previous mm - fdr protocol with single antenna sources and destinations in combating distortion noises . moreover , the bound provides us important insight on the design of practical transceiver scheme to mitigate the distortion noises , when multiple antennas are available at sources and destinations ( see section v ) . as @xmath90 , we set @xmath91 , replace @xmath92 ( @xmath93 ) with @xmath94 , and let @xmath95 $ ] and @xmath96 = { \left [ { { \mathbf{t}}_{s,1}^t\left [ u \right ] , \cdots , { \mathbf{t}}_{s , k}^t\left [ u \right ] } \right]^t}$ ] . we consider a block - fading channel with coherence time @xmath97 ( symbol times ) , where @xmath98 are used for uplink training of each source / destination , and the remaining @xmath99 are used for data transmission . the upper bound on achievable rate is obtained by assuming perfect knowledge of @xmath100 , @xmath101 and @xmath102 can be provided with pilot signals , and meanwhile , the muis ( i.e. , the second terms of right - hand side of ( [ 2 - 6 ] ) and ( [ 2 - 7 ] ) ) and ei term @xmath103 $ ] can be somehow cancelled . this gives us the following upper bound on end - to - end achievable rate @xmath104 where @xmath105 and @xmath106 denote the achievable rates of @xmath107 and @xmath108 channels , respectively . according to ( [ 2 - 6 ] ) , ( [ 2 - 7 ] ) and the assumptions in the above , we have @xmath109 } \right\|}^2 } } \right ] } + 1 } } } \right ) \end{aligned}\ ] ] where @xmath110 is given as @xmath111 with @xmath112{\bf t}^h_s[u]\right]$ ] , @xmath113{\bf t}^h_r[u]\right]$ ] and @xmath114{\bf r}^h_s[u]\right]$ ] denoting the covariance matrices of distortion noises @xmath115 $ ] , @xmath116 $ ] and @xmath117 $ ] , respectively . _ 1 ) achievable rate of @xmath118 channel : _ since @xmath119 is only related with @xmath105 , it can be optimized separately to maximize @xmath105 . from ( [ 3 - 0 - 2 ] ) , the optimization problem is equivalent to the generalized rayleigh quotient problem , which can be solved as @xmath120 . the resultant upper bound on @xmath105 ( denoted by @xmath121 ) can be expressed as @xmath122 where @xmath123 is defined as @xmath124 and is independent of @xmath100 . the second step follows from the matrix inversion lemma @xcite . [ t1 ] assuming that the assumptions * a1 * and * a2 * hold . as @xmath125 , the achievable rate of @xmath118 channel in the large-@xmath30 regime is bounded as @xmath126 where @xmath127 , @xmath128 is determined by the following fixed - point algorithm with @xmath129 @xmath130 see appendix - b . the convergence of the fixed - point algorithm in theorem [ t1 ] has been proved in @xcite . _ 2 ) achievable rate of @xmath108 channel : _ from ( [ 3 - 0 - 2 ] ) , the achievable rates of @xmath118 and @xmath108 channels are coupled through @xmath113{\bf t}^h_r[u]\right]= { \rm diag}\left(\mathbb e\left[{\bf w}_t { \bf \lambda}_r{\bf w}_t^h\right]\right)$ ] , which makes the design of @xmath131 very challenging . however , with the following theorem we show that this coupling disappears in the large-@xmath30 regime , which allows a simple upper bound for achievable rate of @xmath108 channel . [ t2 ] assuming that the assumptions * a1 * and * a2 * hold . as @xmath125 , the achievable rate of @xmath118 channel is independent of @xmath131 . the achievable rate of @xmath108 channel in the large-@xmath30 regime is bounded as ( [ 3 - 3 - 1 ] ) , which is achieved by the eigen bf @xmath132 . see appendix - c . @xmath133 ' '' '' with theorems [ t1 ] and [ t2 ] , the upper bound of end - to - end achievable in the large-@xmath30 regime can be expressed as @xmath134 [ rm1 ] the bound in ( [ 3 - 4 ] ) is derived by assuming linear processing and fixed transmit powers at sources and relay , which is in general not capacity achieving . one exception is when @xmath135 and @xmath31 . from theorem [ t2 ] , the design of receive and transmit bf matrices is decoupled as @xmath31 . moreover , the distortion noises are assumed to be circularly symmetric complex gaussian distributed and independent of the desired signal . thus , based on the results in @xcite , linear processing along with power control is sufficient to achieve the capacity . when @xmath135 , the terms @xmath136 and @xmath137 in ( [ 3 - 3 ] ) vanish . by neglecting the low order term ( in @xmath32 ) , the bound in ( [ 3 - 4 ] ) reduces to @xmath138 , which is independent of @xmath76 and @xmath139 . thus , ( [ 3 - 4 ] ) is the upper bound for capacity of @xmath140 channel as @xmath135 and @xmath31 . different from remark [ rm1 ] , we consider the general case in which @xmath20 can be either fixed or scales with @xmath32 . with theorem [ t1 ] and theorem [ t2 ] , the effect of hardware impairments is still hard to analysis since @xmath141 is not in closed - form . to make the analysis tractable , we assume @xmath142 @xmath143 . using eigenvalue decomposition @xmath144 on @xmath145 in ( [ 3 - 3 ] ) , we have @xmath146 by examining ( [ 3 - 5 ] ) and the expression of @xmath147 in ( [ 3 - 3 ] ) , @xmath148 is an @xmath149 term . note that this is valid for arbitrary @xmath150 . thus , we can conclude that @xmath151 . replacing @xmath152 in @xmath153 with the symbol @xmath149 and substituting the result into ( [ 3 - 2 ] ) , @xmath105 can be written as ( [ 3 - 6 ] ) . @xmath154 ' '' '' for convenience , we further assume @xmath155 . note that this assumption does not effect the basis conclusions of the analysis . by neglecting the low order terms ( in @xmath32 ) in ( [ 3 - 6 ] ) and ( [ 3 - 3 - 1 ] ) , the upper bound on end - to - end achievable rate reduces to @xmath156 one can make several observations from ( [ 3 - 7 ] ) : \1 ) as @xmath157 , the effect of hardware impairments at the relay and ei disappears in the large-@xmath30 regime and the upper bound converges to that in remark [ rm1 ] . the end - to - end achievable rate is limited by distortion noises caused by hardware impairments at sources and destinations . the explanation is that the transmit distortion can be viewed as an interference signal the same channel as data of @xmath22 from ( [ 2 - 5 ] ) . thus , it can not be suppressed after combining by @xmath119 . the power of distortion due to receive imperfection of @xmath23 is proportional to that of desired receive signal ( after eigen bf ) and also can not be reduced by the transmit bf scheme of relay . thus , there is a finite ceiling on end - to - end achievable rate as @xmath31 . in fact , this poses a major limitation on mm - fdr with single antenna sources and destinations , which degrades the gain of massive array of relay greatly . \2 ) as @xmath20 scales with @xmath32 , the effect of hardware impairments at the relay and ei is not negligible . since the ei channel is typically stronger than desired channels , the @xmath107 channel becomes the bottleneck of end - to - end achievable rate . this indicates that it is of great importance to suppress ei to avoid the bottleneck effect . in this section , we consider a general model where sources and destinations are equipped with an arbitrary number of antennas . we first present a low complexity hardware impairments aware transceiver scheme ( referred to as hia scheme ) . then the achievable rate of mm - fdr with proposed hia scheme ( hia - mm - fdr ) is analyzed in the large-@xmath30 regime and the effect of hardware impairments is discussed . the optimal transceiver scheme for mm - fdr to optimize achievable rate is hard to find since the transmissions of @xmath158 and @xmath159 channels are deeply coupled due to the presence of ei and distortion noises . even without ei and distortion noises , the problem has been proved np - hard @xcite . as @xmath102 is an @xmath160 matrix , the learning of @xmath102 requires training sequence with length no less than @xmath32 . this task is prohibitive as the duration of channel coherent time is limited . thus , we assume that passive ei cancellation @xcite has been used and instantaneous ei channel is not available . a low complexity hia scheme is proposed to mitigate ei ( after passive cancellation ) and distortion noises . as will be shown in simulations , the hia scheme along with passive ei cancellation provides satisfactory gain compared to hdr . _ 1 ) transceiver design at the relay : _ as the variance of ei channel is typically stronger than desired channels , it is of great importance to control the ei power . from section iv - b , ei suppression is also important in reducing distortion noise due to hardware impairments of relay . we consider a two - stage bf scheme at the relay . the receive and transmit bf matrices are expressed as @xmath161 and @xmath162 . _ outer bf matrices : _ the outer bf matrices @xmath163 ( @xmath164 ) and @xmath165 ( @xmath166 ) are designed to suppress ei . @xmath167 and @xmath168 are design parameters . due to the lack of @xmath102 , @xmath169 and @xmath170 are allowed to depend only on the statistical knowledge of ei channels . with ( [ 2 - 6 ] ) , we design an average ei power minimization problem , i.e. , @xmath171 \end{aligned}\ ] ] in ( [ 4 - 1 ] ) , the design of @xmath169 and @xmath170 is coupled with the inner bf matrix , which makes closed - form solution inaccessible . thus , it is of great interest to find new target function to decouple the problem . for such , using cauchy - schwarz inequality twice , an upper bound on average ei power can be obtained as @xmath172 then the problem ( [ 4 - 1 ] ) can be rewritten approximately as @xmath173 \\ & \bigg . + { { \nu _ r}\sum\limits_{l = 1}^k { { e_{r , k } } } \mathbb e\left [ { { \text{tr}}\left ( { { \mathbf{p}}_r^h{{\mathbf{h}}_{ei}}{\mathbf{h}}_{ei}^h{{\mathbf{p}}_r } } \right ) } \right ] } \bigg\ } \\ & = { \text{arg}}\mathop { { { \min}}}\limits_{{{\mathbf{p}}_r},{{\mathbf{p}}_t } } \left\ { { \text{tr}}\left ( { { { \mathbf{\underline{w}}}_t}{\lambda _ r}{\mathbf{\underline{w}}}_t^h } \right){\text{tr}}\left ( { { \mathbf{p}}_t^h{{{\mathbf{\tilde c}}}_{ei}}{{\mathbf{p}}_t } } \right)\right . \\ & \bigg . \times{\text{tr}}\left ( { { \mathbf{p}}_r^h{{\mathbf{c}}_{ei}}{{\mathbf{p}}_r } } \right)+ { \nu _ r}{n_t}\sum\limits_{l = 1}^k { { e_{r , k } } } { \text{tr}}\left ( { { \mathbf{p}}_r^h{{\mathbf{c}}_{ei}}{{\mathbf{p}}_r } } \right ) \bigg\ } \end{aligned}\ ] ] the multiplicative term @xmath174 has been removed since it has no effect on the optimal solution . the second step follows from a similar derivation of ( [ a3 ] ) . from ( [ 4 - 3 ] ) , the columns of @xmath175 ( @xmath176 ) are composed of the eigenvectors corresponding to the @xmath167 ( @xmath168 ) smallest eigenvalues of @xmath177 ( @xmath178 ) , i.e. , @xmath179 \\ & { \bf \dot{p}}_t = \left [ { { { \mathbf{u}}_{{n_t } - { a_t } + 1}}\left ( { { { { \mathbf{\tilde c}}}_{ei } } } \right),{{\mathbf{u}}_{{n_t } - { a_t } + 2}}\left ( { { { { \mathbf{\tilde c}}}_{ei } } } \right ) , \cdots , { { \mathbf{u}}_{{n_t}}}\left ( { { { { \mathbf{\tilde c}}}_{ei } } } \right ) } \right ] \end{aligned}\ ] ] in ( [ 4 - 3 - 1 ] ) , @xmath167 and @xmath168 can be viewed as parameters to balance the allowable ei power and available degree of freedom ( dof ) for data transmission , which should be optimized with respect to specific metric . this will be considered in section vi . _ inner bf matrix : _ the inner bf matrices @xmath180 and @xmath181 are designed to realize the multi - user communication . @xmath182 and @xmath183 can be designed with different criteria , e.g. , maximizing the desired signal power which corresponding to the eigen bf , or minimizing the mui which corresponding to the zero - forcing ( zf ) scheme @xcite . we adopt the latter one since the zf scheme is known to approach the asymptotic limit ( in @xmath32 ) of achievable rate faster as the number of relay antennas increases @xcite . for given bf vectors at sources and destinations , define effective channel vectors of @xmath107 and @xmath108 channels as @xmath184 and @xmath185 , and let @xmath186 $ ] and @xmath187 $ ] , the inner bf matrices can be written as @xmath188 where @xmath189 is a diagonal normalized matrix with @xmath190_{l , l}={{\mathbf{e}}_l^h{{\left ( { { \mathbf{\underline{h}}}_{rd}^h{{\mathbf{\underline{h}}}_{rd } } } \right)}^ { - 1}}{{\mathbf{e}}_l}}$ ] . in practice , @xmath191 and @xmath192 should be estimated in order to compute ( [ 4 - 4 ] ) . this will be considered in section v - b . note that ( [ 4 - 4 ] ) requires @xmath193 so that @xmath167 and @xmath168 can be selected to ensure the invertibility of @xmath194 and @xmath195 . _ 2 ) transceiver design at the sources and destinations : _ similar to @xcite , we assume no instantaneous knowledge of channels at sources and destinations . this is reasonable since the amount of feedback could be very huge and unaffordable in mm - fdr . however , it is assumed that the local statistical knowledge of channels ( i.e. , @xmath196 for @xmath22 and @xmath197 for @xmath23 ) can be obtained . as observed from ( [ 3 - 7 ] ) , when @xmath125 , the achievable rates of @xmath118 channel and @xmath108 channel are limited by the transmit distortion noise at @xmath22 and receive distortion noise at @xmath23 . this motivates us to design @xmath75 and @xmath87 to suppress these negative factors with the antenna arrays of @xmath22 and @xmath23 . _ design of @xmath75 : _ intuitively , according to ( [ 2 - 5 ] ) we can design the following problem @xmath198}}{{\mathbb e\left [ { { \rm{tr}}\left ( { { \bf{h}}_{sr , k}{\bf{\theta } } _ { s , k}^\textsf{t}{{\bf{h}}_{sr , k}^h } } \right ) } \right]}}\\ & = \arg \mathop { \max}\limits_{\|{{\bf{p}}_{s , k}}\|=1 } \frac{{{\bf{p}}_{s , k}^h{{{\bf{\tilde c}}}_{sr , k}}{{\bf{p}}_{s , k}}}}{{{\nu _ { s , k}}{\rm{tr}}\left ( { { { { \bf{\tilde c}}}_{sr , k}}{\rm{diag}}\left ( { { \bf{p}}_{s , k}^h{{\bf{p}}_{s , k } } } \right ) } \right ) } } \\ & = \arg \mathop { \max}\limits_{\|{{\bf{p}}_{s , k}}\|=1 } \frac{1}{{{\nu _ { s , k}}}}{\bf{p}}_{s , k}^h{{{\bf{\tilde c}}}_{sr , k}}{{\bf{p}}_{s , k } } \end{aligned}\ ] ] where @xmath199 is the covariance matrix of @xmath200 , i.e. , @xmath201{\bf t}^h_{s , k}[u]]= \nu_{s , k}{\rm diag}\left({\bf p}_{s , k } { \bf p}^h_{s , k}\right)$ ] . the second step follows from a similar derivation with that in ( [ a3 ] ) and the last step is based on assumption * a2*. from ( [ 2 - 5 ] ) , the right - hand side of ( [ 4 - 5 - 0 ] ) can be interpreted as the average signal to transmit distortion noise ratio at @xmath22 . the solution of problem ( [ 4 - 5 - 0 ] ) is @xmath202 _ design of @xmath87 : _ similarly , based on ( [ 2 - 5 ] ) , the received bf vector at @xmath23 can be derived by solving the problem @xmath203}}{{{\mu _ { d , k}}{\mathbf{p}}_{d , k}^h{\text{diag}}\left ( { \mathbb{e}\left [ { { \mathbf{h}}_{rd , k}^h{{\mathbf{h}}_{rd , k } } } \right ] } \right){{\mathbf{p}}_{d , k}}}}$ ] , which results in @xmath204 during a training phase , each source / destination transmits pilot sequence sequentially which allows the relay to compute the estimates of channels . with hia scheme , it is sufficient to estimate the effective channel vectors @xmath205 and @xmath206 . this allows a reduced dimension estimation scheme . for brevity , we consider the estimation of @xmath205 . let @xmath207 and @xmath208 denote the pilot sequence and power of each pilot symbol . @xmath209 is multiplied by @xmath210 and transmitted by @xmath22 . the received pilot matrix at the relay is expressed as @xmath211 where @xmath212 denotes the transmit distortion noise of @xmath22 , whose @xmath13th column has distribution ( based on model ( [ 2 - 3 ] ) ) @xmath213 . @xmath214 denotes receive distortion noise at the relay . with model ( [ 2 - 4 ] ) , the @xmath13th column of @xmath215 has distribution @xmath216 ) } ) $ ] . @xmath217 is the awgn matrix , whose elements are i.i.d and distributed as @xmath42 . by multiplying both side of ( [ 4 - 5 ] ) with @xmath218 , we have @xmath219 from ( [ 4 - 6 ] ) , as a merit of hia scheme , the total length of pilot sequences to obtain the estimates of all effective channels can be as less as @xmath220 ( when @xmath98 is set to 1 ) . [ t3 ] the lmmes estimator of effective channel @xmath205 can be expressed as @xmath221 where @xmath222 is the covariance matrix of @xmath205 , which can be expressed as @xmath223 . @xmath224 is given by @xmath225 and @xmath226 the real effective channel @xmath227 can be decomposed as @xmath228 with @xmath229 denoting the estimation error . @xmath230 and @xmath231 are uncorrelated whose covariance matrices can be expressed as @xmath232 and @xmath233 , respectively . a sketch of the proof is presented in appendix - d . different from that with ideal hardware @xcite , the lmmse estimator is not equivalent to mmse estimator , since the received pilot signal is corrupted by the term @xmath234 ( due to transmit imperfection of @xmath22 ) , which is not independent with the channel to be estimated . there might exist non - linear estimator that results in smaller mse . however , the difference should be small since the distortion noises are relatively weak . the lmmse estimator of @xmath206 can be obtained using the same approach . for further analysis , we define @xmath235 and @xmath236 as the covariance matrix and lmmse estimates of @xmath206 , respectively . moreover , we define @xmath237 as the covariance matrix of @xmath236 , where @xmath238 is given by @xmath239 this subsection analyzes the achievable rate of hia - mm - fdr . the achievable rate expressions are derived based on the bounding technique in @xcite . _ 1 ) achievable rate of @xmath240 channel : _ by treating @xmath241}s_k[u]$ ] as the desired signal at @xmath242 channel , and approximating the remaining terms in ( [ 2 - 6 ] ) , i.e. , @xmath243 \!-\!\sqrt{e_{s , k}}{\mathbb e [ { { \bf{w}}_{r , k}^h{{\bf{h}}_{sr , k}}{{\bf{{\dot{p}}}}_{s , k } } } ] } s_k[u]$ ] , using the worst - case uncorrelated additive gaussian noise with the same variance , the rate in ( [ 4 - 11 ] ) is achievable on @xmath244 channel , where @xmath245 , @xmath246 , @xmath247 and @xmath248 denote the ei , mui , effective distortion due to transmit imperfection of sources and effective distortion due to receive imperfection of relay respectively , which are given by @xmath249 } \right|}^2}}}{{{e_{s , k}}{\rm{var}}\left ( { { \bf{w}}_{r , k}^h{{\bf{h}}_{sr , k}}{{\bf{{\dot{p}}}}_{s , k } } } \right){+ } \mathbb e\left [ { { { \textsf{ei}}}{{{}}_k } } \right ] + \mathbb e\left [ { { \rm{\textsf{mui}}}{{{}}_{r , k } } } \right ] + \mathbb e\left [ { { { \textsf{d}}}_{r , k}^{{\textsf{t } } } } \right ] + \mathbb e\left [ { { { \textsf{d}}}_{r , k}^{{\textsf{r } } } } \right ] + \mathbb e\left [ { { { \left\| { { \bf{w}}_{r , k } } \right\|}^2 } } \right ] } } } \right)\ ] ] ' '' '' @xmath250{\mathbf{t}}_r^h\left [ u \right ] } \right ] } \right){\bf{h}}_{ei}^h{{\bf{w}}_{r , k } } \\ { { \textsf{mui}}_{r , k } } = \sum\limits_{j = 1,j \ne k}^k { { e_{s , j}}{{\left| { { \bf{w}}_{r , k}^h{{\bf{h}}_{sr , j}}{{\bf{{\dot{p}}}}_{s , j } } } \right|}^2 } } \\ \textsf{d}_{r , k}^\textsf{t } = \sum\limits_{j = 1}^k { { \bf{w}}_{r , k}^h{{\bf{h}}_{sr , j } } \mathbb e\left [ { { { \bf{t}}_{s , j}}[u]{\bf{t}}_{s , j}^h[u ] } \right]{\bf{h}}_{sr , j}^h{{\bf{w}}_{r , k } } } \\ \textsf{d}_{r , k}^\textsf{r } = { \bf{w}}_{r , k}^h \mathbb e\left [ { { { \mathbf{r}}_r}\left [ u \right]{\mathbf{r}}_r^h\left [ u \right ] } \right]{{\bf{w}}_{r , k } } \end{array}\ ] ] wherein , the bf matrices at the relay are given by @xmath251 where @xmath252 $ ] ( @xmath93 ) and @xmath253 is a diagonal normalized matrix with @xmath254_{l , l}={{\mathbf{e}}_l^h{{\left ( { { \mathbf{\underline{\hat{h}}}}_{rd}^h{{\mathbf{\underline{\hat{h}}}}_{rd } } } \right)}^ { - 1}}{{\mathbf{e}}_l}}$ ] . note that we have replaced the effective channels in inner bf matrices with their estimates obtained in theorem [ t3 ] . [ t4 ] assume that the assumptions * a1 * and * a2 * hold , and @xmath167 is selected so that @xmath255 . with hia scheme , the achievable rate of @xmath118 channel in the large-@xmath30 regime is given by @xmath256 , where the powers of desired signal and awgn are @xmath257 } \right|}^2 } } } = e_{s , k}$ ] and @xmath258}= { ( { { \rm{tr } } ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } ) } ) ^ { - 1 } } $ ] , and @xmath259 = \frac{{\beta _ { ei}}{\rm{tr}}\left ( { { { { \bf{\tilde c}}}_{ei}}{{\bf \omega } _ r } } \right){\rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k}}{\bf{{\dot{p}}}}_r^h{{\bf{c}}_{ei}}{{\bf{{\dot{p}}}}_r } } \right)}{\left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^2 } \\ & \mathbb e[{{\textsf{mui}}_{r , k } } ] = \sum_{j = 1,j \ne k}^k \frac{{e_{s , j}}{\rm{tr}}\left ( { \left ( { { { \bf{\underline{c}}}_{sr , j } } - { { { \bf{\hat { \underline{c}}}}}_{sr , j } } } \right){{{\bf{\hat { \underline{c}}}}}_{sr , k } } } \right)}{\left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^ { 2}}\\ & \mathbb e[\textsf{d}_{r , k}^\textsf{t } ] = \frac{{\nu _ { s , k}}{e_{s , k } } g_{kk } + \sum\limits_{j = 1,j \ne k}^k { { \nu _ { s , j}}{e_{s , j } } g_{kj } { { } } } } { \left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^ { 2}}\\ & \mathbb e \left[\textsf{d}_{r , k}^\textsf{r } \right ] = \frac{\mu _ r } { { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } } \\ & \times { { \sum\limits_{j = 1}^k { { e_{s , j}}{\beta _ { sr , j}}{\rm{tr}}\left ( { { { { \bf{\tilde c}}}_{sr , j}}{{\bf{\omega } } _ { s , j } } } \right ) } + { \beta _ { ei}}{\rm{tr}}\left ( { { { { \bf{\tilde c}}}_{ei}}{{\bf{\omega } } _ r } } \right ) + 1 } } \end{aligned}\ ] ] where @xmath260 is defined as @xmath261 $ ] and is given by lemma [ lm1 ] in appendix - a . @xmath262 and @xmath263 are given by ( [ d10 ] ) and ( [ d12 ] ) in the appendix - e . @xmath264 and @xmath265 are expressed as @xmath266 the scaling behaviors of above factors are shown in table [ tab1 ] . see appendix - e . [ cols= " < , < , < " , ] several important results on the effect of ei , mui and hardware impairments on achievable rate of @xmath118 channel can be obtained from theorem [ t4 ] and table [ tab1 ] . _ effect of ei : _ with the proposed hia scheme , the scaling behavior of ei is high correlated with @xmath267 and @xmath268 . with the assumption * a1 * , we have @xmath269 and @xmath270 . in fact , under specific spatial correlation model ( e.g. , the physical channel model with a fixed number of angular bins @xcite ) , it is possible to make @xmath271 and @xmath272 by selecting proper @xmath167 and @xmath168 . in this case , the power of ei can decrease faster than @xmath273 in the large-@xmath30 regime . note that this is in contrast with the result in @xcite , which shown that the power of ei decreases exactly with @xmath273 . _ effect of mui : _ the scaling behavior of mui confirms well with the classic result in mm - fdr with perfect hardware @xcite , i.e. , the mui diminishes as @xmath274 if @xmath275 . _ effect of transmit imperfection of sources : _ ( i ) when @xmath25 is fixed , an interesting observation is that the term @xmath276 , which is widely convinced @xmath277 terms with perfect hardware , scales as @xmath278 . as a result , the achievable rate of @xmath118 channel is limited by the joint effect of @xmath276 and the effective distortion due to the transmit imperfection of sources , i.e. , @xmath247 . in fact , the variation of scaling behavior of @xmath276 can also be viewed as a `` negative effect '' of transmit imperfection of sources . to see this , inserting the expression of @xmath260 ( given by lemma [ lm1 ] in appendix - a ) into ( [ 4 - 13 ] ) and letting @xmath274 , we can obtain @xmath279 , which demonstrates that @xmath280 scales as @xmath281 as long as @xmath282 . ( ii ) when @xmath25 scales with @xmath28 and @xmath283 , it is seen that the proposed hia scheme reduces the effect of transmit imperfection at sources by a factor of @xmath284 . the reason is that the transmit distortion is pre - suppressed by the transmit bf vector @xmath285 . _ effect of receive imperfection of the relay : _ by comparing the result in table [ tab1 ] and @xmath286 , it is seen that the hia scheme does not change the scaling behavior of distortion noise caused by receive imperfection of the relay . however , the receive distortion noise power is still reduced , since ei ( which is a main contributor of receive distortion noise ) is suppressed . _ 2 ) achievable rate of @xmath108 channel : _ according to ( [ 2 - 7 ] ) and the bounding technique in @xcite , the achievable rate of @xmath287 channel can be expressed as @xmath288 } \right|}^2}}}{{{e_{r , k}}{\mathop{\rm var } } \left ( { { \bf{{\dot{p}}}}_{d , k}^h{\bf{h}}_{rd , k}^h{{\bf{w}}_{t , k } } } \right ) + \mathbb e\left [ { { \textsf{mui}}_{d , k } } \right ] + \mathbb e\left [ { \textsf{d}_{d , k}^\textsf{t } } \right ] + \textsf{d}_{d , k}^\textsf{r}}+1 } } \right)\end{array } \end{aligned}\ ] ] where @xmath289 , @xmath290 and @xmath291 denote the mui , effective distortion due to transmit imperfection of relay and effective distortion due to receive imperfection of destinations , respectively , which are given by @xmath292 [ t5 ] assume that assumptions * a1 * and * a2 * hold , and @xmath168 is selected so that @xmath293 . with hia scheme , the achievable rate of @xmath108 channel in the large-@xmath30 regime is given by @xmath294 , where the power of desired signal is @xmath295 } \right|^2 } = { e_{r , k } } { { { \rm{tr } } ( { { \bf{\hat { \underline{c}}}}_{rd , k}^ { } } ) } } $ ] , and @xmath296 = \sum\limits_{j = 1,j \ne k}^k \frac{{e_{r , j}}{\rm{tr}}\left ( { \left ( { { { \bf{\underline{c}}}_{rd , k } } - { { { \bf{\hat { \underline{c}}}}}_{rd , k } } } \right){{{\bf{\hat { \underline{c}}}}}_{rd , j } } } \right ) } { { { { \rm{tr}}\left ( { { \bf{\hat { \underline{c}}}}_{rd , j}^ { } } \right ) } } } \\ & \mathbb e\left [ \textsf{d}_{d , k}^\textsf{t } \right ] = { \nu _ r}\sum\limits_{l = 1}^k \frac{{e_{r , l}}{\rm{tr}}\left ( { { { \bf{c}}_{rd , k}}{\rm{diag}}\left ( { { { \bf{{\dot{p}}}}_t}{{{\bf{\hat { \underline{c}}}}}_{rd , l}}{\bf{{\dot{p}}}}_t^h } \right ) } \right)}{\left({\lambda _ 1}\left ( { { { { \bf{\tilde c}}}_{rd , k } } } \right)\right)^{-1 } { { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{rd , l } } } \right ) } } } \\ & \textsf{d}_{d , k}^\textsf{r } = \frac{{{\mu _ { d , k}}{e_{r , k}}}}{{{\text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{rd , k } } } \right)}}\left ( { \frac{{{\mathbf{{\dot{p}}}}_{d , k}^h{\text{diag}}\left ( { { { \mathbf{{\dot{p}}}}_{d , k}}{\mathbf{{\dot{p}}}}_{d , k}^h } \right){{\mathbf{{\dot{p}}}}_{d , k}}}}{{{{\left ( { { \text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{rd , k } } } \right ) } \right)}^ { - 2 } } } } } \right . \\ & + \frac { \frac{{{\mu _ r}}}{\tau } \left ( { \frac{1}{{{e_t } } } { + \frac{1}{{{\mu _ r}{e_t}}}}+ { \beta _ { rd , k}}\left ( { { \lambda _ 1}\left ( { { { { \mathbf{\tilde c}}}_{rd , k } } } \right ) + { \nu _ { d , k } } } \right ) } \right ) } { \left({{\text{tr}}\left ( { { \mathbf{\underline{c}}}_{rd , k}^3{\mathbf{\gamma } } _ { rd , k}^2 } \right)}\right)^{-1}{{\lambda _ 1}\left ( { { { { \mathbf{\tilde c}}}_{rd , k } } } \right ) } } \\ & \left . { + \sum\limits_{j = 1,j \ne k}^k { \frac{{{e_{r , j}}{\beta _ { rd , j}}}}{{{e_{r , k}}}}{\text{tr}}\left ( { { { \mathbf{c}}_{rd , k}}{{\mathbf{{\dot{p}}}}_t}{{{\mathbf{\hat { \underline{c}}}}}_{rd , j}}{\mathbf{{\dot{p}}}}_t^h } \right ) } + \frac{{{\text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{rd , k } } } \right)}}{{{e_{r , k } } } } } \right ) \end{aligned}\ ] ] @xmath297 is defined as @xmath298 $ ] and is given by lemma [ lm1 ] in appendix - a . the scaling behaviors of above factors are shown in table [ tab2 ] . the proof is similar with that for theorem [ t4 ] , and thus it is neglected . from theorem [ t5 ] and table [ tab2 ] , one can observe a strong similarity between the effects of hardware impairments on @xmath287 channel and that on @xmath118 channel . in particular , the effective distortion due to the transmit imperfection of relay @xmath290 is suppressed by the large transmit array of the relay , and it diminishes as @xmath299 and @xmath300 . moreover , when @xmath26 is fixed , the achievable rate of @xmath287 channel is limited by the term @xmath301 and distortion noise caused by receive imperfection of @xmath23 . however , as @xmath26 scales with @xmath29 and @xmath300 , these factors are suppressed by @xmath302 . this is because that applying @xmath303 can be viewed as a post - suppression on receive distortion noise . with hia scheme , the following end - to - end rate for @xmath304 channel is achievable in the large-@xmath30 regime @xmath305 different from that for upper bound in the last section , @xmath306 can not be expressed in a simple form like ( [ 3 - 7 ] ) . however , based on table [ tab1 ] and table [ tab2 ] , one can simply deduce that @xmath307 will converge to a finite ceiling if @xmath25 and @xmath26 are fixed . however , with hia scheme , the end - to - end achievable rate grows without bound as @xmath308 if @xmath25 and @xmath26 scale with @xmath309 and @xmath310 . this is the same to the situation with ideal hardware @xcite . based on the achievable rate expressions , we discuss the hardware design of sources , destinations and relay in this subsection . although the achievable rate expressions hold for arbitrary @xmath311 , we will restrict our analysis to @xmath157 . this is a fundamental condition to ensure the benefit of massive mimo , i.e. , the mui is suppressed by a large surplus of degrees of freedom @xcite . in mm - fdr with hardware impairments , it is also essential to control the number of s - d pairs ( by , e.g. , some user scheduling scheme ) to suppress the ei and distortions . as @xmath25 and @xmath26 are fixed and small ( due to e.g. , size limitation ) , from theorem [ t4 ] and [ t5 ] , it is of great importance to reduce @xmath312 , @xmath313 and @xmath314 in order to alleviate the `` ceiling effect '' on achievable rate . this indicates that it is beneficial to use high - quality hardware at sources and destinations in this case . note that , different from the relay side , the use of high - quality hardware at sources and destinations is affordable , since the increased cost scales with @xmath25 and @xmath26 , instead of @xmath32 . when @xmath25 and @xmath26 scale with @xmath32 , from theorem [ t4 ] and [ t5 ] , the distortion noise powers due to hardware impairments at sources and destinations are lower order terms when compared with desired signal powers . this means that the sources and destinations can decrease the hardware quality to some extent without degrading the performance . later an example will be given to show how to achieve this . similar to that for sources and destinations when @xmath25 and @xmath26 scale with @xmath32 , the distortion noises caused by hardware impairments at the relay are lower order terms when compared with desired signal powers , which makes it possible to decrease the hardware quality without hurting the performance greatly . _ example : _ to get a clear insight , we consider the achievable rate of @xmath315 channel given by theorem [ t4 ] . we assume that there is only one s - d pair and let @xmath316 . when @xmath25 scales with @xmath28 , the achievable rate of @xmath107 channel reduces to @xmath317 where @xmath318 and @xmath319 are positive constants independent of @xmath320 , @xmath321 and @xmath322 . in ( [ 4 - 19 ] ) , we have replaced @xmath25 with the product of @xmath28 and some constant . assume we wish to achieve rate @xmath323 . according to ( [ 4 - 19 ] ) , we have ( as @xmath324 ) @xmath325 @xmath326 . the result is encouraging since it implies that we can increase @xmath327 and @xmath328 linearly as @xmath28 increase ( with @xmath25 scaling with @xmath28 ) without degrading the achievable rate . as a result , inexpensive mm - fdr is possible . in this section , we propose a low - complexity jdpo algorithm to maximize the se ( defined as the sum of all destinations achievable rates ) of hia - mm - fdr , subject to the maximum power constrains . the achievable rate expressions obtained in section v - c are utilized in the proposed algorithm . the algorithm needs only statistical knowledge of channels . therefore , it can be computed offline at a central node ( e.g. , relay ) and then broadcasts to the other nodes . let @xmath329 and @xmath330 be the maximum transmit power constraints at @xmath22 and relay , respectively , the se optimization problem can be formulated as follows @xmath331 where @xmath332 and @xmath333 . according to theorem 4 and theorem 5 , we rewrite @xmath334 and @xmath335 as @xmath336 and @xmath337 , where @xmath338 and @xmath339 denote the effective received sinrs of @xmath107 and @xmath108 channels respectively , which can be expressed as @xmath340 where @xmath341 @xmath342 with the above results and the formula @xmath343 @xmath344 ( @xmath345 is defined as @xmath346 ) , ( [ 5 - 2 ] ) can be rewritten as @xmath347 by using some algebra manipulations on the inequality constraints , @xmath348 , @xmath349 can be rewritten as @xmath350 problem ( [ 5 - 4 ] ) is a combinatorial optimization problem which is in general np hard . to solve ( [ 5 - 4 ] ) , we propose a jdpo algorithm to find suboptimal solution . our strategy is as follows . first , using the similar approach as that in @xcite , we show that the power control problem with fixed @xmath167 and @xmath168 can be approximated as a geometric programming ( gp ) and solved efficiently using the convex optimization tools @xcite . then we present a heuristic approach based on sequential optimization to solve @xmath167 and @xmath168 . _ power control with fixed @xmath167 and @xmath168 : _ from ( [ 5 - 4 ] ) and ( [ 5 - 5 ] ) , the inequality constraints are posynomial functions @xcite . as shown in @xcite , for any @xmath351 , @xmath352 can be approximated using a monomial function @xmath353 near a point @xmath354 , where @xmath355 and @xmath356 . by using this , the target function of ( [ 5 - 4 ] ) can be approximated as a monomial function , i.e. , @xmath357 . as a result , the solution for power control problem ( with fixed @xmath167 and @xmath168 ) can be obtained by solving several gps . the algorithm is summarized in algorithm [ alg1 ] . . * compute @xmath358 and @xmath359 . * solve the following gp problem + l _ e_s,1 , , e_s , k;e_r,1 , , e_r , k ; _ k = 1^k _ k^ - 1_k^ - _ k s.t . c_1 ~c_5 if @xmath360 or @xmath361 , stop . otherwise , set @xmath362 and go back to step 2 . [ alg1 ] _ optimization of @xmath167 and @xmath168 : _ an obvious approach to obtain @xmath167 and @xmath168 is via 2-d search . however , the complexity is very high since the power control problem described in the above should be computed for each @xmath363 . in this work , we present a heuristic approach to find suboptimal solution . in this approach , @xmath167 ( @xmath168 ) is searched over a subset @xmath364 ( @xmath365 ) , whose elements are sampled from @xmath366 ( @xmath367 ) . for example , @xmath364 can be selected as @xmath368 . the motivation of this idea is that the target function does not change dramatically as @xmath167 and @xmath168 increase / decrease by a small step as observed from the numerical results . moreover , instead of 2-d search , @xmath366 and @xmath367 are optimized sequentially over the sampled subsets @xmath364 and @xmath365 via 1-d search . the process repeats several times ( much less than @xmath369 ) to improve the solution . a summary of the jdpo algorithm is presented in algorithm [ alg2 ] . the algorithm converges to a local optimum since the target function is improved in each iteration . the complexity of algorithm [ alg2 ] is upper bounded as @xmath370 , where @xmath371 is the complexity to solve gp when @xmath372 and @xmath373 . usually , gp is solved using inner point method with polynomial time . the exact expression of @xmath371 is quite difficult and related with the structure of the problem . some insights on @xmath371 can be found in [ 37 , sec . 11.5 ] . * initialization : * set an initial @xmath168 , i.e. , @xmath374 , and a maximum repeat times @xmath375 . select the subsets @xmath364 and @xmath365 . . update @xmath376 . [ alg2 ] [ rm2 ] _ ( jdpo for ee optimization ) _ instead of the se optimization , we can also formulate an ee ( defined as the se divided by total transmit power @xcite ) optimization problem subject to a target se @xmath377 @xmath378 using ( [ 5 - 3 ] ) , the above problem is equivalent to @xmath379 @xmath380 and @xmath381 . with the technique in @xcite , the equality constraint can be converted to a monomial , and the power control problem ( with fixed @xmath167 and @xmath168 ) becomes a gp . thus , ( [ 5 - 6 ] ) can be solving by using a similar jdpo algorithm as that in algorithm [ alg2 ] . this section presents the simulation results to verify the analyses in the previous sections . throughout this section , we set @xmath382 , @xmath383 , @xmath384 for convenience . the correlation matrices of desired channels are generated with the exponential correlation model @xcite @xmath385_{l , j } } = \left\ { { \begin{array}{*{20}{l } } { r_{i , k}^{j - l},l \le j}\\ { { { \left ( { r_{i , k}^{l - j } } \right)}^*},l > j } \end{array } } \right.,i \in \left\ { { sr , rd } \right\}\ ] ] the model approximates the property of ula , where the correlation between adjacent antennas is @xmath386 $ ] and the phase of @xmath387 describes the angle of arrival / departure as seen from the array . @xcite shows how to map some of the parameters of ula to this model . the correlation matrices of ei channel @xmath388 and @xmath389 are generated similarly with a parameter @xmath390 . for convenience , we let @xmath391 , @xmath392 . the phases of @xmath387 and @xmath390 are uniformly selected from @xmath393 $ ] . we assume that 25@xmath39435db ei cancellation can be provided by passive ei suppression techniques ( more than 40db cancellation has been reported by using such techniques for infrastructure node @xcite ) . the variances of ei channel ( after passive cancellation ) and desired channels are selected as @xmath395$]db . with the path loss model in @xcite , the above range corresponds to the setup with the distances from sources / destinations to relay varying from 250 m to 500 m and 10 m segregation between relay transmit and receive arrays . this subsection considers the effect of hardware impairments on se of mm - fdr . the channel coherent time is set to @xmath396 and the length of pilot sequence is @xmath397 . , @xmath398 , @xmath399db , @xmath400db , @xmath401 , @xmath402 , @xmath403 , @xmath404 , @xmath405.,width=321 ] .,width=321 ] fig . [ fig1a ] shows the se of mm - fdr with single antenna at sources / destinations with different levels of hardware impairments . the ses based on transceiver scheme in section iv ( which achieves the upper bound when @xmath125 ) and hia scheme ( with @xmath406 ) are simulated . from fig . [ fig1a ] , the se is more sensitive to hardware impairments at sources and destinations . when @xmath407 , the se approaches to a finite ceiling quickly as the number of relay antennas increases . similar results can be observed when sources and destinations are equipped with multiple but fixed number of antennas in fig . [ fig1b ] . however , the result changes when @xmath25 and @xmath26 scale with @xmath32 and hia scheme is used . from fig . [ fig1b ] , as @xmath408 and @xmath409 , similar ses are achieved as sources / destinations or relay employ low - quality hardware , and no performance ceiling appears . this demonstrates the validness of hia scheme . at last , it is seen that the asymptotic results in theorem [ t1 ] & [ t2 ] and theorem [ t4 ] & [ t5 ] match well with exact results . , @xmath398 , @xmath410db , @xmath400db , @xmath411 , @xmath412db , @xmath402 , @xmath403 , @xmath404 , @xmath405 , @xmath408 , @xmath409.,width=321 ] as the performance is affected by impairments of both transmit and receive rf chains at the relay , we compare the effect of transmit and receive imperfections on se in fig . we assume that the sources and destinations use high - quality hardware ( @xmath413 ) . from the figure , the effect of receive imperfection is more detrimental . the reason is that , with ei at the relay , the power of distortion noise caused by receive imperfection is much stronger than that due to transmit imperfection . however , the performance difference decreases as @xmath28 increases from @xmath414 to @xmath415 , since the power of effective receive distortion scales as @xmath273 ( see table [ tab1 ] ) . this implies that use relatively higher - quality hardware or more antennas at the receive side of relay is beneficial . using a similar setup , one can obtain a parallel conclusion for destination , i.e. , the receive imperfection is more harmful than transmit imperfection . this is because that the transmit imperfection of destination only induces larger channel estimator errors , which is a part of the received signal at the destination . thus , based on model ( [ 2 - 4 ] ) , the distortion noise due to receive imperfection will be more detrimental . , where @xmath416 , @xmath417 , @xmath410db , @xmath400db , @xmath411 , @xmath412db , @xmath418 , @xmath419 , @xmath420 , @xmath421 , @xmath404 , @xmath405.,width=321 ] fig . [ fig4 ] shows the se as a function of the number of s - d pairs @xmath20 , where the length of pilot sequence of each source or destination is set to 1 . when @xmath396 , the figure reveals that the se will not increase without bound as @xmath20 increases since se is ultimately limited by the channel coherent time . in fact , as the number of s - d pair approaches to @xmath422 , the se converges to zero since all time is allocated to pilot phase . meanwhile , the se per s - d pair is a strictly decreasing function of @xmath20 . similar result can be observed as the channel coherent time is very long ( @xmath423 ) . this is because that the mui , ei and distortion noises become limiting factors when @xmath20 is large as seen from the scaling behaviors in table [ tab1 ] and table [ tab2 ] . this implies that the number of s - d pairs should be limited in order to ensure a se guarantee for each s - d pair , and this number decreases as the quality of hardware degrades . in this subsection , the se of hia - mm - fdr is compared with the massive mimo hdr ( mm - hdr ) @xcite and mm - fdr with zf - based transceiver ( zf - mm - fdr ) @xcite . the channel coherent time is set to @xmath396 and the length of pilot sequence is @xmath397 . the power of pilot symbol is @xmath400db . for hia - mm - fdr , the maximum power constraints at sources and relay are set to @xmath424db and @xmath425 . without jdpo , we set @xmath426 , @xmath427db . moreover , @xmath167 and @xmath168 are set to @xmath404 and @xmath428 . when jdpo is applied , @xmath76 , @xmath139 , @xmath167 and @xmath168 are determined by algorithm [ alg2 ] . the maximum repeat time @xmath375 of jdpo is set to 3 and the elements of subset @xmath429 are picked uniformly from @xmath430 with step @xmath431 . @xmath432 is obtained with the similar approach . the above parameters are chosen so that the performance loss due to the suboptimal search approach to obtain @xmath167 and @xmath168 is negligible . , @xmath411 , @xmath412db , @xmath408 , @xmath409 , @xmath418 , @xmath419.,width=321 ] fig . [ fig4 ] simulates the number of relay antennas required to achieve 3bit / s / hz se per s - d pair ( which ideally can support the 64-qam transmission with 1/2 channel code ) with different levels of hardware quality , where we set @xmath416 and @xmath417 . the figure reveals a tradeoff between the number of antennas and hardware quality , that is ( as discussed in section v - d ) , by increasing @xmath28 and @xmath29 , we can reduce @xmath433 and @xmath434 linearly without degrading the se . meanwhile , it is seen that the proposed hia - mm - fdr reduces the required number of relay antennas significantly when compared with zf - mm - fdr . moreover , the hia - mm - fdr outperforms the mm - hdr for large @xmath433 and @xmath434 . this is because that the distortion noises become the main limiting factor in this case when compared with ei . , @xmath408 , @xmath409 , @xmath411 , @xmath412db , @xmath418 , @xmath419 , @xmath435 , @xmath436.,width=321 ] the effect of asymmetric numbers of transmit and receive antennas at the relay is shown in fig . [ fig5 ] , where we set @xmath437 . it is shown that allocating more antennas to the receive side of relay is beneficial for hia - mm - fdr ( without jdpo ) and zf - mm - fdr . the reason is that , in the considered setup , the limiting factors of system performance are ei and distortion noise due to receive imperfection at the relay . from theorem [ t4 ] , these factors can be suppressed by receive antenna array of relay . however , when jdpo is applied , it is optimal to set @xmath438 . this is because that the jdpo algorithm in fact tries to balance the achievable rates of each hop by power control and adjusting @xmath167 and @xmath168 . this makes it unnecessary to allocate more antennas to receive side . , where @xmath398 , @xmath408 , @xmath409 , @xmath411 , @xmath418 , @xmath419 , @xmath435 , @xmath436.,width=321 ] fig . [ fig6 ] shows the se of hia - mm - fdr as a function of variance for ei channel @xmath439 . as expected , there exists a switching point between hia - mm - fdr and mm - hdr as @xmath439 increases . by increasing @xmath28 and @xmath29 , the constraint on @xmath439 for hia - mm - fdr to achieve a performance gain relaxes , which indicates that hia - mm - fdr becomes more attractive when the number of relay antennas is large . moreover , fig . [ fig6 ] demonstrates that the proposed jdpo algorithm can reduce the constraint on @xmath439 significantly . in particular , as @xmath440 , a se gain of 7.5bit / s / hz can be achieved by hia - mm - fdr when compared to mm - hdr as @xmath439 is @xmath441db . , @xmath442db , @xmath408 , @xmath409 , @xmath402 , @xmath403 , @xmath435 , @xmath436.,width=321 ] fig . [ fig7 ] considers the se - ee tradeoff of different schemes . the large - scale fading coefficients of channels are set to @xmath443 which is a realization generated with the model in @xcite . the ees of hia - mm - fdr and zf - mm - fdr are optimized by solving the problem in remark [ rm2 ] with jdpo algorithm and [ 12 , algorithm 1 ] , respectively . it is observed that hia - mm - fdr achieves better se - ee tradeoff when compared with zf - mm - fdr . the gain is mainly due to the optimization of @xmath167 and @xmath168 . this reason is that , if we fix @xmath372 and @xmath373 , the jdpo algorithm for ee optimization is similar to [ 12 , algorithm 1 ] . the only difference is that the power for each steam at the relay , i.e. , @xmath139 , is optimized in jdpo algorithm and [ 12 , algorithm 1 ] optimizes only the total power of relay . this paper considers the transceiver design of mm - fdr with hardware impairments . a low complexity hia scheme is proposed to mitigate the distortion noises by exploiting the statistical knowledge of channels and antenna arrays at sources and destinations . a joint degree of freedom and power optimization algorithm is presented to further optimize the se of hia - mm - fdr . the analytic results demonstrate that the proposed scheme can mitigate the ceiling effect " appears in traditional mm - fdr protocol , if the numbers of antennas at sources and destinations can scale with that at the relay . moreover , simulation results show that the hia - mm - fdr outperforms mm - fdr with traditional transceiver scheme . [ lm1 ] let @xmath444 and @xmath445 be the lmmse estimates of @xmath446 and @xmath447 , respectively . define @xmath448 $ ] and @xmath449 $ ] . in the large-(@xmath450 ) regime , we have @xmath451 the expression of @xmath452 can be obtained by replacing @xmath453 , @xmath454 , @xmath455 , @xmath312 and @xmath285 in ( [ x1 ] ) with @xmath456 , @xmath457 , @xmath458 , @xmath313 and @xmath303 , respectively . we show the proof for @xmath260 and the derivation for @xmath297 is similar . according to the uncorrlation between @xmath459 and @xmath460 , we have @xmath461 = \mathbb e [ { { { | { { \mathbf{{\underline{h}}}}_{sr , k}^h{{{\mathbf{\hat { \underline{h}}}}}_{sr , k } } } |}^2 } } ] - { ( { { \text{tr } } ( { { { { \mathbf{\hat { \underline{c}}}}}_{sr , k } } } ) } ) ^2}$ ] . with ( [ 4 - 7 ] ) and ( [ 4 - 8 ] ) , @xmath462 $ ] can be written as @xmath463 = \mathbb e\left [ { { { \left| { { \mathbf{{\underline{h}}}}_{sr , k}^h{{\mathbf{{\underline{c}}}}_{sr , k}}{{\mathbf{\gamma } } _ { sr , k}}{{\mathbf{{\underline{h}}}}_{sr , k } } } \right|}^2 } } \right ] \\ & + { \left ( { \frac{1}{{\tau { e_t } } } } \right)^2 } \mathbb e\left [ { { { \left| { { \mathbf{{\underline{h}}}}_{sr , k}^h{{\mathbf{{\underline{c}}}}_{sr , k}}{{\mathbf{\gamma } } _ { sr , k}}{\mathbf{{\dot{p}}}}_r^h{{\mathbf{h}}_{sr , k}}{\mathbf{t}}_{sr , k}^{}\phi _ { } ^ * } \right|}^2 } } \right ] \\ & + { \left ( { \frac{1}{{\tau { e_t } } } } \right)^2 } \mathbb e\left [ { { { \left| { { \mathbf{{\underline{h}}}}_{sr , k}^h{{\mathbf{{\underline{c}}}}_{sr , k}}{{\mathbf{\gamma } } _ { sr , k}}{\mathbf{{\dot{p}}}}_r^h{{\mathbf{r}}_{sr , k}}\phi _ { } ^ * } \right|}^2 } } \right ] \\ & + { \left ( { \frac{1}{{\tau { e_t } } } } \right)^2}\mathbb e\left [ { { { \left| { { \mathbf{{\underline{h}}}}_{sr , k}^h{{\mathbf{\underline{c}}}_{sr , k}}{{\mathbf{\gamma } } _ { sr , k}}{\mathbf{{\dot{p}}}}_r^h{{\mathbf{n}}_{sr , k}}\phi _ { } ^ * } \right|}^2 } } \right ] \\ \end{aligned}\ ] ] in the large-(@xmath450 ) regime , the result in [ 17 , lemma 4 ( iv ) ] shows that @xmath464 = { \left ( { { \text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^2}$ ] . replacing @xmath465 with @xmath466 and using the following large @xmath28 approximation ( obtained based on [ 17 , lemma 4 ( ii ) ] ) @xmath467 it can be shown that @xmath468 \\ & = { \nu _ { s , k}}\tau e_t^2\frac{{{\bf{{\dot{p}}}}_{s , k}^h{\bf{\tilde c}}_{sr , k}^{}{\rm{diag}}\left ( { { \bf{{\dot{p}}}}_{s , k}^h{{\bf{{\dot{p}}}}_{s , k } } } \right){\bf{\tilde c}}_{sr , k}^{}{{\bf{{\dot{p}}}}_{s , k}}}}{{\left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^{-2}}{{{\left ( { { \lambda _ 1}\left ( { { { { \bf{\tilde c}}}_{sr , k } } } \right ) } \right)}^2 } } } \end{aligned}\ ] ] using the similar approach on the remaining terms in ( [ x3 ] ) , lemma [ lm1 ] is obtained . [ lm2 ] let @xmath469 and @xmath470 ( @xmath471 ) be the covariance matrices of effective channel @xmath472 and its estimates @xmath473 given by theorem [ t3 ] . if @xmath474 and @xmath475 , we have @xmath476 and @xmath477 . moreover , @xmath478 and @xmath479 ( i.e. , @xmath167 and @xmath168 scale with @xmath480 and @xmath481 , respectively ) . we present the proof for @xmath482 and @xmath167 . using definition of @xmath483 in theorem [ t3 ] and eigenvalue decomposition @xmath484 , we have @xmath485 where @xmath486 and @xmath487 are positive constants independent of @xmath167 . from ( [ x6 ] ) , @xmath488 is bounded as @xmath489 in ( [ x7 ] ) , the upper bound is an @xmath480 term . with the definition of @xmath222 in the theorem [ t3 ] , we have @xmath490 the last step is based on assumption * a1*. since @xmath491 , we can conclude that @xmath492 ( otherwise @xmath493 ) . thus , the lower bound in ( [ x7 ] ) also scales with @xmath480 . moreover , based on the definition of @xmath222 in the theorem [ t3 ] , we have @xmath494 where @xmath495 is the @xmath13th column of @xmath496 . from ( [ x9 ] ) , @xmath167 must scale @xmath480 . otherwise , @xmath497 based on assumption * a1*. to facilitate analysis , we first derive the covariance matrices of distortion noises @xmath115 $ ] and @xmath117 $ ] . based on the model ( [ 2 - 3 ] ) , we have @xmath498 . moreover , define @xmath499 = { \left [ { { \mathbf{x}}_{s,1}^t\left [ u \right ] , \cdots , { \mathbf{x}}_{s , k}^t\left [ u \right ] } \right]^t}$ ] , @xmath500 can be expressed based on model ( [ 2 - 4 ] ) as @xmath501 } \right ) \\ & + { \mu _ r}{\text{diag}}\left ( { \mathbb{e}\left [ { { { \mathbf{h}}_{ei}}{{\mathbf{\omega } } _ r}{\mathbf{h}}_{ei}^h } \right ] } \right ) + { \mu _ r}{{\mathbf{i}}_{{n_r } } } \end{aligned}\ ] ] where @xmath502 + { { \bf{t}}_s}\left [ u \right ] } \right){{\left ( { { \bf{s}}\left [ u \right ] + { { \bf{t}}_s}\left [ u \right ] } \right)}^h } } \right ] \\ \hspace{1.7em } = { \rm { diag}}\left ( { \left ( { 1 + { \nu _ { s,1 } } } \right){e_{s,1 } } , \cdots , \left ( { 1 + { \nu _ { s , k } } } \right){e_{s , k } } } \right)\\ { { \bf{\omega } } _ r } = \mathbb{e}\left [ { \left ( { { { \bf{x}}_r}\left [ u \right ] + { { \bf{t}}_r}\left [ u \right ] } \right){{\left ( { { { \bf{x}}_r}\left [ u \right ] + { { \bf{t}}_r}\left [ u \right ] } \right)}^h } } \right ] \\ \hspace{1.7em } = \mathbb{e}\left [ { { { \mathbf{x}}_r}\left [ u \right]{\mathbf{x}}_r^h\left [ u \right ] } \right ] + { \mathbf{\theta } } _ r^\textsf{t } \end{array}\ ] ] for further analysis , we approximate @xmath503{\mathbf{x}}_r^h\left [ u \right ] } \right]$ ] in ( [ a1 - 1 ] ) with @xmath504{\mathbf{x}}_r^h\left [ u \right ] } \right ] ) = \frac{1}{\nu_r } { \mathbf{\theta } } _ r^\textsf{t}$ ] . note that the approximation results in a new upper bound on @xmath121 ( the new bound will also be referred to as @xmath121 for convenience ) . straight - forward computations yield to @xmath505 & = \sum_{l = 1}^k { { \nu _ { s , l}}{e_{s , l}}\mathbb e\left [ { { { \mathbf{h}}_{sr , l}}{\mathbf{h}}_{sr , l}^h } \right ] } \\ & = \sum_{l = 1}^k { { ( 1+\nu _ { s , l})}{e_{s , l}}\beta_{sr , l}{{\mathbf{c}}_{sr , l } } } \end{aligned}\ ] ] following the spatial correlation model of @xmath506 in section iii and the approximation under ( [ a1 - 1 ] ) , the expression of @xmath507}$ ] can be obtained as in ( [ a3 ] ) , where the third step follows from the formula @xmath508 and the fourth step is due to the fact that the elements of @xmath509 are i.i.d with distribution @xmath42 . inserting ( [ a2 ] ) , ( [ a3 ] ) into ( [ a1 ] ) , the expression of @xmath500 in ( [ 3 - 3 ] ) is obtained . @xmath510 & \approx \left(1+\frac{1}{\nu_r}\right){\beta _ { ei } } \mathbb e\left [ { { \mathbf{c}}_{ei}^{1/2}{{\mathbf{x}}_{ei}}{\mathbf{\tilde c}}_{ei}^{1/2}{{\mathbf{\theta } } _ r^\textsf{t}}{\mathbf{\tilde c}}_{ei}^{1/2}{\mathbf{x}}_{ei}^h{\mathbf{c}}_{ei}^{1/2 } } \right ] \hfill \\ & = \left(1+\frac{1}{\nu_r}\right){\beta _ { ei}}\sum\limits_{l = 1}^{{n_t } } { \mathbb{e}\left [ { { \mathbf{c}}_{ei}^{1/2}{{\mathbf{x}}_{ei}}{\mathbf{\tilde c}}_{ei}^{1/2}{{\left ( { { \mathbf{\theta } } _ r^\textsf{t } } \right)}^{1/2}}{{\mathbf{e}}_l}{\mathbf{e}}_l^t{{\left ( { { \mathbf{\theta } } _ r^\textsf{t } } \right)}^{1/2}}{\mathbf{\tilde c}}_{ei}^{1/2}{\mathbf{x}}_{ei}^h{\mathbf{c}}_{ei}^{1/2 } } \right ] } \hfill \\ & = \left(1+\frac{1}{\nu_r}\right){\beta _ { ei}}\sum\limits_{l = 1}^{{n_t } } { \mathbb{e}\left [ { { { \left\| { { { \left ( { { \mathbf{\tilde c}}_{ei}^{1/2}{{\left ( { { \mathbf{\theta } } _ r^\textsf{t } } \right)}^{1/2}}{{\mathbf{e}}_l } } \right)}^t } \otimes { \mathbf{c}}_{ei}^{1/2}{\text{vec}}\left ( { { { \mathbf{x}}_{ei } } } \right ) } \right\|}^2 } } \right ] } \hfill \\ & = \left(1+\frac{1}{\nu_r}\right){\beta _ { ei}}\sum\limits_{l = 1}^{{n_t } } { { { \left\| { { { \left ( { { \mathbf{\tilde c}}_{ei}^{1/2}{{\left ( { { \mathbf{\theta } } _ r^\textsf{t } } \right)}^{1/2}}{{\mathbf{e}}_l } } \right)}^t } \otimes { \mathbf{c}}_{ei}^{1/2 } } \right\|}^2}}\hfill = \left(1+\frac{1}{\nu_r}\right){\beta _ { ei}}{\text{tr}}\left ( { { { { \mathbf{\tilde c}}}_{ei}}{\mathbf{\theta } } _ r^\textsf{t } } \right){\mathbf{c}}_{ei } \hfill \\ \end{aligned}\ ] ] then we derive the asymptotic bound in theorem [ t1 ] . by replacing @xmath511 in @xmath512 with @xmath513 , we approximate @xmath512 as @xmath514 the approximation results in a new upper bound on @xmath121 , since it reduces the power of residual ei after combining by @xmath119 ( the new bound will also be referred to as @xmath121 for convenience ) . based on [ 17 , lemma 4 ( ii ) ] , we can replace @xmath515 with its deterministic equivalence in the large-@xmath29 regime @xmath516 . finally , the expression of @xmath121 in theorem [ t1 ] can be obtained by first applying [ 17 , lemma 4 ( ii ) ] on @xmath517 , and then using [ 17 , theorem 1 ] . since the scenario of interest is the large-@xmath30 regime , we need only consider the effect of @xmath131 on @xmath518 ( given by theorem [ t1 ] ) , more precisely , the term @xmath519 . with the assumption * a2 * , we have @xmath520 . thus , using the model ( [ 2 - 3 ] ) , we have @xmath521}{\mathbf{x}}_r^h[u ] } \right ] } \right ) } \right ) = { \nu _ r}\sum_{l=1}^k e_{r , l}$ ] , which is independent of @xmath131 . then we derive the upper bound in theorem [ t2 ] . using the model ( [ 2 - 4 ] ) , the power of received distortion at the relay can be derived as @xmath522 } \right\|}^2 } } \right ] = { \mu _ { d , k}}\left ( { { n_t}{\beta _ { rd , k}}{e_{r , k } } + { \nu _ r}{n_t}{\beta _ { rd , k}}\sum_{l = 1}^k { { e_{r , l } } } + 1 } \right)$ ] . moreover , in the large @xmath29 regime , the term @xmath523 in ( [ 3 - 0 - 2 ] ) approaches to @xmath524 [ 17 , lemma 4 ( ii ) ] . therefore , it is sufficient to design @xmath131 to maximize the numerator of ( [ 3 - 0 - 2 ] ) , which is exactly the eigen bf scheme . the resultant upper bound on achievable rate can be obtained by applying [ 17 , lemma 4 ( ii ) ] and assumption * a2 * on the numerator of ( [ 3 - 0 - 2 ] ) . with the expression of lmmse estimator @xcite , we have @xmath525 { ( { \mathbb{e } [ { { { { \bf{\tilde { z}}}}_{sr , k}}{\bf { \tilde{z}}}_{sr , k}^h } ] } ) ^ { - 1}}{{{\bf { \tilde{z}}}}_{sr , k}}$ ] . using the independence between @xmath526 and distortion noises , we have @xmath527 & = \mathbb e\left [ { { \bf{{\dot{p}}}}_r^h{{\bf{h}}_{sr , k}}{{\bf{{\dot{p}}}}_{s , k}}{\bf{{\dot{p}}}}_{s , k}^h{\bf{h}}_{sr , k}^h{{\bf{{\dot{p}}}}_r } } \right]\\ & = { \beta _ { sr , k}}{\lambda _ 1}\left ( { { { { \bf{\tilde c}}}_{sr , k } } } \right){\bf{{\dot{p}}}}_r^h{{\bf{c}}_{sr , k}}{{\bf{{\dot{p}}}}_r } \\ & \buildrel \delta \over = { { \bf{\underline{c}}}_{sr , k } } \end{aligned}\ ] ] where the second step follows from a similar derivation with that in ( [ a3 ] ) . moreover , @xmath528}$ ] can be expressed as @xmath529= { { \bf{\underline{c}}}_{sr , k}}\\ & + { \left ( { \frac{1}{{\tau { e_t } } } } \right)^2}\mathbb{e}\left [ { { \mathbf{{\dot{p}}}}_r^h{{\mathbf{h}}_{sr , k}}{{\mathbf{t}}_{sr , k}}{\phi ^*}{\phi ^t}{\mathbf{t}}_{sr , k}^h{\mathbf{h}}_{sr , k}^h{{\mathbf{{\dot{p}}}}_r } } \right]\\ & + { \left ( { \frac{1}{{\tau { e_t } } } } \right)^2}\mathbb{e}\left [ { { \mathbf{{\dot{p}}}}_r^h{{\mathbf{r}}_{sr , k}}{\phi ^*}{\phi ^t}{\mathbf{r}}_{sr , k}^h{{\mathbf{{\dot{p}}}}_r } } \right ] \\ & + { \left ( { \frac{1}{{\tau { e_t } } } } \right)^2}\mathbb{e}\left [ { { \mathbf{{\dot{p}}}}_r^h{{\mathbf{n}}_{sr , k}}{\phi ^*}{\phi ^t}{\mathbf{n}}_{sr , k}^h{{\mathbf{{\dot{p}}}}_r } } \right ] \end{aligned}\ ] ] according to the independence between @xmath191 and @xmath530 , the second term of right - hand side of ( [ c2 ] ) can be rewritten as @xmath531 \\ & = { \nu _ { s , k}}\tau e_t^2\mathbb{e}\left [ { { \mathbf{{\dot{p}}}}_r^h{{\mathbf{h}}_{sr , k}}{\text{diag}}\left ( { { { \mathbf{{\dot{p}}}}_{s , k}}{\mathbf{{\dot{p}}}}_{s , k}^h } \right){\mathbf{h}}_{sr , k}^h{{\mathbf{{\dot{p}}}}_r } } \right]\\ & = { \nu _ { s , k}}\tau e_t^2 \beta_{sr , k } { \rm{tr}}\left ( { { { { \bf{\tilde c}}}_{sr , k}}{\rm{diag}}\left ( { { { \bf{{\dot{p}}}}_{s , k}}{\bf{{\dot{p}}}}_{s , k}^h } \right ) } \right){\bf{{\dot{p}}}}_r^h{{\bf{c}}_{sr , k}}{{\bf{{\dot{p}}}}_r } \end{aligned}\ ] ] using the similar approach on the remaining terms of ( [ c2 ] ) , and substituting the resultant expression of ( [ c2 ] ) and ( [ c1 ] ) into the expression of lmmse estimator , the result in theorem [ t3 ] is obtained . using ( [ 4 - 7])-([4 - 8 ] ) , the second order statistics of @xmath532 and @xmath229 in theorem [ t3 ] can be easily verified . note that @xmath535_{l , j } } = { \bf{\hat { \underline{h}}}}_{i , l}^h{{{\bf{\hat { \underline{h}}}}}_{i , j } } = { \mathbf{x}}_{i , l}^h{\mathbf{\hat { \underline{c}}}}_{i , l}^{1/2}{\mathbf{\hat { \underline{c}}}}_{i , j}^{1/2}{{\mathbf{x}}_{i , j}}$ ] ( @xmath471 ) . based on lemma [ lm2 ] , the dimensions of @xmath536 and @xmath537 ( i.e. , @xmath167 and @xmath168 ) approach to infinity as @xmath538 . thus , according to [ 17 , lemma 4 ( ii ) and ( iii ) ] , in the large-(@xmath450 ) regime , we can deduce @xmath539 _ ( a ) asymptotic expression : _ based on the model ( [ 2 - 3 ] ) , we have @xmath541{\mathbf{t}}_r^h\left [ u \right ] } \right ] } = { \nu _ r}{\rm{diag}}\left ( { \mathbb e\left [ { { { \bf{w}}_t}{\lambda _ r}{\bf{w}}_t^h } \right ] } \right)$ ] . since @xmath44 and @xmath542 are independent , @xmath245 can be rewritten as @xmath543 = \mathbb e [ { \bf{w}}_{r , k}^h{{\bf{h}}_{ei}}{{\bf{\omega } } _ r}{\bf{h}}_{ei}^h{{\bf{w}}_{r , k } } ] = \mathbb e [ { { { \| { { \bf{e}}_k^t { { ( { { \bf{\hat { \underline{h}}}}_{sr}^h{{{\bf{\hat { \underline{h}}}}}_{sr } } } ) } ^ { - 1}}{{{\bf{\hat { \underline{h}}}}}_{sr}}{{\bf{{\dot{p}}}}_r^h}{{\bf{h}}_{ei}}{\bf{\omega } } _ r^{1/2 } } \|^2 } } } ] $ ] , where @xmath544 is defined as @xmath545 + { { \bf{t}}_r}\left [ u \right ] } \right){{\left ( { { { \bf{x}}_r}\left [ u \right ] + { { \bf{t}}_r}\left [ u \right ] } \right)}^h } } ] = \mathbb e\left [ { { { \bf{w}}_t}{\lambda _ r}{\bf{w}}_t^h } \right ] + { \nu _ r}{\rm{diag}}\left ( { \mathbb e\left [ { { { \bf{w}}_t}{\lambda _ r}{\bf{w}}_t^h } \right ] } \right)$ ] . inserting ( [ 4 - 12 - 1 ] ) into the expression of @xmath544 and using ( [ d1 ] ) , we can obtain @xmath546 in the large-@xmath30 regime , using ( [ d1 ] ) , we have @xmath547 & = \frac{\mathbb e\left [ { { { { \bf{\hat { \underline{h}}}}}_{sr , k}}{{\bf{{\dot{p}}}}_r^h}{{\bf{h}}_{ei}}{{\bf{\omega } } _ r}{\bf{h}}_{ei}^h{{\bf{{\dot{p}}}}_r}{\bf{\hat { \underline{h}}}}_{sr , k}^h } \right]}{\left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^ { 2 } } \\ & = \frac{\mathbb e\left [ { { \rm{tr}}\left ( { { \bf{\hat { \underline{c}}}}_{sr , k}^{1/2}{{\bf{{\dot{p}}}}_r^h}{{\bf{h}}_{ei}}{{\bf{\omega } } _ r}{\bf{h}}_{ei}^h{{\bf{{\dot{p}}}}_r}{\bf{\hat { \underline{c}}}}_{sr , k}^{1/2 } } \right ) } \right]}{\left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^ { 2 } } \end{aligned}\ ] ] the second step is based on the property @xmath548 and independence between @xmath549 and @xmath44 . finally , using a similar derivation with ( [ a3 ] ) , @xmath543 $ ] in theorem [ t4 ] is obtained . _ ( b ) scaling behavior : _ substituting ( [ d6 ] ) into the expression of @xmath543 $ ] in theorem [ t4 ] and neglecting the terms that have no effect on the scaling behavior , we have @xmath550 = { \beta _ { ei}}{\left ( { { \text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^ { - 2}}\sum\limits_{l = 1}^k { { e_{r , l}}{{\left ( { { \text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{rd , l } } } \right ) } \right)}^ { - 1 } } } \\ \times { \text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{rd , l}}{\mathbf{{\dot{p}}}}_t^h{{{\mathbf{\tilde c}}}_{ei}}{{\mathbf{{\dot{p}}}}_t } } \right){\text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{sr , k}}{\mathbf{{\dot{p}}}}_r^h{{\mathbf{c}}_{ei}}{{\mathbf{{\dot{p}}}}_r } } \right ) \hfill \\ \le { \beta _ { ei}}\sum\limits_{l = 1}^k \frac{{{e_{r , l}}{\lambda _ { { n_r } - { a_r } + 1}}\left ( { { { \mathbf{c}}_{ei } } } \right ) } { \lambda _ { { n_t } - { a_t } + 1}}\left ( { { { { \mathbf{\tilde c}}}_{ei } } } \right ) } { { { \text{tr}}\left ( { { { { \mathbf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } } \end{aligned}\ ] ] where the second step follows from the definition of @xmath496 and @xmath551 . based on lemma [ lm2 ] , @xmath552 . thus we can conclude that @xmath543 \le { \cal o}({\lambda _ { { n_r } - { a_r } + 1}}\left ( { { { \mathbf{c}}_{ei } } } \right ) { \lambda _ { { n_t } - { a_t } + 1 } } ( { { { { \mathbf{\tilde c}}}_{ei } } } ) kn_r^ { - 1})$ ] . _ ( a ) asymptotic expression : _ based on the model ( [ 2 - 3 ] ) , the covariance matrix of @xmath553 can be derived as @xmath554{\mathbf{t}}_{s , j}^h\left [ u \right ] } \right ] } = { \nu _ { s , j}}{e_{s , j}}{\rm{diag}}\left ( { { { \bf{{\dot{p}}}}_{s , j}}{\bf{{\dot{p}}}}_{s , j}^h } \right)$ ] . thus , in the large-@xmath30 regime , using ( [ d1 ] ) @xmath534 $ ] can be expressed as @xmath555 = { \left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^ { - 2}}{\nu _ { s , j}}{e_{s , k}}\\ & \times\underbrace { \mathbb e\left [ { { \bf{\hat { \underline{h}}}}_{sr , k}^h{\bf{{\dot{p}}}}_r^h{{\bf{h}}_{sr , k}}{\rm{diag}}\left ( { { { \bf{{\dot{p}}}}_{s , k}}{\bf{{\dot{p}}}}_{s , k}^h } \right){\bf{h}}_{sr , k}^h{{\bf{{\dot{p}}}}_r}{{{\bf{\hat { \underline{h}}}}}_{sr , k } } } \right]}_{{g_{kk } } } \\ & + { \left ( { { \rm{tr}}\left ( { { { { \bf{\hat { \underline{c}}}}}_{sr , k } } } \right ) } \right)^ { - 2}}\sum\limits_{j = 1,j \ne k}^k { \nu _ { s , j}}{e_{s , j } } \\ & \times\underbrace { \mathbb e\left [ { { \bf{\hat { \underline{h}}}}_{sr , k}^h{\bf{{\dot{p}}}}_r^h{{\bf{h}}_{sr , j}}{\rm{diag}}\left ( { { { \bf{{\dot{p}}}}_{s , j}}{\bf{{\dot{p}}}}_{s , j}^h } \right){\bf{h}}_{sr , j}^h{{\bf{{\dot{p}}}}_r}{{{\bf{\hat { \underline{h}}}}}_{sr , k } } } \right]}_{{g_{kj } } } \end{aligned}\ ] ] according to the independence between @xmath556 and @xmath557 when @xmath558 , @xmath559 can be derived as @xmath560 moreover , substituting ( [ 4 - 7 ] ) and ( [ 4 - 8 ] ) in theorem [ t3 ] into ( [ d9 ] ) and using the large-@xmath28 approximation in ( [ x5 ] ) , it is straightforward but tedious to show that @xmath262 can be derived as @xmath561 where @xmath562 . inserting ( [ d10 ] ) and ( [ d12 ] ) into ( [ d9 ] ) , the expression of @xmath534 $ ] is obtained . _ ( b ) scaling behavior : _ using the similar approach as that for @xmath540 $ ] , it can be shown that , as @xmath25 is fixed , the first and second terms on the right - hand side of ( [ d9 ] ) scale as @xmath149 and @xmath273 , respectively . thus , @xmath534 = { \cal o}(1)$ ] since @xmath563 . as @xmath25 scales with @xmath28 , it is shown in @xcite that the eigenvectors of @xmath564 form a unitary dft matrix in the large @xmath25 regime . in this case , we have @xmath565 . applying this property and lemma [ lm2 ] on ( [ d12 ] ) , it can be verified that the first term on the right - hand side of ( [ d9 ] ) scales as @xmath566 as @xmath25 scales with @xmath28 . this completes the proof . h. a. suraweera , i. krikidis , g. zheng , c. yuen and p. j. smith , `` low - complexity end - to - end performance optimization in mimo full - duplex relay systems , '' _ ieee trans . wireless commun . _ , vol . 13 , pp . 913 - 927 , feb . 2014 . c. y. a. shang , p. j. smith , g. k. woodward and h. a. suraweera , `` linear transceivers for full duplex mimo relays , '' in proc australian communications theory workshop ( ausctw 2014 ) , sydney , australia , feb . 2014 , pp . d. nguyen , l .- n . tran , p. pirinen , and m. latva - aho , `` precoding for full duplex multiuser mimo systems : spectral and energy efficiency maximization , '' _ ieee trans . signal process . 16 , pp . 4038 - 4050 , aug . 2013 . a. sabharwal , p. schniter , d. guo , d. w. bliss , s. rangarajan , and r. wichman , `` in - band full - duplex wireless : challenges and opportunities , '' _ ieee j. sel . areas commun . 1637 - 1652 , sep . 2014 . h. q. ngo , h. a. suraweera , m. matthaiou , and e. g. larsson , `` multipair full - duplex relaying with massive arrays and linear processing , '' _ ieee j. sel . areas commun . _ , 9 , pp . 1721 - 1737 , sep . 2014 . j. zhang , c. k. wen , s. jin , x. gao , k. wong , `` on capacity of large - scale mimo multiple access channels with distributed sets of correlated antennas '' _ ieee j. sel . areas commun . 133 - 148 , feb . 2013 . h. yin , d. gesbert , l. cottatellucci , `` dealing with interference in distributed large - scale mimo systems : a statistical approach , '' _ ieee j. sel . topics signal process . _ , vol . 8 , no . 942 - 953 , aug . 2014 . g. santella and f. mazzenga , `` a hybrid analytical - simulation procedure for performance evaluation in m - qam - ofdm schemes in presence of nonlinear distortions , '' _ ieee trans . 142 - 151 , feb . 1998 . h. suzuki , t. tran , i. b. collings , and et al , `` transmitter noise effect on the performance of a mimo - ofdm hardware implementation achieving improved coverage , '' _ ieee j. sel . areas commun . _ , vol . 26 , pp . 867 - 876 , aug . 2008 . j. w. silverstein and z. d. bai , `` on the empirical distribution of eigenvalues of a class of large dimensional random matrices , '' _ journal of multivariate analysis _ , 175 - 192 , 1995 . s. wagner , r. couillet , m. debbah , and d. t. m. slock , `` large system analysis of linear precoding in correlated miso broadcast channels under limited feedback , '' _ ieee trans . inf . theory _ 58 , no . 7 , pp . 4509 - 4537 , jul . 2012 . e. bj@xmath567rnson , p. zetterberg , m. bengtsson , and b. ottersten , `` capacity limits and multiplexing gains of mimo channels with transceiver impairments , '' _ ieee commun . 91 - 94 , 2013 . p. c. weeraddana , m. codreanu , m. latva - aho , and a. ephremides , `` resource allocation for cross - layer utility maximization in wireless networks , '' _ ieee trans . 2790 - 2809 , 2011 . e. bj@xmath567rnson , d. hammarwall , and b. ottersten , `` exploiting quantized channel norm feedback through conditional statistics in arbitrarily correlated mimo systems , '' _ ieee trans . signal process . 4027 - 4041 , 2009 . xiaochen xia received his b.e . degree in electronic science and technology from tianjin university ( tju ) in 2010 , and m.s . degree in communication and information system from pla university of science and technology ( plaust ) , in 2013 . he is currently working toward the ph.d . degree in institution of communications engineering , plaust . his research interests include relaying network , full - duplex communication , network coding , mimo techniques . he received the 2013 excellent master degree dissertation award of jiangsu province , china . dongmei zhang received the b.s . degree in institute of communications engineering , nanjing , china in 1994 , and the m.s . degree in communications and information system from the institute of communications engineering , nanjing , china in 1997 . she is currently an associate professor with the pla university of science and technology , nanjing , china . her research interests include broadband wireless communications , cognitive radio , wireless resource allocation . kui xu received the b.s . degree and ph.d . degree from the pla university of science and technology ( plaust ) , nanjing , china in 2004 and 2009 . he is currently a lecturer in the plaust . his research interests include broadband wireless communications , signal processing for communications and wireless communication networks . he is the author of about 50 papers in refereed journals and conference proceedings and holds 5 patents in china . xu currently served on the technical program committee of the ieee wcsp 2014 . he received the ursi young scientists award in 2014 and the 2010 ten excellent doctor degree dissertation award of plaust . wenfeng ma received the b.s . degree in microwave communication from the institute of communications engineering , nanjing , china in 1995 , and the m.s . degree in communications and information system from the institute of communications engineering , nanjing , china in 1998 , and the ph.d . degree in communications and information system from the pla university of science and technology , nanjing , china , in 2002 . he is currently an associate professor with the pla university of science and technology , nanjing , china . his research interests include wireless communication networks , broadband wireless communications . youyun xu was graduated from shanghai jiao tong university with ph.d . degree in information and communication engineering in 1999 . he is currently a professor with nanjing institute of communication engineering , china . he is also a part - time professor with the institute of wireless communication technology of shanghai jiao tong university ( sjtu ) , china . professor xu has more than 20 years professional experience of teaching and researching in communication theory and engineering . now , his research interests are focusing on new generation wireless mobile communication system ( imt - advanced and related ) , advanced channel coding and modulation techniques , multi - user information theory and radio resource management , wireless sensor networks , cognitive radio networks , etc .

the production of heavy quarkonium at hadron colliders provides particular challenges and opportunity for insight into the theory of quantum chromodynamics ( qcd ) as its mechanisms of production operate at the boundary of the perturbative and non - perturbative regimes . despite being among the most studied of the bound - quark systems , there is still no clear understanding of the mechanisms in the production of quarkonium states like the @xmath0 that can consistently explain both the production cross - section and spin - alignment measurements in @xmath7 , heavy - ion and hadron - hadron collisions ( see review articles@xcite and references therein ) . data obtained by the large hadron collider ( lhc ) collaborations can help to test existing theoretical models of both quarkonium production and @xmath8-production in a new energy regime , at higher transverse momenta and in wider rapidity ranges than have previously been studied . furthermore , quarkonium production in proton - proton collisions plays a key role as a reference point to understand heavy ion collisions and to understand the interplay between production and suppression mechanisms in such collisions@xcite . this paper presents a measurement of the inclusive @xmath0 production cross - section and the production fraction @xmath9 of non - prompt @xmath0 ( produced via the decay of a @xmath1-hadron ) to inclusively - produced @xmath0 ( hereafter referred to as the _ non - prompt fraction _ ) : @xmath10{\textrm{inclusive } } j/\psi x^{\prime\prime})}\ ] ] in the decay channel @xmath11 as a function of both @xmath0 transverse momentum and rapidity in @xmath12 collisions at the lhc at a centre - of - mass energy of 7 tev and with an integrated luminosity of up to 2.3 pb@xmath3 . the fraction has the advantage that acceptances and many efficiencies are the same for the numerator and denominator , and so systematic effects are reduced . the results of these analyses are compared to those made by the cms collaboration @xcite with 314 nb@xmath13 of integrated luminosity and those from the cdf collaboration@xcite where appropriate . from these measurements , the prompt @xmath0 production cross - section ( @xmath14 , produced directly from the proton - proton collisions or from decays of heavier charmonium states like the @xmath15 or @xmath16 ) , and the non - prompt ( @xmath17 ) @xmath0 production cross - section , are extracted . these results are compared to corresponding predictions made by the colour evaporation model @xcite , fixed - order next - to - leading log ( fonll ) @xcite and colour singlet nnlo@xmath6 calculations @xcite . further details of the results of measurements presented here may be found in reference @xcite . in this section , the collection and processing of the data used in the paper are outlined . this involves a description of the most relevant subsystems of the atlas detector@xcite : the trigger system , the muon system and the inner tracking detector . also specified are the triggers used and the offline data processing , in particular the selection of candidate muons . the atlas detector covers almost the full solid angle around the collision point with layers of tracking detectors , calorimeters and muon chambers . for the measurements presented in this paper , the trigger system , the inner detector tracking devices ( i d ) and the muon spectrometer ( ms ) are of particular importance . the i d covers the pseudorapidity range @xmath18 2.5 . it consists of a silicon pixel detector , a silicon strip detector ( sct ) and a transition radiation tracker ( trt ) . these detectors are located at a radial distance from the beam axis between 50.5 mm and 1066 mm and are immersed in a 2 t solenoidal magnetic field . the i d barrel consists of 3 pixel layers , 4 layers of double - sided silicon strip modules and 73 layers of trt straws . the i d end - cap has @xmath19 pixel layers , @xmath20 layers of silicon strips and @xmath21 layers of trt straws . the ms is located inside a toroidal magnetic field which provides 2.5 tm of bending power in the barrel and 5 tm in the end - caps . it consists of four detectors using different technologies and is divided into a barrel region ( @xmath22 ) and two end - cap regions ( @xmath23 ) . precise muon measurements are made using monitored drift tube chambers ( mdt ) in both the barrel and end - cap sections and using cathode strip chambers ( csc ) in the end - caps ; fast triggers are obtained from resistive plate chambers ( rpc ) in the barrel and thin gap chambers ( tgc ) in the end - caps . the chambers are arranged in three layers , so high- particles leave at least three measurement points with a lever arm of several metres . the atlas detector has a three - level trigger system : level 1 ( l1 ) , level 2 ( l2 ) and the event filter ( ef ) . for the measurements presented here , the trigger relies on the minimum bias trigger scintillators ( mbts ) and the muon trigger chambers . the mbts are mounted in front of each liquid argon endcap calorimeter cryostat at @xmath24 m and are segmented into eight sectors in azimuth and two rings in pseudorapidity ( @xmath25 and @xmath26 ) . the mbts trigger is configured to require two hits above threshold from either side of the detector . a dedicated muon trigger at the ef level is required to confirm the candidate events chosen for these measurements . this is initiated by the mbts l1 trigger and searches for the presence of at least one track in the entire ms . this trigger is referred to as the ef minimum bias trigger ; it has an adjustable threshold on the reconstructed muon @xmath4 above which events are accepted and can be prescaled to accept a pre - determined fraction of events meeting the trigger condition . the l1 muon trigger is based on rpcs for the barrel and tgcs for the end - caps @xcite . it seeks hit coincidences within different rpc or tgc detector layers inside programmed geometrical windows which define the muon candidate @xmath4 , then selects candidates above six programmable thresholds and provides a rough estimate of their positions @xcite . for the earlier data used in this analysis , the muon trigger corresponds to the lowest threshold trigger which requires a simple two - layer time coincidence within a region of @xmath27.1 in @xmath28-@xmath29 . no further geometrical constraint is applied . as the instantaneous luminosity of the collider increases , the trigger requirement switches from the ef minimum bias trigger to the l1 muon trigger . later data periods make use of triggers seeded by this l1 trigger but with additional @xmath4 cuts applied at the ef stage ( these are referred to henceforth as the ef muon triggers ) . muon identification and reconstruction extends to @xmath30 , covering a range from 1 gev up to more than 1 tev . `` standalone ms tracks '' are constructed entirely based on the signal hits collected in the ms . the track parameters are obtained from the ms track and are extrapolated to the interaction point , taking into account multiple scattering and energy loss in the traversed material . in this analysis , two categories of reconstructed muons are then defined : * * muons from combined reconstruction : * the _ combined _ muon reconstruction relies on a statistical combination of both a standalone ms track and an i d track . due to i d coverage , the combined reconstruction covers @xmath31 . * * muons from i d track tagging : * a _ tagged _ muon is formed by ms track segments which are not formed into a complete ms track , but which are matched to i d tracks extrapolated to the ms . such a reconstructed muon adopts the measured parameters of the associated i d track . in this paper , the muon tagging is limited to @xmath32 , in order to ensure high quality tracking and a reduction of fake muon candidates . the muon track helix parameters are taken from the i d measurement alone , since the ms does not add much to the precision in the lower momentum range relevant for the @xmath0 measurements presented here . proton - proton collision data , at a centre - of - mass energy of 7 tev , are included in this analysis if taken during stable beam periods and when the ms , i d and magnet systems were collecting data of a sufficiently high quality to be suitable for physics analysis . monte carlo samples are used for determining acceptance corrections , as part of the trigger efficiency studies and in systematic cross - checks . they are generated using pythia 6 @xcite and tuned using the atlas mc09 tune@xcite which uses the mrst lo@xmath6 parton distribution functions @xcite . the passage of the generated particles through the detector is simulated with geant4 @xcite and the data are fully reconstructed with the same software that is used to process the data from the detector . for the signal @xmath0 monte carlo ( used to derive the kinematic acceptance corrections ) , the pythia implementation of prompt @xmath0 production sub - processes in the nrqcd colour octet mechanism framework @xcite is used . prompt @xmath0 production includes _ direct _ production from the hard interaction , as well as charmonium feed - down from excited states . these _ prompt _ production modes are distinct from _ non - prompt _ production that is characterised by the production of @xmath0 via the decay of a @xmath1-hadron . all samples are generated with polar and azimuthal isotropy in the decay of the @xmath0 ( the default in pythia ) and are reweighted at the particle level according to their respective angular dependencies in order to describe a number of different spin - alignment scenarios ( see section [ sec : acceptance ] ) . the @xmath0 spin - alignment is not measured in this analysis , so the reweighted mc samples are used to provide an uncertainty band on the measurement of the production cross - section , determined by the maximum variation in acceptance across the full allowed range of @xmath0 spin alignment . the analyses presented in this paper make use of the triggers described in section [ section : atlastrig ] . for the inclusive cross - section , in a given data taking period an event is retained or discarded based on the decision of a single specific trigger , without reference to any other triggers . for data from the initial running with lower instantaneous luminosity , the l1 muon trigger is used . during later periods , with higher instantaneous luminosity , a more selective ef muon trigger with a 4 gev threshold is required , and eventually , this is increased to a 6 gev threshold . the sample collected by these triggers and passing the data quality selections corresponds to an integrated luminosity of @xmath33 . for the measurement of the @xmath34 non - prompt fraction ( see equation [ eqn : fraction ] ) , two additional triggers are employed , and rather than using a single trigger to veto or accept events , several triggers are used simultaneously such that any one of them having fired results in the event being included . from the initial period , events triggering either the l1 muon trigger or the ef minimum bias trigger are used ( whereas only the l1 muon trigger is used for the cross section ) . for intermediate instantaneous luminosities the l1 muon trigger is used alone since the ef minimum bias trigger is highly prescaled at this stage . for the highest instantaneous luminosities , events are accepted which pass any of the ef muon triggers with thresholds of 4 , 6 or 10 gev . during the runs with the highest instantaneous luminosities , the triggers with @xmath35 and @xmath36 gev are prescaled ; however , the @xmath37 threshold trigger is not . the inclusion of this unprescaled trigger along with the addition of the ef minimum bias trigger for the @xmath34 non - prompt fraction measurement results in a slightly higher integrated luminosity of @xmath38 . to veto cosmic rays , events passing the trigger selection are required to have at least three tracks associated with the same reconstructed primary vertex . the three tracks must each have at least one hit in the pixel system and at least six hits in the sct . each remaining event is required to contain at least one pair of reconstructed muons . only muons associated with i d tracks that have at least one hit in the pixels and six in the sct are accepted . di - muon pairs with opposite charges are considered to be @xmath0 candidates if at least one combined muon is present in the pair . at least one reconstructed muon candidate is required to match a muon trigger ( that is , at least one muon from the @xmath0 candidate should have fired the trigger ) . for the early data , when the trigger is essentially based on the l1 muon trigger , at least one of the offline muons is required to match the trigger muon candidate to within @xmath39 at the ms plane ; for the later data taking , where the ef muon trigger is used , the offline and trigger muons are required to match within @xmath40 . the two i d tracks from each pair of muons passing these selections are fitted to a common vertex @xcite . no constraints are applied in the fit and a very loose vertex quality requirement , which retains over @xmath41 of the candidates , is used . for the @xmath34 non - prompt fraction analysis , where lifetime information is an important element of the fit , additional requirements are made on the @xmath11 candidates . the probability of the fit to the @xmath0 vertex is required to be greater than @xmath42 . for this measurement @xmath0 candidates are rejected if the two muon candidate tracks were used to build different primary vertices in the offline reconstruction ( so that there is an ambiguity as to which primary vertex to use in the lifetime calculation ) . this rejects fewer than @xmath43 of the @xmath0 candidates . this selection is not applied for the cross - section analysis . the measurement of the inclusive differential cross - section is determined as @xmath45 where @xmath46 is the @xmath0 yield in a given @xmath47 bin after continuum background subtraction and correction for detector efficiency , bin migration and acceptance effects , @xmath48 is the integrated luminosity of the data sample and @xmath49 and @xmath50 are the @xmath4 and rapidity bin widths . the probability @xmath51 that a @xmath52 decay is reconstructed depends on the kinematics of the decay , as well as the muon reconstruction and trigger efficiencies . in order to recover the true number @xmath46 of such decays produced in the collisions , a weight @xmath53 is applied to each observed @xmath0 candidate , defined as the inverse of that probability and calculated as follows : @xmath54 where @xmath55 is the kinematic acceptance , @xmath56 is a correction factor for bin migrations due to finite detector resolution , @xmath57 is the i d tracking efficiency and @xmath58 is the single - muon offline reconstruction efficiency . here @xmath59 and @xmath60 are the transverse momenta and pseudorapidities of the positive and negative muons from the @xmath0 decay . the trigger efficiency @xmath61 for a given @xmath0 candidate is calculated from single - muon trigger efficiencies @xmath62 as follows : @xmath63 the resultant weighted invariant mass peak is then fitted ( see section [ sec : mlfit ] ) to extract @xmath46 . the kinematic acceptance @xmath64 is the probability that the muons from a @xmath0 with transverse momentum @xmath4 and rapidity @xmath65 fall into the fiducial volume of the detector . this is calculated using generator - level monte carlo , applying cuts on the momenta and pseudorapidities of the muons to emulate the detector geometry . global cuts of @xmath66gev for @xmath67 are supplemented by finer @xmath4 thresholds in slices of @xmath28 to ensure that regions of the detector where the values of offline and trigger efficiencies are so low as to be compatible with zero within the uncertainties ( approximately 10% ) are excluded from the analysis . the acceptance also depends on the spin - alignment of the @xmath0 , which is not known for lhc conditions . the general angular distribution for the decay @xmath52 in the @xmath0 decay frame is given by : @xmath68 where @xmath69 is the angle between the direction of the positive muon momentum in the @xmath0 decay frame and the @xmath0 line of flight , while @xmath70 is defined as the angle between the @xmath0 production and decay planes in the lab frame ( see figure [ fig : coordinates ] , reference @xcite and references therein ) . spin - alignment angles , in the @xmath0 decay frame . @xmath69 is the angle between the direction of the positive muon in that frame and the direction of @xmath0 in the laboratory frame , which is directed along the @xmath71-axis . @xmath70 is the angle between the @xmath0 production ( @xmath72 ) plane and its decay plane formed by the direction of the @xmath0 and the lepton @xmath73 ( from @xcite ) . , scaledwidth=40.0% ] a large number of possible combinations of the coefficients @xmath74 have been studied , including some with @xmath75 . five extreme cases have been identified that lead to the biggest variation of acceptance within the kinematics of the atlas detector and define an envelope in which the results may vary under all possible polarisation assumptions : 1 . isotropic distribution , independent of @xmath69 and @xmath70 , with @xmath76 , labelled as flat " . this is used as the main ( central ) hypothesis . 2 . full longitudinal alignment with @xmath77 , labelled as long " . 3 . transverse alignment with @xmath78 , labelled as @xmath79 . 4 . transverse alignment with @xmath80 , labelled as @xmath81 . transverse alignment with @xmath82 , labelled as @xmath83 . two - dimensional acceptance maps are produced in bins of @xmath4 and @xmath65 of the @xmath0 , for each of these five scenarios , and are illustrated in figure [ fig : acc_2d ] . the maps are obtained by reweighting the flat distribution at the generator level using equation [ eqn : spinalign ] . the central value for the cross - section measurement is obtained using the flat distribution , and the measurement is repeated using the other scenarios to provide an envelope of maximum variation , which is stated as a separate uncertainty . the measured efficiency and acceptance corrected @xmath0 @xmath4 distribution is parameterised in each rapidity slice by a smooth analytic function smeared with a gaussian distribution , with resolution derived from the data . this function is integrated numerically over each analysis bin , both with and without smearing applied , and the ratio of the two integrals is assigned as the correction factor . the effects of this correction are minimal at low @xmath4 and at low rapidities ( around 0.1% ) but increase at higher @xmath4 and at higher rapidities ( reflecting the decreasing momentum resolution ) to a maximum of approximately 3% . the offline single muon reconstruction efficiencies are obtained from data using a tag and probe method @xcite , where muons are paired with i d tracks ( probes " ) of opposite charge . the pairs are divided into two categories : those in which the probe is reconstructed as a muon ( `` matched '' ) and those in which it is not ( `` unmatched '' ) . both sets of pairs are binned according to the @xmath4 and @xmath28 of the probe . in each of these bins , the muon reconstruction efficiency is obtained as the ratio of the number of @xmath0 candidates in the peak of the matched distribution to the total number of candidates in the two mass distributions . the efficiency is extracted as a parameter of a simultaneous fit to both distributions . the dependence of the offline reconstruction efficiency on the muon charge is well described by mc within the acceptance . this procedure is repeated separately for combined and tagged muons . at higher @xmath4 ( for muons with @xmath4 above 6 gev ) , the efficiency determination is supported by additional tag and probe @xmath84 data@xcite for improved precision in the efficiency plateau region . a hybrid monte carlo and data - derived ( tag and probe ) scheme is used to provide trigger efficiencies for the analysis with finer binning than would be possible with the available data statistics . this is necessary to avoid significant biases that would otherwise appear in the analysis with coarsely binned efficiencies across rapidly - changing efficiency regions . due to significant charge dependence at low @xmath4 and high pseudorapidity , separate trigger efficiency maps are produced for positive and negative muons . fully simulated samples of prompt @xmath85 decays are used to populate the @xmath0 @xmath47 plane , using a fine binning . for each bin , the probability of a muon activating the trigger is determined . the derived efficiencies are then reweighted to match the data efficiencies in the reconstructed bins in cases where discrepancies exist between the data and monte carlo , and uncertainties from data are assigned . muon reconstruction efficiencies have been determined relative to reconstructed i d tracks . inner detector tracks associated to muons and having the selection cuts used in this analysis have a reconstruction efficiency @xmath57 of @xmath86 per track ( with no significant pseudorapidity or @xmath4 dependence observed within the phase space probed with this analysis ) , which is applied as an additional correction to the @xmath0 candidate yields . the distribution of reconstructed @xmath0 candidates over the candidate @xmath87 plane is shown in figure [ fig : yieldmap ] . the majority of @xmath0 candidates are reconstructed in intermediate-@xmath4 , high-@xmath65 areas , as at lower @xmath4 values the acceptance of the detector is limited . candidates ( in the invariant mass interval @xmath88 gev ) as a function of @xmath0 @xmath4 and rapidity.,title="fig:",scaledwidth=55.0% ] + candidates ( in the invariant mass interval @xmath88 gev ) as a function of @xmath0 @xmath4 and rapidity.,title="fig:",scaledwidth=42.0%,scaledwidth=33.0% ] candidates ( in the invariant mass interval @xmath88 gev ) as a function of @xmath0 @xmath4 and rapidity.,title="fig:",scaledwidth=55.0% ] + the inclusive @xmath0 production cross - section is determined in four slices of @xmath0 rapidity : @xmath89 and @xmath90 . in figure [ fig : jpsimass ] , the invariant mass distributions for all oppositely charged muon pairs passing the selection for the differential cross - section measurement are shown , before acceptance and efficiency corrections , for the four rapidity slices . table [ tab : jpsimass ] presents the results of the combined signal and background fits . in these fits the @xmath0 and @xmath91(2s ) peaks are represented by gaussians , while the background is described by a quadratic polynomial . [ fig : jpsimass1 ] candidates used in the cross - section analysis , corresponding to an integrated luminosity of @xmath92 pb@xmath3 . the points are data , and the uncertainties indicated are statistical only . the solid lines are the result of the fit described in the text.,title="fig:",scaledwidth=45.0% ] [ fig : jpsimass2 ] candidates used in the cross - section analysis , corresponding to an integrated luminosity of @xmath92 pb@xmath3 . the points are data , and the uncertainties indicated are statistical only . the solid lines are the result of the fit described in the text.,title="fig:",scaledwidth=45.0% ] [ fig : jpsimass3 ] candidates used in the cross - section analysis , corresponding to an integrated luminosity of @xmath92 pb@xmath3 . the points are data , and the uncertainties indicated are statistical only . the solid lines are the result of the fit described in the text.,title="fig:",scaledwidth=45.0% ] [ fig : jpsimass4 ] candidates used in the cross - section analysis , corresponding to an integrated luminosity of @xmath92 pb@xmath3 . the points are data , and the uncertainties indicated are statistical only . the solid lines are the result of the fit described in the text.,title="fig:",scaledwidth=45.0% ] .fitted mass , resolution and yields of @xmath0 candidates reconstructed in four @xmath0 rapidity bins . all uncertainties quoted are statistical only . the shift in mass away from the world average in the highest rapidity bin reflects the few - per - mille uncertainty in the tracking @xmath4 scale at the extreme ends of the detector . [ cols=">,^,^,^,^ " , ] g. aad@xmath93 , b. abbott@xmath94 , j. abdallah@xmath95 , a.a . abdelalim@xmath96 , a. abdesselam@xmath97 , o. abdinov@xmath98 , b. abi@xmath99 , m. abolins@xmath100 , h. abramowicz@xmath101 , h. abreu@xmath102 , e. acerbi@xmath103 , b.s . acharya@xmath104 , d.l . adams@xmath105 , t.n . addy@xmath106 , j. adelman@xmath107 , m. aderholz@xmath108 , s. adomeit@xmath109 , p. adragna@xmath110 , t. adye@xmath111 , s. aefsky@xmath112 , j.a . aguilar - saavedra@xmath113@xmath114 , m. aharrouche@xmath115 , s.p . ahlen@xmath116 , f. ahles@xmath93 , a. ahmad@xmath117 , m. ahsan@xmath118 , g. aielli@xmath119 , t. akdogan@xmath120 , t.p.a . kesson@xmath121 , g. akimoto@xmath122 , a.v . akimov @xmath123 , a. akiyama@xmath124 , m.s . alam@xmath125 , m.a . alam@xmath126 , s. albrand@xmath127 , m. aleksa@xmath128 , i.n . aleksandrov@xmath129 , m. aleppo@xmath103 , f. alessandria@xmath130 , c. alexa@xmath131 , g. alexander@xmath101 , g. alexandre@xmath96 , t. alexopoulos@xmath132 , m. alhroob@xmath133 , m. aliev@xmath134 , g. alimonti@xmath130 , j. alison@xmath135 , m. aliyev@xmath98 , p.p . allport@xmath136 , s.e . allwood - spiers@xmath137 , j. almond@xmath138 , a. aloisio@xmath139 , r. alon@xmath140 , a. alonso@xmath121 , m.g . alviggi@xmath139 , k. amako@xmath141 , p. amaral@xmath128 , c. amelung@xmath112 , v.v . ammosov@xmath142 , a. amorim@xmath143@xmath144 , g. amors@xmath145 , n. amram@xmath101 , c. anastopoulos@xmath146 , t. andeen@xmath147 , c.f . anders@xmath133 , k.j . anderson@xmath148 , a. andreazza@xmath103 , v. andrei@xmath149 , m - l . andrieux@xmath127 , x.s . anduaga@xmath150 , a. angerami@xmath147 , f. anghinolfi@xmath128 , n. anjos@xmath143 , a. annovi@xmath151 , a. antonaki@xmath152 , m. antonelli@xmath151 , s. antonelli@xmath153 , a. antonov@xmath154 , j. antos@xmath155 , f. anulli@xmath156 , s. aoun@xmath157 , l. aperio bella@xmath158 , r. apolle@xmath97 , g. arabidze@xmath100 , i. aracena@xmath159 , y. arai@xmath141 , a.t.h . arce@xmath160 , j.p . archambault@xmath161 , s. arfaoui@xmath128@xmath162 , j - f . arguin@xmath163 , e. arik@xmath120@xmath164 , m. arik@xmath120 , a.j . armbruster@xmath165 , o. arnaez@xmath115 , c. arnault@xmath102 , a. artamonov@xmath166 , g. artoni@xmath167 , d. arutinov@xmath133 , s. asai@xmath122 , r. asfandiyarov@xmath168 , s. ask@xmath169 , b. sman@xmath170 , l. asquith@xmath171 , k. assamagan@xmath105 , a. astbury@xmath172 , a. astvatsatourov@xmath173 , g. atoian@xmath107 , b. aubert@xmath158 , b. auerbach@xmath107 , e. auge@xmath102 , k. augsten@xmath174 , m. aurousseau@xmath158 , n. austin@xmath136 , r. avramidou@xmath132 , d. axen@xmath175 , c. ay@xmath176 , g. azuelos@xmath177@xmath178 , y. azuma@xmath122 , m.a . baak@xmath128 , g. baccaglioni@xmath130 , c. bacci@xmath179 , a.m. bach@xmath163 , h. bachacou@xmath180 , k. bachas@xmath128 , g. bachy@xmath128 , m. backes@xmath96 , m. backhaus@xmath133 , e. badescu@xmath131 , p. bagnaia@xmath167 , s. bahinipati@xmath181 , y. bai@xmath182 , d.c . bailey@xmath183 , t. bain@xmath183 , j.t . baines@xmath111 , o.k . baker@xmath107 , m.d . baker@xmath105 , s. baker@xmath184 , f. baltasar dos santos pedrosa@xmath128 , e. banas@xmath185 , p. banerjee@xmath177 , sw . banerjee@xmath172 , d. banfi@xmath128 , a. bangert@xmath186 , v. bansal@xmath172 , h.s . bansil@xmath187 , l. barak@xmath140 , s.p . baranov@xmath123 , a. barashkou@xmath129 , a. barbaro galtieri@xmath163 , t. barber@xmath169 , e.l . barberio@xmath188 , d. barberis@xmath189 , m. barbero@xmath133 , d.y . bardin@xmath129 , t. barillari@xmath108 , m. barisonzi@xmath190 , t. barklow@xmath159 , n. barlow@xmath169 , b.m . barnett@xmath111 , r.m . barnett@xmath163 , a. baroncelli@xmath191 , a.j . barr@xmath97 , f. barreiro@xmath192 , j. barreiro guimares da costa@xmath193 , p. barrillon@xmath102 , r. bartoldus@xmath159 , a.e . barton@xmath194 , d. bartsch@xmath133 , v. bartsch@xmath195 , r.l . bates@xmath137 , l. batkova@xmath196 , j.r . batley@xmath169 , a. battaglia@xmath197 , m. battistin@xmath128 , g. battistoni@xmath130 , f. bauer@xmath180 , h.s . bawa@xmath159@xmath198 , b. beare@xmath183 , t. beau@xmath199 , p.h . beauchemin@xmath97 , r. beccherle@xmath200 , p. bechtle@xmath201 , h.p . beck@xmath197 , m. beckingham@xmath93 , k.h . becks@xmath190 , a.j . beddall@xmath202 , a. beddall@xmath202 , s. bedikian@xmath107 , v.a . bednyakov@xmath129 , c.p . bee@xmath157 , m. begel@xmath105 , s. behar harpaz@xmath203 , p.k . behera@xmath204 , m. beimforde@xmath108 , c. belanger - champagne@xmath205 , p.j . bell@xmath96 , w.h . bell@xmath96 , g. bella@xmath101 , l. bellagamba@xmath206 , f. bellina@xmath128 , g. bellomo@xmath103 , m. bellomo@xmath207 , a. belloni@xmath193 , o. beloborodova@xmath208 , k. belotskiy@xmath154 , o. beltramello@xmath128 , s. ben ami@xmath203 , o. benary@xmath101 , d. benchekroun@xmath209 , c. benchouk@xmath157 , m. bendel@xmath115 , b.h . benedict@xmath210 , n. benekos@xmath211 , y. benhammou@xmath101 , d.p . benjamin@xmath160 , m. benoit@xmath102 , j.r . bensinger@xmath112 , k. benslama@xmath212 , s. bentvelsen@xmath213 , d. berge@xmath128 , e. bergeaas kuutmann@xmath201 , n. berger@xmath158 , f. berghaus@xmath172 , e. berglund@xmath96 , j. beringer@xmath163 , k. bernardet@xmath157 , p. bernat@xmath184 , r. bernhard@xmath93 , c. bernius@xmath105 , t. berry@xmath126 , f. bertinelli@xmath128 , f. bertolucci@xmath214 , m.i . besana@xmath103 , n. besson@xmath180 , s. bethke@xmath108 , w. bhimji@xmath215 , r.m . bianchi@xmath128 , m. bianco@xmath216 , o. biebel@xmath109 , s.p . bieniek@xmath184 , j. biesiada@xmath163 , m. biglietti@xmath179 , h. bilokon@xmath151 , m. bindi@xmath153 , s. binet@xmath102 , a. bingul@xmath202 , c. bini@xmath167 , c. biscarat@xmath217 , u. bitenc@xmath93 , k.m . black@xmath116 , r.e . blair@xmath171 , j .- b . blanchard@xmath102 , g. blanchot@xmath128 , c. blocker@xmath112 , j. blocki@xmath185 , a. blondel@xmath96 , w. blum@xmath115 , u. blumenschein@xmath176 , g.j . bobbink@xmath213 , v.b . bobrovnikov@xmath208 , a. bocci@xmath160 , c.r . boddy@xmath97 , m. boehler@xmath201 , j. boek@xmath190 , n. boelaert@xmath218 , s. bser@xmath184 , j.a . bogaerts@xmath128 , a. bogdanchikov@xmath208 , a. bogouch@xmath219@xmath164 , c. bohm@xmath220 , v. boisvert@xmath126 , t. bold@xmath210@xmath221 , v. boldea@xmath131 , m. bona@xmath110 , v.g . bondarenko@xmath154 , m. boonekamp@xmath180 , g. boorman@xmath126 , c.n . booth@xmath146 , p. booth@xmath146 , s. bordoni@xmath199 , c. borer@xmath197 , a. borisov@xmath142 , g. borissov@xmath194 , i. borjanovic@xmath222 , s. borroni@xmath167 , k. bos@xmath213 , d. boscherini@xmath206 , m. bosman@xmath95 , h. boterenbrood@xmath213 , d. botterill@xmath111 , j. bouchami@xmath177 , j. boudreau@xmath223 , e.v . bouhova - thacker@xmath194 , c. boulahouache@xmath223 , c. bourdarios@xmath102 , n. bousson@xmath157 , a. boveia@xmath148 , j. boyd@xmath128 , i.r . boyko@xmath129 , n.i . bozhko@xmath142 , i. bozovic - jelisavcic@xmath224 , j. bracinik@xmath187 , a. braem@xmath128 , e. brambilla@xmath216 , p. branchini@xmath191 , g.w . brandenburg@xmath193 , a. brandt@xmath225 , g. brandt@xmath134 , o. brandt@xmath176 , u. bratzler@xmath226 , b. brau@xmath227 , j.e . brau@xmath228 , h.m . braun@xmath190 , b. brelier@xmath183 , j. bremer@xmath128 , r. brenner@xmath205 , s. bressler@xmath203 , d. breton@xmath102 , n.d . brett@xmath97 , p.g . bright - thomas@xmath187 , d. britton@xmath137 , f.m . brochu@xmath169 , i. brock@xmath133 , r. brock@xmath100 , t.j . brodbeck@xmath194 , e. brodet@xmath101 , f. broggi@xmath130 , c. bromberg@xmath100 , g. brooijmans@xmath147 , w.k . brooks@xmath229 , g. brown@xmath138 , e. brubaker@xmath148 , p.a . bruckman de renstrom@xmath185 , d. bruncko@xmath155 , r. bruneliere@xmath93 , s. brunet@xmath230 , a. bruni@xmath206 , g. bruni@xmath206 , m. bruschi@xmath206 , t. buanes@xmath231 , f. bucci@xmath96 , j. buchanan@xmath97 , n.j . buchanan@xmath181 , p. buchholz@xmath232 , r.m . buckingham@xmath97 , a.g . buckley@xmath215 , s.i . buda@xmath131 , i.a . budagov@xmath129 , b. budick@xmath233 , v. bscher@xmath115 , l. bugge@xmath234 , d. buira - clark@xmath97 , e.j . buis@xmath213 , o. bulekov@xmath154 , m. bunse@xmath235 , t. buran@xmath234 , h. burckhart@xmath128 , s. burdin@xmath136 , t. burgess@xmath231 , s. burke@xmath111 , e. busato@xmath236 , p. bussey@xmath137 , c.p . buszello@xmath205 , f. butin@xmath128 , b. butler@xmath159 , j.m . butler@xmath116 , c.m . buttar@xmath137 , j.m . butterworth@xmath184 , w. buttinger@xmath169 , t. byatt@xmath184 , s. cabrera urbn@xmath145 , m. caccia@xmath103 , d. caforio@xmath153 , o. cakir@xmath237 , p. calafiura@xmath163 , g. calderini@xmath199 , p. calfayan@xmath109 , r. calkins@xmath238 , l.p . caloba@xmath239 , r. caloi@xmath167 , d. calvet@xmath236 , s. calvet@xmath236 , r. camacho toro@xmath236 , a. camard@xmath199 , p. camarri@xmath119 , m. cambiaghi@xmath240 , d. cameron@xmath234 , j. cammin@xmath133 , s. campana@xmath128 , m. campanelli@xmath184 , v. canale@xmath139 , f. canelli@xmath148 , a. canepa@xmath241 , j. cantero@xmath192 , l. capasso@xmath139 , m.d.m . capeans garrido@xmath128 , i. caprini@xmath131 , m. caprini@xmath131 , d. capriotti@xmath108 , m. capua@xmath242 , r. caputo@xmath117 , c. caramarcu@xmath131 , r. cardarelli@xmath243 , t. carli@xmath128 , g. carlino@xmath244 , l. carminati@xmath103 , b. caron@xmath241 , s. caron@xmath93 , c. carpentieri@xmath93 , g.d . carrillo montoya@xmath168 , a.a . carter@xmath110 , j.r . carter@xmath169 , j. carvalho@xmath143@xmath245 , d. casadei@xmath233 , m.p . casado@xmath95 , m. cascella@xmath214 , c. caso@xmath189@xmath164 , a.m. castaneda hernandez@xmath168 , e. castaneda - miranda@xmath168 , v. castillo gimenez@xmath145 , n.f . castro@xmath143 , g. cataldi@xmath246 , f. cataneo@xmath128 , a. catinaccio@xmath128 , j.r . catmore@xmath194 , a. cattai@xmath128 , g. cattani@xmath119 , s. caughron@xmath100 , d. cauz@xmath247 , a. cavallari@xmath167 , p. cavalleri@xmath199 , d. cavalli@xmath130 , m. cavalli - sforza@xmath95 , v. cavasinni@xmath214 , a. cazzato@xmath216 , f. ceradini@xmath179 , a.s . cerqueira@xmath239 , a. cerri@xmath128 , l. cerrito@xmath110 , f. cerutti@xmath151 , s.a . cetin@xmath248 , f. cevenini@xmath139 , a. chafaq@xmath209 , d. chakraborty@xmath238 , k. chan@xmath181 , b. chapleau@xmath249 , j.d . chapman@xmath169 , j.w . chapman@xmath165 , e. chareyre@xmath199 , d.g . charlton@xmath187 , v. chavda@xmath138 , s. cheatham@xmath194 , s. chekanov@xmath171 , s.v . chekulaev@xmath241 , g.a . chelkov@xmath129 , m.a . chelstowska@xmath250 , c. chen@xmath251 , h. chen@xmath105 , l. chen@xmath181 , s. chen@xmath252 , t. chen@xmath252 , x. chen@xmath168 , s. cheng@xmath182 , a. cheplakov@xmath129 , v.f . chepurnov@xmath129 , r. cherkaoui el moursli@xmath253 , v. chernyatin@xmath105 , e. cheu@xmath254 , s.l . cheung@xmath183 , l. chevalier@xmath180 , f. chevallier@xmath180 , g. chiefari@xmath139 , l. chikovani@xmath255 , j.t . childers@xmath149 , a. chilingarov@xmath194 , g. chiodini@xmath246 , m.v . chizhov@xmath129 , g. choudalakis@xmath148 , s. chouridou@xmath186 , i.a . christidi@xmath184 , a. christov@xmath93 , d. chromek - burckhart@xmath128 , j. chudoba@xmath256 , g. ciapetti@xmath167 , k. ciba@xmath257 , a.k . ciftci@xmath237 , r. ciftci@xmath237 , d. cinca@xmath236 , v. cindro@xmath258 , m.d . ciobotaru@xmath210 , c. ciocca@xmath153 , a. ciocio@xmath163 , m. cirilli@xmath165 , m. ciubancan@xmath131 , a. clark@xmath96 , p.j . clark@xmath215 , w. cleland@xmath223 , j.c . clemens@xmath157 , b. clement@xmath127 , c. clement@xmath170 , r.w . clifft@xmath111 , y. coadou@xmath157 , m. cobal@xmath247 , a. coccaro@xmath189 , j. cochran@xmath251 , p. coe@xmath97 , j.g . cogan@xmath159 , j. coggeshall@xmath211 , e. cogneras@xmath217 , c.d . cojocaru@xmath161 , j. colas@xmath158 , a.p . colijn@xmath213 , c. collard@xmath102 , n.j . collins@xmath187 , c. collins - tooth@xmath137 , j. collot@xmath127 , g. colon@xmath227 , r. coluccia@xmath216 , g. comune@xmath100 , p. conde muio@xmath143 , e. coniavitis@xmath97 , m.c . conidi@xmath95 , m. consonni@xmath250 , s. constantinescu@xmath131 , c. conta@xmath240 , f. conventi@xmath244@xmath259 , j. cook@xmath128 , m. cooke@xmath163 , b.d . cooper@xmath184 , a.m. cooper - sarkar@xmath97 , n.j . cooper - smith@xmath126 , k. copic@xmath147 , t. cornelissen@xmath189 , m. corradi@xmath206 , f. corriveau@xmath249@xmath260 , a. cortes - gonzalez@xmath211 , g. cortiana@xmath108 , g. costa@xmath130 , m.j . costa@xmath145 , d. costanzo@xmath146 , t. costin@xmath148 , d. ct@xmath128 , l. courneyea@xmath172 , g. cowan@xmath126 , c. cowden@xmath169 , b.e . cox@xmath138 , k. cranmer@xmath233 , f. crescioli@xmath214 , m. cristinziani@xmath133 , g. crosetti@xmath242 , r. crupi@xmath216 , s. crp - renaudin@xmath127 , c. cuenca almenar@xmath107 , t. cuhadar donszelmann@xmath146 , s. cuneo@xmath189 , m. curatolo@xmath151 , c.j . curtis@xmath187 , p. cwetanski@xmath230 , h. czirr@xmath232 , z. czyczula@xmath234 , s. dauria@xmath137 , m. donofrio@xmath136 , a. dorazio@xmath167 , a. da rocha gesualdi mello@xmath239 , p.v.m . da silva@xmath239 , c. da via@xmath138 , w. dabrowski@xmath257 , a. dahlhoff@xmath93 , t. dai@xmath165 , c. dallapiccola@xmath227 , s.j . dallison@xmath111@xmath164 , m. dam@xmath218 , m. dameri@xmath189 , d.s . damiani@xmath186 , h.o . danielsson@xmath128 , r. dankers@xmath213 , d. dannheim@xmath108 , v. dao@xmath96 , g. darbo@xmath200 , g.l . darlea@xmath261 , c. daum@xmath213 , j.p . dauvergne @xmath128 , w. davey@xmath188 , t. davidek@xmath262 , n. davidson@xmath188 , r. davidson@xmath194 , m. davies@xmath177 , a.r . davison@xmath184 , e. dawe@xmath263 , i. dawson@xmath146 , j.w . dawson@xmath171@xmath164 , r.k . daya@xmath264 , k. de@xmath225 , r. de asmundis@xmath244 , s. de castro@xmath153 , p.e . de castro faria salgado@xmath105 , s. de cecco@xmath199 , j. de graat@xmath109 , n. de groot@xmath250 , p. de jong@xmath213 , c. de la taille@xmath102 , h. de la torre@xmath192 , b. de lotto@xmath247 , l. de mora@xmath194 , l. de nooij@xmath213 , m. de oliveira branco@xmath128 , d. de pedis@xmath156 , p. de saintignon@xmath127 , a. de salvo@xmath156 , u. de sanctis@xmath247 , a. de santo@xmath195 , j.b . de vivie de regie@xmath102 , s. dean@xmath184 , d.v . dedovich@xmath129 , j. degenhardt@xmath135 , m. dehchar@xmath97 , m. deile@xmath109 , c. del papa@xmath247 , j. del peso@xmath192 , t. del prete@xmath214 , a. dellacqua@xmath128 , l. dellasta@xmath103 , m. della pietra@xmath244@xmath259 , d. della volpe@xmath139 , m. delmastro@xmath128 , p. delpierre@xmath157 , n. delruelle@xmath128 , p.a . delsart@xmath127 , c. deluca@xmath117 , s. demers@xmath107 , m. demichev@xmath129 , b. demirkoz@xmath95 , j. deng@xmath210 , s.p . denisov@xmath142 , d. derendarz@xmath185 , j.e . derkaoui@xmath265 , f. derue@xmath199 , p. dervan@xmath136 , k. desch@xmath133 , e. devetak@xmath117 , p.o . deviveiros@xmath183 , a. dewhurst@xmath111 , b. dewilde@xmath117 , s. dhaliwal@xmath183 , r. dhullipudi@xmath105@xmath266 , a. di ciaccio@xmath119 , l. di ciaccio@xmath158 , a. di girolamo@xmath128 , b. di girolamo@xmath128 , s. di luise@xmath179 , a. di mattia@xmath100 , b. di micco@xmath179 , r. di nardo@xmath119 , a. di simone@xmath119 , r. di sipio@xmath153 , m.a . diaz@xmath267 , f. diblen@xmath202 , e.b . diehl@xmath165 , h. dietl@xmath108 , j. dietrich@xmath93 , t.a . dietzsch@xmath149 , s. diglio@xmath102 , k. dindar yagci@xmath264 , j. dingfelder@xmath133 , c. dionisi@xmath167 , p. dita@xmath131 , s. dita@xmath131 , f. dittus@xmath128 , f. djama@xmath157 , r. djilkibaev@xmath233 , t. djobava@xmath255 , m.a.b . do vale@xmath239 , a. do valle wemans@xmath143 , t.k.o . doan@xmath158 , m. dobbs@xmath249 , r. dobinson @xmath128@xmath164 , d. dobos@xmath235 , e. dobson@xmath128 , m. dobson@xmath210 , j. dodd@xmath147 , o.b . dogan@xmath120@xmath164 , c. doglioni@xmath97 , t. doherty@xmath137 , y. doi@xmath141@xmath164 , j. dolejsi@xmath262 , i. dolenc@xmath258 , z. dolezal@xmath262 , b.a . dolgoshein@xmath154@xmath164 , t. dohmae@xmath122 , m. donadelli@xmath268 , m. donega@xmath135 , j. donini@xmath127 , j. dopke@xmath190 , a. doria@xmath244 , a. dos anjos@xmath168 , m. dosil@xmath95 , a. dotti@xmath214 , m.t . dova@xmath150 , j.d . dowell@xmath187 , a.d . doxiadis@xmath213 , a.t . doyle@xmath137 , z. drasal@xmath262 , j. drees@xmath190 , n. dressnandt@xmath135 , h. drevermann@xmath128 , c. driouichi@xmath218 , m. dris@xmath132 , j.g . drohan@xmath184 , j. dubbert@xmath108 , t. dubbs@xmath186 , s. dube@xmath163 , e. duchovni@xmath140 , g. duckeck@xmath109 , a. dudarev@xmath128 , f. dudziak@xmath251 , m. dhrssen @xmath128 , i.p . duerdoth@xmath138 , l. duflot@xmath102 , m - a . dufour@xmath249 , m. dunford@xmath128 , h. duran yildiz@xmath269 , r. duxfield@xmath146 , m. dwuznik@xmath257 , f. dydak @xmath128 , d. dzahini@xmath127 , m. dren@xmath173 , w.l . ebenstein@xmath160 , j. ebke@xmath109 , s. eckert@xmath93 , s. eckweiler@xmath115 , k. edmonds@xmath115 , c.a . edwards@xmath126 , i. efthymiopoulos@xmath96 , w. ehrenfeld@xmath201 , t. ehrich@xmath108 , t. eifert@xmath128 , g. eigen@xmath231 , k. einsweiler@xmath163 , e. eisenhandler@xmath110 , t. ekelof@xmath205 , m. el kacimi@xmath158 , m. ellert@xmath205 , s. elles@xmath158 , f. ellinghaus@xmath115 , k. ellis@xmath110 , n. ellis@xmath128 , j. elmsheuser@xmath109 , m. elsing@xmath128 , r. ely@xmath163 , d. emeliyanov@xmath111 , r. engelmann@xmath117 , a. engl@xmath109 , b. epp@xmath270 , a. eppig@xmath165 , j. erdmann@xmath176 , a. ereditato@xmath197 , d. eriksson@xmath220 , j. ernst@xmath125 , m. ernst@xmath105 , j. ernwein@xmath180 , d. errede@xmath211 , s. errede@xmath211 , e. ertel@xmath115 , m. escalier@xmath102 , c. escobar@xmath145 , x. espinal curull@xmath95 , b. esposito@xmath151 , f. etienne@xmath157 , a.i . etienvre@xmath180 , e. etzion@xmath101 , d. evangelakou@xmath176 , h. evans@xmath230 , l. fabbri@xmath153 , c. fabre@xmath128 , k. facius@xmath218 , r.m . fakhrutdinov@xmath142 , s. falciano@xmath156 , a.c . falou@xmath102 , y. fang@xmath168 , m. fanti@xmath103 , a. farbin@xmath225 , a. farilla@xmath191 , j. farley@xmath117 , t. farooque@xmath183 , s.m . farrington@xmath97 , p. farthouat@xmath128 , d. fasching@xmath168 , p. fassnacht@xmath128 , d. fassouliotis@xmath152 , b. fatholahzadeh@xmath183 , a. favareto@xmath103 , l. fayard@xmath102 , s. fazio@xmath242 , r. febbraro@xmath236 , p. federic@xmath196 , o.l . fedin@xmath271 , i. fedorko@xmath128 , w. fedorko@xmath100 , m. fehling - kaschek@xmath93 , l. feligioni@xmath157 , d. fellmann@xmath171 , c.u . felzmann@xmath188 , c. feng@xmath272 , e.j . feng@xmath148 , a.b . fenyuk@xmath142 , j. ferencei@xmath155 , j. ferland@xmath177 , b. fernandes@xmath143@xmath144 , w. fernando@xmath273 , s. ferrag@xmath137 , j. ferrando@xmath97 , v. ferrara@xmath201 , a. ferrari@xmath205 , p. ferrari@xmath213 , r. ferrari@xmath207 , a. ferrer@xmath145 , m.l . ferrer@xmath151 , d. ferrere@xmath96 , c. ferretti@xmath165 , a. ferretto parodi@xmath189 , m. fiascaris@xmath148 , f. fiedler@xmath115 , a. filipi@xmath258 , a. filippas@xmath132 , f. filthaut@xmath250 , m. fincke - keeler@xmath172 , m.c.n . fiolhais@xmath143@xmath245 , l. fiorini@xmath95 , a. firan@xmath264 , g. fischer@xmath201 , p. fischer @xmath133 , m.j . fisher@xmath273 , s.m . fisher@xmath111 , j. flammer@xmath128 , m. flechl@xmath93 , i. fleck@xmath232 , j. fleckner@xmath115 , p. fleischmann@xmath274 , s. fleischmann@xmath190 , t. flick@xmath190 , l.r . flores castillo@xmath168 , m.j . flowerdew@xmath108 , f. fhlisch@xmath149 , m. fokitis@xmath132 , t. fonseca martin@xmath197 , d.a . forbush@xmath275 , a. formica@xmath180 , a. forti@xmath138 , d. fortin@xmath241 , j.m . foster@xmath138 , d. fournier@xmath102 , a. foussat@xmath128 , a.j . fowler@xmath160 , k. fowler@xmath186 , h. fox@xmath194 , p. francavilla@xmath214 , s. franchino@xmath240 , d. francis@xmath128 , t. frank@xmath140 , m. franklin@xmath193 , s. franz@xmath128 , m. fraternali@xmath240 , s. fratina@xmath135 , s.t . french@xmath169 , r. froeschl@xmath128 , d. froidevaux@xmath128 , j.a . frost@xmath169 , c. fukunaga@xmath226 , e. fullana torregrosa@xmath128 , j. fuster@xmath145 , c. gabaldon@xmath128 , o. gabizon@xmath140 , t. gadfort@xmath105 , s. gadomski@xmath96 , g. gagliardi@xmath189 , p. gagnon@xmath230 , c. galea@xmath109 , e.j . gallas@xmath97 , m.v . gallas@xmath128 , v. gallo@xmath197 , b.j . gallop@xmath111 , p. gallus@xmath256 , e. galyaev@xmath118 , k.k . gan@xmath273 , y.s . gao@xmath159@xmath198 , v.a . gapienko@xmath142 , a. gaponenko@xmath163 , f. garberson@xmath107 , m. garcia - sciveres@xmath163 , c. garca@xmath145 , j.e . garca navarro@xmath96 , r.w . gardner@xmath148 , n. garelli@xmath128 , h. garitaonandia@xmath213 , v. garonne@xmath128 , j. garvey@xmath187 , c. gatti@xmath151 , g. gaudio@xmath207 , o. gaumer@xmath96 , b. gaur@xmath232 , l. gauthier@xmath180 , i.l . gavrilenko@xmath123 , c. gay@xmath175 , g. gaycken@xmath133 , j - c . gayde@xmath128 , e.n . gazis@xmath132 , p. ge@xmath272 , c.n.p . gee@xmath111 , d.a.a . geerts@xmath213 , ch . geich - gimbel@xmath133 , k. gellerstedt@xmath170 , c. gemme@xmath200 , a. gemmell@xmath137 , m.h . genest@xmath109 , s. gentile@xmath167 , m. george@xmath176 , s. george@xmath126 , p. gerlach@xmath190 , a. gershon@xmath101 , c. geweniger@xmath149 , p. ghez@xmath158 , n. ghodbane@xmath236 , b. giacobbe@xmath206 , s. giagu@xmath167 , v. giakoumopoulou@xmath152 , v. giangiobbe@xmath214 , f. gianotti@xmath128 , b. gibbard@xmath105 , a. gibson@xmath183 , s.m . gibson@xmath128 , g.f . gieraltowski@xmath171 , l.m . gilbert@xmath97 , m. gilchriese@xmath163 , v. gilewsky@xmath276 , d. gillberg@xmath161 , a.r . gillman@xmath111 , d.m . gingrich@xmath181@xmath178 , j. ginzburg@xmath101 , n. giokaris@xmath152 , r. giordano@xmath139 , f.m . giorgi@xmath134 , p. giovannini@xmath108 , p.f . giraud@xmath180 , d. giugni@xmath130 , p. giusti@xmath206 , b.k . gjelsten@xmath234 , l.k . gladilin@xmath277 , c. glasman@xmath192 , j. glatzer@xmath93 , a. glazov@xmath201 , k.w . glitza@xmath190 , g.l . glonti@xmath129 , j. godfrey@xmath263 , j. godlewski@xmath128 , m. goebel@xmath201 , t. gpfert@xmath278 , c. goeringer@xmath115 , c. gssling@xmath235 , t. gttfert@xmath108 , s. goldfarb@xmath165 , d. goldin@xmath264 , t. golling@xmath107 , s.n . golovnia@xmath142 , a. gomes@xmath143@xmath144 , l.s . gomez fajardo@xmath201 , r. gonalo@xmath126 , j. goncalves pinto firmino da costa@xmath201 , l. gonella@xmath133 , a. gonidec@xmath128 , s. gonzalez@xmath168 , s. gonzlez de la hoz@xmath145 , m.l . gonzalez silva@xmath279 , s. gonzalez - sevilla@xmath96 , j.j . goodson@xmath117 , l. goossens@xmath128 , p.a . gorbounov@xmath166 , h.a . gordon@xmath105 , i. gorelov@xmath280 , g. gorfine@xmath190 , b. gorini@xmath128 , e. gorini@xmath216 , a. goriek@xmath258 , e. gornicki@xmath185 , s.a . gorokhov@xmath142 , v.n . goryachev@xmath142 , b. gosdzik@xmath201 , m. gosselink@xmath213 , m.i . gostkin@xmath129 , m. gouanre@xmath158 , i. gough eschrich@xmath210 , m. gouighri@xmath209 , d. goujdami@xmath209 , m.p . goulette@xmath96 , a.g . goussiou@xmath275 , c. goy@xmath158 , i. grabowska - bold@xmath210@xmath221 , v. grabski@xmath281 , p. grafstrm@xmath128 , c. grah@xmath190 , k - j . grahn@xmath282 , f. grancagnolo@xmath246 , s. grancagnolo@xmath134 , v. grassi@xmath117 , v. gratchev@xmath271 , n. grau@xmath147 , h.m . gray@xmath128 , j.a . gray@xmath117 , e. graziani@xmath191 , o.g . grebenyuk@xmath271 , d. greenfield@xmath111 , t. greenshaw@xmath136 , z.d . greenwood@xmath105@xmath266 , i.m . gregor@xmath201 , p. grenier@xmath159 , e. griesmayer@xmath283 , j. griffiths@xmath275 , n. grigalashvili@xmath129 , a.a . grillo@xmath186 , s. grinstein@xmath95 , p.l.y . gris@xmath236 , y.v . grishkevich@xmath277 , j .- f . grivaz@xmath102 , j. grognuz@xmath128 , m. groh@xmath108 , e. gross@xmath140 , j. grosse - knetter@xmath176 , j. groth - jensen@xmath121 , m. gruwe@xmath128 , k. grybel@xmath232 , v.j . guarino@xmath171 , d. guest@xmath107 , c. guicheney@xmath236 , a. guida@xmath216 , t. guillemin@xmath158 , s. guindon@xmath176 , h. guler@xmath249@xmath284 , j. gunther@xmath256 , b. guo@xmath183 , j. guo@xmath147 , a. gupta@xmath148 , y. gusakov@xmath129 , v.n . gushchin@xmath142 , a. gutierrez@xmath177 , p. gutierrez@xmath94 , n. guttman@xmath101 , o. gutzwiller@xmath168 , c. guyot@xmath180 , c. gwenlan@xmath97 , c.b . gwilliam@xmath136 , a. haas@xmath159 , s. haas@xmath128 , c. haber@xmath163 , r. hackenburg@xmath105 , h.k . hadavand@xmath264 , d.r . hadley@xmath187 , p. haefner@xmath108 , f. hahn@xmath128 , s. haider@xmath128 , z. hajduk@xmath185 , h. hakobyan@xmath281 , j. haller@xmath176 , k. hamacher@xmath190 , p. hamal@xmath285 , a. hamilton@xmath96 , s. hamilton@xmath286 , h. han@xmath182 , l. han@xmath287 , k. hanagaki@xmath288 , m. hance@xmath135 , c. handel@xmath115 , p. hanke@xmath149 , c.j . hansen@xmath205 , j.r . hansen@xmath218 , j.b . hansen@xmath218 , j.d . hansen@xmath218 , p.h . hansen@xmath218 , p. hansson@xmath159 , k. hara@xmath289 , g.a . hare@xmath186 , t. harenberg@xmath190 , d. harper@xmath165 , r.d . harrington@xmath116 , o.m . harris@xmath275 , k. harrison@xmath187 , j. hartert@xmath93 , f. hartjes@xmath213 , t. haruyama@xmath141 , a. harvey@xmath106 , s. hasegawa@xmath290 , y. hasegawa@xmath291 , s. hassani@xmath180 , m. hatch@xmath128 , d. hauff@xmath108 , s. haug@xmath197 , m. hauschild@xmath128 , r. hauser@xmath100 , m. havranek@xmath133 , b.m . hawes@xmath97 , c.m . hawkes@xmath187 , r.j . hawkings@xmath128 , d. hawkins@xmath210 , t. hayakawa@xmath124 , d hayden@xmath126 , h.s . hayward@xmath136 , s.j . haywood@xmath111 , e. hazen@xmath116 , m. he@xmath272 , s.j . head@xmath187 , v. hedberg@xmath121 , l. heelan@xmath225 , s. heim@xmath100 , b. heinemann@xmath163 , s. heisterkamp@xmath218 , l. helary@xmath158 , m. heldmann@xmath93 , m. heller@xmath102 , s. hellman@xmath170 , c. helsens@xmath95 , r.c.w . henderson@xmath194 , m. henke@xmath149 , a. henrichs@xmath176 , a.m. henriques correia@xmath128 , s. henrot - versille@xmath102 , f. henry - couannier@xmath157 , c. hensel@xmath176 , t. hen@xmath190 , y. hernndez jimnez@xmath145 , r. herrberg@xmath134 , a.d . hershenhorn@xmath203 , g. herten@xmath93 , r. hertenberger@xmath109 , l. hervas@xmath128 , n.p . hessey@xmath213 , a. hidvegi@xmath220 , e. hign - rodriguez@xmath145 , d. hill@xmath171@xmath164 , j.c . hill@xmath169 , n. hill@xmath171 , k.h . hiller@xmath201 , s. hillert@xmath133 , s.j . hillier@xmath187 , i. hinchliffe@xmath163 , e. hines@xmath135 , m. hirose@xmath288 , f. hirsch@xmath235 , d. hirschbuehl@xmath190 , j. hobbs@xmath117 , n. hod@xmath101 , m.c . hodgkinson@xmath146 , p. hodgson@xmath146 , a. hoecker@xmath128 , m.r . hoeferkamp@xmath280 , j. hoffman@xmath264 , d. hoffmann@xmath157 , m. hohlfeld@xmath115 , m. holder@xmath232 , a. holmes@xmath97 , s.o . holmgren@xmath220 , t. holy@xmath174 , j.l . holzbauer@xmath100 , y. homma@xmath124 , l. hooft van huysduynen@xmath233 , c. horn@xmath159 , s. horner@xmath93 , k. horton@xmath97 , j - y . hostachy@xmath127 , t. hott@xmath108 , s. hou@xmath292 , m.a . houlden@xmath136 , a. hoummada@xmath209 , j. howarth@xmath138 , d.f . howell@xmath97 , i. hristova @xmath201 , j. hrivnac@xmath102 , i. hruska@xmath256 , t. hrynova@xmath158 , p.j . hsu@xmath107 , s .- c . hsu@xmath163 , g.s . huang@xmath94 , z. hubacek@xmath174 , f. hubaut@xmath157 , f. huegging@xmath133 , t.b . huffman@xmath97 , e.w . hughes@xmath147 , g. hughes@xmath194 , r.e . hughes - jones@xmath138 , m. huhtinen@xmath128 , p. hurst@xmath193 , m. hurwitz@xmath163 , u. husemann@xmath201 , n. huseynov@xmath129@xmath293 , j. huston@xmath100 , j. huth@xmath193 , g. iacobucci@xmath244 , g. iakovidis@xmath132 , m. ibbotson@xmath138 , i. ibragimov@xmath232 , r. ichimiya@xmath124 , l. iconomidou - fayard@xmath102 , j. idarraga@xmath102 , m. idzik@xmath257 , p. iengo@xmath158 , o. igonkina@xmath213 , y. ikegami@xmath141 , m. ikeno@xmath141 , y. ilchenko@xmath264 , d. iliadis@xmath294 , d. imbault@xmath199 , m. imhaeuser@xmath190 , m. imori@xmath122 , t. ince@xmath133 , j. inigo - golfin@xmath128 , p. ioannou@xmath152 , m. iodice@xmath191 , g. ionescu@xmath158 , a. irles quiles@xmath145 , k. ishii@xmath141 , a. ishikawa@xmath124 , m. ishino@xmath141 , r. ishmukhametov@xmath264 , c. issever@xmath97 , s. istin@xmath120 , y. itoh@xmath290 , a.v . ivashin@xmath142 , w. iwanski@xmath185 , h. iwasaki@xmath141 , j.m . izen@xmath118 , v. izzo@xmath244 , b. jackson@xmath135 , j.n . jackson@xmath136 , p. jackson@xmath159 , m.r . jaekel@xmath128 , v. jain@xmath230 , k. jakobs@xmath93 , s. jakobsen@xmath218 , j. jakubek@xmath174 , d.k . jana@xmath94 , e. jankowski@xmath183 , e. jansen@xmath184 , a. jantsch@xmath108 , m. janus@xmath133 , g. jarlskog@xmath121 , l. jeanty@xmath193 , k. jelen@xmath257 , i. jen - la plante@xmath148 , p. jenni@xmath128 , a. jeremie@xmath158 , p. je@xmath218 , s. jzquel@xmath158 , m.k . jha@xmath206 , h. ji@xmath168 , w. ji@xmath115 , j. jia@xmath117 , y. jiang@xmath287 , m. jimenez belenguer@xmath201 , g. jin@xmath287 , s. jin@xmath182 , o. jinnouchi@xmath295 , m.d . joergensen@xmath218 , d. joffe@xmath264 , l.g . johansen@xmath231 , m. johansen@xmath170 , k.e . johansson@xmath220 , p. johansson@xmath146 , s. johnert@xmath201 , k.a . johns@xmath254 , k. jon - and@xmath170 , g. jones@xmath138 , r.w.l . jones@xmath194 , t.w . jones@xmath184 , t.j . jones@xmath136 , o. jonsson@xmath128 , c. joram@xmath128 , p.m. jorge@xmath143@xmath144 , j. joseph@xmath163 , x. ju@xmath212 , v. juranek@xmath256 , p. jussel@xmath270 , v.v . kabachenko@xmath142 , s. kabana@xmath197 , m. kaci@xmath145 , a. kaczmarska@xmath185 , p. kadlecik@xmath218 , m. kado@xmath102 , h. kagan@xmath273 , m. kagan@xmath193 , s. kaiser@xmath108 , e. kajomovitz@xmath203 , s. kalinin@xmath190 , l.v . kalinovskaya@xmath129 , s. kama@xmath264 , n. kanaya@xmath122 , m. kaneda@xmath122 , t. kanno@xmath295 , v.a . kantserov@xmath154 , j. kanzaki@xmath141 , b. kaplan@xmath107 , a. kapliy@xmath148 , j. kaplon@xmath128 , d. kar@xmath278 , m. karagoz@xmath97 , m. karnevskiy@xmath201 , k. karr@xmath171 , v. kartvelishvili@xmath194 , a.n . karyukhin@xmath142 , l. kashif@xmath168 , a. kasmi@xmath264 , r.d . kass@xmath273 , a. kastanas@xmath231 , m. kataoka@xmath158 , y. kataoka@xmath122 , e. katsoufis@xmath132 , j. katzy@xmath201 , v. kaushik@xmath254 , k. kawagoe@xmath124 , t. kawamoto@xmath122 , g. kawamura@xmath115 , m.s . kayl@xmath213 , v.a . kazanin@xmath208 , m.y . kazarinov@xmath129 , s.i . kazi@xmath188 , j.r . keates@xmath138 , r. keeler@xmath172 , r. kehoe@xmath264 , m. keil@xmath176 , g.d . kekelidze@xmath129 , m. kelly@xmath138 , j. kennedy@xmath109 , m. kenyon@xmath137 , o. kepka@xmath256 , n. kerschen@xmath128 , b.p . kerevan@xmath258 , s. kersten@xmath190 , k. kessoku@xmath122 , c. ketterer@xmath93 , m. khakzad@xmath161 , f. khalil - zada@xmath98 , h. khandanyan@xmath211 , a. khanov@xmath99 , d. kharchenko@xmath129 , a. khodinov@xmath117 , a.g . kholodenko@xmath142 , a. khomich@xmath149 , t.j . khoo@xmath169 , g. khoriauli@xmath133 , n. khovanskiy@xmath129 , v. khovanskiy@xmath166 , e. khramov@xmath129 , j. khubua@xmath255 , g. kilvington@xmath126 , h. kim@xmath225 , m.s . kim@xmath181 , p.c . kim@xmath159 , s.h . kim@xmath289 , n. kimura@xmath296 , o. kind@xmath134 , b.t . king@xmath136 , m. king@xmath124 , r.s.b . king@xmath97 , j. kirk@xmath111 , g.p . kirsch@xmath97 , l.e . kirsch@xmath112 , a.e . kiryunin@xmath108 , d. kisielewska@xmath257 , t. kittelmann@xmath223 , a.m. kiver@xmath142 , h. kiyamura@xmath124 , e. kladiva@xmath155 , j. klaiber - lodewigs@xmath235 , m. klein@xmath136 , u. klein@xmath136 , k. kleinknecht@xmath115 , m. klemetti@xmath249 , a. klier@xmath140 , a. klimentov@xmath105 , r. klingenberg@xmath235 , e.b . klinkby@xmath218 , t. klioutchnikova@xmath128 , p.f . klok@xmath250 , s. klous@xmath213 , e .- e . kluge@xmath149 , t. kluge@xmath136 , p. kluit@xmath213 , s. kluth@xmath108 , e. kneringer@xmath270 , j. knobloch@xmath128 , e.b.f.g . knoops@xmath157 , a. knue@xmath176 , b.r . ko@xmath160 , t. kobayashi@xmath122 , m. kobel@xmath278 , b. koblitz@xmath128 , m. kocian@xmath159 , a. kocnar@xmath285 , p. kodys@xmath262 , k. kneke@xmath128 , a.c . knig@xmath250 , s. koenig@xmath115 , s. knig@xmath93 , l. kpke@xmath115 , f. koetsveld@xmath250 , p. koevesarki@xmath133 , t. koffas@xmath128 , e. koffeman@xmath213 , f. kohn@xmath176 , z. kohout@xmath174 , t. kohriki@xmath141 , t. koi@xmath159 , t. kokott@xmath133 , g.m . kolachev@xmath208 , h. kolanoski@xmath134 , v. kolesnikov@xmath129 , i. koletsou@xmath130 , j. koll@xmath100 , d. kollar@xmath128 , m. kollefrath@xmath93 , s.d . kolya@xmath138 , a.a . komar@xmath123 , j.r . komaragiri@xmath263 , t. kondo@xmath141 , t. kono@xmath201@xmath297 , a.i . kononov@xmath93 , r. konoplich@xmath233@xmath298 , n. konstantinidis@xmath184 , a. kootz@xmath190 , s. koperny@xmath257 , s.v . kopikov@xmath142 , k. korcyl@xmath185 , k. kordas@xmath294 , v. koreshev@xmath142 , a. korn@xmath163 , a. korol@xmath208 , i. korolkov@xmath95 , e.v . korolkova@xmath146 , v.a . korotkov@xmath142 , o. kortner@xmath108 , s. kortner@xmath108 , v.v . kostyukhin@xmath133 , m.j . kotamki@xmath128 , s. kotov@xmath108 , v.m . kotov@xmath129 , c. kourkoumelis@xmath152 , v. kouskoura@xmath294 , a. koutsman@xmath213 , r. kowalewski@xmath172 , t.z . kowalski@xmath257 , w. kozanecki@xmath180 , a.s . kozhin@xmath142 , v. kral@xmath174 , v.a . kramarenko@xmath277 , g. kramberger@xmath258 , o. krasel@xmath235 , m.w . krasny@xmath199 , a. krasznahorkay@xmath233 , j. kraus@xmath100 , a. kreisel@xmath101 , f. krejci@xmath174 , j. kretzschmar@xmath136 , n. krieger@xmath176 , p. krieger@xmath183 , k. kroeninger@xmath176 , h. kroha@xmath108 , j. kroll@xmath135 , j. kroseberg@xmath133 , j. krstic@xmath222 , u. kruchonak@xmath129 , h. krger@xmath133 , z.v . krumshteyn@xmath129 , a. kruth@xmath133 , t. kubota@xmath122 , s. kuehn@xmath93 , a. kugel@xmath299 , t. kuhl@xmath190 , d. kuhn@xmath270 , v. kukhtin@xmath129 , y. kulchitsky@xmath219 , s. kuleshov@xmath229 , c. kummer@xmath109 , m. kuna@xmath157 , n. kundu@xmath97 , j. kunkle@xmath135 , a. kupco@xmath256 , h. kurashige@xmath124 , m. kurata@xmath289 , y.a . kurochkin@xmath219 , v. kus@xmath256 , w. kuykendall@xmath275 , m. kuze@xmath295 , p. kuzhir@xmath276 , o. kvasnicka@xmath256 , j. kvita@xmath128 , r. kwee@xmath134 , a. la rosa@xmath128 , l. la rotonda@xmath242 , l. labarga@xmath192 , j. labbe@xmath158 , s. lablak@xmath209 , c. lacasta@xmath145 , f. lacava@xmath167 , h. lacker@xmath134 , d. lacour@xmath199 , v.r . lacuesta@xmath145 , e. ladygin@xmath129 , r. lafaye@xmath158 , b. laforge@xmath199 , t. lagouri@xmath192 , s. lai@xmath93 , e. laisne@xmath127 , m. lamanna@xmath128 , c.l . lampen@xmath254 , w. lampl@xmath254 , e. lancon@xmath180 , u. landgraf@xmath93 , m.p.j . landon@xmath110 , h. landsman@xmath203 , j.l . lane@xmath138 , c. lange@xmath201 , a.j . lankford@xmath210 , f. lanni@xmath105 , k. lantzsch@xmath128 , v.v . lapin@xmath142@xmath164 , s. laplace@xmath199 , c. lapoire@xmath133 , j.f . laporte@xmath180 , t. lari@xmath130 , a.v . larionov @xmath142 , a. larner@xmath97 , c. lasseur@xmath128 , m. lassnig@xmath128 , w. lau@xmath97 , p. laurelli@xmath151 , a. lavorato@xmath97 , w. lavrijsen@xmath163 , p. laycock@xmath136 , a.b . lazarev@xmath129 , a. lazzaro@xmath103 , o. le dortz@xmath199 , e. le guirriec@xmath157 , c. le maner@xmath183 , e. le menedeu@xmath180 , m. leahu@xmath128 , a. lebedev@xmath251 , c. lebel@xmath177 , t. lecompte@xmath171 , f. ledroit - guillon@xmath127 , h. lee@xmath213 , j.s.h . lee@xmath300 , s.c . lee@xmath292 , l. lee@xmath107 , m. lefebvre@xmath172 , m. legendre@xmath180 , a. leger@xmath96 , b.c . legeyt@xmath135 , f. legger@xmath109 , c. leggett@xmath163 , m. lehmacher@xmath133 , g. lehmann miotto@xmath128 , x. lei@xmath254 , m.a.l . leite@xmath268 , r. leitner@xmath262 , d. lellouch@xmath140 , j. lellouch@xmath199 , m. leltchouk@xmath147 , v. lendermann@xmath149 , k.j.c . leney@xmath301 , t. lenz@xmath190 , g. lenzen@xmath190 , b. lenzi@xmath180 , k. leonhardt@xmath278 , s. leontsinis@xmath132 , c. leroy@xmath177 , j - r . lessard@xmath172 , j. lesser@xmath220 , c.g . lester@xmath169 , a. leung fook cheong@xmath168 , j. levque@xmath157 , d. levin@xmath165 , l.j . levinson@xmath140 , m.s . levitski@xmath142 , m. lewandowska@xmath116 , g.h . lewis@xmath233 , m. leyton@xmath134 , b. li@xmath157 , h. li@xmath168 , s. li@xmath287 , x. li@xmath165 , z. liang@xmath264 , z. liang@xmath97@xmath302 , b. liberti@xmath243 , p. lichard@xmath128 , m. lichtnecker@xmath109 , k. lie@xmath211 , w. liebig@xmath231 , r. lifshitz@xmath203 , j.n . lilley@xmath187 , c. limbach@xmath133 , a. limosani@xmath188 , m. limper@xmath204 , s.c . lin@xmath292@xmath303 , f. linde@xmath213 , j.t . linnemann@xmath100 , e. lipeles@xmath135 , l. lipinsky@xmath256 , a. lipniacka@xmath231 , t.m . liss@xmath211 , d. lissauer@xmath105 , a. lister@xmath96 , a.m. litke@xmath186 , c. liu@xmath161 , d. liu@xmath292@xmath304 , h. liu@xmath165 , j.b . liu@xmath165 , m. liu@xmath287 , s. liu@xmath181 , y. liu@xmath287 , m. livan@xmath240 , s.s.a . livermore@xmath97 , a. lleres@xmath127 , s.l . lloyd@xmath110 , e. lobodzinska@xmath201 , p. loch@xmath254 , w.s . lockman@xmath186 , s. lockwitz@xmath107 , t. loddenkoetter@xmath133 , f.k . loebinger@xmath138 , a. loginov@xmath107 , c.w . loh@xmath175 , t. lohse@xmath134 , k. lohwasser@xmath93 , m. lokajicek@xmath256 , j. loken @xmath97 , v.p . lombardo@xmath130 , r.e . long@xmath194 , l. lopes@xmath143@xmath144 , d. lopez mateos@xmath147@xmath305 , m. losada@xmath306 , p. loscutoff@xmath163 , f. lo sterzo@xmath167 , m.j . losty@xmath241 , x. lou@xmath118 , a. lounis@xmath102 , k.f . loureiro@xmath306 , j. love@xmath116 , p.a . love@xmath194 , a.j . lowe@xmath159@xmath198 , f. lu@xmath182 , j. lu@xmath181 , l. lu@xmath264 , h.j . lubatti@xmath275 , c. luci@xmath167 , a. lucotte@xmath127 , a. ludwig@xmath278 , d. ludwig@xmath201 , i. ludwig@xmath93 , j. ludwig@xmath93 , f. luehring@xmath230 , g. luijckx@xmath213 , d. lumb@xmath93 , l. luminari@xmath156 , e. lund@xmath234 , b. lund - jensen@xmath282 , b. lundberg@xmath121 , j. lundberg@xmath170 , j. lundquist@xmath218 , m. lungwitz@xmath115 , a. lupi@xmath214 , g. lutz@xmath108 , d. lynn@xmath105 , j. lys@xmath163 , e. lytken@xmath121 , h. ma@xmath105 , l.l . ma@xmath168 , j.a . macana goia@xmath177 , g. maccarrone@xmath151 , a. macchiolo@xmath108 , b. maek@xmath258 , j. machado miguens@xmath143 , d. macina@xmath96 , r. mackeprang@xmath218 , r.j . madaras@xmath163 , w.f . mader@xmath278 , r. maenner@xmath299 , t. maeno@xmath105 , p. mttig@xmath190 , s. mttig@xmath201 , p.j . magalhaes martins@xmath143@xmath245 , l. magnoni@xmath128 , e. magradze@xmath255 , c.a . magrath@xmath250 , y. mahalalel@xmath101 , k. mahboubi@xmath93 , g. mahout@xmath187 , c. maiani@xmath167 , c. maidantchik@xmath239 , a. maio@xmath143@xmath144 , s. majewski@xmath105 , y. makida@xmath141 , n. makovec@xmath102 , p. mal@xmath254 , pa . malecki@xmath185 , p. malecki@xmath185 , v.p . maleev@xmath271 , f. malek@xmath127 , u. mallik@xmath204 , d. malon@xmath171 , s. maltezos@xmath132 , v. malyshev@xmath208 , s. malyukov@xmath129 , r. mameghani@xmath109 , j. mamuzic@xmath224 , a. manabe@xmath141 , l. mandelli@xmath130 , i. mandi@xmath258 , r. mandrysch@xmath134 , j. maneira@xmath143 , p.s . mangeard@xmath100 , i.d . manjavidze@xmath129 , a. mann@xmath176 , p.m. manning@xmath186 , a. manousakis - katsikakis@xmath152 , b. mansoulie@xmath180 , a. manz@xmath108 , a. mapelli@xmath128 , l. mapelli@xmath128 , l. march @xmath192 , j.f . marchand@xmath128 , f. marchese@xmath119 , m. marchesotti@xmath128 , g. marchiori@xmath199 , m. marcisovsky@xmath256 , a. marin@xmath116@xmath164 , c.p . marino@xmath230 , f. marroquim@xmath239 , r. marshall@xmath138 , z. marshall@xmath147@xmath305 , f.k . martens@xmath183 , s. marti - garcia@xmath145 , a.j . martin@xmath107 , b. martin@xmath128 , b. martin@xmath100 , f.f . martin@xmath135 , j.p . martin@xmath177 , ph . martin@xmath127 , t.a . martin@xmath187 , b. martin dit latour@xmath96 , m. martinez@xmath95 , v. martinez outschoorn@xmath193 , a.c . martyniuk@xmath138 , m. marx@xmath138 , f. marzano@xmath156 , a. marzin@xmath94 , l. masetti@xmath115 , t. mashimo@xmath122 , r. mashinistov@xmath123 , j. masik@xmath138 , a.l . maslennikov@xmath208 , m. ma@xmath235 , i. massa@xmath153 , g. massaro@xmath213 , n. massol@xmath158 , a. mastroberardino@xmath242 , t. masubuchi@xmath122 , m. mathes@xmath133 , p. matricon@xmath102 , h. matsumoto@xmath122 , h. matsunaga@xmath122 , t. matsushita@xmath124 , c. mattravers@xmath97@xmath307 , j.m . maugain@xmath128 , s.j . maxfield@xmath136 , d.a . maximov@xmath208 , e.n . may@xmath171 , a. mayne@xmath146 , r. mazini@xmath292 , m. mazur@xmath133 , m. mazzanti@xmath130 , e. mazzoni@xmath214 , s.p . mc kee@xmath165 , a. mccarn@xmath211 , r.l . mccarthy@xmath117 , t.g . mccarthy@xmath161 , n.a . mccubbin@xmath111 , k.w . mcfarlane@xmath106 , j.a . mcfayden@xmath146 , h. mcglone@xmath137 , g. mchedlidze@xmath255 , r.a . mclaren@xmath128 , t. mclaughlan@xmath187 , s.j . mcmahon@xmath111 , r.a . mcpherson@xmath172@xmath260 , a. meade@xmath227 , j. mechnich@xmath213 , m. mechtel@xmath190 , m. medinnis@xmath201 , r. meera - lebbai@xmath94 , t. meguro@xmath288 , r. mehdiyev@xmath177 , s. mehlhase@xmath218 , a. mehta@xmath136 , k. meier@xmath149 , j. meinhardt@xmath93 , b. meirose@xmath121 , c. melachrinos@xmath148 , b.r . mellado garcia@xmath168 , l. mendoza navas@xmath306 , z. meng@xmath292@xmath304 , a. mengarelli@xmath153 , s. menke@xmath108 , c. menot@xmath128 , e. meoni@xmath95 , k.m . mercurio@xmath193 , p. mermod@xmath97 , l. merola@xmath139 , c. meroni@xmath130 , f.s . merritt@xmath148 , a. messina@xmath128 , j. metcalfe@xmath280 , a.s . mete@xmath251 , s. meuser@xmath133 , c. meyer@xmath115 , j - p . meyer@xmath180 , j. meyer@xmath274 , j. meyer@xmath176 , t.c . meyer@xmath128 , w.t . meyer@xmath251 , j. miao@xmath272 , s. michal@xmath128 , l. micu@xmath131 , r.p . middleton@xmath111 , p. miele@xmath128 , s. migas@xmath136 , l. mijovi@xmath201 , g. mikenberg@xmath140 , m. mikestikova@xmath256 , b. mikulec@xmath96 , m. miku@xmath258 , d.w . miller@xmath159 , r.j . miller@xmath100 , w.j . mills@xmath175 , c. mills@xmath193 , a. milov@xmath140 , d.a . milstead@xmath170 , d. milstein@xmath140 , a.a . minaenko@xmath142 , m. miano@xmath145 , i.a . minashvili@xmath129 , a.i . mincer@xmath233 , b. mindur@xmath257 , m. mineev@xmath129 , y. ming@xmath212 , l.m . mir@xmath95 , g. mirabelli@xmath156 , l. miralles verge@xmath95 , a. misiejuk@xmath126 , j. mitrevski@xmath186 , g.y . mitrofanov@xmath142 , v.a . mitsou@xmath145 , s. mitsui@xmath141 , p.s . miyagawa@xmath138 , k. miyazaki@xmath124 , j.u . mjrnmark@xmath121 , t. moa@xmath170 , p. mockett@xmath275 , s. moed@xmath193 , v. moeller@xmath169 , k. mnig@xmath201 , n. mser@xmath133 , s. mohapatra@xmath117 , b. mohn@xmath231 , w. mohr@xmath93 , s. mohrdieck - mck@xmath108 , a.m. moisseev@xmath142@xmath164 , r. moles - valls@xmath145 , j. molina - perez@xmath128 , l. moneta@xmath96 , j. monk@xmath184 , e. monnier@xmath157 , s. montesano@xmath103 , f. monticelli@xmath150 , s. monzani@xmath153 , r.w . moore@xmath181 , g.f . moorhead@xmath188 , c. mora herrera@xmath96 , a. moraes@xmath137 , a. morais@xmath143@xmath144 , n. morange@xmath180 , g. morello@xmath242 , d. moreno@xmath115 , m. moreno llcer@xmath145 , p. morettini@xmath200 , m. morii@xmath193 , j. morin@xmath110 , y. morita@xmath141 , a.k . morley@xmath128 , g. mornacchi@xmath128 , m - c . morone@xmath96 , s.v . morozov@xmath154 , j.d . morris@xmath110 , h.g . moser@xmath108 , m. mosidze@xmath255 , j. moss@xmath273 , r. mount@xmath159 , e. mountricha@xmath132 , s.v . mouraviev@xmath123 , e.j.w . moyse@xmath227 , m. mudrinic@xmath224 , f. mueller@xmath149 , j. mueller@xmath223 , k. mueller@xmath133 , t.a . mller@xmath109 , d. muenstermann@xmath235 , a. muijs@xmath213 , a. muir@xmath175 , y. munwes@xmath101 , k. murakami@xmath141 , w.j . murray@xmath111 , i. mussche@xmath213 , e. musto@xmath139 , a.g . myagkov@xmath142 , m. myska@xmath256 , j. nadal@xmath95 , k. nagai@xmath289 , k. nagano@xmath141 , y. nagasaka@xmath308 , a.m. nairz@xmath128 , y. nakahama@xmath102 , k. nakamura@xmath122 , i. nakano@xmath309 , g. nanava@xmath133 , a. napier@xmath286 , m. nash@xmath184@xmath307 , n.r . nation@xmath116 , t. nattermann@xmath133 , t. naumann@xmath201 , g. navarro@xmath306 , h.a . neal@xmath165 , e. nebot@xmath192 , p.yu . nechaeva@xmath123 , a. negri@xmath240 , g. negri@xmath128 , s. nektarijevic@xmath96 , a. nelson@xmath251 , s. nelson@xmath159 , t.k . nelson@xmath159 , s. nemecek@xmath256 , p. nemethy@xmath233 , a.a . nepomuceno@xmath239 , m. nessi@xmath128@xmath310 , s.y . nesterov@xmath271 , m.s . neubauer@xmath211 , a. neusiedl@xmath115 , r.m . neves@xmath233 , p. nevski@xmath105 , p.r . newman@xmath187 , r.b . nickerson@xmath97 , r. nicolaidou@xmath180 , l. nicolas@xmath146 , b. nicquevert@xmath128 , f. niedercorn@xmath102 , j. nielsen@xmath186 , t. niinikoski@xmath128 , a. nikiforov@xmath134 , v. nikolaenko@xmath142 , k. nikolaev@xmath129 , i. nikolic - audit@xmath199 , k. nikolopoulos@xmath105 , h. nilsen@xmath93 , p. nilsson@xmath225 , y. ninomiya @xmath122 , a. nisati@xmath156 , t. nishiyama@xmath124 , r. nisius@xmath108 , l. nodulman@xmath171 , m. nomachi@xmath288 , i. nomidis@xmath294 , h. nomoto@xmath122 , m. nordberg@xmath128 , b. nordkvist@xmath170 , p.r . norton@xmath111 , j. novakova@xmath262 , m. nozaki@xmath141 , m. noika@xmath201 , l. nozka@xmath285 , i.m . nugent@xmath241 , a .- e . nuncio - quiroz@xmath133 , g. nunes hanninger@xmath133 , t. nunnemann@xmath109 , e. nurse@xmath184 , t. nyman@xmath128 , b.j . obrien@xmath215 , s.w . oneale@xmath187@xmath164 , d.c . oneil@xmath263 , v. oshea@xmath137 , f.g . oakham@xmath161@xmath178 , h. oberlack@xmath108 , j. ocariz@xmath199 , a. ochi@xmath124 , s. oda@xmath122 , s. odaka@xmath141 , j. odier@xmath157 , h. ogren@xmath230 , a. oh@xmath138 , s.h . oh@xmath160 , c.c . ohm@xmath170 , t. ohshima@xmath290 , h. ohshita@xmath291 , t.k . ohska@xmath141 , t. ohsugi@xmath311 , s. okada@xmath124 , h. okawa@xmath210 , y. okumura@xmath290 , t. okuyama@xmath122 , m. olcese@xmath200 , a.g . olchevski@xmath129 , m. oliveira@xmath143@xmath245 , d. oliveira damazio@xmath105 , e. oliver garcia@xmath145 , d. olivito@xmath135 , a. olszewski@xmath185 , j. olszowska@xmath185 , c. omachi@xmath124 , a. onofre@xmath143@xmath312 , p.u.e . onyisi@xmath148 , c.j . oram@xmath241 , g. ordonez@xmath250 , m.j . oreglia@xmath148 , f. orellana@xmath96 , y. oren@xmath101 , d. orestano@xmath179 , i. orlov@xmath208 , c. oropeza barrera@xmath137 , r.s . orr@xmath183 , e.o . ortega@xmath212 , b. osculati@xmath189 , r. ospanov@xmath135 , c. osuna@xmath95 , g. otero y garzon@xmath279 , j.p ottersbach@xmath213 , m. ouchrif@xmath265 , f. ould - saada@xmath234 , a. ouraou@xmath180 , q. ouyang@xmath182 , m. owen@xmath138 , s. owen@xmath146 , a. oyarzun@xmath229 , o.k . ye@xmath231 , v.e . ozcan@xmath120 , n. ozturk@xmath225 , a. pacheco pages@xmath95 , c. padilla aranda@xmath95 , e. paganis@xmath146 , f. paige@xmath105 , k. pajchel@xmath234 , s. palestini@xmath128 , d. pallin@xmath236 , a. palma@xmath143@xmath144 , j.d . palmer@xmath187 , y.b . pan@xmath168 , e. panagiotopoulou@xmath132 , b. panes@xmath267 , n. panikashvili@xmath165 , s. panitkin@xmath105 , d. pantea@xmath131 , m. panuskova@xmath256 , v. paolone@xmath223 , a. paoloni@xmath119 , a. papadelis@xmath220 , th.d . papadopoulou@xmath132 , a. paramonov@xmath171 , w. park@xmath105@xmath313 , m.a . parker@xmath169 , f. parodi@xmath189 , j.a . parsons@xmath147 , u. parzefall@xmath93 , e. pasqualucci@xmath156 , a. passeri@xmath191 , f. pastore@xmath179 , fr . pastore@xmath128 , g. psztor @xmath96@xmath314 , s. pataraia@xmath168 , n. patel@xmath300 , j.r . pater@xmath138 , s. patricelli@xmath139 , t. pauly@xmath128 , m. pecsy@xmath196 , m.i . pedraza morales@xmath168 , s.v . peleganchuk@xmath208 , h. peng@xmath168 , r. pengo@xmath128 , a. penson@xmath147 , j. penwell@xmath230 , m. perantoni@xmath239 , k. perez@xmath147@xmath305 , t. perez cavalcanti@xmath201 , e. perez codina@xmath95 , m.t . prez garca - esta@xmath145 , v. perez reale@xmath147 , i. peric@xmath133 , l. perini@xmath103 , h. pernegger@xmath128 , r. perrino@xmath246 , p. perrodo@xmath158 , s. persembe@xmath237 , v.d . peshekhonov@xmath129 , o. peters@xmath213 , b.a . petersen@xmath128 , j. petersen@xmath128 , t.c . petersen@xmath218 , e. petit@xmath157 , a. petridis@xmath294 , c. petridou@xmath294 , e. petrolo@xmath156 , f. petrucci@xmath179 , d. petschull@xmath201 , m. petteni@xmath263 , r. pezoa@xmath229 , a. phan@xmath188 , a.w . phillips@xmath169 , p.w . phillips@xmath111 , g. piacquadio@xmath128 , e. piccaro@xmath110 , m. piccinini@xmath153 , a. pickford@xmath137 , s.m . piec@xmath201 , r. piegaia@xmath279 , j.e . pilcher@xmath148 , a.d . pilkington@xmath138 , j. pina@xmath143@xmath144 , m. pinamonti@xmath247 , a. pinder@xmath97 , j.l . pinfold@xmath181 , j. ping@xmath252 , b. pinto@xmath143@xmath144 , o. pirotte@xmath128 , c. pizio@xmath103 , r. placakyte@xmath201 , m. plamondon@xmath172 , w.g . plano@xmath138 , m .- a . pleier@xmath105 , a.v . pleskach@xmath142 , a. poblaguev@xmath105 , s. poddar@xmath149 , f. podlyski@xmath236 , l. poggioli@xmath102 , t. poghosyan@xmath133 , m. pohl@xmath96 , f. polci@xmath127 , g. polesello@xmath207 , a. policicchio@xmath275 , a. polini@xmath206 , j. poll@xmath110 , v. polychronakos@xmath105 , d.m . pomarede@xmath180 , d. pomeroy@xmath112 , k. pomms@xmath128 , l. pontecorvo@xmath156 , b.g . pope@xmath100 , g.a . popeneciu@xmath131 , d.s . popovic@xmath222 , a. poppleton@xmath128 , x. portell bueso@xmath93 , r. porter@xmath210 , c. posch@xmath116 , g.e . pospelov@xmath108 , s. pospisil@xmath174 , i.n . potrap@xmath108 , c.j . potter@xmath195 , c.t . potter@xmath228 , g. poulard@xmath128 , j. poveda@xmath168 , r. prabhu@xmath184 , p. pralavorio@xmath157 , s. prasad@xmath193 , r. pravahan@xmath225 , s. prell@xmath251 , k. pretzl@xmath197 , l. pribyl@xmath128 , d. price@xmath230 , l.e . price@xmath171 , m.j . price@xmath128 , p.m. prichard@xmath136 , d. prieur@xmath223 , m. primavera@xmath246 , k. prokofiev@xmath233 , f. prokoshin@xmath229 , s. protopopescu@xmath105 , j. proudfoot@xmath171 , x. prudent@xmath278 , h. przysiezniak@xmath158 , s. psoroulas@xmath133 , e. ptacek@xmath228 , j. purdham@xmath165 , m. purohit@xmath105@xmath313 , p. puzo@xmath102 , y. pylypchenko@xmath234 , j. qian@xmath165 , z. qian@xmath157 , z. qin@xmath201 , a. quadt@xmath176 , d.r . quarrie@xmath163 , w.b . quayle@xmath168 , f. quinonez@xmath267 , m. raas@xmath250 , v. radescu@xmath315 , b. radics@xmath133 , t. rador@xmath120 , f. ragusa@xmath103 , g. rahal@xmath217 , a.m. rahimi@xmath273 , d. rahm@xmath105 , s. rajagopalan@xmath105 , s. rajek@xmath235 , m. rammensee@xmath93 , m. rammes@xmath232 , m. ramstedt@xmath170 , k. randrianarivony@xmath161 , p.n . ratoff@xmath194 , f. rauscher@xmath109 , e. rauter@xmath108 , m. raymond@xmath128 , a.l . read@xmath234 , d.m . rebuzzi@xmath240 , a. redelbach@xmath274 , g. redlinger@xmath105 , r. reece@xmath135 , k. reeves@xmath118 , a. reichold@xmath213 , e. reinherz - aronis@xmath101 , a. reinsch@xmath228 , i. reisinger@xmath235 , d. reljic@xmath222 , c. rembser@xmath128 , z.l . ren@xmath292 , a. renaud@xmath102 , p. renkel@xmath264 , b. rensch@xmath218 , m. rescigno@xmath156 , s. resconi@xmath130 , b. resende@xmath180 , p. reznicek@xmath109 , r. rezvani@xmath183 , r. richter@xmath108 , e. richter - was@xmath185@xmath316 , m. ridel@xmath199 , s. rieke@xmath115 , m. rijpstra@xmath213 , m. rijssenbeek@xmath117 , a. rimoldi@xmath240 , l. rinaldi@xmath206 , r.r . rios@xmath264 , i. riu@xmath95 , g. rivoltella@xmath103 , f. rizatdinova@xmath99 , e. rizvi@xmath110 , s.h . robertson@xmath249@xmath260 , a. robichaud - veronneau@xmath96 , d. robinson@xmath169 , j.e.m . robinson@xmath184 , m. robinson@xmath228 , a. robson@xmath137 , j.g . rocha de lima@xmath238 , c. roda@xmath214 , d. roda dos santos@xmath128 , s. rodier@xmath192 , d. rodriguez@xmath306 , y. rodriguez garcia@xmath134 , a. roe@xmath176 , s. roe@xmath128 , o. rhne@xmath234 , v. rojo@xmath125 , s. rolli@xmath286 , a. romaniouk@xmath154 , v.m . romanov@xmath129 , g. romeo@xmath279 , d. romero maltrana@xmath267 , l. roos@xmath199 , e. ros@xmath145 , s. rosati@xmath275 , m. rose@xmath126 , g.a . rosenbaum@xmath183 , e.i . rosenberg@xmath251 , p.l . rosendahl@xmath231 , l. rosselet@xmath96 , v. rossetti@xmath95 , e. rossi@xmath139 , l.p . rossi@xmath200 , l. rossi@xmath103 , m. rotaru@xmath131 , i. roth@xmath140 , j. rothberg@xmath275 , i. rottlnder@xmath133 , d. rousseau@xmath102 , c.r . royon@xmath180 , a. rozanov@xmath157 , y. rozen@xmath203 , x. ruan@xmath102 , i. rubinskiy@xmath201 , b. ruckert@xmath109 , n. ruckstuhl@xmath213 , v.i . rud@xmath277 , g. rudolph@xmath270 , f. rhr@xmath254 , f. ruggieri@xmath179 , a. ruiz - martinez@xmath251 , e. rulikowska - zarebska@xmath257 , v. rumiantsev@xmath276@xmath164 , l. rumyantsev@xmath129 , k. runge@xmath93 , o. runolfsson@xmath133 , z. rurikova@xmath93 , n.a . rusakovich@xmath129 , d.r . rust@xmath230 , j.p . rutherfoord@xmath254 , c. ruwiedel@xmath163 , p. ruzicka@xmath256 , y.f . ryabov@xmath271 , v. ryadovikov@xmath142 , p. ryan@xmath100 , m. rybar@xmath262 , g. rybkin@xmath102 , n.c . ryder@xmath97 , s. rzaeva@xmath98 , a.f . saavedra@xmath300 , i. sadeh@xmath101 , h.f - w . sadrozinski@xmath186 , r. sadykov@xmath129 , f. safai tehrani@xmath167 , h. sakamoto@xmath122 , g. salamanna@xmath213 , a. salamon@xmath243 , m. saleem@xmath94 , d. salihagic@xmath108 , a. salnikov@xmath159 , j. salt@xmath145 , b.m . salvachua ferrando@xmath171 , d. salvatore@xmath242 , f. salvatore@xmath195 , a. salvucci@xmath132 , a. salzburger@xmath128 , d. sampsonidis@xmath294 , b.h . samset@xmath234 , h. sandaker@xmath231 , h.g . sander@xmath115 , m.p . sanders@xmath109 , m. sandhoff@xmath190 , p. sandhu@xmath183 , t. sandoval@xmath169 , r. sandstroem@xmath213 , s. sandvoss@xmath190 , d.p.c . sankey@xmath111 , a. sansoni@xmath151 , c. santamarina rios@xmath249 , c. santoni@xmath236 , r. santonico@xmath119 , h. santos@xmath143 , j.g . saraiva@xmath143@xmath144 , t. sarangi@xmath168 , e. sarkisyan - grinbaum@xmath225 , f. sarri@xmath214 , g. sartisohn@xmath190 , o. sasaki@xmath141 , t. sasaki@xmath141 , n. sasao@xmath317 , i. satsounkevitch@xmath219 , g. sauvage@xmath158 , j.b . sauvan@xmath102 , p. savard@xmath183@xmath178 , v. savinov@xmath223 , d.o . savu@xmath128 , p. savva @xmath132 , l. sawyer@xmath105@xmath266 , d.h . saxon@xmath137 , l.p . says@xmath236 , c. sbarra@xmath153 , a. sbrizzi@xmath153 , o. scallon@xmath177 , d.a . scannicchio@xmath210 , j. schaarschmidt@xmath102 , p. schacht@xmath108 , u. schfer@xmath115 , s. schaepe@xmath133 , s. schaetzel@xmath315 , a.c . schaffer@xmath102 , d. schaile@xmath109 , r.d . schamberger@xmath117 , a.g . schamov@xmath208 , v. scharf@xmath149 , v.a . schegelsky@xmath271 , d. scheirich@xmath165 , m.i . scherzer@xmath163 , c. schiavi@xmath189 , j. schieck@xmath109 , m. schioppa@xmath242 , s. schlenker@xmath128 , j.l . schlereth@xmath171 , e. schmidt@xmath93 , m.p . schmidt@xmath107@xmath164 , k. schmieden@xmath133 , c. schmitt@xmath115 , m. schmitz@xmath133 , a. schning@xmath315 , m. schott@xmath128 , d. schouten@xmath263 , j. schovancova@xmath256 , m. schram@xmath249 , c. schroeder@xmath115 , n. schroer@xmath299 , s. schuh@xmath128 , g. schuler@xmath128 , j. schultes@xmath190 , h .- c . schultz - coulon@xmath149 , h. schulz@xmath134 , j.w . schumacher@xmath133 , m. schumacher@xmath93 , b.a . schumm@xmath186 , ph . schune@xmath180 , c. schwanenberger@xmath138 , a. schwartzman@xmath159 , ph . schwemling@xmath199 , r. schwienhorst@xmath100 , r. schwierz@xmath278 , j. schwindling@xmath180 , w.g . scott@xmath111 , j. searcy@xmath228 , e. sedykh@xmath271 , e. segura@xmath95 , s.c . seidel@xmath280 , a. seiden@xmath186 , f. seifert@xmath278 , j.m . seixas@xmath239 , g. sekhniaidze@xmath244 , d.m . seliverstov@xmath271 , b. sellden@xmath220 , g. sellers@xmath136 , m. seman@xmath155 , n. semprini - cesari@xmath153 , c. serfon@xmath109 , l. serin@xmath102 , r. seuster@xmath108 , h. severini@xmath94 , m.e . sevior@xmath188 , a. sfyrla@xmath128 , e. shabalina@xmath176 , m. shamim@xmath228 , l.y . shan@xmath182 , j.t . shank@xmath116 , q.t . shao@xmath188 , m. shapiro@xmath163 , p.b . shatalov@xmath166 , l. shaver@xmath254 , c. shaw@xmath137 , k. shaw@xmath247 , d. sherman@xmath107 , p. sherwood@xmath184 , a. shibata@xmath233 , s. shimizu@xmath128 , m. shimojima@xmath318 , t. shin@xmath106 , a. shmeleva@xmath123 , m.j . shochet@xmath148 , d. short@xmath97 , m.a . shupe@xmath254 , p. sicho@xmath256 , a. sidoti@xmath167 , a. siebel@xmath190 , f. siegert@xmath93 , j. siegrist@xmath163 , dj . sijacki@xmath222 , o. silbert@xmath140 , j. silva@xmath143@xmath144 , y. silver@xmath101 , d. silverstein@xmath159 , s.b . silverstein@xmath220 , v. simak@xmath174 , o. simard@xmath180 , lj . simic@xmath222 , s. simion@xmath102 , b. simmons@xmath184 , m. simonyan@xmath218 , p. sinervo@xmath183 , n.b . sinev@xmath228 , v. sipica@xmath232 , g. siragusa@xmath115 , a.n . sisakyan@xmath129 , s.yu . sivoklokov@xmath277 , j. sjlin@xmath170 , t.b . sjursen@xmath231 , l.a . skinnari@xmath163 , k. skovpen@xmath208 , p. skubic@xmath94 , n. skvorodnev@xmath112 , m. slater@xmath187 , t. slavicek@xmath174 , k. sliwa@xmath286 , t.j . sloan@xmath194 , j. sloper@xmath128 , v. smakhtin@xmath140 , s.yu . smirnov@xmath154 , l.n . smirnova@xmath277 , o. smirnova@xmath121 , b.c . smith@xmath193 , d. smith@xmath159 , k.m . smith@xmath137 , m. smizanska@xmath194 , k. smolek@xmath174 , a.a . snesarev@xmath123 , s.w . snow@xmath138 , j. snow@xmath94 , j. snuverink@xmath213 , s. snyder@xmath105 , m. soares@xmath143 , r. sobie@xmath172@xmath260 , j. sodomka@xmath174 , a. soffer@xmath101 , c.a . solans@xmath145 , m. solar@xmath174 , j. solc@xmath174 , e. soldatov@xmath154 , u. soldevila@xmath145 , e. solfaroli camillocci@xmath167 , a.a . solodkov@xmath142 , o.v . solovyanov@xmath142 , j. sondericker@xmath105 , n. soni@xmath181 , v. sopko@xmath174 , b. sopko@xmath174 , m. sorbi@xmath103 , m. sosebee@xmath225 , a. soukharev@xmath208 , s. spagnolo@xmath216 , f. span@xmath147 , r. spighi@xmath206 , g. spigo@xmath128 , f. spila@xmath167 , e. spiriti@xmath191 , r. spiwoks@xmath128 , m. spousta@xmath262 , t. spreitzer@xmath183 , b. spurlock@xmath225 , r.d . denis@xmath137 , t. stahl@xmath232 , j. stahlman@xmath135 , r. stamen@xmath149 , e. stanecka@xmath128 , r.w . stanek@xmath171 , c. stanescu@xmath191 , s. stapnes@xmath234 , e.a . starchenko@xmath142 , j. stark@xmath127 , p. staroba@xmath256 , p. starovoitov@xmath276 , a. staude@xmath109 , p. stavina@xmath196 , g. stavropoulos@xmath163 , g. steele@xmath137 , p. steinbach@xmath278 , p. steinberg@xmath105 , i. stekl@xmath174 , b. stelzer@xmath263 , h.j . stelzer@xmath201 , o. stelzer - chilton@xmath241 , h. stenzel@xmath173 , k. stevenson@xmath110 , g.a . stewart@xmath137 , j.a . stillings@xmath133 , t. stockmanns@xmath133 , m.c . stockton@xmath128 , k. stoerig@xmath93 , g. stoicea@xmath131 , s. stonjek@xmath108 , p. strachota@xmath262 , a.r . stradling@xmath225 , a. straessner@xmath278 , j. strandberg@xmath165 , s. strandberg@xmath170 , a. strandlie@xmath234 , m. strang@xmath273 , e. strauss@xmath159 , m. strauss@xmath94 , p. strizenec@xmath155 , r. strhmer@xmath274 , d.m . strom@xmath228 , j.a . strong@xmath126@xmath164 , r. stroynowski@xmath264 , j. strube@xmath111 , b. stugu@xmath231 , i. stumer@xmath105@xmath164 , j. stupak@xmath117 , p. sturm@xmath190 , d.a . soh@xmath292@xmath302 , d. su@xmath159 , hs . subramania@xmath181 , y. sugaya@xmath288 , t. sugimoto@xmath290 , c. suhr@xmath238 , k. suita@xmath124 , m. suk@xmath262 , v.v . sulin@xmath123 , s. sultansoy@xmath319 , t. sumida@xmath128 , x. sun@xmath127 , j.e . sundermann@xmath93 , k. suruliz@xmath104 , s. sushkov@xmath95 , g. susinno@xmath242 , m.r . sutton@xmath146 , y. suzuki@xmath141 , yu.m . sviridov@xmath142 , s. swedish@xmath175 , i. sykora@xmath196 , t. sykora@xmath262 , b. szeless@xmath128 , j. snchez@xmath145 , d. ta@xmath213 , k. tackmann@xmath128 , a. taffard@xmath210 , r. tafirout@xmath241 , a. taga@xmath234 , n. taiblum@xmath101 , y. takahashi@xmath290 , h. takai@xmath105 , r. takashima@xmath320 , h. takeda@xmath124 , t. takeshita@xmath291 , m. talby@xmath157 , a. talyshev@xmath208 , m.c . tamsett@xmath105 , j. tanaka@xmath122 , r. tanaka@xmath102 , s. tanaka@xmath321 , s. tanaka@xmath141 , y. tanaka@xmath318 , k. tani@xmath124 , n. tannoury@xmath157 , g.p . tappern@xmath128 , s. tapprogge@xmath115 , d. tardif@xmath183 , s. tarem@xmath203 , f. tarrade@xmath105 , g.f . tartarelli@xmath130 , p. tas@xmath262 , m. tasevsky@xmath256 , e. tassi@xmath242 , m. tatarkhanov@xmath163 , c. taylor@xmath184 , f.e . taylor@xmath322 , g.n . taylor@xmath188 , w. taylor@xmath323 , m. teixeira dias castanheira@xmath110 , p. teixeira - dias@xmath126 , k.k . temming@xmath93 , h. ten kate@xmath128 , p.k . teng@xmath292 , s. terada@xmath141 , k. terashi@xmath122 , j. terron@xmath192 , m. terwort@xmath201@xmath297 , m. testa@xmath151 , r.j . teuscher@xmath183@xmath260 , c.m . tevlin@xmath138 , j. thadome@xmath190 , j. therhaag@xmath133 , t. theveneaux - pelzer@xmath199 , m. thioye@xmath107 , s. thoma@xmath93 , j.p . thomas@xmath187 , e.n . thompson@xmath227 , p.d . thompson@xmath187 , p.d . thompson@xmath183 , a.s . thompson@xmath137 , e. thomson@xmath135 , m. thomson@xmath169 , r.p . thun@xmath165 , t. tic@xmath256 , v.o . tikhomirov@xmath123 , y.a . tikhonov@xmath208 , c.j.w.p . timmermans@xmath250 , p. tipton@xmath107 , f.j . tique aires viegas@xmath128 , s. tisserant@xmath157 , j. tobias@xmath93 , b. toczek@xmath257 , t. todorov@xmath158 , s. todorova - nova@xmath286 , b. toggerson@xmath210 , j. tojo@xmath141 , s. tokr@xmath196 , k. tokunaga@xmath124 , k. tokushuku@xmath141 , k. tollefson@xmath100 , m. tomoto@xmath290 , l. tompkins@xmath163 , k. toms@xmath280 , a. tonazzo@xmath179 , g. tong@xmath182 , a. tonoyan@xmath231 , c. topfel@xmath197 , n.d . topilin@xmath129 , i. torchiani@xmath128 , e. torrence@xmath228 , e. torr pastor@xmath145 , j. toth@xmath157@xmath314 , f. touchard@xmath157 , d.r . tovey@xmath146 , d. traynor@xmath110 , t. trefzger@xmath274 , j. treis@xmath133 , l. tremblet@xmath128 , a. tricoli@xmath128 , i.m . trigger@xmath241 , s. trincaz - duvoid@xmath199 , t.n . trinh@xmath199 , m.f . tripiana@xmath150 , n. triplett@xmath251 , w. trischuk@xmath183 , a. trivedi@xmath105@xmath313 , b. trocm@xmath127 , c. troncon@xmath130 , m. trottier - mcdonald@xmath263 , a. trzupek@xmath185 , c. tsarouchas@xmath128 , j.c - l . tseng@xmath97 , m. tsiakiris@xmath213 , p.v . tsiareshka@xmath219 , d. tsionou@xmath158 , g. tsipolitis@xmath132 , v. tsiskaridze@xmath93 , e.g. tskhadadze@xmath255 , i.i . tsukerman@xmath166 , v. tsulaia@xmath223 , j .- w . tsung@xmath133 , s. tsuno@xmath141 , d. tsybychev@xmath117 , a. tua@xmath146 , j.m . tuggle@xmath148 , m. turala@xmath185 , d. turecek@xmath174 , i. turk cakir@xmath324 , e. turlay@xmath213 , r. turra@xmath103 , p.m. tuts@xmath147 , a. tykhonov@xmath258 , m. tylmad@xmath170 , m. tyndel@xmath111 , d. typaldos@xmath187 , h. tyrvainen@xmath128 , g. tzanakos@xmath152 , k. uchida@xmath133 , i. ueda@xmath122 , r. ueno@xmath161 , m. ugland@xmath231 , m. uhlenbrock@xmath133 , m. uhrmacher@xmath176 , f. ukegawa@xmath289 , g. unal@xmath128 , d.g . underwood@xmath171 , a. undrus@xmath105 , g. unel@xmath210 , y. unno@xmath141 , d. urbaniec@xmath147 , e. urkovsky@xmath101 , p. urrejola@xmath267 , g. usai@xmath225 , m. uslenghi@xmath240 , l. vacavant@xmath157 , v. vacek@xmath174 , b. vachon@xmath249 , s. vahsen@xmath163 , c. valderanis@xmath108 , j. valenta@xmath256 , p. valente@xmath156 , s. valentinetti@xmath153 , s. valkar@xmath262 , e. valladolid gallego@xmath145 , s. vallecorsa@xmath203 , j.a . valls ferrer@xmath145 , h. van der graaf@xmath213 , e. van der kraaij@xmath213 , r. van der leeuw@xmath213 , e. van der poel@xmath213 , d. van der ster@xmath128 , b. van eijk@xmath213 , n. van eldik@xmath227 , p. van gemmeren@xmath171 , z. van kesteren@xmath213 , i. van vulpen@xmath213 , w. vandelli@xmath128 , g. vandoni@xmath128 , a. vaniachine@xmath171 , p. vankov@xmath201 , f. vannucci@xmath199 , f. varela rodriguez@xmath128 , r. vari@xmath156 , e.w . varnes@xmath254 , d. varouchas@xmath163 , a. vartapetian@xmath225 , k.e . varvell@xmath300 , v.i . vassilakopoulos@xmath106 , f. vazeille@xmath236 , g. vegni@xmath103 , j.j . veillet@xmath102 , c. vellidis@xmath152 , f. veloso@xmath143 , r. veness@xmath128 , s. veneziano@xmath156 , a. ventura@xmath216 , d. ventura@xmath275 , m. venturi@xmath93 , n. venturi@xmath197 , v. vercesi@xmath207 , m. verducci@xmath275 , w. verkerke@xmath213 , j.c . vermeulen@xmath213 , a. vest@xmath278 , m.c . vetterli@xmath263@xmath178 , i. vichou@xmath211 , t. vickey@xmath301@xmath325 , g.h.a . viehhauser@xmath97 , s. viel@xmath175 , m. villa@xmath153 , m. villaplana perez@xmath145 , e. vilucchi@xmath151 , m.g . vincter@xmath161 , e. vinek@xmath128 , v.b . vinogradov@xmath129 , m. virchaux@xmath180@xmath164 , s. viret@xmath236 , j. virzi@xmath163 , a. vitale @xmath153 , o. vitells@xmath140 , m. viti@xmath201 , i. vivarelli@xmath93 , f. vives vaque@xmath95 , s. vlachos@xmath132 , m. vlasak@xmath174 , n. vlasov@xmath133 , a. vogel@xmath133 , p. vokac@xmath174 , g. volpi@xmath151 , m. volpi@xmath95 , g. volpini@xmath130 , h. von der schmitt@xmath108 , j. von loeben@xmath108 , h. von radziewski@xmath93 , e. von toerne@xmath133 , v. vorobel@xmath262 , a.p . vorobiev@xmath142 , v. vorwerk@xmath95 , m. vos@xmath145 , r. voss@xmath128 , t.t . voss@xmath190 , j.h . vossebeld@xmath136 , a.s . vovenko@xmath142 , n. vranjes@xmath222 , m. vranjes milosavljevic@xmath222 , v. vrba@xmath256 , m. vreeswijk@xmath213 , t. vu anh@xmath115 , r. vuillermet@xmath128 , i. vukotic@xmath102 , w. wagner@xmath190 , p. wagner@xmath135 , h. wahlen@xmath190 , j. wakabayashi@xmath290 , j. walbersloh@xmath235 , s. walch@xmath165 , j. walder@xmath194 , r. walker@xmath109 , w. walkowiak@xmath232 , r. wall@xmath107 , p. waller@xmath136 , c. wang@xmath160 , h. wang@xmath168 , j. wang@xmath292 , j. wang@xmath272 , j.c . wang@xmath275 , r. wang@xmath280 , s.m . wang@xmath292 , a. warburton@xmath249 , c.p . ward@xmath169 , m. warsinsky@xmath93 , p.m. watkins@xmath187 , a.t . watson@xmath187 , m.f . watson@xmath187 , g. watts@xmath275 , s. watts@xmath138 , a.t . waugh@xmath300 , b.m . waugh@xmath184 , j. weber@xmath235 , m. weber@xmath111 , m.s . weber@xmath197 , p. weber@xmath176 , a.r . weidberg@xmath97 , p. weigell@xmath108 , j. weingarten@xmath176 , c. weiser@xmath93 , h. wellenstein@xmath112 , p.s . wells@xmath128 , m. wen@xmath151 , t. wenaus@xmath105 , s. wendler@xmath223 , z. weng@xmath292@xmath302 , t. wengler@xmath128 , s. wenig@xmath128 , n. wermes@xmath133 , m. werner@xmath93 , p. werner@xmath128 , m. werth@xmath210 , m. wessels@xmath149 , k. whalen@xmath161 , s.j . wheeler - ellis@xmath210 , s.p . whitaker@xmath116 , a. white@xmath225 , m.j . white@xmath188 , s. white@xmath105 , s.r . whitehead@xmath97 , d. whiteson@xmath210 , d. whittington@xmath230 , f. wicek@xmath102 , d. wicke@xmath190 , f.j . wickens@xmath111 , w. wiedenmann@xmath168 , m. wielers@xmath111 , p. wienemann@xmath133 , c. wiglesworth@xmath136 , l.a.m . wiik@xmath93 , p.a . wijeratne@xmath184 , a. wildauer@xmath145 , m.a . wildt@xmath201@xmath297 , i. wilhelm@xmath262 , h.g . wilkens@xmath128 , j.z . will@xmath109 , e. williams@xmath147 , h.h . williams@xmath135 , w. willis@xmath147 , s. willocq@xmath227 , j.a . wilson@xmath187 , m.g . wilson@xmath159 , a. wilson@xmath165 , i. wingerter - seez@xmath158 , s. winkelmann@xmath93 , f. winklmeier@xmath128 , m. wittgen@xmath159 , m.w . wolter@xmath185 , h. wolters@xmath143@xmath245 , g. wooden@xmath97 , b.k . wosiek@xmath185 , j. wotschack@xmath128 , m.j . woudstra@xmath227 , k. wraight@xmath137 , c. wright@xmath137 , b. wrona@xmath136 , s.l . wu@xmath168 , x. wu@xmath96 , y. wu@xmath287 , e. wulf@xmath147 , r. wunstorf@xmath235 , b.m . wynne@xmath215 , l. xaplanteris@xmath132 , s. xella@xmath218 , s. xie@xmath93 , y. xie@xmath182 , c. xu@xmath287 , d. xu@xmath146 , g. xu@xmath182 , b. yabsley@xmath300 , m. yamada@xmath141 , a. yamamoto@xmath141 , k. yamamoto@xmath251 , s. yamamoto@xmath122 , t. yamamura@xmath122 , j. yamaoka@xmath160 , t. yamazaki@xmath122 , y. yamazaki@xmath124 , z. yan@xmath116 , h. yang@xmath165 , u.k . yang@xmath138 , y. yang@xmath230 , y. yang@xmath182 , z. yang@xmath170 , s. yanush@xmath276 , w - m . yao@xmath163 , y. yao@xmath163 , y. yasu@xmath141 , g.v . ybeles smit@xmath212 , j. ye@xmath264 , s. ye@xmath105 , m. yilmaz@xmath326 , r. yoosoofmiya@xmath223 , k. yorita@xmath296 , r. yoshida@xmath171 , c. young@xmath159 , s. youssef@xmath116 , d. yu@xmath105 , j. yu@xmath225 , j. yu@xmath252@xmath327 , l. yuan@xmath182@xmath328 , a. yurkewicz@xmath117 , v.g . zaets @xmath142 , r. zaidan@xmath204 , a.m. zaitsev@xmath142 , z. zajacova@xmath128 , yo.k . zalite @xmath271 , l. zanello@xmath167 , p. zarzhitsky@xmath264 , a. zaytsev@xmath208 , c. zeitnitz@xmath190 , m. zeller@xmath107 , p.f . zema@xmath128 , a. zemla@xmath185 , c. zendler@xmath133 , a.v . zenin@xmath142 , o. zenin@xmath142 , t. eni@xmath196 , z. zenonos@xmath214 , s. zenz@xmath163 , d. zerwas@xmath102 , g. zevi della porta@xmath193 , z. zhan@xmath272 , d. zhang@xmath287 , h. zhang@xmath100 , j. zhang@xmath171 , x. zhang@xmath272 , z. zhang@xmath102 , l. zhao@xmath233 , t. zhao@xmath275 , z. zhao@xmath287 , a. zhemchugov@xmath129 , s. zheng@xmath182 , j. zhong@xmath292@xmath329 , b. zhou@xmath165 , n. zhou@xmath210 , y. zhou@xmath292 , c.g . zhu@xmath272 , h. zhu@xmath201 , y. zhu@xmath168 , x. zhuang@xmath109 , v. zhuravlov@xmath108 , d. zieminska@xmath230 , b. zilka@xmath196 , r. zimmermann@xmath133 , s. zimmermann@xmath133 , s. zimmermann@xmath93 , m. ziolkowski@xmath232 , r. zitoun@xmath158 , l. ivkovi@xmath147 , v.v . zmouchko@xmath142@xmath164 , g. zobernig@xmath168 , a. zoccoli@xmath153 , y. zolnierowski@xmath158 , a. zsenei@xmath128 , m. zur nedden@xmath134 , v. zutshi@xmath238 , l. zwalinski@xmath128 . @xmath330 university at albany , albany ny , united states of america + @xmath331 department of physics , university of alberta , edmonton ab , canada + @xmath332 @xmath333department of physics , ankara university , ankara ; @xmath334department of physics , dumlupinar university , kutahya ; @xmath335department of physics , gazi university , ankara ; @xmath336division of physics , tobb university of economics and technology , ankara ; @xmath337turkish atomic energy authority , ankara , turkey + @xmath338 lapp , cnrs / in2p3 and universit de savoie , annecy - le - vieux , france + @xmath339 high energy physics division , argonne national laboratory , argonne il , united states of america + @xmath340 department of physics , university of arizona , tucson az , united states of america + @xmath341 department of physics , the university of texas at arlington , arlington tx , united states of america + @xmath342 physics department , university of athens , athens , greece + @xmath343 physics department , national technical university of athens , zografou , greece + @xmath344 institute of physics , azerbaijan academy of sciences , baku , azerbaijan + @xmath345 institut de fsica daltes energies and universitat autnoma de barcelona and icrea , barcelona , spain + @xmath346 @xmath333institute of physics , university of belgrade , belgrade ; @xmath334vinca institute of nuclear sciences , belgrade , serbia + @xmath347 department for physics and technology , university of bergen , bergen , norway + @xmath348 physics division , lawrence berkeley national laboratory and university of california , berkeley ca , united states of america + @xmath349 department of physics , humboldt university , berlin , germany + @xmath350 albert einstein center for fundamental physics and laboratory for high energy physics , university of bern , bern , switzerland + @xmath351 school of physics and astronomy , university of birmingham , birmingham , united kingdom + @xmath352 @xmath333department of physics , bogazici university , istanbul ; @xmath334division of physics , dogus university , istanbul ; @xmath335department of physics engineering , gaziantep university , gaziantep ; @xmath336department of physics , istanbul technical university , istanbul , turkey + @xmath353 @xmath333infn sezione di bologna ; @xmath334dipartimento di fisica , universit di bologna , bologna , italy + @xmath354 physikalisches institut , university of bonn , bonn , germany + @xmath355 department of physics , boston university , boston ma , united states of america + @xmath356 department of physics , brandeis university , waltham ma , united states of america + @xmath357 @xmath333universidade federal do rio de janeiro coppe / ee / if , rio de janeiro ; @xmath334instituto de fisica , universidade de sao paulo , sao paulo , brazil + @xmath358 physics department , brookhaven national laboratory , upton ny , united states of america + @xmath359 @xmath333national institute of physics and nuclear engineering , bucharest ; @xmath334university politehnica bucharest , bucharest ; @xmath335west university in timisoara , timisoara , romania + @xmath360 departamento de fsica , universidad de buenos aires , buenos aires , argentina + @xmath361 cavendish laboratory , university of cambridge , cambridge , united kingdom + @xmath362 department of physics , carleton university , ottawa on , canada + @xmath363 cern , geneva , switzerland + @xmath364 enrico fermi institute , university of chicago , chicago il , united states of america + @xmath365 @xmath333departamento de fisica , pontificia universidad catlica de chile , santiago ; @xmath334departamento de fsica , universidad tcnica federico santa mara , valparaso , chile + @xmath366 @xmath333institute of high energy physics , chinese academy of sciences , beijing ; @xmath334department of modern physics , university of science and technology of china , anhui ; @xmath335department of physics , nanjing university , jiangsu ; @xmath336high energy physics group , shandong university , shandong , china + @xmath367 laboratoire de physique corpusculaire , clermont universit and universit blaise pascal and cnrs / in2p3 , aubiere cedex , france + @xmath368 nevis laboratory , columbia university , irvington ny , united states of america + @xmath369 niels bohr institute , university of copenhagen , kobenhavn , denmark + @xmath370 @xmath333infn gruppo collegato di cosenza ; @xmath334dipartimento di fisica , universit della calabria , arcavata di rende , italy + @xmath371 faculty of physics and applied computer science , agh - university of science and technology , krakow , poland + @xmath372 the henryk niewodniczanski institute of nuclear physics , polish academy of sciences , krakow , poland + @xmath373 physics department , southern methodist university , dallas tx , united states of america + @xmath374 physics department , university of texas at dallas , richardson tx , united states of america + @xmath375 desy , hamburg and zeuthen , germany + @xmath376 institut fr experimentelle physik iv , technische universitt dortmund , dortmund , germany + @xmath377 institut fr kern- und teilchenphysik , technical university dresden , dresden , germany + @xmath378 department of physics , duke university , durham nc , united states of america + @xmath379 supa - school of physics and astronomy , university of edinburgh , edinburgh , united kingdom + @xmath380 fachhochschule wiener neustadt , wiener neustadt , austria + @xmath381 infn laboratori nazionali di frascati , frascati , italy + @xmath382 fakultt fr mathematik und physik , albert - ludwigs - universitt , freiburg i.br . , germany + @xmath383 section de physique , universit de genve , geneva , switzerland + @xmath384 @xmath333infn sezione di genova ; @xmath334dipartimento di fisica , universit di genova , genova , italy + @xmath385 institute of physics and hep institute , georgian academy of sciences and tbilisi state university , tbilisi , georgia + @xmath386 ii physikalisches institut , justus - liebig - universitt giessen , giessen , germany + @xmath387 supa - school of physics and astronomy , university of glasgow , glasgow , united kingdom + @xmath388 ii physikalisches georg - august - universitt , gttingen , germany + @xmath389 laboratoire de physique subatomique et de cosmologie , universit joseph fourier and cnrs / in2p3 and institut national polytechnique de grenoble , grenoble , france + @xmath390 department of physics , hampton university , hampton va , united states of america + @xmath391 laboratory for particle physics and cosmology , harvard university , cambridge ma , united states of america + @xmath392 @xmath333kirchhoff - institut fr physik , ruprecht - karls - universitt heidelberg , heidelberg ; @xmath334physikalisches institut , ruprecht - karls - universitt heidelberg , heidelberg ; @xmath335ziti institut fr technische informatik , ruprecht - karls - universitt heidelberg , mannheim , germany + @xmath393 faculty of science , hiroshima university , hiroshima , japan + @xmath394 faculty of applied information science , hiroshima institute of technology , hiroshima , japan + @xmath395 department of physics , indiana university , bloomington in , united states of america + @xmath396 institut fr astro- und teilchenphysik , leopold - franzens - universitt , innsbruck , austria + @xmath397 university of iowa , iowa city ia , united states of america + @xmath398 department of physics and astronomy , iowa state university , ames ia , united states of america + @xmath399 joint institute for nuclear research , jinr dubna , dubna , russia + @xmath400 kek , high energy accelerator research organization , tsukuba , japan + @xmath401 graduate school of science , kobe university , kobe , japan + @xmath402 faculty of science , kyoto university , kyoto , japan + @xmath403 kyoto university of education , kyoto , japan + @xmath404 instituto de fsica la plata , universidad nacional de la plata and conicet , la plata , argentina + @xmath405 physics department , lancaster university , lancaster , united kingdom + @xmath406 @xmath333infn sezione di lecce ; @xmath334dipartimento di fisica , universit del salento , lecce , italy + @xmath407 oliver lodge laboratory , university of liverpool , liverpool , united kingdom + @xmath408 department of physics , joef stefan institute and university of ljubljana , ljubljana , slovenia + @xmath409 department of physics , queen mary university of london , london , united kingdom + @xmath410 department of physics , royal holloway university of london , surrey , united kingdom + @xmath411 department of physics and astronomy , university college london , london , united kingdom + @xmath412 laboratoire de physique nuclaire et de hautes energies , upmc and universit paris - diderot and cnrs / in2p3 , paris , france + @xmath413 fysiska institutionen , lunds universitet , lund , sweden + @xmath414 departamento de fisica teorica c-15 , universidad autonoma de madrid , madrid , spain + @xmath415 institut fr physik , universitt mainz , mainz , germany + @xmath416 school of physics and astronomy , university of manchester , manchester , united kingdom + @xmath417 cppm , aix - marseille universit and cnrs / in2p3 , marseille , france + @xmath418 department of physics , university of massachusetts , amherst ma , united states of america + @xmath419 department of physics , mcgill university , montreal qc , canada + @xmath420 school of physics , university of melbourne , victoria , australia + @xmath421 department of physics , the university of michigan , ann arbor mi , united states of america + @xmath422 department of physics and astronomy , michigan state university , east lansing mi , united states of america + @xmath423 @xmath333infn sezione di milano ; @xmath334dipartimento di fisica , universit di milano , milano , italy + @xmath424 b.i . stepanov institute of physics , national academy of sciences of belarus , minsk , republic of belarus + @xmath425 national scientific and educational centre for particle and high energy physics , minsk , republic of belarus + @xmath426 department of physics , massachusetts institute of technology , cambridge ma , united states of america + @xmath427 group of particle physics , university of montreal , montreal qc , canada + @xmath428 p.n . lebedev institute of physics , academy of sciences , moscow , russia + @xmath429 institute for theoretical and experimental physics ( itep ) , moscow , russia + @xmath430 moscow engineering and physics institute ( mephi ) , moscow , russia + @xmath431 skobeltsyn institute of nuclear physics , lomonosov moscow state university , moscow , russia + @xmath432 fakultt fr physik , ludwig - maximilians - universitt mnchen , mnchen , germany + @xmath433 max - planck - institut fr physik ( werner - heisenberg - institut ) , mnchen , germany + @xmath434 nagasaki institute of applied science , nagasaki , japan + @xmath435 graduate school of science , nagoya university , nagoya , japan + @xmath436 @xmath333infn sezione di napoli ; @xmath334dipartimento di scienze fisiche , universit di napoli , napoli , italy + @xmath437 department of physics and astronomy , university of new mexico , albuquerque nm , united states of america + @xmath438 institute for mathematics , astrophysics and particle physics , radboud university nijmegen / nikhef , nijmegen , netherlands + @xmath439 nikhef national institute for subatomic physics and university of amsterdam , amsterdam , netherlands + @xmath440 department of physics , northern illinois university , dekalb il , united states of america + @xmath441 budker institute of nuclear physics ( binp ) , novosibirsk , russia + @xmath442 department of physics , new york university , new york ny , united states of america + @xmath443 ohio state university , columbus oh , united states of america + @xmath444 faculty of science , okayama university , okayama , japan + @xmath445 homer l. dodge department of physics and astronomy , university of oklahoma , norman ok , united states of america + @xmath446 department of physics , oklahoma state university , stillwater ok , united states of america + @xmath447 palack university , rcptm , olomouc , czech republic + @xmath448 center for high energy physics , university of oregon , eugene or , united states of america + @xmath449 lal , univ . paris - sud and cnrs / in2p3 , orsay , france + @xmath450 graduate school of science , osaka university , osaka , japan + @xmath451 department of physics , university of oslo , oslo , norway + @xmath452 department of physics , oxford university , oxford , united kingdom + @xmath453 @xmath333infn sezione di pavia ; @xmath334dipartimento di fisica nucleare e teorica , universit di pavia , pavia , italy + @xmath454 department of physics , university of pennsylvania , philadelphia pa , united states of america + @xmath455 petersburg nuclear physics institute , gatchina , russia + @xmath456 @xmath333infn sezione di pisa ; @xmath334dipartimento di fisica e. fermi , universit di pisa , pisa , italy + @xmath457 department of physics and astronomy , university of pittsburgh , pittsburgh pa , united states of america + @xmath458 @xmath333laboratorio de instrumentacao e fisica experimental de particulas - lip , lisboa , portugal ; @xmath334departamento de fisica teorica y del cosmos and cafpe , universidad de granada , granada , spain + @xmath459 institute of physics , academy of sciences of the czech republic , praha , czech republic + @xmath460 faculty of mathematics and physics , charles university in prague , praha , czech republic + @xmath461 czech technical university in prague , praha , czech republic + @xmath462 state research center institute for high energy physics , protvino , russia + @xmath463 particle physics department , rutherford appleton laboratory , didcot , united kingdom + @xmath464 physics department , university of regina , regina sk , canada + @xmath465 ritsumeikan university , kusatsu , shiga , japan + @xmath466 @xmath333infn sezione di roma i ; @xmath334dipartimento di fisica , universit la sapienza , roma , italy + @xmath467 @xmath333infn sezione di roma tor vergata ; @xmath334dipartimento di fisica , universit di roma tor vergata , roma , italy + @xmath468 @xmath333infn sezione di roma tre ; @xmath334dipartimento di fisica , universit roma tre , roma , italy + @xmath469 @xmath333facult des sciences ain chock , rseau universitaire de physique des hautes energies - universit hassan ii , casablanca ; @xmath334centre national de lenergie des sciences techniques nucleaires , rabat ; @xmath335universit cadi ayyad , facult des sciences semlalia dpartement de physique , b.p . 2390 marrakech 40000 ; @xmath336facult des sciences , universit mohamed premier and lptpm , oujda ; @xmath337facult des sciences , universit mohammed v , rabat , morocco + @xmath470 dsm / irfu ( institut de recherches sur les lois fondamentales de lunivers ) , cea saclay ( commissariat a lenergie atomique ) , gif - sur - yvette , france + @xmath471 santa cruz institute for particle physics , university of california santa cruz , santa cruz ca , united states of america + @xmath472 department of physics , university of washington , seattle wa , united states of america + @xmath473 department of physics and astronomy , university of sheffield , sheffield , united kingdom + @xmath474 department of physics , shinshu university , nagano , japan + @xmath475 fachbereich physik , universitt siegen , siegen , germany + @xmath476 department of physics , simon fraser university , burnaby bc , canada + @xmath477 slac national accelerator laboratory , stanford ca , united states of america + @xmath478 @xmath333faculty of mathematics , physics & informatics , comenius university , bratislava ; @xmath334department of subnuclear physics , institute of experimental physics of the slovak academy of sciences , kosice , slovak republic + @xmath479 @xmath333department of physics , university of johannesburg , johannesburg ; @xmath334school of physics , university of the witwatersrand , johannesburg , south africa + @xmath480 @xmath333department of physics , stockholm university ; @xmath334the oskar klein centre , stockholm , sweden + @xmath481 physics department , royal institute of technology , stockholm , sweden + @xmath482 department of physics and astronomy , stony brook university , stony brook ny , united states of america + @xmath483 department of physics and astronomy , university of sussex , brighton , united kingdom + @xmath484 school of physics , university of sydney , sydney , australia + @xmath485 institute of physics , academia sinica , taipei , taiwan + @xmath486 department of physics , technion : israel inst . of technology , haifa , israel + @xmath487 raymond and beverly sackler school of physics and astronomy , tel aviv university , tel aviv , israel + @xmath488 department of physics , aristotle university of thessaloniki , thessaloniki , greece + @xmath489 international center for elementary particle physics and department of physics , the university of tokyo , tokyo , japan + @xmath490 graduate school of science and technology , tokyo metropolitan university , tokyo , japan + @xmath491 department of physics , tokyo institute of technology , tokyo , japan + @xmath492 department of physics , university of toronto , toronto on , canada + @xmath493 @xmath333triumf , vancouver bc ; @xmath334department of physics and astronomy , york university , toronto on , canada + @xmath494 institute of pure and applied sciences , university of tsukuba , ibaraki , japan + @xmath495 science and technology center , tufts university , medford ma , united states of america + @xmath496 centro de investigaciones , universidad antonio narino , bogota , colombia + @xmath497 department of physics and astronomy , university of california irvine , irvine ca , united states of america + @xmath498 @xmath333infn gruppo collegato di udine ; @xmath334ictp , trieste ; @xmath335dipartimento di fisica , universit di udine , udine , italy + @xmath499 department of physics , university of illinois , urbana il , united states of america + @xmath500 department of physics and astronomy , university of uppsala , uppsala , sweden + @xmath501 instituto de fsica corpuscular ( ific ) and departamento de fsica atmica , molecular y nuclear and departamento de ingenier electrnica and instituto de microelectrnica de barcelona ( imb - cnm ) , university of valencia and csic , valencia , spain + @xmath502 department of physics , university of british columbia , vancouver bc , canada + @xmath503 department of physics and astronomy , university of victoria , victoria bc , canada + @xmath504 waseda university , tokyo , japan + @xmath505 department of particle physics , the weizmann institute of science , rehovot , israel + @xmath506 department of physics , university of wisconsin , madison wi , united states of america + @xmath507 fakultt fr physik und astronomie , julius - maximilians - universitt , wrzburg , germany + @xmath508 fachbereich c physik , bergische universitt wuppertal , wuppertal , germany + @xmath509 department of physics , yale university , new haven ct , united states of america + @xmath510 yerevan physics institute , yerevan , armenia + @xmath511 domaine scientifique de la doua , centre de calcul cnrs / in2p3 , villeurbanne cedex , france + @xmath512 also at laboratorio de instrumentacao e fisica experimental de particulas - lip , lisboa , portugal + @xmath513 also at faculdade de ciencias and cfnul , universidade de lisboa , lisboa , portugal + @xmath514 also at cppm , aix - marseille universit and cnrs / in2p3 , marseille , france + @xmath515 also at triumf , vancouver bc , canada + @xmath516 also at department of physics , california state university , fresno ca , united states of america + @xmath517 also at faculty of physics and applied computer science , agh - university of science and technology , krakow , poland + @xmath518 also at department of physics , university of coimbra , coimbra , portugal + @xmath519 also at universit di napoli parthenope , napoli , italy + @xmath520 also at institute of particle physics ( ipp ) , canada + @xmath521 also at louisiana tech university , ruston la , united states of america + @xmath522 also at group of particle physics , university of montreal , montreal qc , canada + @xmath523 also at institute of physics , azerbaijan academy of sciences , baku , azerbaijan + @xmath524 also at institut fr experimentalphysik , universitt hamburg , hamburg , germany + @xmath525 also at manhattan college , new york ny , united states of america + @xmath526 also at school of physics and engineering , sun yat - sen university , guanzhou , china + @xmath527 also at academia sinica grid computing , institute of physics , academia sinica , taipei , taiwan + @xmath528 also at high energy physics group , shandong university , shandong , china + @xmath529 also at california institute of technology , pasadena ca , united states of america + @xmath530 also at particle physics department , rutherford appleton laboratory , didcot , united kingdom + @xmath531 also at section de physique , universit de genve , geneva , switzerland + @xmath532 also at departamento de fisica , universidade de minho , braga , portugal + @xmath533 also at department of physics and astronomy , university of south carolina , columbia sc , united states of america + @xmath534 also at kfki research institute for particle and nuclear physics , budapest , hungary + @xmath535 also at institute of physics , jagiellonian university , krakow , poland + @xmath536 also at department of physics , oxford university , oxford , united kingdom + @xmath537 also at dsm / irfu ( institut de recherches sur les lois fondamentales de lunivers ) , cea saclay ( commissariat a lenergie atomique ) , gif - sur - yvette , france + @xmath538 also at laboratoire de physique nuclaire et de hautes energies , upmc and universit paris - diderot and cnrs / in2p3 , paris , france + @xmath539 also at department of physics , nanjing university , jiangsu , china + @xmath540 deceased

hierarchical models of structure formation , in which galaxies formed at high redshift and rapidly coalesced into groups and clusters through accretion and mergers , have become a compelling paradigm for understanding galaxy evolution . in such dynamically rich environments , galaxies are subject to an array of physical processes that affect their evolution . these physical processes form two broad classes : ( 1 ) tidal interactions such as those induced by major mergers ( e.g. , lavery & henry 1988 ) , off - axis galaxy collisions ( mller et al . 1989 ) , galaxy harassment ( moore et al . 1996 ) or galaxy fly - bys near the core of the cluster potential ( byrd & valtonen 1990 ) , and ( 2 ) gas - gas interactions , notably ram pressure by the intracluster gas ( icm ) on the galaxy s interstellar medium ( ism ) , due to the galaxy s motion through the surrounding icm ( gunn & gott 1972 ) , or ism - ism interactions produced in galaxy - galaxy collisions ( kenney et al . these gas - dynamical processes may be enhanced by turbulence and viscous effects ( nulsen 1982 ; quilis et al . 2000 ) or inhomogeneities and bulk motions in the icm gas ( kenney et al . 2004 ) . combes ( 2004 ) and references therein provide a recent overview . no model for galaxy evolution can be complete without a detailed understanding of the gas - dynamics of these interactions . similarly no model for the evolution of the intracluster medium can be complete without understanding the feedback of gas and energy from the galaxy into the icm by these gravitational and hydrodynamical stripping processes . the actions of tidal forces are identified by the appearance of disturbed stellar morphologies , such as stretched stellar tails ( see e.g. gnedin 2003 , vollmer 2003 ) ; while the characteristic signatures of hydrodynamic stripping ( see e.g. stevens et al . 1999 ; toniazzo & schindler 2001 ; acreman et al . 2003 and references therein ) are imprinted on the structure of the hot x - ray emitting gas in and near the galaxy . notably the ram - pressure experienced by galaxies moving through the dense icm can produce sharp surface brightness edges ( `` cold fronts '' ) along the leading interface between the galaxy and the icm ( acreman et al . 2003 ; machacek et al . these cold fronts are contact discontinuities where a sharp rise in surface brightness ( gas density ) is accompanied by a corresponding drop in gas temperature , and are the galaxy - sized analogues of similar subcluster - scale features identified near the cores of rich clusters ( see e.g. markevitch et al . 2000 , vikhlinin et al . 2001 ; mazzotta et al . the higher pressure found in the cold front compared to that in the cluster gas is balanced by the ram pressure caused by the motion of the cloud ( in our case , galaxy ) through the surrounding icm . tails of stripped galaxy gas have been found to extend @xmath18 to @xmath19kpc behind the galaxy before fading into the background emission of the ambient cluster gas ( see e.g. forman et al . 1979 , white et al . 1991 , rangarajan et al . 1995 for m86 ; irwin & sarazin 1996 , biller et al . 2004 for ngc 4472 ; sakelliou et al . 1996 for 4c34.16 ; wang et al . 2004 for c153 ; machacek et al . 2005a , scharf et al . 2005 for ngc 1404 ) . detailed studies of these x - ray features in nearby systems ( see , for example , biller et al . 2004 ; machacek et al . 2005a , 2005b ; sun & vikhlinin 2005 ; sun , jerius & jones 2005 ) , made possible by the high angular resolution of _ chandra _ and _ xmm - newton _ , not only reveal the nature of the gas - dynamics of the stripping process , but also are one of the few ways to constrain the three dimensional motion of the galaxy as it passes through the group or cluster core ( merrifield 1998 ; dosaj et al . 2002 ) . ngc 4552 ( @xmath20,@xmath21 , j2000 ) is an elliptical galaxy located @xmath22 east of m87 in subcluster a of the virgo cluster . measurement of ngc 4552 s distance modulus using surface brightness fluctuations ( tonry et al . 2001 ) places it at nearly the same luminosity distance as m87 , while ngc 4552 s low line of sight velocity ( @xmath23kms@xmath15 , ned ; smith et al . 2000 ) compared to that of m87 ( @xmath24kms@xmath15 , ned ; smith et al . 2000 ) implies that ngc 4552 is moving supersonically through the virgo icm with at least @xmath25kms@xmath15towards us relative to m87 . thus ngc 4552 is a likely candidate for ram - pressure stripping by the cluster gas . since ngc 4552 is a member of the large sample of nearby elliptical galaxies used to establish correlations between x - ray emission and other ism tracers ( see e.g. osullivan et al . 2001 and references therein ) , its global x - ray properties are well studied . early observations with the _ einstein _ observatory ( forman , jones & tucker 1985 ; canizares , fabbiano & trinchieri 1987 ; roberts et al . 1991 ; kim , fabbiano , & trinchieri 1992 ; eskridge , fabbiano , & kim 1995 ) focused on the measurement of ngc 4552 s total x - ray luminosity to establish the @xmath26 @xmath27 relation . subsequent _ rosat _ observations of ngc 4552 were used to study the possible dependence of this relation on environment ( brown & bregman 2000 ; osullivan et al . fits to the mean x - ray spectrum for ngc 4552 using _ rosat _ resulted in mean temperatures for the x - ray gas in the galaxy of @xmath28kev depending on the spectral model ( davis & white 1996 ; brown & bregman 2000 ; matsushita 2001 ) . using _ asca _ data , matsushita , ohashi , & makishima ( 2000 ) modelled the mean spectrum of ngc 4552 using variable abundance thermal plasma models including an additional bremsstrahlung component in the spectral models to account for the contribution of unresolved point sources in low spatial resolution data . they found mean temperatures @xmath29kev for ngc 4552 , in agreement with the _ rosat _ studies , but with generally higher abundances ( @xmath30 ) . most recent studies , across a wide range of wavelengths , have concentrated on the nuclear properties of ngc 4552 . a brief review of prior work on the nucleus can be found in machacek et al . ( 2006 ; hereafter known as paper ii ) , where we present our results from chandra observations of the central regions of ngc 4552 that show shocks close to the center of ngc 4552 that were produced by recent nuclear activity . in this paper we use _ chandra _ x - ray observations to focus on the properties of gas in the outer regions of ngc 4552 where hydrodynamic gas stripping is occuring . our discussion is organized as follows : in [ sec : obs ] we describe the observations and our data reduction and processing procedures . in [ sec : ramresult ] we present the background subtracted , exposure corrected image of ngc 4552 showing the prominent x - ray emission features of the system : a sharp surface brightness edge to the north , horns of emission extending away to the south from the surface brightness edge , a ram pressure stripped x - ray tail , a bright nucleus , and bright ringlike features ( discussed in detail in paper ii ) near the galaxy center . we then discuss our analysis method and main results , including the determination of the gas density and temperature , for gas stripping from the outer regions of the galaxy . we use these results to constrain the velocity of ngc 4552 through the virgo icm . in [ sec : conclude ] we summarize our results . unless otherwise indicated all errors correspond to @xmath31 confidence levels and coordinates are j2000 . taking the distance to the dominant elliptical m87 as representative of the distance to subcluster a of the virgo cluster containing ngc 4552 , the luminosity distance to the cluster is @xmath32mpc ( tonry et al . 2001 ) and @xmath33 corresponds to a distance scale of @xmath34pc . for our analysis , we used a @xmath0ks observation of the elliptical galaxy ngc 4552 in virgo taken with _ chandra _ on 2001 april 22 using the advanced ccd imaging spectrometer array ( acis , garmire et al . 1992 ; bautz et al . 1998 ) with acis - s ( chip s3 ) at the aimpoint . the focal plane temperature of the instrument was @xmath35c throughout the observation . the data were analyzed using the standard x - ray processing packages , ciao version @xmath2 , ftools and xspec version @xmath36 . filtering removed events with bad grades ( @xmath37 , @xmath38 , and @xmath39 ) and those with significant flux in border pixels of the @xmath40 event islands ( vfaint mode ) , as well as any events falling on hot pixels . use of vfaint mode improves background rejection by as much as a factor of @xmath41 in the s3 chip for soft x - ray energies ( @xmath42kev ) important to this observation . the data were reprocessed and response files created using the most recent gain tables and instrumental corrections . these included correcting for the time - dependent declining efficiency of the acis detector due to the buildup of contaminants on the optical filter ( plucinsky et al . 2003 ) , which is important at energies below @xmath43kev , and for the slow secular drift ( tgain ) of the average pha values for photons of fixed energy . periods of anomalously high background ( flares ) were identified and removed from the data , along with periods of anomalously low count rates at the beginning and end of the observation . this resulted in a useful exposure time of @xmath44s . backgrounds for use in the imaging analyses and spectral measurements of the virgo cluster gas were created from the @xmath45ks period d source free dataset ( aciss_d_7_bg_evt_271103 ) appropriate for the date of observation and instrument configuration .. identical cleaning , energy and spatial filters were applied to source and background data throughout . we checked the standard normalization of the source free background , set by the ratio of the exposure times , by comparing count rates between the source and background files in the @xmath46kev energy range , where particle background dominates . we found the standard normalization was @xmath47 too high and renormalized the source free data by a factor of @xmath48 to correct for this difference . point sources were identified in the @xmath49 field of the s3 chip in the @xmath50kev energy band using a multiscale wavelet decomposition algorithm set with a @xmath51 detection threshold . the resulting @xmath52 source identifications were excluded from the spectral and surface brightness analyses that follow . in figure [ fig : image ] we present the @xmath1kev _ chandra _ image of the diffuse emission in ngc 4552 overlaid with x - ray surface brightness contours showing the x - ray features of interest for our analysis . point sources were removed from the image and the point source regions filled with the local average emission level using ciao tool _ dmfilth_. the image was then background subtracted , corrected for telescope vignetting and detector response using exposure maps created with standard ciao tools . first , two narrow band ( @xmath53 and @xmath54kev ) images were background subtracted and exposure corrected using monoenergetic instrument maps of @xmath55 and @xmath43kev , respectively . the fluxed narrow band images were then summed and smoothed with a @xmath33 gaussian kernel to produce the image in figure 1 . we chose conservative gaussian smoothing for our images to minimize potential numerical artifacts from the smoothing algorithm . the @xmath33 smoothing scale represents a compromise between the need for high spatial resolution to map sharp or narrow features and the need to smooth over larger scales to highlight faint extended emission features . we see a sharp , flattened discontinuity ( edge ) in the surface brightness @xmath2kpc ( @xmath56 ) north of ngc 4552 s center . horns of emission extend @xmath57kpc ( @xmath58 ) to the southeast and southwest away from each side of the flattened edge . an x - ray tail extends @xmath6kpc ( @xmath59 ) to the south - southeast before becoming indistinguishable from the ambient icm . in fig . [ fig : hehwtsig ] we show projections across the horns and tail in a @xmath60kev image of the diffuse emission ( binned to pixel size @xmath61 but without background subtraction ) , demonstrating that these features are highly significant . we see in figure [ fig : image ] an enhanced region of x - ray emission in the southwest quadrant of the galaxy , near the beginning of the tail , that we denote as the southern lump sl . in figure [ fig : panda ] we show the @xmath60kev emission in the galaxy for radii @xmath62 in @xmath63 equal sectors numbered counterclockwise from @xmath37 . as shown in the right panel of figure [ fig : panda ] , x - ray emission from the outer radii of the galaxy is very asymmetric . enhanced soft emission occurs both in the northen galaxy regions ( sectors @xmath64 ) , behind the leading edge of the galaxy , and in the southern galaxy regions ( sectors @xmath65 ) near the beginning of the tail . the peak in the soft emission for these regions occurs in the southwest quadrant of the galaxy in sectors @xmath66 , at the position of the southern lump ( sl ) . thus we see that the southern lump sl , as well as the horns and the tail , is also highly significant . disturbed x - ray morphologies may arise from tidal interactions caused by a recent or ongoing merger , winds of outflowing material due to a recent starburst , or outflows from a central agn , as well as from ram pressure stripping of the galaxy ism due to the motion of the galaxy through the ambient virgo icm . for ngc 4552 these first three scenarios are disfavored . tidal ( gravitational ) interactions act on both stars and gas , producing characteristic stellar tidal streams or tails ( gnedin 2003 ; combes 2004 ) . there is no evidence for disturbed morphology in the stellar distribution of ngc 4552 or for significant extragalactic stellar light that might result , if the observed features were of tidal origin ( bettoni et al . 2003 ) . a second possibility , given the classification of ngc 4552 as a transition nucleus ( ho et al . 1997 ) , is that the distorted x - ray morphology may be due to winds from a compact nuclear starburst . however , the transition nucleus classification for ngc 4552 is weak , due to @xmath67 uncertainties in the measurement of the h@xmath68 line emission ( ho et al . 1997 ) , and the radio spectrum of ngc 4552 is inconsistent with that expected from such a starburst . the nucleus of ngc 4552 exhibits a compact core radio source with a flat spectrum , whose peak brightness temperature of @xmath69k ( filho et al . 2004 ) is several orders of magnitude above the @xmath70k upper limit ( condon et al . 1991 ) for the brightness temperature of a compact nuclear starburst . in addition , starburst winds tend to produce conical outflows with associated optical line ( h@xmath71 ) emission from the interaction between the supernova driven ejecta and the surrounding ism ( see , e.g. , strickland et al . 2000 for ngc 253 and cecil , bland - hawthorn & veilleux 2002 for ngc 3079 ) , which are not seen in ngc 4552 . a third possibility is that agn outbursts could be responsible for the distorted x - ray morphology seen at large radii in ngc 4552 . the nucleus of ngc 4552 does harbor a supermassive black hole ( filho et al . 2004 ) and does undergo outbursts , as evidenced by the ring structures seen in figure [ fig : image ] and analyzed in detail in paper ii . however , the x - ray signatures of such outbursts are residual bright - rimmed cavities and buoyant bubbles , as found in m87 ( e.g. , forman et al . 2005 ) , and/or x - ray edges corresponding to shocks driven into the ambient medium by the outburst ( e.g. , fabian et al . 2003 for ngc 1275 in perseus and nulsen et al . 2005 for hydra a ) . this is quite different from the ` flattened leading edge - trailing tail ' morphology seen in figure [ fig : image ] . in contrast , the qualitative correspondence between the main x - ray features shown in figure [ fig : image ] , i.e. the leading edge , horns , and tail , and those found in simulations ( e.g. see stevens et al . 1999 ; toniazzo & schindler 2001 ; acreman et al . 2003 ) of ram - pressure stripping of elliptical galaxies moving through surrounding cluster gas is striking . however , since none of these simulations model the specific galaxy , orbital , or cluster characteristics of ngc 4552 in virgo , only a qualitative , not quantitative , comparison of these simulations to our observation should be made . for supersonic stripping , the galaxy ism is initially pushed back and fanned out ( stevens et al . 1999 , model 1b , figure 2 ; toniazzo & schindler 2001 , figure 3 ; acreman et al . 2003 , figure 2c , @xmath72gyr slice ) causing the edge between the galaxy gas and the icm to flatten , as is seen for ngc 4552 ( fig . [ fig : image ] ) . this sharp surface brightness discontinuity , @xmath2kpc north of the galaxy s center , coupled with a gas tail in the opposite direction , fixes the galaxy s direction of motion in the plane of the sky . irregular filaments of stripped material also are seen in simulation images extending back from the leading edge , similar to the horns we observe in ngc 4552 . these filaments signal the onset of kelvin - helmholtz instabilities , where stripping is occurring at the boundary between galaxy and icm gas . in the simulation images , galaxy gas , once stripped and decelerated , forms a distinctive tail of emission behind the galaxy , again similar to what we see in figure [ fig : image ] . the three - dimensional simulations of toniazzo & schindler ( 2001 ) demonstrate that the tail is in general not axisymmetric with respect to the direction of motion , but may appear angular or curved , as is also found in our image . thus the most likely explanation for the origin of the features shown in figure [ fig : image ] is ram - pressure stripping of galaxy gas due to the motion of ngc 4552 through the virgo cluster icm . in order to investigate quantitatively these complex gas - dynamical processes and test our ram - pressure - stripping hypothesis , we first determine the spectral properties and densities of hot gas in the surrounding icm and in ngc 4552 s main emission features . spectral extraction regions were chosen to isolate , as much as possible given our limited photon statistics , emission from the individual features . the geometries of these spectral extraction regions are listed in table [ tab : specregs ] and are shown in figure [ fig : specimage ] . these regions are he and hw for the eastern and western horns extending from the northern surface brightness edge to either side of the galaxy , oa for galaxy emission at large radii but inside the northern edge , sl for the region ( lump ) of enhanced emission in the southwestern quadrant of ngc 4552 , and t for the tail of emission extending from the galaxy to the south . to determine source existence , one compares the source plus background counts in a region to that expected from the background alone . we also can determine how well the flux can be measured for each feature . we compute both these statistics for regions he , hw , sl , and t. the background regions we used to compute these measures are listed in table [ tab : bgsigregs ] , and the respective areas and observed counts are given in table [ tab : fluxsigdat ] . we note that the background region ( sl_bg ) used for region sl , that is embedded in the galaxy , was a neighboring region that included the surrounding galaxy emission . neighboring virgo cluster emission regions were used as local backgrounds to compute these statistics for features he , hw , and t , since those regions extend outward from ngc 4552 into the virgo cluster emission . for each region we find that the cumulative poisson probability that the observed or greater number of counts could have arisen from a fluctuation of the expected background is @xmath73 . thus the detection of these features is highly significant . the precision to which the flux can be measured for each feature is given in table [ tab : fluxsigdat ] , as the net source counts and the total uncertainty in each measurement . for the analysis below , since the spectra for gas in ngc 4552 and in the x - ray tail are soft , we restrict the spectral fit energy range for the tail ( t ) , horns ( he+hw ) and galaxy regions outside the rings to @xmath74kev , where the source count rates are above background . we initially separated the outer annular region ( oa ) for ngc 4552 into two separate rectangular regions , one just inside the northern leading edge ( ng ) and one to the south near the beginning of the tail ( sg ) , to look for temperature differences between these two regions . however , since we found no statistically significant difference between the spectral fits for these regions ( see table [ tab : allspectra ] ) , we combined them into the annular region ( oa ) shown , in order to improve statistics and sharpen our measurement of the temperature and abundance for galaxy gas outside the central region . we chose a @xmath75 rectangular region ( vn ) centered @xmath76 north of ngc 4552 , comprising approximately @xmath77 of the area of the s3 chip , to determine the spectrum of the virgo cluster gas in the vicinity of ngc 4552 . we fit the spectrum for the virgo icm ( region vn ) over the full @xmath78kev energy band , where the s3 detector has good efficiency . we fix the hydrogen absorbing column at @xmath79@xmath80 , the galactic value ( dickey & lockman 1990 ) and allow the temperature , abundance and normalization to vary . we find the temperature of the virgo cluster icm to be @xmath81kev with metal abundance @xmath82 . we then allowed the absorbing column to vary and found no suggestion for increased absorption . our results are in excellent agreement with previous spectral analyses of the virgo cluster icm using data from _ ginga _ ( takano et al . 1989 ) , _ rosat _ ( bhringer et al . 1994 ) , and _ asca _ ( ohashi et al . 1998 , shibata et al . 2001 ) , that find , with galactic hydrogen absorption , temperatures @xmath83kev and metal abundances of @xmath84 , at projected distances ( @xmath85kpc from m87 ) comparable to that of ngc 4552 . we note that the temperature and x - ray emissivity of the virgo icm are relatively insensitive to the metal abundance over the abundance range ( @xmath86 ) determined by our fit . if we allow the abundance to vary over this range , we find variations in the virgo icm temperature of @xmath87kev , within the @xmath31 confidence level for the canonical fit , and variations in the inferred @xmath1kev x - ray emissivity for the cluster gas of @xmath88 . region vn also was used as a local background region for galaxy source regions at large radii ( ng , sg , oa , sl ) , the horns ( hw+he ) and the tail ( t ) , to remove contamination from virgo cluster emission . for spectral fitting , counts were grouped both by using predefined bins resulting in channels of approximately logarithmic width , and by using a minimum @xmath89 counts per bin . both methods gave consistent spectral model fits . the spectra for all of the above regions are well represented by single - temperature apec or vapec thermal plasma models ( smith et al . 2001 ) corrected for absorption using wisconsin photo - electric cross - sections ( morrison & mccammon 1983 ) . our results for these regions are summarized in table [ tab : allspectra ] . we checked that our results were not sensitive to the position of the local background region ( vn ) on the detector by recomputing the spectral fits using a second @xmath90 circular region @xmath91 east of ngc 4552 ( ve in table [ tab : specregs ] ) and found no significant differences in the fitted parameters . we also fit the spectra to these regions using blank sky backgrounds ( as for vn ) and a two component apec + apec model . we fixed the temperature and abundance of one apec component at our best fit values for the virgo cluster emission ( @xmath92kev and @xmath93 , see table [ tab : allspectra ] ) , while allowing the temperature and abundance of the second apec component and both component normalizations to vary , and found consistent results . thus our results are not biased by our choice of vn as the local background . since the models agree within their @xmath31 cl , we present only those results ( using background region vn ) for the tail and galaxy emission at large radii in table [ tab : allspectra ] . for fixed galactic absorption , we find the temperature and abundance of the outer galaxy gas in region oa to be @xmath94kev and @xmath95 , a metallicity marginally higher than that for the virgo icm , but with large uncertainty . to check the stability of the fit , we allowed the absorbing column to vary and found no suggestion for increased absorption above the galactic value . to assess the effect of any residual unresolved x - ray binaries on our results , we also fit the spectrum for region oa , galaxy emission within the northern edge , over the @xmath96kev energy range with an absorbed two component model consisting of an apec component for emission from the diffuse gas and a @xmath38kev bremsstrahlung component to model unresolved x - ray binaries ( kraft et al . we fix the hydrogen absorbing column at its galactic value , while allowing the temperature and abundance in the apec component and both component normalizations to vary . we find that the fitted temperature ( @xmath97kev ) and abundance ( @xmath98 ) for the thermal component in this two component model agree , within their @xmath31 confidence limits , with the single apec model results listed in table [ tab : allspectra ] . the bremsstrahlung ( unresolved binary ) component contributes only @xmath99 to the @xmath1kev luminosity of region oa . this is consistent with theoretical expectations for the contribution of x - ray binaries to the flux in region oa found by integrating the average x - ray luminosity function for lmxb s ( gilfanov , 2004 ) below our observed @xmath100 kev point source detection threshold ( @xmath101ergs@xmath15@xmath80 ) in that region . thus , the effects of residual unresolved x - ray binaries on the measured properties of the diffuse gas in region oa do not bias the single temperature apec model results for the properties of the galaxy gas which are important to our analysis . we find that the temperature of the ism on the galaxy side of the ( leading ) northern edge ( @xmath102kev ) is a factor @xmath103 lower than the temperature of the gas in the ambient virgo icm ( @xmath104kev ) . as shown in figure [ fig : image ] , this temperature drop corresponds to a sharp increase in surface brightness ( and thus gas density as quantified in [ sec : sboutprofile ] ) . this is consistent with the interpretation of the northern edge as a cold front , the leading edge of the galaxy ism as it undergoes ram pressure stripping , in broad agreement with expectations from simulations ( e.g. see stevens et al . 1999 , figure 2 , model 1b ; acreman et al . 2003 , figures 10a & c and 11b & d , @xmath105gyr time slice ) . figures [ fig : image ] and [ fig : panda ] show a bright region ( lump ) of emission in the outer southwest quadrant of the galaxy . we use a @xmath106 circular region centered at ( @xmath107 , @xmath108 ) surrounding the lump ( region sl in figure [ fig : specimage ] ) to isolate the spectrum for this region . although we find , using an apec model with galactic absorption , a temperature @xmath109kev and abundance @xmath110 formally consistent with the properties of galaxy gas elsewhere outside the central rings in ngc 4552 , the @xmath111 is large . the fit is improved ( @xmath112 ) by using an absorbed vapec model with a higher iron abundance , while fixing all other abundances at @xmath113 , the mean abundance for the galaxy ism in region oa . the hydrogen absorbing column remained fixed at the galactic value . we include this fit in table [ tab : allspectra ] . in this vapec model , the temperature of the gas in the southern lump is @xmath114kev , consistent with previous results for gas at large radii , but the iron abundance is increased to near solar values ( @xmath115 ) . this may suggest that the southwest quadrant of the galaxy has either experienced a recent episode of star formation itself or has had metals transported from star forming activity near the galaxy center ( expected for a transition type nucleus ) to larger radii in the galaxy as a result of outflows . the regions he and hw for the horn - like features are faint , with region he and hw containing only @xmath116 and @xmath117 net source counts , respectively , in the @xmath74kev energy band . in order to improve statistics , we combine these regions and fit their average spectrum ( he+hw in table [ tab : allspectra ] ) . we restrict our models to those with three or fewer free parameters . neither an absorbed bremsstrahlung model nor absorbed power law model , each with all parameters free to vary , can fully describe the data , giving @xmath118 of @xmath119 and @xmath120 , respectively . a thermal apec model with galactic absorption provides a better fit to the data with temperature @xmath121kev and abundance @xmath122 ( @xmath123 ) . the probability of obtaining this value of @xmath124 or higher , given @xmath125 degrees of freedom , is @xmath126 ( @xmath127 confidence ) . the @xmath124 for the model fit to the combined spectrum for the horns might suggest that these two regions are not homogeneous in their spectral properties . we investigate this possibility by fitting the spectrum for each horn individually . using absorbed apec thermal plasma models with fixed abundance and galactic absorption , we find the temperature of gas in the eastern horn he to be @xmath128kev ( @xmath129 ) and in the western horn hw to be @xmath130kev ( @xmath131 ) for @xmath132 abundance , taken from the fit to the combined horn regions he+hw ( see table [ tab : allspectra ] ) . these temperatures vary by @xmath133 when the abundance is varied between @xmath134 to @xmath135 ( see table [ tab : metalscool ] ) . although suggestive , the fitted temperatures agree within their @xmath31 confidence limits . thus with these data , we can not conclude that the temperatures of the eastern and western horns are significantly different . however , if such a temperature difference were to be confirmed in a deeper exposure , it would not be surprising . the horns are spatially separated and most likely correspond to different stripping events . temperature differences between gas in the two horns could arise from hydrodynamical effects caused by local temperature and pressure gradients in the different turbulent eddies responsible for the stripping process in each region . the main conclusion that can be drawn from the current data is that the temperature of the gas in the horns is cool , similar to that measured throughout the outer regions of ngc 4552 ( see table [ tab : allspectra ] . these temperatures agree within their @xmath31 confidence limits with the temperature ( @xmath136kev ) for gas in the outer radii of ngc 4552 ( region oa ) . in contrast , they are @xmath137 times lower than the @xmath138kev temperature of the surrounding virgo cluster icm . this is the key property of the gas in the horns that identifies it as galaxy ( and not cluster ) gas , likely in the process of being stripped . these results are consistent with simulations of ram - pressure stripping ( see , e.g. stevens et al . 1999 , figure 2 , model 1b ) , where irregular filaments due to the onset of hydrodynamical instabilities during stripping are cooler and denser than the surrounding icm . in figure [ fig : edgeimage ] we show the angular sector used to construct the @xmath1kev surface brightness profile to the north across the flattened surface brightness edge between galaxy gas and the virgo cluster icm , as a function of the mean distance from the center of ngc 4552 . the profile is constructed from elliptical annuli constrained to lie within the angular sector centered on the galaxy nucleus and extending from @xmath139 to @xmath140 measured clockwise from east , that was chosen to exclude the horns of emission extending southward from each side of the northern leading edge . the elliptical annuli are concentric to a ` bounding ' ellipse , with semi - major ( -minor ) axes of @xmath141 ( @xmath142 ) and position angle of @xmath143 , that traces the shape of the northern surface brightness edge within the angular sector . the radial extent of each elliptical annulus varies logarithmically inward and outward from the bounding ellipse with logarithmic step size @xmath144 . in the left panel of figure [ fig : sbprof ] , we show the @xmath1kev surface brightness profile as a function of the mean distance from the center of ngc 4552 to the north across the flattened surface brightness edge between galaxy gas and the virgo cluster icm . two sharp surface brightness discontinuities ( denoted by vertical dashed lines ) are evident in figure [ fig : sbprof ] : the outer , ` leading ' edge is at the northern interface between the galaxy cold front and virgo cluster gas , and the inner discontinuity is at the position of the rim of the bright northern ring . we analyse the leading edge of the cold front in this paper and the central rings in paper ii . the observed surface brightness profile is proportional to the product of the x - ray emissivity and the square of the electron density , integrated along the line of sight . by fitting the surface brightness profile , we are able to determine the shape of the density distribution , the location of the edge , and the ratio of electron densities on either side of the edge , averaged over the respective profile bins . if the density is slowly varying over the profile bin , as is the case for a cold front , this measured density ratio is a good measure of the physical density discontinuity across the edge . as was done for other cold fronts ( markevitch et al . 2000 ; vikhlinin et al . 2001 ; machacek et al . 2005a ) , we assume a spherically symmetric power law distribution of the form @xmath145 for the electron density in ngc 4552 inside the leading edge of the cold front ( but outside the central rings ) , where @xmath146 and @xmath147 are the normalization and power law index for the electron density of ngc 4552 in the cold front , respectively . the radial distance @xmath148 and location @xmath149 of the leading surface brightness discontinuity are measured from the optical center of ngc 4552 , i.e. the center of the bounding ellipse . we model the electron density in the virgo cluster icm ( within subcluster a ) using a spherically symmetric , isothermal @xmath68 model centered on m87 with @xmath150 , and core radius @xmath151 taken from a fit by schindler et al . ( 1999 ) to _ rosat _ all - sky survey data of the virgo icm out to a radius of @xmath152 from m87 . we normalize the cluster @xmath68 model using the surface brightness @xmath153kpc away from ngc 4552 , in the undisturbed virgo icm ( cluster free stream region ) . we find the central electron density ( density @xmath68 model normalization ) for the virgo icm to be @xmath154@xmath5 , in agreement with previous results ( schindler et al . 1999 ) . however , to model the electron density , at the interface between galaxy gas in the cold front and the virgo icm , requires additional assumptions about the three - dimensional geometry of ngc 4552 relative to the cluster center . we assume that the distance of m87 is representative of the cluster center , at least for subcluster a containing both m87 and ngc 4552 . measurements of distance moduli @xmath155 for ngc 4552 ( @xmath156 ) and m87 ( @xmath157 ) , using i - band surface brightness fluctuations ( tonry et al . 2001 ) , agree within uncertainties , suggesting the two galaxies are at comparable distances . we thus assume that the @xmath158 projected distance between ngc 4552 and m87 is representative of their true physical separation . the electron density of the virgo icm , just outside the leading edge , is then @xmath159@xmath5 . we then use a multivariate minimization algorithm to fit the surface brightness profile across the leading edge , allowing the location of the edge ( @xmath149 ) , the power law index of the electron density inside the edge ( @xmath147 ) , and the discontinuity ( @xmath160 ) to be free parameters . the discontinuity ( @xmath160 ) is given by @xmath161 where @xmath162(@xmath163 ) and @xmath164(@xmath165 ) are the x - ray emissivity ( electron density ) inside ( @xmath166 ) and outside ( @xmath167 ) the leading edge of the cold front . we find the best fit position of the leading , northern edge ( cold front ) to be @xmath168kpc from the galaxy center , the slope of the electron density inside the cold front to be @xmath169 , and the discontinuity @xmath160 ( given by equation [ eq : jump ] ) to be @xmath170 . the uncertainties on the discontinuity reflect the @xmath31 cl uncertainties in the slope found for @xmath147 . while the surface brightness discontinuity @xmath160 depends only on the slope of the power law distribution given by the fit to the surface brightness profile , we see from equation [ eq : jump ] that the ratio of the electron densities inside and outside the leading edge is sensitive , through the x - ray emissivity , to the spectral properties of the gas . in table [ tab : metalscool ] we show the dependence of the spectral fit parameters on the galaxy abundances for all the galaxy regions of interest and list the parameters for the nearby virgo cluster icm for comparison . as this table shows , the temperature found for gas in each region is insensitive to the abundance over the range of interest ( @xmath171 ) . on the other hand , since metal lines dominate the cooling function for temperatures @xmath172kev ( tucker & gould 1966 ; raymond & smith 1977 ; smith et al . 2001 ) , the spectral model normalization , and thus x - ray emissivity , vary by a factor of @xmath173 over the same range . as shown for the infalling elliptical galaxy ngc 1404 in the fornax cluster ( machacek et al . 2005a ) , this may introduce significant uncertainties in modeling the electron density in the galaxy from the surface brightness discontinuity between galaxy and cluster gas . we use @xmath1kev x - ray emissivities given in table [ tab : metalscool ] derived from our spectral fits , the cluster icm density outside the cold front ( @xmath174@xmath5 ) determined from the cluster icm @xmath68-model , and the measured discontinuity ( @xmath170 ) in equation [ eq : jump ] to infer the electron density @xmath175 inside the cold front . in table [ tab : properties ] we show the effect of the uncertainty in ngc 4552 s abundance ( and thus emissivity ) on the ratio of electron densities inside and outside the leading edge ( cold front ) and the propagation of that uncertainty through the analysis of the galaxy motion that follows . since we found no strong metallicity gradients within the galaxy ( see table [ tab : allspectra ] ) , we adopt a common mean metallicity for gas in the whole galaxy of @xmath176 , determined from a spectral fit to the galaxy emission within a radius of @xmath177 , excluding resolved point sources and the nucleus . from table [ tab : metalscool ] , the @xmath1 kev emissivity for @xmath178kev galaxy gas at large radii ( region oa ) with abundance @xmath179 is @xmath180erg@xmath181s@xmath15 . we take the @xmath1kev x - ray emissivity for @xmath93 virgo icm gas ( also from table [ tab : metalscool ] ) to be @xmath182erg@xmath183s@xmath15 . the density ratio across the cold front is then found to be @xmath184 , and the electron density at @xmath185kpc , inside ngc 4552 in the cold front , is @xmath186@xmath5 . following vikhlinin et al . ( 2001 ) , we use the gas temperature and density ratios across the leading edge in ngc 4552 to calculate the pressure ratio @xmath187 between galaxy gas in the cold front inside the leading edge and undisturbed virgo cluster gas in the free - stream region . we assume the galaxy gas in ngc 4552 just inside the edge is in pressure equilibrium with cluster gas at the stagnation point just outside the edge , where the relative velocity between ngc 4552 and the cluster gas is zero . thus @xmath187 is a measure of the pressure ratio between cluster gas at the stagnation point and in the free - stream region , and can be used to determine the mach number @xmath188 of ngc 4552 s motion relative to the icm ( landau & lifschitz 1959 ; vikhlinin et al . 2001 ) . in table [ tab : properties ] we summarize the dependencies of this velocity analysis for ngc 4552 on the metallicity of the galaxy gas over the abundance range @xmath189 . it is important to note that , because the motion lies on the supersonic , steeply rising branch of the pressure ratio versus mach number curve ( eq . @xmath41 in vikhlinin et al . 2001 ) , the mach number remains well constrained , despite the uncertainties in the density and pressure ratios caused by uncertainties in ngc 4552 s poorly constrained metallicity . we find that ngc 4552 is moving at mach @xmath190 through the virgo cluster gas . given the speed of sound in @xmath138kev gas of @xmath191kms@xmath15 , the speed of ngc 4552 relative to the icm is @xmath192kms@xmath15 . taking @xmath193kms@xmath15 , the relative radial velocity between ngc 4552 and m87 ( ned , smith et al . 2000 ) , as the relative radial velocity between ngc 4552 and the surrounding virgo icm , we find the component of velocity in the plane of the sky to be @xmath194kms@xmath15and the inclination angle of the motion with respect to the plane of the sky to be @xmath195 towards the observer . with the motion of ngc 4552 through the icm constrained by the properties of the leading cold front , we can examine the properties of the hot gas in the tail . from table [ tab : allspectra ] , we see that the temperature ( @xmath196kev ) of x - ray - emitting gas in the tail is consistent ( given the large uncertainties ) with the temperature of gas in the outer region of the galaxy oa ( @xmath94 ) and in the horns ( @xmath121 ) , where stripping is expected to be occurring . the gas temperature in the tail is @xmath173 times lower than that of the virgo icm , proving that the tail is composed predominantly of ram - pressure - stripped galaxy gas rather than virgo icm gas concentrated into a wake . although poorly constrained due to our limited statistics , the metallicity of gas in the tail ( @xmath197 ) is also consistent with that in the galaxy . the @xmath1kev luminosity for hot gas in the tail ( from region t ) is @xmath198ergs@xmath15 . we estimate the mean electron density @xmath199 for gas in the tail from the xspec apec spectral model normalization @xmath200 , where @xmath201 @xmath202 is the angular size distance to the source , @xmath203 is the source redshift , @xmath199 and @xmath204 are the mean electron and hydrogen number densities , respectively , @xmath205 is the gas filling factor , and @xmath206 is the volume of the tail emission region . all quantities are in cgs units . if we assume gas in the tail uniformly fills a cylindrical volume ( corresponding to the spectral extraction region t ) of radius @xmath207kpc ( @xmath208 ) and length @xmath209 , where @xmath210 is the inclination angle of the motion with respect to the plane of the sky , we find a mean electron density for hot gas in the tail of @xmath211@xmath5 . as shown in table [ tab : metalscool ] and reflected by the large uncertainties , the apec model normalization @xmath200 , and thus the inferred mean electron density , are strongly dependent on the metallicity of the tail . assuming the `` best fit '' model for the tail ( @xmath212 , @xmath213kev , @xmath214 ) , the mean electron density in the tail ( region t ) is @xmath215@xmath5 . thus the tail is cooler and denser than the surrounding icm , as expected from simulations ( see , e.g. stevens et al . 1999 , figure 2 , model 1b ) . in figure [ fig : tailicmproj ] we compare the surface brightness in a rectangular region of width @xmath38kpc projected onto the tail s central axis in @xmath43kpc bins along the length of the tail ( filled circles ) to the projection of the surface brightness of the surrounding virgo icm in a parallel region of the same dimensions , orientation , and bin size ( open squares ) . we define the beginning of the tail ( and @xmath216 of the projection in figure [ fig : tailicmproj ] ) at the point @xmath173kpc south of ngc 4552 s center , where the tail visually appears distinct from the galaxy s more spherical , hot gas halo ( see figure [ fig : image ] ) . the solid and long - dashed lines denote the best linear fits to the projected surface brightness of the tail and virgo icm , respectively . as expected from the @xmath68-model for the virgo icm ( schindler et al . 1999 ) , the projected surface brightness of the virgo icm at the distance of the tail from m87 is very flat , consistent with being constant . in contrast the projected surface brightness of the tail declines by a factor @xmath103 before becoming indistinguishable from the surrounding virgo icm . defining the length of the observed tail , in the plane of the sky , as the distance from where the tail appears distinct from ngc 4552 to where it fades into the virgo icm , we estimate this length to be @xmath217kpc ( from figure [ fig : tailicmproj ] ) , corresponding to a physical length ( given @xmath218 ) of @xmath219kpc . using our derived mean density of @xmath220@xmath5and assuming uniform filling of a cylindrical volume of radius @xmath207kpc and physical length @xmath221kpc , we estimate the gas mass in the observed tail to be @xmath222 . we find the mean thermal pressure for gas in the tail to be @xmath223kev@xmath5 , where the errors reflect the @xmath31 cl combined uncertainties due to the measured tail gas temperature and metallicity and ngc 4552 s inclination angle @xmath210 . we estimate the mean pressure of the surrounding icm using the virgo icm @xmath68-model evaluated at the midpoint of the cylindrical emission region ( t ) . since the perpendicular ( @xmath210 dependent ) component of the distance between the midpoint of region t and the virgo cluster center ( taken to be m87 ) is small ( @xmath224 for @xmath225 ) compared to the transverse component ( @xmath22 ) , the @xmath210 dependence of the cluster gas electron density can be ignored . we find the mean thermal pressure for undisturbed virgo cluster gas at the location of the tail to be @xmath226kev@xmath5 . although the uncertainties are large , these data suggest that the thermal pressure of the undisturbed icm alone may not be sufficient to establish pressure equilibrium with the stripped gas in the tail . this is in agreement with simulations that find the stripped ism may not be in pressure equilibrium with the ambient medium , but that gas in the tail expands while it is removed ( toniazzo & schindler 2001 ; heinz et al . 2003 ) . if we assume the gas , once heated and stripped from the galaxy , is free to expand adiabatically at approximately the tail sound speed ( @xmath227kms@xmath15for @xmath228kev gas ) , we can use the initial tail radius @xmath229 to estimate the time @xmath230 needed for gas in the tail to reach pressure equilibrium with the undisturbed icm ( machacek et al . 2005b ) , i.e. @xmath231 for a monatomic ideal gas with adiabatic index @xmath232 , where @xmath233 and @xmath234 are the initial and final thermal pressures in the tail . assuming thermal pressures dominate , the final tail pressure @xmath234 is approximately the pressure of the surrounding icm . using the best apec model fit parameters for the tail gas , @xmath235 , and initial radius @xmath236kpc ( from region t ) , we find the time for the tail to expand to equilibrium to be @xmath6myr . given ngc 4552 s transverse velocity of @xmath237kms , gas at the end of the observable tail ( @xmath217kpc downstream in the plane of the sky ) has had @xmath153myr more time to expand than gas now found near the beginning of the tail , and so the southern end of the tail may be nearly in pressure equilibrium with the icm . as the gas density in the tail decreases due to adiabatic expansion , we would expect the tail to broaden and fade . from conservation of entropy , the mean density and temperature of the tail gas at pressure equilibrium would be @xmath238@xmath5and @xmath239kev , with an expected corresponding decrease in surface brightness by a factor @xmath153 . although our statistics are too limited to directly observe a temperature decline in the tail , the observed rapid fading of the tail , by a factor @xmath240 from figure [ fig : tailicmproj ] , is broadly consistent with that expected from adiabatic expansion . for supersonic motion one would expect a bow shock to form in the virgo icm in front of the galaxy . however , the bow shock is likely to be the least visible of the three features ( cold front , tail , bow shock ) expected from supersonic ram pressure stripping ( stevens et al . the primary reason for this is that the shock front is narrow and the observed properties are very sensitive to projection effects . while the rankine - hugonoit conditions ( @xmath241 landau & lifshitz 1959 ) predict a peak density and temperature jump of @xmath242 and @xmath243 at the shock front for mach @xmath138 motion through the cluster gas , the observed temperature and density jumps will be much lower , primarily due to integration along the line of sight that crosses mostly through unshocked cluster gas , and to projection effects caused by the fact that the motion of ngc 4552 is not in the plane of the sky , but at an inclination angle of @xmath244 out of the plane towards the observer . following vikhlinin et al . ( 2001 ) we estimate the approximate location of the bow shock in front of ngc 4552 from the mach number and the location ( @xmath245 ) of the body sonic point , i.e. the point on the body where the flow velocity equals the sound speed of the cluster gas . @xmath246 is the coordinate of the body sonic point measured along the axis of symmetry and @xmath247 is the radial distance of the body sonic point from the axis of symmetry ( see vikhlinin et al . 2001 , figure 9 ) . as vikhlinin et al . ( 2001 ) describe , we derive the shock position using moeckel s ( 1949 ) figure 7 , that shows the ratio of the shock detachment distance @xmath248 , where @xmath249 is the location of the bow shock along the symmetry axis , versus the radial coordinate of the body sonic point @xmath247 as a function of the mach number . for a mach @xmath138 shock , @xmath250 . thus if the body sonic point is known , the calculation of the shock position is well determined . if the body being stripped has a well defined shoulder , as in a3667 ( vikhlinin et al . 2001 ) , the body sonic point is at the location of the shoulder . however , in ngc 4552 ( see figure [ fig : image ] ) , the ` horns ' make the precise location of the shoulder difficult to discern . instead we use figure 9 of moeckel ( 1949 ) to find the body sonic point for mach @xmath138 motion as that point on the body surface where the angle , @xmath251 , of the tangent line from the symmetry axis to the body surface is @xmath252 . again the precise shape of the body surface is uncertain due to the irregular ` horns ' . we estimate the ` body surface ' of the galaxy by a sphere centered on the galaxy nucleus with a radius of @xmath2kpc ( the radial distance to the leading edge ) . this approximately spherical shape is consistent with the distribution of optical light and broadly consistent with the galaxy gas excluding the horns and the tail . we then find @xmath253kpc , the shock detachment distance @xmath254 , and the distance @xmath255 , between the position of the leading edge @xmath256 and the body sonic point along the symmetry axis , of @xmath257kpc . rewriting the shock detachment distance @xmath258 , we expect the bow shock to intersect the symmetry axis at @xmath257kpc outside the leading edge or equivalently at @xmath259kpc from ngc 4552 s center . it is interesting to note that the surface brightness outside the leading edge , shown in figure [ fig : sbprof ] , does increase by a factor @xmath83 between @xmath260kpc , broadly consistent with this picture . while suggestive , a much deeper observation is needed to establish whether or not this rise is actually evidence for the expected bow shock . the density and thus surface brightness are also expected to rise in the stagnation region as the gas behind the shock slows and is adiabatically compressed in front of the leading edge ( vikhlinin et al . 2001 ) . again a rise in the surface brightness is observed in figure [ fig : sbprof ] in front of the northern edge ( @xmath261kpc ) . however , this observed rise in surface brightness also could partly be due to galaxy gas being stripped by instabilities to form the horns . the expected temperature rise in the gas behind the shock front , as seen , e.g. , in simulation slices by stevens et al . ( 1999 ) , would not be observable in our data due to dilution by unshocked gas projected along the line of sight and our limited statistics over the small region ( mach angle @xmath262 ) between the shock front and the galaxy . in this work we analysed a @xmath0ks _ chandra _ observation of the elliptical galaxy ngc 4552 in the virgo cluster and found x - ray evidence for ram - pressure stripping in the outer regions of the galaxy , due to its supersonic motion through the virgo icm . we find the following : 1 . the @xmath1kev x - ray emission shows the classic features of a cold front and ram - pressure stripping as a gas rich galaxy moves supersonically within a rich cluster . we see : ( 1 ) a sharp , flattened leading surface brightness edge located @xmath2kpc from the center of ngc 4552 due to the ram - pressure of the virgo icm , ( 2 ) two horns of emission extending @xmath7kpc to either side of the edge that are composed of galaxy gas in the process of being stripped due to the onset of kelvin - helmholtz instabilities , and ( 3 ) a tail of emission extending @xmath6kpc behind ngc 4552 opposite the leading edge . galaxy gas inside the leading edge is cool ( @xmath263kev ) compared to the surrounding @xmath138kev virgo icm . the surface brightness distribution inside the leading edge is well fit by a power law density distribution @xmath264@xmath5 ( for @xmath265 ) , where @xmath266kpc is the position of the edge . the density ratio between gas inside the cold front and the surrounding virgo icm , found from fitting the surface brightness profile , is @xmath267 for galaxy ism metal abundances from @xmath134 to @xmath268 . 3 . assuming ngc 4552 and m87 both lie in the plane of the sky , the resulting pressure ratio ( @xmath12 ) between the free - streaming icm and cluster gas at the stagnation point implies that ngc 4552 is moving supersonically ( mach @xmath16 ) through the cluster gas with speed @xmath14kms@xmath15at an angle @xmath269 towards us with respect to the plane of the sky . 4 . the properties of the x - ray tail behind ngc 4552 are consistent with it being composed primarily of ram - pressure stripped galaxy gas . the tail is cool , with mean temperature @xmath196kev , and denser than the virgo icm , with @xmath270@xmath5 for @xmath271 . although the errors are large , mostly due to uncertainties in the tail gas metallicity , the mean thermal pressure of gas in the tail suggests that it is over - pressured with respect to the ambient icm . the subsequent adiabatic expansion of gas in the tail may partially explain the rapid fading of the tail with distance from the galaxy . this work is supported in part by nasa grant go3 - 4176a and the smithsonian institution . this work has made use 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 . we wish to thank maxim markevitch for the use of his edge analysis codes . acreman , d. , stevens , i. , ponman , t. , & sakelliou , i. 2003 , , 341 , 1333 bautz , m.w . , et al . 1998 , x - ray optics , instruments and missions , ed . hoover & a.b . walker , proc . spie , 3444 , 210 bettoni , d. , galletta , g. & garcia - burill0 , s. 2003 , a & a , 405 , 5 biller , b.a . , jones , c. , forman , w.r . , kraft , r. & ensslin , t. 2004 , , 613 , 238 bhringer , h. , briel , u. g. , schwarz , r. a. , voges , w. , hartner , g. , & trmper , j. 1994 , nature , 368 , 828 brown , b. & bregman , j. 2000 , , 539 , 592 byrd , g. & valtonen , m. 1990 , , 350 , 89 canizares , c. , fabbiano , g. & trinchieri , g. 1987 , , 312 , 503 cecil , g . , bland - hawthorn , j. & veilleux , s 2002 , , 576 , 745 combes , f. 2004 , in iau symposium 217 : recycling intergalactic and interstellar matter , ed . duc , j.braine , & e. brinks , asp conference series , vol . 217 ( san franscisco , astronomical society of the pacific ) , 440 condon , j. , huang , z .- p . , yin , q.f . , thuan , t.x . 1991 , , 378 , 65 davis , d & white , r. 1996 , , 470 , l35 dickey , j.m . & lockman , f.j . , 1990 , araa , 28 , 215 dosaj , a. , forman , c. , forman , w.r . , markevitch , m. , & vikhlinin , a. 2002 , baas , 34 , 712 eskridge , p. , fabbiano , g. & kim , d .- w . 1995 , , 97 , 141 fabian , a.c . , sanders , j.s . , allen , s.w . , crawford , c.s . , iwasawa , k. , johnstone , r.m . , schmidt , r.w . , & taylor , g.b . 2003 , , 344 , l43 filho , m.e.,fraternali , f. , markoff , s. , nagar , n.m . , barthel , p.d . , ho , l.d . , and yuan , f. 2004 , a & a , 418 , 429 forman , w. , jones , c. & tucker , w. 1985 , , 293 , 102 forman , w. , schwarz , j. , jones , c. , liller , w. , & fabian , a. 1979 , , 234 , l27 forman , w. , nulsen , p. , heinz , s. , owen , f. , eilek , j. , vikhlinin , a. , markevitch , m. , kraft , r. , churazov , e. , jones , c. 2005 , , in press , eprint astro - ph/0312576 garmire , g.p . et al . 1992 , aiaa , space programs and technologies conference , march 24 - 27 , 1992 ( paper 92 - 1473 ; new york : aiaa ) gilfanov , m. 2004 , , 349 , 146 gnedin , o.j . 2003 , , 582,141 gunn , j. , & gott , j. 1972 , , 176 , 1 heinz , s. , churazov , e. , forman , w. , jones , c. , & briel , u.g . ( 2003 ) , , 346 , 13 ho , l.c . , filippenko , a.v . & sargent , w.l.w . 1997 , , 112 , 315 irwin , j.a . & sarazin , c.l . 1996 , , 471 , 683 kenney , j. , et al . 1995 , , 438 , 135 kenney , j. , van gorkom , j.h . & vollmer , b. 2004 , astro - ph/0403103 kim , d .- w . , fabbiano , g. , & trinchieri , g. 1992 , , 80 , 645 kraft , r. , kregenow , j.m . , forman , w.r . , jones , c. , & murray , s.s . 2005 , , 560 , 675 landau , l.d . & lifshitz , e.m . 1959 , fluid mechanics ( london : pergamon ) , chapter ix lavery , r. & henry , j.p . 1988 , , 330 , 596 machacek , m. , dosaj , a. , forman , w. , jones , c. , markevitch , m. , vikhlinin , a. , warmflash , a. , & kraft , r. 2005a , , 621 , 663 machacek , m. , nulsen , p. , stirbat , l. , jones , c. & forman , w.r . 2005b , , 630 , 280 machacek , m. , nulsen , p.e.j . , jones , c. , & forman , w.r . 2006 , , submitted ( paper ii ) markevitch , m. et al . 2000 , , 541 , 542 matsushita , k 2001 , , 547 , 693 matsushita , k. , ohashi , t. & makishima , k. 2000 , pasj , 52 , 685 mazzotta , p. , markevitch , m. , vikhlinin , a. , forman , w.r.,david , l. & vanspeybroeck , l. 2001 , , 555 , 205 merrifield , m.r . 1998 , , 294 , 347 moore , b. , katz , n. , lake , g. , dressler , a. , & oemler , a. 1996 , nature , 379 , 613 moekel , w.e . 1949 , approximate method for predicting form and location of detached shock waves ahead of plane or axially symmetric bodies , naca technical note 1921 , available online at http://naca.larc.nasa.gov/reports/1949/naca-tn-1921 morrison , r. & mccammon , d. 1983 , , 270 , 119 mller , m. , mair , g. , & hillebrandt , w. 1989 , , 216 , 19 nulsen , p.e.j . 1982 , , 198 , 1007 nulsen , p.e.j . , mcnamara , b.r . , wise , m.w . , & david , l.p . 2005 , , 628 , 629 ohashi , t. et al . in iau symposium 188 , the hot universe , ed . k koyama , s. kitamoto , & m. itoh ( dordrecht : kluwer ) , 317 osullivan , e. , forbes , d.a . & ponman , t. j. 2001 , mnras , 328 , 461 plucinsky , p.p . , schulz , n.s . , marshall , h.l . , grant , c.e . , chartas , g. , sanwal , d. , teter , m. , vikhlinin , a.a . , edgar , r.j . , wise , m.w . , allen , g.e . , virani , s.n . , depasquale , j.m . & raley , m.t . 2003 , in proceedings of the spie x - ray and gamma - ray telescopes and instruments for astronomy , ed . trluemper & h. tannenbaum , proceedings of the spie , 4851 , 89 quilis , v. , moore , b. , & bower , r. 2000 , science,288 , 1617 rangarajan , f.v.n . , white , d.a . , ebeling , h. & fabian , a. 1995 , mnras , 277 , 1047 raymond , j.c . & smith , b.w . 1977 , , 35 , 419 roberts , m. , hogg , d. , bregman , j. , forman , w.r . & jones , c. 1991 , , 75 , 751 sakelliou , i. , merrifield , m.r . , & mchardy , i.m . 1996 , , 283 , 673 scharf , c.a.,zurek , d.r . , & bureau , m. 2005 , , 633 , 154 schindler , s. , binggeli , b. , bhringer , h. 1999 , a & a , 343 , 420 shibata , r. , matsushita , k. , yamasaki , n.y . , ohashi , t. , ishida , m. , kikuchi , k , bhringer , h. , & matsumoto , h. 2001 , apj , 549 , 228 smith , r.j . , lucey , j.r . , hudson , m.j . , schlegel , d.j . & davies , r.l . 2000 , , 313 , 469 smith , r.k . , brickhouse , n.s . , liedahl , d. a. & raymond , j.d . 2001 , , 556 , l91 stevens , i.r . , acreman , d. & ponman , t.j . 1999 , , 310 , 663 strickland , d , heckman , t. , weaver , k. , & dahlem , m. 2000 , , 120 , 2965 sun , m. & vikhlinin , a. 2005 , , 621 , 718 sun , m. , jerius , d. & jones , c. 2005 , , submitted , eprint astroph/0501501 takano , s. et al . , 1989 , nature , 340 , 289 toniazzo , t. , & schindler , s. 2001 , , 325 , 509 tonry , j.l . , dressler , a. , blakeslee , j.p . , ajhar , e.a . , fletcher , a.b . , luppino , g.a . metzger , m.r . & moore , c.b . 2001 , , 546 , 681 tucker , w. & gould , r. 1966 , , 144 , 244 vikhlinin , a. , markevitch , m. & murray , s.s . 2001 , , 551 , 160 vollmer , b. 2003 , , 398 , 525 wang , q.d . , owen , f. , ledlow , m. , & keel , w. 2004 , in proc . iau colloq . 195 , outskirts of galaxy clusters : intense life in the suburbs , ed . a. diaferio ( cambridge : cambridge univ . press ) , 78 white , d. , fabian , a. , forman , w. , jones , c. & stern , c. 1991 , , 375 , 35 ccccc vn & rectangular & @xmath272 , @xmath273 & @xmath274,@xmath275 & @xmath276 + ve & circular & @xmath277 , @xmath278 & @xmath279 & + ng & rectangular & @xmath280 , @xmath281 & @xmath221 , @xmath282 & @xmath283 + sg & rectangular & @xmath280 , @xmath284 & @xmath285 , @xmath286 & @xmath283 + oa & annular & @xmath287 , @xmath288 & @xmath289 , @xmath290 & + sl & circular & @xmath291 , @xmath292 & @xmath293 & + hw & elliptical & @xmath294 , @xmath295 & @xmath296 , @xmath297 & @xmath298 + he & elliptical & @xmath299 , @xmath300 & @xmath301 , @xmath18 & @xmath302 + t & rectangular & @xmath299 , @xmath303 & @xmath304 , @xmath305 & @xmath306 + cccc he_bg & circular & @xmath307 , @xmath308 & @xmath309 + hw_bg & circular & @xmath307 , @xmath308 & @xmath309 + sl_bg & elliptical & @xmath280 , @xmath281 & @xmath310 , @xmath63 + t_bg & circular & @xmath311 , @xmath312 & @xmath313 + ccccccc he & @xmath314 & @xmath315 & he_bg & @xmath316 & @xmath317 & @xmath318 + hw & @xmath319 & @xmath320 & hw_bg & @xmath316 & @xmath317 & @xmath321 + sl & @xmath322 & @xmath323 & sl_bg & @xmath324 & @xmath325 & @xmath326 + t & @xmath327 & @xmath328 & t_bg & @xmath329 & @xmath330 & @xmath331 + ccccc vn & @xmath332 & @xmath81 & @xmath333 & @xmath334 + ng & @xmath335 & @xmath336 & @xmath337 & @xmath338 + sg & @xmath339 & @xmath340 & @xmath341 & @xmath342 + oa & @xmath343 & @xmath344 & @xmath345 & @xmath346 + sl & @xmath347 & @xmath348 & @xmath349 & @xmath350 + sl@xmath351 & @xmath347 & @xmath352 & @xmath353 & @xmath354 + he+hw & @xmath355 & @xmath356 & @xmath357&@xmath358 + he & @xmath359 & @xmath360 & @xmath361 & @xmath362 + hw & @xmath363 & @xmath364 & @xmath361 & @xmath365 + t & @xmath366 & @xmath228 & @xmath367 & @xmath368 + t & @xmath369 & @xmath228 & @xmath370 & @xmath371 + cccccc vn & & & & & + & @xmath372 & @xmath373 & @xmath374 & @xmath375 & @xmath376 + & @xmath134 & @xmath377 & @xmath378 & @xmath379 & @xmath376 + & @xmath380 & @xmath381 & @xmath382 & @xmath383 & @xmath384 + oa & & & & & + & @xmath134 & @xmath385 & @xmath310 & @xmath386 & @xmath387 + & @xmath388 & @xmath389 & @xmath390 & @xmath55 & @xmath391 + & @xmath386 & @xmath97 & @xmath392 & @xmath393 & @xmath394 + & @xmath395 & @xmath396 & @xmath397 & @xmath398 & @xmath399 + & @xmath400 & @xmath401 & @xmath402 & @xmath105 & @xmath403 + & @xmath393 & @xmath404 & @xmath405 & @xmath406 & @xmath407 + t & & & & & + & @xmath134 & @xmath408 & @xmath409 & @xmath395 & @xmath410 + & @xmath388 & @xmath228 & @xmath411 & @xmath393 & @xmath371 + & @xmath386 & @xmath412 & @xmath242 & @xmath398 & @xmath413 + & @xmath395 & @xmath414 & @xmath415 & @xmath416 & @xmath417 + & @xmath400 & @xmath418 & @xmath105 & @xmath419 & @xmath420 + & @xmath393 & @xmath421 & @xmath422 & @xmath138 & @xmath423 + he+hw & & & & & + & @xmath372 & @xmath424 & @xmath425 & @xmath380 & @xmath426 + & @xmath134 & @xmath427 & @xmath428 & @xmath386 & @xmath429 + & @xmath380 & @xmath430 & @xmath242 & @xmath431 & @xmath432 + & @xmath388 & @xmath430 & @xmath433 & @xmath55 & @xmath434 + he & & & & & + & @xmath372 & @xmath435 & @xmath2 & @xmath380 & @xmath436 + & @xmath134 & @xmath360 & @xmath415 & @xmath386 & @xmath362 + & @xmath380 & @xmath437 & @xmath43 & @xmath395 & @xmath438 + & @xmath388 & @xmath439 & @xmath398 & @xmath400 & @xmath440 + & & & & & + hw & & & & & + & @xmath372 & @xmath441 & @xmath243 & @xmath388 & @xmath442 + & @xmath134 & @xmath364 & @xmath43 & @xmath395 & @xmath365 + & @xmath380 & @xmath443 & @xmath144 & @xmath400 & @xmath444 + & @xmath388 & @xmath443 & @xmath445 & @xmath393 & @xmath446 + cccccccc @xmath134 & @xmath447 & @xmath134 & @xmath448 & @xmath449 & @xmath450 & @xmath451 & @xmath301 + @xmath388 & @xmath282 & @xmath134 & @xmath452 & @xmath243 & @xmath453 & @xmath454 & @xmath455 + @xmath386 & @xmath456 & @xmath134 & @xmath457 & @xmath138 & @xmath458 & @xmath459 & @xmath460 + @xmath395 & @xmath460 & @xmath134 & @xmath461 & @xmath415 & @xmath462 & @xmath463 & @xmath290 + @xmath400 & @xmath464 & @xmath134 & @xmath465 & @xmath433 & @xmath466 & @xmath467 & @xmath468 + @xmath393 & @xmath301 & @xmath134 & @xmath405 & @xmath406 & @xmath469 & @xmath470 & @xmath282 +

it has long been conjectured , on both theoretical and observational grounds , that the outermost regions of the solar system may contain a large population of small icy bodies . for example , on the basis of theoretical considerations of the genesis of the planetary system from the primordial solar nebula , @xcite suggested that a remnant population of icy planetesimals left over from the planet - building process may exist at the current epoch beyond the orbit of neptune . @xcite and @xcite speculated on a massive comet belt as the source of unexplained perturbations of neptune s orbit [ although this argument must now be discarded as the post-_voyager _ revisions in the planetary ephemeris no longer show any unexplained residuals in neptune s motion ( @xcite ) ] . @xcite analyzed the orbital plane perturbations of comet p / halley and concluded that any comet belt between 40 au and 50 au has a total mass less than @xmath1 . this limit still allows for very large numbers perhaps @xmath2 of cometary bodies . more recently , it has been suggested that the observed short period comets with orbital periods @xmath3 yr the `` jupiter family '' comets originate in a belt of low - inclination bodies just beyond the orbit of neptune , between 35 au and 50 au [ @xcite ] . the older hypothesis that short period comets originate in a population of near - parabolic oort cloud comets which are perturbed into shorter orbits by the giant planets appears unlikely : @xcite have shown that the orbital element distribution of the observed short period comets is inconsistent with a source in the nearly isotropic oort cloud but is compatible with a disk - like source in a trans - neptune comet belt , which they call the `` kuiper belt '' . a possible member of the kuiper belt was first discovered in 1992 at a distance of 41 au from the sun [ 1992 qb@xmath4 , reported in @xcite ] , and several additional discoveries have been reported since suggesting a potential population of @xmath5 bodies larger than @xmath6 km [ @xcite ] . the dynamical structure of this population has been the subject of several recent theoretical studies [ @xcite ] . the first two of these [ @xcite and @xcite ] , studied the long term stability of test particles in low - eccentricity and low - inclination orbits beyond neptune , subject only to the gravitational perturbations of the giant planets in their present orbital configuration . they found orbital instability on short timescales ( @xmath7 yr ) interior to 33 - 34 au , an intricate structure of interspersed regions of stability and instability in the semimajor axis range of 34 au to 43 au , and substantially stable orbits beyond 43 au . the intricacy of the dynamical structure appears to be particularly acute near the locations of orbital resonances with neptune . any primordial trans - neptune population of planetesimals was undoubtedly subject to dynamical evolution during the planet formation process , and the initial conditions assumed in the above studies are not necessarily representative of the state of the kuiper belt at the end of planet formation , as acknowledged in @xcite . the nature of the `` dynamical sculpting '' of the kuiper belt that would have occurred during the late stages of planet formation was the subject of a study by @xcite . this study concludes that the giant planets orbits would have evolved significantly and the kuiper belt would have been sculpted into a highly non - uniform distribution early in solar system history : most of the primordial small bodies in the region beyond neptune s orbit and up to approximately 50 au heliocentric distance would have been swept into narrow regions of orbital resonances with neptune , particularly the 3:2 and the 2:1 orbital resonances which are located at semimajor axes of approximately 39.4 au and 47.8 au , respectively . the orbital inclinations of most of these objects would remain low ( typically @xmath8 ) , but the eccentricities would be excited to significant values , typically @xmath9 . this structure would be largely preserved to the present epoch . pluto ( and its satellite charon ) have long been known to reside in the 3:2 neptune resonance ; the resonance libration protects this pair from close encounters with neptune , despite their neptune - crossing orbit ( with eccentricity @xmath10 ) . several of the newly - discovered kuiper belt objects are also likely residents of this resonance [ @xcite ] . ( however , note that the orbital parameters of most of these objects remain rather poorly constrained pending future follow - up observations . ) in any case , it is clear that the locations of neptune s orbital resonances are particularly interesting with regard to the long term dynamics of the kuiper belt . in the region between neptune s orbit and approximately 50 au heliocentric distance , by far the most important orbital perturbations are due to neptune alone , although the n - body effects of all the planets particularly secular resonance effects also surely play a significant role over the age of the solar system . in this paper , i use the simplest possible dynamical model namely , the circular , planar restricted three - body model to determine the basic phase space structure near the locations of neptune s orbital resonances exterior to neptune s orbit . this model does not reflect the complexities that arise with ( i ) non - zero orbital inclination of the test particle , ( ii ) a realistic representation of neptune s motion ( its time - varying orbit ) , and ( iii ) the perturbations of the other planets . some of these unmodeled effects can lead to important variations in the dynamics . in particular , as mentioned above , secular resonances in certain regions of phase space can drive the inclinations and eccentricities to large amplitudes and further complicate the dynamics near mean motion resonances . nevertheless , this model is a reasonable first approximation because ( a ) the evidence from short period comets suggests a low - inclination source population , ( b ) in the trans - neptunian space , there are no first order orbital resonances with any other giant planets , and ( c ) neptune s eccentricity and inclination do not exceed @xmath11 and @xmath12 , respectively , over billion year timescales under the effects of the mutual gravitational perturbations of the planets . the advantage of this simple model is that it is tractable : it allows for the calculation of 2-dimensional surfaces - of - section in which the phase space structure in particular , the properties of resonance regions can be easily visualized . this provides for relatively straightforward estimation of the widths of resonances and the determination of several interesting dynamical properties of resonant orbits , such as the libration periods and their dependence on orbital eccentricity and libration amplitude . to a considerable degree , most of these properties are preserved in the more realistic situation . in short , this model provides a reasonable baseline " for the phase space structure in the trans - neptune region . the rest of this paper is organized as follows . section 2 describes the technical details of the planar , circular restricted 3-body model for the sun , neptune and a kuiper belt object , including a description of the information that may be gleaned from surfaces - of - section . section 3 provides a detailed look at the phase space structure near the major neptune resonances of interest for the kuiper belt . section 4 summarizes and discusses the results . the classical planar circular restricted three - body problem is a particular case of the general gravitational three - body problem of masses @xmath13 defined by the following restrictions : 1 . the motion of all three bodies takes place in a common plane ; 2 . the third body , @xmath14 , has zero mass ; therefore it does not influence the motion of @xmath15 and @xmath16 ; and 3 . the masses @xmath15 and @xmath16 describe circular orbits about their common center of mass . in the context of this paper , @xmath15 represents the sun and @xmath16 represents neptune . the system is made non - dimensional by the following choice of units : * the unit of mass is taken to be @xmath17 ; * the unit of length is chosen to be , @xmath18 , the constant separation between @xmath15 and @xmath16 ( the mean heliocentric distance of neptune ) ; * the unit of time is chosen such that the orbital period of @xmath15 and @xmath16 about their center of mass is @xmath19 . then the universal constant of gravitation , @xmath20 , and the masses of the sun and neptune are @xmath21 and @xmath22 , respectively , where @xmath23 . for the sun - neptune system , we have @xmath24 , and this value is used throughout this paper . in a reference frame with axes @xmath25 rotating with @xmath15 and @xmath16 and with origin at their center - of - mass , the sun and neptune have fixed coordinates , @xmath26 and @xmath27 , respectively , and the third ( massless ) body has the following equations of motion : @xmath28 where @xmath29 and @xmath30 are the test particle s distance to the sun and to neptune , respectively : @xmath31^{1/2 } , \nonumber\\ & & \\ r_2 & = & [ ( x-1+\mu)^2+y^2]^{1/2}. \nonumber\end{aligned}\ ] ] these equations of motion admit a constant of the motion , the _ jacobi integral _ , given by @xmath32 no other constant of the motion is known . for future reference , we note that the jacobi integral can be expressed in terms of the test particle s semimajor axis , @xmath33 , and eccentricity , @xmath34 : @xmath35 _ surfaces - of - section _ the motion of the test particle takes place on a three - dimensional subspace ( defined by a particular value of @xmath36 ) embedded in the four - dimensional phase space , @xmath37 . the usual two - dimensional surface - of - section ( s - o - s ) is defined by @xmath38 and the coordinates on the section are @xmath39 . the geometrical interpretation is straightforward : we plot the x coordinate and velocity of the test particle at every conjunction with neptune . for orbits exterior to neptune ( strictly speaking , with mean motion smaller than neptune s ) , this occurs every time the test particle is aligned with the sun and neptune and is on the _ opposite _ side of the sun from neptune ; for orbits interior to neptune ( with mean motion larger than neptune s ) , this occurs every time the test particle is aligned with the sun and neptune and is on the _ same _ side of the sun as neptune . for the solar system beyond neptune , only the exterior orbits are of interest . in the s - o - s so defined , _ periodic _ orbits of the test particle appear as discrete points . the successive crossings of this surface by a _ quasiperiodic _ orbit lie on a closed smooth curve , while a _ chaotic _ orbit fills a two - dimensional area . as the regions in the neighborhood of orbital resonances are of particular interest here , the following remarks are in order . * the location of a @xmath40 orbital period resonance is defined by kepler s relation between the orbital period and the semimajor axis : @xmath41 where @xmath42 and @xmath43 are integers . * in general , at any resonance , one finds a family of stable periodic orbits that can be parametrized by the orbital eccentricity ( or , equivalently , the jacobi integral , if we use eqn . [ eq : ares ] in eqn . [ eq : c ] ) such that the semimajor axis equals @xmath44 for all orbits in this family , and the eccentricity ranges from zero to some maximum value , @xmath45 ; for @xmath46 the periodic orbit becomes unstable . * in some neighborhood , @xmath47 of @xmath44 , near each resonance , there exist stable quasiperiodic orbits that librate with finite amplitude about the exact resonant orbit ; the half - width @xmath48 depends upon the mass parameter @xmath22 , the resonance itself ( @xmath44 ) and the jacobi integral @xmath36 . let us consider the neptune - pluto 3:2 resonance as an illustrative example . if pluto s inclination is neglected , and if one assumes a fixed circular orbit for neptune , then the jacobi integral @xmath49 for the sun+neptune+pluto restricted three - body problem . figure 1(a ) is a s - o - s in the neighborhood of the 3:2 neptune resonance for this value of @xmath36 . this section shows examples of all three types of orbits . the center of the smooth curves lying near @xmath50 corresponds to the periodic orbit whose period is exactly 3/2 that of neptune . the smooth curves are quasiperiodic orbits surrounding this exact periodic orbit and represent test particle orbits that librate about the exact resonance ; the size of a smooth curve bears a direct relationship to the amplitude of libration . such librating orbits are phase - protected from having close encounters with neptune even if the orbit is neptune - crossing . it can be seen that , in general , the regions of quasiperiodic orbits are surrounded by chaotic orbits , beyond a certain libration amplitude , the smooth curves dissolve into a chaotic zone . embedded within the `` chaotic sea '' are other stable zones or chains - of - islands that represent other nearby higher order and secondary resonances . the origin of the chaos is due to the overlap of the higher - order and secondary resonances . it is perhaps more meaningful to visualize the resonance zone in variables that can be more directly translated into `` resonance width '' . the latter is traditionally ( and somewhat loosely ) defined as the range in semimajor axis where the orbital perturbations are large . in the restricted three - body problem with @xmath51 , this is the range in which the amplitude of the orbital parameter variations of the test particle is not simply linearly proportional to @xmath22 , but is @xmath52 with @xmath53 . in general , in the neighborhood of any resonance , we find regions of stable , librating orbits surrounded by a substantial chaotic zone where test particles may suffer close encounters with neptune . in this paper , i will define the resonance half - width @xmath48 as the maximum amplitude of the semimajor axis variations of _ stably librating _ resonant orbits . the value of @xmath48 will be estimated from the s - o - s as described below . the phase space structure seen in the @xmath39 plane can also be seen in other ( generalized ) canonical coordinates . for example , in plane polar coordinates the equivalent s - o - s @xmath54 in the inertial reference frame is @xmath55 with the same section condition as in eqn . [ eq : sos ] . the s - o - s condition can be rewritten in terms of polar coordinates as follows : @xmath56 where @xmath57 is the longitude of neptune , and @xmath58 is the mean motion of the test particle . in other words , the @xmath55 plane is equivalent to the @xmath39 plane with points plotted at every conjunction of the test particle with neptune . the @xmath55 s - o - s is shown in figure 1(b ) . carrying this one step further by means of a canonical transformation to delaunay variables , @xmath59 ) where @xmath60 are the mean anomaly and mean longitude and @xmath61 ( cf . chapter xvii in @xcite and chapter 10 in @xcite ) , this same section can also be seen in the @xmath62 plane . the @xmath62 s - o - s is shown in figure 1(c ) . it is evident from the latter figure that for pluto - like resonance - locked orbits ( for the particular value of @xmath36 chosen in this s - o - s ) the mean anomaly at conjunctions with neptune librates about @xmath63 and its maximum libration amplitude is @xmath64 . orbits with larger amplitudes are chaotic . in addition to the maximum amplitude of @xmath65 , the resonance libration region is bounded in @xmath66 as well . now , if we note that ( i ) the jacobi integral provides a relationship between @xmath33 and @xmath34 ( see eqn . ( [ eq : c ] ) ) , and ( ii ) for a particular orbital resonance defined by the ratio @xmath40 of the orbital periods of the test particle and neptune , the librating orbits all have a phase - averaged semimajor axis @xmath67 , then any s - o - s can be labeled uniquely either by the value of @xmath36 or , in the neighborhood of a particular orbital resonance , by the value of @xmath68 ( with the understanding that @xmath67 ) . for example , for the s - o - s in figure 1 , we have @xmath49 ; near the 3:2 resonance , @xmath69 ; therefore , from eqn . ( [ eq : c ] ) we have @xmath70 . thus , the phase space structure in the neighborhood of a particular orbital resonance can be systematically studied in a set of surfaces - of - section where all of them have a common value of @xmath71 but various values of @xmath68 ( hence different values of @xmath36 ) . this allows a visualization of the phase space near an orbital resonance as a function of the mean orbital eccentricity . finally , in figure 1(d ) i have plotted the values of @xmath33 against @xmath65 for the same points represented in figures 1(a c ) . _ note that this is not a surface - of - section . _ one might call it a `` pseudo - surface - of - section '' , for @xmath33 is not a canonical variable . however , the constancy of @xmath36 imposes a relationship between @xmath33 and @xmath34 ( cf . [ eq : c ] ) which ensures that this plot looks very similar to the @xmath62 s - o - s in figure 1(b ) and the stable libration regions and the chaotic regions are easily distinguished . the reason for choosing this non - canonical variable is that the resonance width @xmath48 of the stable libration zone can be readily determined from this figure . therefore , the @xmath72 plane the pseudo - surface - of - section will be used in the next section for a systematic look at the properties of the phase space in the neighborhood of several neptune resonances . the libration timescale in the libration zone and the timescale for orbital instability in the chaotic zone are of particular interest for the dynamics of kuiper belt objects . these libration timescales have been determined for individual resonances and are also given in the next section . typically , the timescale for orbital instability inside the chaotic zone surrounding a resonance is only a few times the libration period . however , close to the boundary between the libration zone and the chaotic zone , this timescale can become exceedingly large . in particular , for pluto - like orbits with eccentricity @xmath10 in the 3:2 neptune resonance , the timescale for instability in the chaotic zone ( for amplitude of the resonance angle greater than @xmath73 ) is @xmath74 yr , but close to the edge of the libration zone this timescale increases by an order of magnitude or more . this is illustrated in figure 2 . in principle , there is an infinite number of orbital resonances in the three - body problem . however , we will see that only a handful warrant any detailed investigation . low - eccentricity orbits in the immediate neighborhood of a planet are unstable . there is a simple explanation for this orbital instability that follows from chirikov s ( 1979 ) `` resonance overlap criterion '' : first order @xmath75 resonances with @xmath76 overlap completely [ @xcite ] . [ numerical experiments by @xcite show that the numerical coefficient on the right hand side in this relation must be revised to 0.44 ] . complete overlap of neighboring resonances results in the destabilization of all periodic and quasi - periodic orbits in the neighborhood of those resonances . in other words , circular test particle orbits are unstable in the immediate neighborhood @xmath77 of a planet s orbit . for neptune ( @xmath24 ) , this criterion shows that all first order resonances with @xmath78 are completely overlapping , so that circular orbits beyond neptune with orbital radii less than about 33 au are unstable . therefore , beyond the semimajor axis range of about 33 au , at most only seven first order @xmath75 resonances with @xmath79 are isolated from each other ( at least for low eccentricities ) . however , i have found numerically that low - eccentricity orbits at the 8:7 and 7:6 neptune resonances are also chaotic . this is not entirely surprising , for the resonance overlap criterion is quite approximate , and wisdom s scaling law is strictly valid only in the asymptotic limit @xmath80 . therefore , only five first order @xmath75 resonances with @xmath81 are of potential significance for the long - term storage of kuiper belt objects . furthermore , since all second and higher order resonances in the first - order resonance overlap region are also destroyed , it follows that there are only seven second order neptune resonances of potential interest : @xmath82 with @xmath83 . of these , i have found numerically that only three ( 3:1 , 5:3 , and 7:5 ) have resonance widths @xmath84 . therefore , the phase space in the neighborhood of only these three second order resonances will be discussed here . figures 310 show the properties of these first and second order neptune resonances . they are arranged in order of decreasing mean heliocentric distance . each figure has seven parts : the first four [ ( a)(d ) ] show the resonance zones in the pseudo - surfaces - of - section [ plots of @xmath72 ] for @xmath86 and 0.4 , respectively ; note that the vertical scale is the same in all these figures , so that the variations in the extent of the libration and chaotic zones with @xmath68 and with the mean distance of the resonance from neptune can be readily discerned . the next two panels [ ( e ) and ( f ) ] show examples of librating , resonance - locked orbits in the rotating reference frame . the last one , panel ( g ) , shows the period of libration as a function of the mean orbital eccentricity and the amplitude of libration of the resonance angle , @xmath87 . a few words are in order regarding the latter variable . for a @xmath40 resonance the resonance angle is defined as follows : @xmath88 where @xmath89 and @xmath90 are the mean longitude and longitude of perihelion of the test particle , and @xmath91 is the mean longitude of neptune . ( note that @xmath42 is a positive integer ; @xmath43 is a negative integer for interior resonances , positive for exterior resonances ; the latter are the only resonances of interest in this paper . ) @xmath87 is the natural variable that arises in the perturbative analysis of orbital resonances , and it can be used to make a first approximation theory that models the resonant motion of the test particle with a single degree - of - freedom pendulum - like dynamical system [ see @xcite ] . for a test particle locked in a stable exterior resonance , @xmath87 librates about a mean value which is usually @xmath63 . [ there are two exceptions to this : ( i ) for sufficiently high eccentricity , librations of the resonance angle about @xmath92 are also possible ; and ( ii ) the 2:1 and the 3:1 resonances allow for asymmetric librations where the center of libration of the resonance angle is displaced away from @xmath63 . these are discussed further below . ] at conjunctions with neptune , the resonance angle is related to the mean anomaly : @xmath93 . therefore , the librations or chaotic variations of @xmath65 evident in the surfaces - of - section are also reflected in the behavior of @xmath87 . the physical significance of the resonance angle can be seen by noting that when the test particle is at perihelion , @xmath94 , so that @xmath95 thus , the behavior of @xmath87 gives a direct measure of the longitude separation of neptune from the test particle s perihelion . this is obviously a critical quantity , especially for high - eccentricity neptune - crossing orbits , as the stable libration of @xmath87 then ensures that the particle is protected from catastrophic close encounters with neptune . the maximum excursion from the center of libration of @xmath87 then gives a measure of the smallest possible separation between neptune and the test particle s perihelion in stable resonance - protected orbits . the resonance libration period as a function of the amplitude of libration of @xmath87 is shown for various values of @xmath68 in the last panel ( g ) of figures 310 . these curves are rather counter - intuitive . recall that in the oft - used pendulum model for non - linear resonances , the pendulum libration period increases monotonically with amplitude and becomes arbitrarily large close to the separatrix ( the orbit that separates the oscillations from rotations of the pendulum ) . in contrast , in figures 310(g ) , we see that , except for the asymmetric librations at the 2:1 and 3:1 resonances , the libration period in general _ decreases _ with increasing libration amplitude . furthermore , near a @xmath43-th order resonance , the leading order ( in eccentricity ) resonance term in the perturbing potential for the test particle is proportional to the @xmath43-th power of the eccentricity , hence the small - amplitude libration period might be expected to vary as @xmath96 ; this analogy from the pendulum model does not hold either , as evident in figures 310(g ) . thus , the standard non - linear pendulum model should be used with caution for orbital resonances . the general characteristics of the phase space near these first and second order neptune resonances are summarized as follows . ( note that the 2:1 and the 3:1 resonances are exceptional in many respects , and are discussed later in detail . ) * in general , the phase space in the neighborhood of a resonance contains a zone of stably librating , resonance - locked orbits surrounded by a zone of chaotic orbits . the width of the resonance zone ( including the stable as well as the chaotic regions ) increases with @xmath68 ; however , the width of the libration zone , in general , shrinks as the resonance separatrix dissolves into a chaotic layer that increases in thickness with @xmath68 . the typical width of a neptune resonance libration zone is @xmath97 , or about @xmath98 au . * the periodic orbit at exact @xmath40 resonance has a @xmath42-fold symmetry in the rotating frame : it makes @xmath42 perihelion passages during a synodic period . the perihelion ( and aphelion ) longitudes are spaced @xmath99 . the quasi - periodic orbits librating about the exact resonance also exhibit @xmath42-fold symmetry in the rotating frame when traced over a complete libration period . for the stable librators ( resonance angle @xmath87 librating about @xmath63 ) , the opposition with neptune occurs near the test particle s aphelion ; hence these may be called `` aphelion librators '' . * for sufficiently large @xmath68 , a new type of periodic ( resonant ) orbit is possible , where one of the @xmath42 perihelion passages during a synodic period occurs near neptune . in fact , the perihelion loop encompasses neptune . for this orbit and for the quasi - periodic orbits in its neighborhood , the resonance angle @xmath87 librates about @xmath92 . these orbits may be called `` perihelion librators '' . however , these perihelion librators are probably not of practical importance , as unmodeled perturbations are likely to destabilize them . * in general , the libration periods are smaller for resonances located closer to neptune , and the libration periods decrease with libration amplitude . typical small amplitude libration periods are ( 1 - 2)@xmath100 yr . the exceptional cases of the 2:1 and the 3:1 resonances are discussed below . _ the 2:1 neptune resonance _ the synodic period of a test particle near the 2:1 resonance is approximately equal to its orbital period . thus , conjunctions with neptune occur every @xmath101 years . the properties of the phase space near the 2:1 neptune resonance are summarized as follows ( cf . figure 4 ) . * the resonance half - width , @xmath48 , increases from @xmath102 to @xmath103 as @xmath68 increases from @xmath104 to @xmath105 . the phase space at this resonance is largely regular ; there is no discernible chaotic zone until the eccentricity exceeds @xmath106 . * for @xmath68 exceeding a critical value ( which is @xmath107 for the 2:1 neptune resonance ) , there are two types of librating orbits : ( i ) symmetric librators whose perihelia librate ( with large amplitude ) with a mean value @xmath63 away from neptune , and ( ii ) asymmetric librators whose perihelia librate ( with amplitude smaller than @xmath108 ) about a mean value which depends upon @xmath68 . ( surprisingly , the orbit with aphelion exactly at neptune s longitude is unstable : it coincides with the separatrix that divides the phase space into the symmetric and asymmetric libration zones . ) the existence of such asymmetric librations has been noted previously ( @xcite , @xcite ) , but the work here provides the first quantitative analysis for the specific case of neptune s resonances . * figures 4(e , f ) show examples of these two types of orbits . note that for the asymmetric librators , there are two independent centers of libration , offset equally on either side of neptune s longitude . upon inspection of the behavior of the resonance angle @xmath87 , it is found that these centers of perihelion libration ( relative to neptune ) vary from @xmath109 for @xmath110 to @xmath111 for @xmath112 . * the phase space for the symmetric librators shrinks rapidly with increasing @xmath68 , while that for the asymmetric librators increases slightly . * at small values of @xmath68 , there is no evidence for a chaotic zone near this resonance , but for @xmath113 , the largest amplitude librators become chaotic . a thin chaotic zone also appears in the neighborhood of the separatrix that separates the asymmetric librations from the symmetric librations . * the eccentricity variations of the resonance - locked orbits have amplitude @xmath114 . ( indeed , within the planar circular restricted three - body model , an initially circular test particle orbit at the 2:1 resonance has its eccentricity pumped up to only @xmath115 . ) the eccentricity variations on the chaotic orbits are generally of considerably larger magnitude . * figure 4(g ) shows libration amplitude vs. libration period . for the asymmetric ( small amplitude ) librators , the libration period decreases monotonically with @xmath68 and increases with libration amplitude ( reminiscent of the standard non - linear pendulum ) . the libration period becomes indefinitely large at the separatrix between the asymmetric and symmetric librators . the small amplitude libration periods are several tens - of - thousands of years long . _ the 3:1 neptune resonance _ the synodic period of a test particle near the 3:1 resonance is half its orbital period . thus , conjunctions with neptune occur approximately every 248 years . the properties of the phase space near the 3:1 neptune resonance ( cf . figure 3 ) are very similar to that near the 2:1 resonance ; some specifics are noted below . * the resonance half - width , @xmath48 , increases from @xmath116 to @xmath117 as @xmath68 increases from @xmath104 to @xmath105 . the phase space at this resonance is regular ; there is no discernible chaotic zone for this range of eccentricities . * for librating orbits , perihelion and aphelion occur at approximately the same longitude relative to neptune . as in the 2:1 resonance , there are two types of librating orbits the symmetric and asymmetric librators ; however , the asymmetric librations appear only at @xmath68 exceeding @xmath118 . * the small amplitude libration periods are very long ( @xmath119 myr ) for low eccentricity @xmath120 orbits , but considerably shorter ( 30,00080,000 yr ) for higher eccentricities . numerical studies of the stability of low eccentricity , low inclination orbits of small objects in the trans - neptunian kuiper belt subject to the gravitational perturbations of the giant planets have shown that the inner edge of the kuiper belt is at about 34 au heliocentric distance ; beyond 34 au and up to about 42 au there are interspersed regions of stability and instability [ @xcite , @xcite ] . this structure bears a complex correlation with the locations of neptune orbital resonances . other studies [ @xcite , @xcite ] suggest that early in the history of the solar system , the majority of kuiper belt objects were swept into eccentric orbits in narrow zones located at neptune s orbital resonances , and that the regions in - between the resonances would have been largely cleared of residual planetesimals . recent and ongoing observational surveys of the outer solar system indicate the presence of a large population of small bodies beyond neptune [ @xcite ] . these are quite likely the source of short period comets [ @xcite , @xcite ] . furthermore , their orbital distribution is likely to hold clues to the formation and early dynamical evolution of the outer solar system [ @xcite ] . all of these considerations have motivated the present study . using the planar circular restricted three body model ( with the sun , neptune and a test particle ) , i have described in this paper the basic phase space structure in the neighborhood of neptune s exterior orbital resonances . the details may be gleaned from figures 310 . a succinct summary of these results is given in figure 11 which shows the locations and widths of the stable resonance libration zones . in general , these stable zones are bounded by chaotic layers of thickness that generally increases with eccentricity and decreases with distance from neptune . ( as the mean semimajor axis and eccentricity of chaotic orbits is not well defined , it is not possible to represent the chaotic layers in such a figure . ) the planar circular restricted three body model is the simplest dynamical model for the orbital dynamics of small objects in the kuiper belt . although this model may appear oversimplified , it provides a reasonable description and explanation for much of the dynamical behaviors found in the numerical studies mentioned above . for example , the location of the inner edge of this belt at @xmath121(3334 ) au is readily understood in terms of orbital instability induced by overlapping first order neptune resonances ; the additional perturbations due to neptune s non - circular , inclined , and time - varying orbit and the perturbations of the other planets do not significantly change the location of this boundary . the relatively isolated first and second order neptune resonances beyond 34 au ( in semimajor axis ) provide narrow stable libration regions for the long term storage of kuiper belt objects in eccentric ( often neptune - crossing ) orbits . this paper provides first approximation estimates for the locations and widths of these regions , and for the dynamical properties of resonant orbits . it also provides a direct visualization in two - dimensional surfaces - of - section of the global phase space structure ( the chaotic and stable regions ) in the vicinity of orbital resonances a characteristic of stable resonant orbits is the libration of the perihelion about a longitude well removed from neptune s location . typical libration periods are several tens - of - thousands of years . the libration zones are generally surrounded by narrow chaotic zones where orbits are unstable on timescales of a few libration periods , or @xmath122 yr . at the boundary between the stable resonance libration zone and the chaotic zone , the timescale for orbital instability may be exceedingly long . this paper also provides a quantitative analysis of libration periods and their dependence on libration amplitudes and orbital eccentricities at all the major exterior mean motion resonances of neptune . of course , the planar circular restricted three body model does not provide a complete picture . indeed , i expect that it underestimates the extent of the chaotic zones and overestimates the sizes of the stable libration zones near neptune resonances . but perhaps the most important missing element is the effect of secular resonances [ see @xcite ] on the long term dynamics of kuiper belt objects . for example , in some regions , secular resonances produce significant inclination excitation ; in such cases , the assumption of planarity becomes a poor approximation . [ pluto s orbit is a case in point ( @xcite ) . ] in order to discern these effects , one has to build much more elaborate analytical models or resort to extensive numerical integrations . as this work was being completed , i became aware of two preprints [ @xcite and @xcite ] on the same subject . these use different approaches to the problem : the present work has focussed on the phase space near mean motion resonances with neptune only and gives results that are exact but for a highly simplified physical model ; @xcite discuss the dynamics near mean motion resonances as well as secular resonances using primarily approximate semi - analytic models ; and @xcite discuss the dynamics near mean motion resonances , secular resonances as well as non - resonant regions using a purely numerical approach . a comparison of the present work with these two preprints follows . morbidelli ( 1995 ) describe semi - analytic and numerical investigations that include the effects of all four giant planets on the dynamics of test particles in the kuiper belt . a comparison of their results for the locations and widths of mean motion resonances ( see their figure 1 ) with the estimates obtained here ( cf . figure 11 ) shows that their models significantly overestimate the sizes of the resonance libration regions . s estimates of the resonance widths were obtained by averaging the planar , circular restricted problem of the four giant planets plus a test particle over fast variables , retaining only the dependence on the resonance angle . thus , their analysis does not account for the overlapping secondary resonances at the edges of the mean motion resonance libration zones that produce a chaotic layer and cause the libration regions to shrink rapidly with increasing eccentricity . this is the explanation for the differences between their figure 1 and figure 11 here . in addition , morbidelli find a singularity in the averaged hamiltonian that causes a great increase in the resonance width near neptune - crossing values of the eccentricity . however , this is an artifact , and does not appear in the unaveraged problem , as is obvious from the surfaces - of - section shown in the present work . the technique i used here does not make any averaging approximations , but uses the full unaveraged hamiltonian for the circular planar restricted problem , albeit without the secular effects of jupiter , saturn and uranus . _ s models also show that secular effects are negligible near mean motion resonances except in the case of the 3:2 neptune resonance . in the special case of the 3:2 resonance , as mentioned above , @xcite had previously shown the existence of the @xmath123 secular resonance embedded within the mean motion resonance ; in addition , analysis of pluto s orbit had also previously revealed the existence of another resonance characterized by the libration of the argument of perihelion ( @xcite ) . both of these resonances have the effect of exciting the inclination of the test particle . these resonances are not modeled in the present work , but morbidelli describe in some detail the location and widths of both these and other secular resonances . however , note that because most regions near secular resonances are regions of orbital instability , they are not likely to be of importance as reservoirs of kuiper belt objects although they may drive their transport to small heliocentric distances at the present epoch . duncan ( 1995 ) have performed quite extensive numerical integrations of test particle orbits subject to the gravitational perturbations of the four giant planets . they describe their results in terms of the stability ( over their integration time of 14 byr ) of orbits with particular initial semimajor axes , eccentricities and inclinations . their main conclusions are consistent with the results of the present work : namely , that particles with low initial inclination ( @xmath124 ) and initial perihelion distance less than @xmath125 au are unstable on timescales short compared to the age of the solar system , except that particles librating in low order mean motion resonances with neptune remain phase - protected from close encounters with that planet . they have also analyzed the 3:2 neptune resonance in more detail for the particular case with initial eccentricities of 0.2 and found three regimes : stable orbits deep in resonance with small libration amplitudes , @xmath126 ; an intermediate region with libration amplitudes in the range @xmath127 where the timescale for instability is @xmath128 yr ; and highly unstable orbits for libration amplitudes exceeding @xmath129 with stability timescales of @xmath130 yr . similar dynamics is found for all the neptune resonances studied in this paper . in comparison , the present work has shown that the region of stability deep in resonance as well as the highly chaotic region at large amplitudes are both explained by the perturbations of neptune alone ( assumed to be on a circular orbit ) ; the intermediate regime where the timescale for instability is @xmath128 yr is very small in the restricted three body model but evidently expands greatly ( at the expense of the stable libration zone ) in the full n - body model for the outer solar system . [ in this context , we note that if malhotra s ( 1995 ) theory for the orbital distribution of kuiper belt objects is correct , then this intermediate libration amplitude regime which allows a long - term leakage of resonant objects may be the primary source of the jupiter - family short - period comets . ] duncan also find that the resonance - protection fails only at inclinations exceeding @xmath131 . thus , we conclude that the planar circular restricted three body model used in the present work provides a fairly good description of these main trends . a clear advantage of this simple model is that its analysis is quite `` inexpensive '' compared to the months of cpu time expended in the numerical simulations by duncan furthermore , the numerical integration of orbits for the age of the solar system at sufficiently high resolution in the space of initial conditions remains beyond the reach of present - day computers , as acknowledged by duncan thus , the global picture of the phase space structure obtained in the present work provides an inexpensive yet reasonably good baseline for understanding the dynamics in the trans - neptunian solar system , and is particularly useful near mean motion resonances . this research was done while the author was a staff scientist at the lunar and planetary institute which is operated by the universities space research association under contract no . nasw-4574 with the national aeronautics and space administration . partial support for this work was also provided by nasa s origins of solar systems research program under grant no . this paper is lunar and planetary institute contribution 870 . 1 the phase space structure in the neighborhood of the neptune - pluto 3:2 orbital resonance determined from the circular planar restricted three - body model . the same surface - of - section is shown in different variables : ( a ) cartesian variables , @xmath39 ; ( b ) plane polar variables , @xmath55 ; ( c ) delaunay variables ( mean anomaly @xmath65 , canonical momentum @xmath132 ) . ( d ) is a pseudo - surface - of - section since the semimajor axis , @xmath33 , is not a canonical variable . however , qualitatively it is very similar to the surface - of - section ( c ) , and it provides a visualization of the resonance width in terms of the semimajor axis and the amplitude of the perihelion libration . 2 the time variation of the semimajor axis for orbits with @xmath133 near the boundary between the stable libration zone and the chaotic zone at the 3:2 neptune resonance ( cf . figure 1 ) . ( the unit of length is neptune s semimajor axis , and the unit of time is neptune s orbital period , approximately 165 yr . ) the jacobi integral for all four orbits is @xmath49 ; the initial conditions @xmath134 for the four orbits are ( a)(-1.6100,0.0,0.0,0.92442 ) , ( b)(-1.6105,0.0,0.0,0.92508 ) , ( c)(-1.6106,0.0,0.0,0.92521 ) , ( d)(-1.6110,0.0,0.0,0.92574 ) . these orbits are all chaotic and are shown in order of decreasing distance to the stability boundary . the orbits in ( a ) , ( b ) and ( c ) become unstable after 7 , 31 and @xmath135 librations ( @xmath136 , @xmath137 and @xmath138 yr ) , respectively . fig . 3 the 3:1 neptune resonance located at @xmath139 ( 62.6 au ) . the phase space for orbits with ( a ) @xmath140 ; ( b ) @xmath141 ; ( c ) @xmath142 ; ( d ) @xmath143 . the phase space is largely regular at this resonance . the width of the libration zone increases in with @xmath68 . for @xmath144 , the libration zone splits into two zones of asymmetric librations surrounded by a narrow zone of large - amplitude symmetric librations . panel ( e ) shows a symmetrically librating orbit ( with @xmath145 ) in the rotating frame ; panel ( f ) shows an asymmetrically librating orbit ( with @xmath146 ) . the symbols @xmath147 and @xmath148 indicate the fixed locations of the sun and neptune in the rotating frame . panel ( g ) shows the libration period ( in units of neptune s orbital period ) as a function of the maximum excursion of the resonance angle @xmath87 from @xmath63 . note that the center of the asymmetric librators moves away from `` @xmath149 '' with increasing @xmath68 , as indicated by the beginning of the curves labeled by the different values of @xmath68 . the libration period increases with libration amplitude for the asymmetric librators , but decreases with amplitude for the symmetric ( large amplitude ) librators ; at the separatrix on the boundary between the symmetric and asymmetric librations , the libration period becomes indefinitely large ( although this is not quite obvious due to the finite resolution in this figure ) . 4 the 2:1 neptune resonance located at @xmath150 ( 47.8 au ) . the phase space for orbits with ( a ) @xmath151 ; ( b ) @xmath152 ; ( c ) @xmath153 ; ( d ) @xmath154 . the phase space is largely regular at this resonance ; a chaotic zone appears at large amplitude librations for eccentricity exceeding @xmath106 . the width of the libration zone increases in with @xmath68 . for @xmath155 , the libration zone splits into two zones of asymmetric librations surrounded by a narrow zone of large - amplitude symmetric librations . panels ( e ) and ( f ) show a small - amplitude asymmetrically librating orbit and a large amplitude symmetrically librating orbit ( both with @xmath156 ) in the rotating frame . ( the motion of the test particle over a complete libration period is traced in these figures . ) note that in both types of orbits the perihelion occurs away from neptune s longitude . panel ( g ) shows the libration period ( in units of neptune s orbital period ) as a function of the maximum excursion of the resonance angle @xmath87 from @xmath63 . the center of the asymmetric librators is indicated by the beginning of the curves labeled by the different values of @xmath68 . the libration period increases with libration amplitude for the asymmetric librators , but decreases with amplitude for the symmetric ( large amplitude ) librators ; at the separatrix on the boundary between the symmetric and asymmetric librations , the libration period becomes indefinitely large . 5 the 5:3 neptune resonance located at @xmath157 ( 42.3 au ) . the phase space for orbits with ( a ) @xmath158 ; ( b ) @xmath159 ; ( c ) @xmath160 ; ( d ) @xmath161 . the width of the libration zone increases for @xmath68 up to @xmath162 , but then shrinks with increasing @xmath68 as the larger amplitude resonant orbits dissolve into the chaotic zone . note that for @xmath163 , the formerly period-2 exact resonant orbit in the surface - of - section becomes a period-3 orbit ; this is merely a reflection of the fact that at large , neptune - crossing eccentricities the perihelion loop results in two conjunctions with neptune in rapid succession [ cf . panel ( e ) and ( f ) ] . panels ( e ) and ( f ) show the periodic orbit at exact resonance , and an orbit librating with finite amplitude , respectively ( both with @xmath156 ) in the rotating frame . ( the motion of the test particle over a complete libration period is traced in these figures . ) note that in both types of orbits the perihelion occurs away from neptune s longitude . panel ( g ) shows the libration period ( in units of neptune s orbital period ) as a function of the libration amplitude of the resonance angle @xmath87 . the libration period increases with libration amplitude only for small @xmath68 ; it decreases with amplitude for larger @xmath68 . 6 the 3:2 neptune resonance located at @xmath164 ( 39.4 au ) . the phase space for orbits with ( a ) @xmath165 ; ( b ) @xmath166 ; ( c ) @xmath167 ; ( d ) @xmath168 . the width of the main libration zone decreases with increasing @xmath68 , as the chaotic zone surrounding the libration region expands . panel ( e ) shows an example of a stably librating orbit ( with @xmath156 ) in the rotating frame ; note that this is an `` aphelion librator '' : the test particle is near aphelion when passing neptune s longitude , and its perihelion librates about @xmath169 away from neptune . for @xmath163 , a new libration zone appears in which the librating orbits have a perihelion near neptune . this is indicated by the appearance of a chain of two libration islands ( distinct from the main resonance zone ) in panels ( c)(d ) ; these are the `` perihelion librators '' [ cf . panel(f ) ] ; the new libration zone increases with increasing @xmath68 . an example of a perihelion librator is shown in panel ( f ) ; note that one of the two perihelion loops in each synodic period encompasses neptune . panel ( g ) shows the libration period ( in units of neptune s orbital period ) of the aphelion librators as a function of the libration amplitude of the resonance angle @xmath87 . the libration period generally decreases with libration amplitude , except for very small eccentricity ( @xmath170 ) , large - amplitude librators . 7 the 7:5 neptune resonance located at @xmath171 ( 37.7 au ) . the phase space for orbits with ( a ) @xmath172 ; ( b ) @xmath173 ; ( c ) @xmath174 ; ( d ) @xmath175 . the width of the main libration zone ( for aphelion librators ) decreases rapidly with increasing @xmath68 , as the chaotic zone surrounding the libration region expands . in fact , for @xmath176 [ cf . panel ( d ) ] the aphelion libration zone has completely disappeared . a libration zone for perihelion librators appears at @xmath156 and expands slightly with increasing @xmath68 . panels ( e ) and ( f ) show examples of an aphelion librator with @xmath146 , and a perihelion librator with @xmath156 , respectively . panel ( g ) shows the libration period ( in units of neptune s orbital period ) of aphelion librators as a function of the libration amplitude of the resonance angle @xmath87 . the libration period generally decreases with libration amplitude , except for very small eccentricity ( @xmath170 ) , large - ampitude librators . 8 the neighborhood of the neptune 4:3 orbital resonance located at @xmath177 ( 36.5 au ) . the phase space for orbits with ( a ) @xmath178 ; ( b ) @xmath179 ; ( c ) @xmath180 ; ( d ) @xmath181 . the libration zone shrinks rapidly with @xmath68 as the larger amplitude librators dissolve into the chaotic zone . for @xmath182 , the formerly period-1 orbit corresponding to exact 4:3 resonance bifurcates into a period-2 orbit in the surface - of - section ; this is merely a reflection of the fact that for a neptune - crossing value of the eccentricity , the perihelion loop causes two conjunctions with neptune in each synodic period . panel ( e ) and ( f ) show examples of small amplitude librating orbit with @xmath145 and @xmath146 , respectively . panel ( g ) shows the period of libration as a function of the libration amplitude of the resonance angle , @xmath87 ( for @xmath86 and @xmath105 ) . the period as well as the maximum amplitude for stable librations decreases with increasing @xmath68 . 9 the neighborhood of the neptune 5:4 orbital resonance located at @xmath183 . the phase space for orbits with ( a ) @xmath184 ; ( b ) @xmath185 ; ( c ) @xmath186 ; ( d ) @xmath187 . notice that the main libration zone ( for aphelion librators ) shrinks rapidly with @xmath68 . a libration zone for perihelion librators appears for @xmath188 , as indicated by the appearance of a chain of two libration islands in ( c)(d ) ; this zone increases with increasing @xmath68 . panel ( e ) shows the stable periodic orbit with @xmath156 at the 5:4 resonance ( an aphelion librator ) in the rotating frame coordinates ; panel ( f ) shows an example of a perihelion librator with @xmath156 . notice the four - fold symmetry in the rotating frame . panel ( g ) shows the period of libration for the aphelion librators as a function of the libration amplitude of the resonance angle ( for @xmath86 and @xmath105 ) fig . 10 the neighborhood of the neptune 6:5 orbital resonance located at @xmath189 ( 34.0 au ) . the phase space for orbits with ( a ) @xmath190 ; ( b ) @xmath191 ; ( c ) @xmath192 ; ( d ) @xmath193 . notice that the main libration zone ( for aphelion librators ) shrinks rapidly with @xmath68 ; for @xmath194 the center of this libration zone bifurcates into a period-2 orbit ; this is simply a reflection of the fact that the perihelion loop causes two conjunctions with neptune in rapid succession during each synodic period . also for @xmath194 , a new libration zone appears for perihelion librators [ indicated by the appearance of a chain of two libration islands with centers at @xmath195 in ( b)(d ) ] whose extent increases with increasing @xmath68 . the increased complexity of this libration zone in ( d ) is a reflection of the fact that the perihelion loops for large amplitude librations are so large that there are four conjunctions with neptune during each synodic period . panels ( e ) and ( f ) show examples an aphelion librator and of a perihelion librator with @xmath156 are the 6:5 resonance . notice the five - fold symmetry in the rotating frame . panel ( g ) shows the period of libration as a function of the libration amplitude of the resonance angle , @xmath87 ( for @xmath86 and @xmath105 ) for the aphelion librators . 11 the locations and widths of first and second order neptune resonances in the kuiper belt , as determined by the planar , circular restricted three - body model . the shaded region on the extreme left indicates the chaotic zone of `` resonance overlap '' in the vicinity of neptune s orbit . in the region above dotted line , orbits are neptune - crossing ; these orbits are dynamically short - lived ( due to close encounters with neptune ) , unless protected by orbital resonances .

the josephson effect has implications going beyond electrical transport @xcite . in 1965 , maki and griffin predicted that the interplay between cooper pairs and quasiparticles in tunneling events would provide heat currents with quantum coherence @xcite . this phenomenon manifests itself as a phase - dependent component of the heat current flowing through a thermally - biased josephson junction ( jj ) , which should , therefore , enable the realization of heat interference . towards this end , a _ magnetic flux - controllable _ superconducting heat interferometer was recently theoretically conceived @xcite and subsequently realized experimentally @xcite . this achievement served to show , on the one hand , how quantum coherence extends also to dissipative observables such as heat current . on the other , these devices might constitute the building blocks for the implementation of superconducting hybrid _ coherent caloritronic circuits _ consisting of , for instance , thermal modulators , heat transistors , splitters , etc . the purpose of this paper is to review recent advances done within the emerging field of _ coherent caloritronics_. the paper is organized as follows : we dedicate sec . [ sec00 ] to distinguish first between coherent caloritronics and thermoelectric effects also present in temperature - biased jj . in sec . [ sec02 ] we describe the general model used to derive the behavior of the _ heat _ current flowing through the jj . in sec . [ sec1 ] a superconducting heat interferometer is analyzed theoretically and its operation is demonstrated experimentally . in sec . [ sec2 ] we discuss the implementation of a double - loop heat modulator an even more versatile device is theoretically analyzed in sec . it consist of a temperature - biased extended jj that serves to produce coherent diffraction of the thermal currents . in sec . [ sec5 ] the remarkable heat rectification properties of nis and sis junctions are addressed ( where n stands for a normal metal , i represents a thin insulating layer and s and s refer to two different superconducting electrodes ) . this review is enriched with realistic calculations showing the achieved modulation of the electronic temperature of a normal metal electrode placed nearby each of the aforementioned structures . finally , our results are summarized in sec . [ summary ] along with a few final remarks . it is convenient to begin by clearly stating the nature of the phenomena we will deal with along this manuscript . it is our goal to investigate the intriguing quantum behavior of the _ heat _ current @xmath0 flowing through a temperature - biased jj . here , we will concentrate on jjs consisting of two superconducting electrodes separated by a thin insulating layer , i.e. tunnel sis junctions . these studies are not to be confused with thermoelectric effects also present in jjs @xcite . the latter concern the consequences of a thermal bias in the _ electric _ current flowing through the junction and manifests as the generation of a potential difference of thermoelectric nature . specifically , a thermovoltage @xmath1 might arise across the junction when the electronic temperature difference between the superconducting electrodes exceeds a threshold value @xmath2 @xcite . these effects can be understood by assuming that the temperature drop across the junction gives rise to a normal electric current ( @xmath3 ) . as noted by ginzburg @xcite , this current is shorted out in superconducting materials by the generation of a supercurrent ( @xmath4 ) flowing in the opposite direction . as a result , the voltage difference generated across the junction remains zero . if , however , the temperature drop is large enough to generate @xmath5 , @xmath6 being the critical current of the jj , one enters into the dissipative regime leading to the appearance of @xmath1 . the smallness of this effect has complicated considerably its experimental observation , making it measurable only in certain kind of jjs such as the sns type . in this way , kartsovnik _ et al _ observed the generation of @xmath7 v upon temperature differences of the order of @xmath8 k in ta - cu - ta sns sandwiches @xcite . thermoelectric effects on response to an in - plane magnetic field were also studied by ryazanov _ et al _ in ref . specifically , @xmath2 was found to depend on the modulus of the cardinal sine function of the magnetic flux piercing the junction leading to a fraunhofer - like dependence . the influence of thermoelectric phenomena in sis junctions was analyzed in ref . according to these calculations , a temperature gradient of @xmath9 k across a tunnel junction with normal state resistance @xmath10 @xmath11 will generate a cooper pair thermocurrent @xmath12 a. the latter will imply , in turn , the existence of a tiny phase difference between the superconducting electrodes of only @xmath13 rad that can be neglected for the considerations made here . we will turn our attention now to the description of the main equations governing heat transport in jjs . we start by assuming a josephson tunnel junction between two superconducting electrodes @xmath14 and @xmath15 characterized by a phase difference @xmath16 as schematized in fig . [ fig01](a ) . under such circumstances , an _ electric _ supercurrent will flow that depends on @xmath17 . assume now that the jj is electrically open but a thermal gradient is established by heating up the quasiparticles of @xmath14 up to @xmath18 while keeping quasiparticles in @xmath15 at temperature @xmath19 . then , a stationary heat current @xmath20 will flow from the hot to the cold reservoir . and s@xmath21 separated by a thin insulating layer constitute a conventional jj . if a phase difference @xmath22 arises , an electric supercurrent @xmath23 will flow . ( b ) if s@xmath24 and s@xmath21 are kept at temperatures @xmath18 and @xmath19 ( with @xmath25 ) , respectively , a steady - state total heat current @xmath26 will flow from s@xmath24 to s@xmath21 . ] the explicit form of @xmath20 was calculated by maki and griffin in 1965 @xcite giving . @xmath27 equation ( [ heatcurrent ] ) consists of two terms ; @xmath28 is the usual heat flux carried by quasiparticles @xcite and @xmath29 is the phase - dependent part of the heat current , which is peculiar of josephson tunnel junctions . quoting the authors , the latter term `` gives rise to an oscillatory heat flux and is analogous to a similar term in conventional electric transport through josephson junctions '' . we stress the fact that , depending on @xmath30 , the phase - dependent component of the heat current can flow backwards , i.e. , opposite to the thermal gradient . the total heat current goes however from the hot to the cold reservoir , preserving the second principle of thermodynamics @xcite . it is worthwhile to stress as well the absence of any cooper pair - related energy term . this is so for , in a static situation , the superconducting condensate shall carry no entropy @xcite . the quasiparticle term appearing in eq . ( [ heatcurrent ] ) reads @xcite @xmath31 , \label{qqp } \\end{aligned}\ ] ] where @xmath32 $ ] is the quasiparticle bcs normalized density of states in s@xmath33 at temperature @xmath34 ( @xmath35hot , cold ) and @xmath36 is the energy measured from the condensate chemical potential . furthermore , @xmath37 is the temperature - dependent superconducting energy gap and @xmath38 , @xmath39 being the critical temperature of the superconductor , @xmath40 , @xmath41 is the heaviside step function , @xmath42 is the boltzmann constant , @xmath43 is the junction normal - state resistance , and @xmath44 is the electron charge . on the other hand , for the phase - dependent component of the heat current one gets @xcite @xmath45 , \label{qint}\end{aligned}\ ] ] where @xmath46 $ ] is the cooper pair bcs density of states in s@xmath33 at temperature @xmath34 @xcite . this term originates from those energy - carrying processes involving cooper pair tunneling and recombination or destruction of cooper pairs . since the phase difference between the annihilated and created pair is relevant in such a process this gives rise to the cos@xmath30 contribution to the transferred heat . along this manuscript , we will neglect any contribution to thermal transport arising from lattice phonons . moreover , we will assume the latter to be very well thermalized with the substrate phonons that reside at the bath temperature ( @xmath47 ) . this is usually the case in the kind of systems of our interest as the kapitza resistance between a metallic thin film ( @xmath48 nm ) and the substrate is typically negligibly small at low temperatures @xcite . it will be also convenient to provide with the expression for the energy exchanged between electrons and phonons in the superconductor @xcite , @xmath49 , \label{eph}\end{aligned}\ ] ] where @xmath50 $ ] , @xmath51 is the electron - phonon coupling constant , and @xmath52 is the volume of the superconducting electrode . the behavior of @xmath0 can be experimentally revealed by probing the electronic temperature of one of the superconducting electrodes or , alternatively , a normal metal electrode tunnel - connected to the junction . the latter option is more convenient from an experimental point of view as electron thermometry is typically simpler to be performed on normal metals @xcite and will be therefore considered here . we start by theoretically investigating heat exchange between two normal metal electrodes kept at different temperatures and tunnel - coupled to each other through a thermal ` modulator ' @xcite in the form of a dc - squid . heat transport in a similar system is subsequently studied experimentally and found to be phase dependent , in agreement with the original prediction @xcite . the structure we envision is sketched in fig . [ fig11](a ) . it consists of a dc squid composed of two superconductors s@xmath24 and s@xmath21 in thermal equilibrium kept at different temperatures @xmath18 and @xmath19 , respectively . @xmath53 and @xmath54 denote the normal - state resistance and phase drop of jj@xmath55 , respectively . in this specific case the total heat current given in eq . ( [ heatcurrent ] ) flowing from s@xmath24 to s@xmath21 becomes @xmath56 , where @xmath57 , and @xmath58 . for definiteness , we assume that @xmath25 so that the squid is only biased with a temperature drop across the junctions , but the voltage across them vanishes . and s@xmath21 kept at temperature @xmath18 and @xmath19 , respectively , are tunnel coupled so to implement a thermally - biased dc squid . ( b ) interference heat current @xmath59 vs @xmath60 calculated for a few values of @xmath61 . ( c ) total heat current @xmath62 vs @xmath60 calculated for a few values of @xmath18 at @xmath63 assuming @xmath64 . , scaledwidth=90.0% ] by neglecting the geometric inductance of the ring it follows that @xmath65 where @xmath60 is the applied magnetic flux through the loop , @xmath66 is an integer and @xmath67 wb is the flux quantum . for a given @xmath60 , the phases @xmath68 and @xmath69 are determined by the equation @xmath70 , which describes conservation of the supercurrent circulating along the loop , where @xmath71 is the ambegaokar - baratoff critical current @xcite of junction jj@xmath55 . by defining @xmath72 ( with @xmath73 ) as the degree of asymmetry of the squid junctions one gets @xcite @xmath74 and @xmath75 . using the latter equations @xmath59 can be rewritten as @xmath76 , which is analogous to the expression for the total josephson _ electric _ critical current in a dc squid with generic junctions asymmetry @xcite . in particular , for a symmetric squid ( @xmath64 ) we get @xmath77 we remark that , in this specific case , the phase - dependent component of the heat current always flows in the direction opposite to the quasiparticle heat flow , i.e. , from the cold to the hot electrode@xcite . figure [ fig11 ] ( b ) shows @xmath59 vs @xmath60 calculated for a few values of @xmath78 at generic temperatures @xmath18 and @xmath19 such that @xmath79 . as it can be seen , @xmath59 is a periodic function of @xmath60 maximized at integer values of @xmath80 . by increasing @xmath78 leads to a suppression of the modulation amplitude combined with a reduction of the average value of the heat current . eventually , the modulation amplitude is totally suppressed for @xmath81 , as only one junction is driving heat flow through the squid . therefore high junctions symmetry is desired to maximize heat current modulation in the device . for that reason , in the following we will restrict our calculations to the case @xmath64 . figure [ fig11](c ) shows the total heat current @xmath62 vs @xmath60 at @xmath63 for a few values of @xmath18 . @xmath62 is @xmath80-periodic and minimized for integer values of @xmath80 . the experimental implementation of our heat interferometer is shown in fig . [ fig13](a ) . the structure consists of a source and drain copper ( cu ) electrodes tunnel - coupled to a superconducting aluminum ( al ) island defining one half of a dc - squid ( s@xmath24 ) . the other half of the squid ( s@xmath21 ) extends into a large volume lead to insure proper thermalization of its quasiparticles at the bath temperature @xmath47 . s@xmath24 is also contacted by an extra al probe ( s@xmath82 ) via a tunnel junction , enabling independent characterization of the squid . both source and drain are tunnel - coupled to a few external al probes so to realize normal metal - insulator - superconductor ( nis ) junctions , which allow joule heating and thermometry @xcite junctions normal - state resistances are @xmath83 k@xmath11 , @xmath84 k@xmath11 and @xmath85 k@xmath11 , respectively , whereas the resistance of each squid junction is @xmath86 k@xmath11 . the ring area is @xmath87m@xmath88 . finally , nis thermometers exhibit @xmath89 k@xmath11 normal - state resistance each . ] . . ( c ) squid voltage ( @xmath90 ) versus current ( @xmath91 ) characteristics at two representative values of the applied flux and ( d ) @xmath60-dependent experimental pattern of the squid critical current @xmath92 along with the theoretical curve ( solid line ) taken at 240 mk . ] below the critical temperature of al ( @xmath93 k ) josephson coupling allows dissipationless charge transport through the squid . the squid voltage - current characteristics at @xmath94 mk for two representative magnetic - flux values are shown in fig . [ fig13](c ) . in particular , a well - defined josephson current with maximum amplitude of @xmath95 na is observed at @xmath96 . the magnetic - flux pattern of the squid critical current @xmath92 along with the theoretical prediction @xcite is displayed in fig . [ fig13](d ) , and shows a nearly - complete supercurrent modulation , which confirms the good symmetry of the squid . thermal transport and , therefore , heat interference in the structure , arises from heating electrons in the source above lattice temperature ( @xmath47 ) so to elevate the quasiparticles temperature in s@xmath24 ( @xmath18 ) . as the second half of the squid ( s@xmath21 ) is well thermalized at @xmath47 a temperature gradient arises across the squid . @xmath62 will thus manifest itself leading to a @xmath80-periodic modulation of drain electron temperature ( @xmath97 . measured for several @xmath98 values . ( b ) modulation amplitude @xmath99 and average temperature @xmath100 vs. @xmath98 . ( c ) flux - to - temperature transfer function @xmath101 versus @xmath60 measured at a few selected @xmath98 . ( d ) flux modulation of @xmath102 recorded at different @xmath47 and @xmath103 mk . dashed lines in all panels are the theoretical results obtained from the thermal model ( see text ) . , scaledwidth=80.0% ] investigation of heat transport in our system is performed by stabilizing first the cryostat temperature at a desired @xmath47 and heating the source up to a given temperature @xmath98 . for this purpose , one pair of nis junctions in the source is operated as a heater whereas a second pair is used to measure @xmath98 by applying a small dc bias current and recording the corresponding temperature - dependent voltage drop @xcite . analogously , another pair of nis junctions is used to probe the electronic drain temperature ( @xmath102 ) against a slowly - sweeping external magnetic flux . figure [ fig14](a ) shows @xmath102 against @xmath60 measured at 235 mk for increasing values of @xmath98 . notably , @xmath102 is @xmath80-periodic in @xmath60 , as the josephson critical current [ see fig . [ fig13](d ) ] , and maximized at integer values of @xmath80 . as we shall argue , such a temperature modulation is of coherent nature , and stems from magnetic flux - control of @xmath62 , which is a hallmark of the josephson effect . a monotonic enhancement of the average drain temperature over one flux quantum , @xmath100 , is observed when raising @xmath98 , which follows from increased heat flow across the structure . on the other hand , the modulation amplitude @xmath99 , defined as the difference between the maximum and minimum values of @xmath102 , turns out to initially increase and then tends to saturate at larger @xmath98 . in particular , @xmath99 up to @xmath104 mk is observed corresponding to @xmath105 of relative modulation amplitude at 235 mk . the full @xmath98-dependence of @xmath100 and @xmath99 are displayed in fig . [ fig14](b ) and confirm the above described behavior . a relevant figure of merit of the heat interferometer is represented by the flux - to - temperature transfer coefficient , @xmath101 , shown in fig . [ fig14](c ) versus @xmath60 for a few selected @xmath98 . it turns out that @xmath106 exceeding @xmath107 mk/@xmath80 is obtained at @xmath108 mk . the dependence on bath temperature is shown in fig . [ fig14](d ) , which displays @xmath109 at a few increasing @xmath47 for @xmath98 set around 700 mk . besides leading to a monotonic enhancement of @xmath110 , by increasing @xmath47 yields suppression of @xmath99 and smearing of @xmath109 , which can be mainly ascribed to the enhancement of electron - phonon coupling in the drain @xcite as well as to the influence of thermal broadening . @xmath111 mk is still observable at 450 mk , whereas the modulation disappears for @xmath112 mk . we emphasize that the latter is substantially smaller than the temperature setting the disappearance of the josephson effect in the squid ( @xmath93 k ) . to account for our observations we have elaborated a thermal model sketched in fig . [ fig13](b ) . we assume s@xmath24 to exchange heat at power @xmath113 and @xmath114 due to quasiparticle heat conduction with source and drain , respectively , at power @xmath62 with s@xmath21 and @xmath115 with s@xmath82 . both s@xmath21 and s@xmath82 are assumed to be thermalized at @xmath47 . furthermore , drain electrons exchange energy at power @xmath114 with s@xmath24 , and at power @xmath116 with lattice phonons residing at @xmath47 @xcite . the thermal steady - state of the system may be described by the energy - balance equations @xmath117 and @xmath118 , where first equation accounts for thermal budget in s@xmath24 , while the second one describes heat exchange in the drain . @xmath18 and @xmath102 can be determined under given conditions by numerically solving the aforementioned equations [ see eq . ( [ heatcurrent ] ) ] substituting @xmath119 for @xmath120 , @xmath121 for @xmath122 and @xmath123 for @xmath124 . furthermore , @xmath113 can be obtained from eq . ( [ qqp ] ) by doing @xmath125 with @xmath126 and substituting @xmath119 for @xmath127 , @xmath114 can be also obtained from eq . ( [ qqp ] ) by setting @xmath128 with @xmath126 and substituting @xmath119 for @xmath129 . for the normal metal @xmath130 , where @xmath131 wk@xmath132m@xmath133 is the electron - phonon coupling constant for cu @xcite , and @xmath134 m@xmath135 is drain volume . ] . the model neglects heat exchange with photons due to mismatched impedance @xcite , electron - phonon coupling in s@xmath24 owing to its reduced volume and low experimental @xmath47 @xcite , as well as phonon heat current @xcite . theoretical curves have been obtained by setting the structure parameters as extracted from the experiment , and varying @xmath120 between @xmath136 and @xmath137 to match measured data . results from the thermal model are shown in fig . [ fig14 ] ( dashed lines ) . although idealized , this model provides reasonable agreement with our observations , and grasps the relevant physical picture at the origin of heat interference in our system . in this section we envision and theoretically analyze a double - loop heat interferometer @xcite . the latter allows to finely tune and fully balance the heat flux flowing through the device making it an improved _ heat transistor_. additionally , it is much more robust against fabrication deficiencies such as differences between the normal - state resistances of the jjs . thanks to the application of two independent magnetic fluxes , even a quite asymmetric device is able of providing a much more robust temperature modulation compared to the simple single - loop heat interferometer or its complete annihilation . our thermal circuit consists of two superconductors s@xmath24 and s@xmath21 , weak linked forming a double - loop interrupted by three parallel jjs [ see fig . [ fig21](a ) ] . @xmath138 and @xmath139 denote the normal - state resistance and phase drop of junction @xmath140 , with @xmath141 . this structure behaves as a conventional squid pierced by a magnetic flux @xmath142 in which one of the junctions has been replaced by a dc squid . the characteristics of this second `` junction '' can be tuned thanks to the application of a control magnetic flux @xmath143 . the system is temperature biased by setting the temperature in s@xmath24 to be @xmath25 , @xmath19 being the temperature in s@xmath21 . furthermore , the voltage drop across the whole structure is set to zero . under these circumstances , a thermal gradient arises across the junctions and a stationary heat current @xmath144 will flow from s@xmath24 to s@xmath21 , which are in steady - state thermal equilibrium @xcite . is risen up to @xmath145 , which yields a steady - state heat current @xmath144 . source and drain normal - metal electrodes are tunnel - coupled to s@xmath24 . superconducting tunnel junctions operated as heaters and thermometers are connected to source and drain . ( b ) idealized thermal diagram accounting for our setup . arrows indicate the direction of heat flows for @xmath146 . ( c ) density plots showing @xmath147 as a function of the magnetic flux on the main and the control loop , @xmath142 and @xmath143 , respectively . two representative cases have been considered and are schematized on the top part of each plot . ] we shall concentrate on the phase - dependent heat current only given by @xmath148 . on the one hand , neglecting the self - induced flux in the loops , the fluxoid quantization on both rings imposes @xmath149 and @xmath150 , where @xmath66 and @xmath151 are integers . the conservation of the circulating supercurrent in both loops , on the other hand , imposes @xmath152 , where @xmath153 . in writing the previous equation we have established a given current sign convention [ see yellow arrows in fig . [ fig21](a ) ] . furthermore we define @xmath154 , and @xmath155 . combining the aforementioned conditions and using simple trigonometric relations one gets @xmath156 , @xmath157 and @xmath158 , where @xmath159 , @xmath160 , @xmath161 and @xmath162 $ ] . finally , one gets @xmath163 . in fig . [ fig21](c ) we show two density plots of @xmath147 vs. @xmath142 and @xmath143 for two representative cases . in general , the maximum of @xmath147 is always reduced for the cases in which one resistance is different from the others . the straightest choice is @xmath164 . although being the most simple configuration , it enables us to infer most of the characteristics of our thermal interferometer . when analyzing the behavior of @xmath165 vs. @xmath142 for different @xmath143 values one can distinguish between two regimes ; in the first one , defined by @xmath166 [ fig . [ fig22](a ) ] , the mean value and the shape of the curves evolves whereas the amplitude of the modulation , denoted @xmath167 , holds unchanged as we increase the amplitude of the control flux @xmath143 . at @xmath168 one gets @xmath169 |$ ] . this is to say , apart from a small shift equal to @xmath170 , one recovers the same dependence on @xmath142 obtained for the symmetric single - loop heat interferometer . in the second regime , covered by @xmath171 [ fig . [ fig22](b ) ] , @xmath167 decreases linearly with @xmath143 whereas the mean value of @xmath165 remains constant . furthermore , at @xmath172 , the modulation disappears completely and @xmath147 becomes independent of @xmath142 . the aforementioned characteristics are emphasized in figs . [ fig22](c ) and [ fig22](d ) where we plot the transfer function , @xmath173 , for both regimes . similar curves are obtained in the more general case for which @xmath174 . moreover , if the condition @xmath175 is satisfied , it can be demonstrated that @xmath165 can be written as a function of @xmath176 | $ ] where @xmath177 is a shift in @xmath142 . this is to say it is still possible to suppress completely @xmath165 . unlikely to a single - loop heat interferometer , even a quite asymmetric double - loop structure offers therefore the possibility of maximizing or suppressing completely @xmath165 through an appropriate choice of @xmath143 . let us finally analyze how the previously described control over the magnetic flux - to - heat current transfer functions are translated into a realistic situation . for this purpose , a device similar to that of fig . [ fig13](a ) could be envisioned . in this specific case , the single squid loop must be substituted by a double - loop consisting of three parallel - connected josephson junctions and with independent magnetic flux controls . similar devices operating as charge interferometers @xcite have been reported already in the literature proving the feasibility of this structure . vs. @xmath142 plotted for different values of the control flux @xmath143 for @xmath164 . the inset shows the case for @xmath178 . panels ( c ) and ( d ) : transfer function @xmath173 vs. @xmath142 plotted for the same values of @xmath143 as in the top panels . panels ( e ) and ( f ) : @xmath102 modulation calculated using the thermal model depicted in fig . [ fig21](b ) for the same conditions as in panels ( a ) and ( b ) . panels ( e ) and ( f ) : corresponding flux - to - temperature transfer coefficient . ] the device we envision is that schematized in fig . [ fig21](a ) . as it was argued in section [ subsec12 ] heat transport in this structure can be studied by intentionally heating electrons in the source up to @xmath98 yielding a quasiparticle temperature @xmath179 in s@xmath24 , therefore leading to a finite heat current @xmath144 . the latter can be modulated by means of the two control magnetic fluxes @xmath142 and @xmath143 and inferred by measuring the drain electron temperature ( @xmath102 ) . drain temperature can be predicted by solving a couple of thermal balance equations similar to those described in section [ subsec12 ] . new equations will differ in that no additional probe [ i.e. , @xmath180 electrode in fig . [ fig13](b ) ] is considered this time . for this reason it will be convenient to include now the electron - phonon interaction in @xmath14 [ @xmath181 , defined in eq . ( [ eph ] ) ] . in this way , we have now @xmath182 and @xmath118 for s@xmath24 and drain , respectively . k@xmath11 , @xmath183 @xmath11 , @xmath184 m@xmath133 , @xmath185 wk@xmath132m@xmath133 ( typical of cu ) @xcite , @xmath186 m@xmath133 , @xmath187 wk@xmath132m@xmath133 and @xmath188ev , the last two parameters typical of aluminum ( al ) @xcite . ] the results of thermal balance equations for drain temperature are shown in fig . [ fig22](e ) and ( f ) for @xmath164 and the same @xmath143 values as in panels ( a ) and ( b ) . for these calculations we have set @xmath189 mk and @xmath190 mk . as it can be seen , @xmath102 is @xmath142 periodic reaching its minimum value for @xmath191 , as expected , but the appearance of the oscillation depends strongly on @xmath143 . we highlight that , for @xmath192 , the @xmath102 modulation is , under equal conditions , more than twice that obtained with the single thermal interferometer discussed in the previous section [ see inset in fig . [ fig22](f ) ] . this enhancement is related with the overall enhancement of @xmath147 and the fact that @xmath102 does not depend linearly on @xmath144 . in this way , by simultaneously playing with both @xmath142 and @xmath143 , @xmath102 could be modulated from @xmath193 up to @xmath194 mk in overall . differences between the modulation curves obtained at each @xmath143 values are emphasized in fig . [ fig22](g ) and ( h ) where the flux - to - temperature transfer coefficient @xmath195 is plotted . in this section we theoretically analyze heat transport in temperature - biased extended jj showing that the phase - dependent component of thermal flux through the weak - link interferes in the presence of an in - plane magnetic field leading to _ heat diffraction _ @xcite , in analogy to what occurs for the josephson _ electric _ critical current @xcite . in particular , thermal transport is investigated in three prototypical _ electrically - open _ junctions geometries showing that the quantum phase difference across the junction undergoes @xmath196 _ jumps _ in order to minimize the josephson coupling energy . we finally propose how to demonstrate thermal diffraction and to prove the existence of @xmath196 jumps in a realistic microstructure . our system is schematized in fig . [ fig31](a ) , and consists of an extended jj composed of two superconducting electrodes s@xmath24 and s@xmath21 residing at different temperatures @xmath18 and @xmath19 , respectively . s@xmath24 and s@xmath21 are characterized by their london penetration depths @xmath197 and @xmath198 , respectively , and separated by an insulating layer with thickness @xmath199 . an external magnetic field ( @xmath200 ) is applied parallel to a symmetry axis of the junction [ see fig . [ fig31](a ) ] penetrating within a distance @xmath201 in the @xmath202 direction ) must be used , where @xmath203 @xcite . ] . we will consider symmetric jjs in the _ short _ limit and neglect the effect of the edges . plane in the presence of an in - plane magnetic field @xmath200 along the @xmath204 direction . additionally , source and drain normal - metal electrodes are tunnel - coupled to one of the junction electrodes ( s@xmath24 ) . superconducting tunnel junctions operated as heaters and thermometers are connected to source and drain . @xmath205vs . @xmath60 calculated for a rectangular [ ( a ) ] , circular [ ( b ) ] , and annular [ ( c ) ] jj represented as insets . @xmath206 , @xmath207 , @xmath208 and @xmath78 represent the junctions geometrical parameters . ( e ) thermal model describing the main heat exchange mechanisms existing in the structure shown in ( a ) . ] we shall focus onto the phase - dependent component only . to this end we need to determine the phase gradient @xmath209 induced by the application of @xmath200 . by choosing the closed integration contour indicated by the dashed line depicted in fig . [ fig31](a ) it can be shown @xcite that , neglecting screening induced by the josephson current , @xmath210 , where @xmath211 and @xmath212 is the phase difference at @xmath213 . after integration over @xmath204 the phase - dependent component of the heat current can then be written as @xmath214 , where @xmath215 is the heat current density integrated along @xmath204 . this equation resembles the expression for the josephson current given by @xmath216 , where @xmath217 is the supercurrent density integrated along @xmath204 @xcite . in the actual configuration of electrically - open junction , the condition of _ zero _ josephson current for any given value of @xmath200 yields the solution @xmath218 , with @xmath219 . on the other hand , the josephson coupling energy of the junction ( @xmath220 ) can be expressed as @xmath221 where @xmath222 , @xmath6 is the zero - field critical supercurrent . minimization of @xmath220 for any applied @xmath200 imposes the second term of @xmath223 to be always negative , so that @xmath212 will undergo a @xmath196 _ jump _ whenever the integral does contribute to @xmath220 with negative sign . it can be shown that the @xmath196 jumps are present in any junction geometry @xcite . they have energetic origin only . this fact makes them very different from those present in low dimensional superconductors , caused by thermal @xcite and quantum @xcite fluctuations . taking into account the aforementioned @xmath196 _ jumps _ and assuming that the symmetry of the junction is reflected in an analogous symmetry in the heat current , i.e. , @xmath224 , the phase - dependent component of the heat current can be written as @xmath225 . we can now determine the behavior of @xmath205 for the three prototypical junction geometries sketched in fig . ( [ fig31 ] ) . in particular , we shall consider two well - known examples such as the _ rectangular _ [ see fig . [ fig31](b ) ] and _ circular _ [ see fig . [ fig31](c ) ] junction , and the more exotic _ one [ see fig . [ fig31](d ) ] . annular junctions offer the possibility to investigate fluxons dynamics due to the absence of collisions with boundaries ; yet , they provide fluxoid quantization thanks to their geometry , which allows fluxons trapping . we assume that the total phase - dependent heat current is characterized by a uniform distribution . @xmath205 can therefore be calculated for the three considered geometries by following , for instance , refs . @xcite . in particular , for the rectangular junction , the absolute value of the sine cardinal function is obtained , @xmath226 . for the circular geometry one gets the airy diffraction pattern , @xmath227 , where @xmath228 is the bessel function of the first kind , @xmath229 , and @xmath208 is the junction radius . finally , for the annular junction @xcite , the phase - dependent component of the heat current takes the form @xmath230 , where @xmath229 , @xmath231 , @xmath232 is the @xmath66th bessel function of integer order , @xmath208 ( @xmath78 ) is the external ( internal ) radius , and @xmath233 is the number of @xmath66 trapped fluxons in the junction barrier . right panels in fig . [ fig31 ] illustrate the behavior of @xmath205 for the three geometries . in particular , the curve displayed in fig . [ fig31](a ) for the rectangular case shows the well - known fraunhofer diffraction pattern analogous to that produced by light diffraction through a rectangular slit exhibiting minima for integer multiples of @xmath80 . in the case of a circular junction , the flux values where @xmath234 vanishes do not coincide anymore with multiples of @xmath80 , and @xmath234 falls more rapidly than in the rectangular junction case . finally , for an annular junction , the heat current diffraction pattern is strongly @xmath66-dependent providing , apparently , enhanced flexibility to tailor the heat current response . demonstration of diffraction of thermal currents could be achieved in the setup shown in fig . [ fig31](a ) . this is similar to the device already analyzed in section [ subsec22 ] but the in - plane static magnetic field @xmath200 is applied perpendicular to the josephson weak - link . the experimental operation mode will be the same and , again , the drain temperature behavior can be predicted by solving a couple of thermal balance equations accounting for the main heat exchange mechanisms existing in the structure , according to the model shown in fig . [ fig31](e ) . , i.e. , @xmath235 k@xmath11 , @xmath183 @xmath11 , @xmath236 m@xmath133 , @xmath237 wk@xmath132m@xmath133 , @xmath238 m@xmath133 , @xmath239 wk@xmath132m@xmath133 and @xmath188ev . ] vs @xmath60 calculated at @xmath189 mk for several values of source temperature @xmath98 for a structure based on a a rectangular [ ( a ) ] , circular [ ( c ) ] , and annular with @xmath240 [ ( e ) ] , @xmath241 [ ( g ) ] and @xmath242 [ ( i ) ] . bottom panels show the corresponding flux - to temperature transfer function @xmath243 vs @xmath60 . for the annular case we set @xmath244 . ] the results of thermal balance equations for drain temperature and different source temperatures are shown in fig . [ fig33 ] . as expected , @xmath102 shows a response to magnetic flux clearly resembling a fraunhofer or an airy - like diffraction patterns for the rectangular and circular cases , respectively . for the annular case the temperature patterns are drastically different depending on the number of fluxons trapped inside the loop . in all cases , temperature minima appear at the expected @xmath80 positions being the the unequivocal manifestation of the above - described phase _ jumps_. the flux - to - temperature transfer coefficient , @xmath245 , corresponding to each of the above - mentioned cases is plotted in the bottom panels of fig . [ fig33 ] . as it can be seen there , @xmath243 exceeding @xmath246 mk@xmath247 could be achieved at @xmath248 mk for the annular case . moreover , the transfer coefficient clearly demonstrates the non - monotonicity of the amplitude of drain temperature lobes as a function of @xmath98 . in this section we propose and analyze theoretically the performance of two different thermal diodes . one , consisting of a nis junction , and a second one , consisting of a sis josephson tunnel junction @xcite . although never considered in the literature so far for such a purpose @xcite , superconducting tunnel junctions appear particularly well suited for the implementation of electron heat rectifiers . heat transport in such structures is deeply influenced by the strong temperature dependence of the superconducting density of states . in addition , the sis diode allows for the _ in - situ _ fine tuning of the thermal rectification magnitude and direction . we shall start , first of all , by defining a heat rectification parameter @xmath249 . to this end , let us consider a tunnel jj made by two different superconductors , s@xmath24 and s@xmath21 characterized by its energy gap @xmath250 and @xmath251 leading to critical temperatures @xmath252 and @xmath253 , respectively . the electronic temperature in both s@xmath24 and s@xmath21 is kept at fixed @xmath18 and @xmath19 , respectively , and the voltage drop across the junction is set to zero . additionally , @xmath30 denotes the macroscopic phase difference across the junction with normal - state resistance @xmath119 . in the forward thermal bias configuration , a thermal gradient is created by setting @xmath254 , which leads to a total heat current @xmath255 flowing from s@xmath24 to s@xmath21 [ see fig . [ fig41](a ) ] . in the reverse thermal bias configuration , the thermal gradient is inverted so that @xmath256 leading to a heat current @xmath257 flowing from s@xmath21 to s@xmath24 [ see fig . [ fig41](b ) ] . under these hypothesis we define the rectification coefficient as @xmath258 . we start by analyzing the case in which one of the two electrodes is a normal metal , i.e. , a nis junction . for this purpose we can simply set @xmath259 , which leads to the complete suppression of the interference component of the heat current , @xmath260 . this leads to @xmath261 $ ] and @xmath262 $ ] , where @xmath263 is the tunnel resistance of the nis junction . we calculate @xmath249 as a function of @xmath18 for different values of @xmath19 , i.e. , for different temperature gradients established across the weak link . as shown in fig . [ fig42](c ) a maximum positive rectification of @xmath264 is obtained for @xmath265 aproaching @xmath266 . as @xmath18 increases , heat rectification starts to decrease eventually inverting its sign , which implies that heat flux from n to s becomes preferred . furthermore , by increasing @xmath19 leads to a reduction of @xmath249 , which reaches its maximum for larger values of @xmath18 . note that the heat rectification character of the nis diode stems from the presence of two different dos in the materials constituting the junction and the fact that the superconducting dos is strongly temperature - dependent . as a consequence , @xmath249 depends strongly on temperature as well , inverting its sign in the vicinity of @xmath267 , i.e. , when the superconducting dos becomes energy independent . vs. @xmath18 for different values of @xmath19 corresponding to a nis thermal diode . black horizontal line indicates @xmath268 . ( b ) and ( c ) show the results from the thermal model described in fig . [ fig41](d ) corresponding to different fabrication parameters detailed in the legends . black horizontal line indicates @xmath269 . ] we conclude by analyzing the theoretical behavior of a realistic nis - based thermal rectifier . for this purpose we consider the experimental design proposed in fig . [ fig41](c ) . two identical normal metal electrodes , source and drain , are weakly connected one each via resistances @xmath270 to both s and n , respectively . superconducting probes can be tunnel - coupled to these electrodes so to implement sinis thermometers and heaters @xcite . yet , the forward thermal bias configuration can be realized by intentionally increasing the electronic temperature in source electrode up to @xmath271 and probing the temperature in drain electrode @xmath272 . on the reverse configuration , we set @xmath273 and @xmath274 is measured in a similar way . the difference @xmath275 for a given @xmath276 can be used to assess experimentally heat rectification . as we have done in the previous sections , @xmath277 can be computed numerically using the thermal model described in fig . [ fig41](d ) . thermal balance equations are given by @xmath278 , @xmath279 and @xmath280 for the @xmath281 , @xmath282 and drain electrode , respectively$ ] @xcite . @xmath283 can be obtained in a similar way for the reverse configuration by simply exchanging the roles of @xmath284 , @xmath285 , @xmath286 and @xmath287 . as representative parameters we set @xmath288 m@xmath135 . we choose cu for which @xmath289 wk@xmath132m@xmath133 @xcite and al @xmath290 wk@xmath132m@xmath133 @xcite . ] . using realistic parameters and assuming @xmath189 mk , the computed values of @xmath277 vs. @xmath291 are plotted in fig . [ fig42](b ) for different @xmath263 values . remarkably , temperature differences exceeding @xmath292 mk can be obtained . as it can be seen , @xmath263 does not affect much the maximum value of @xmath277 . the choice of @xmath270 , on the other hand , does influence noticeably the appearance and sign of @xmath277 as shown in fig . [ fig42](c ) . the expected temperature differences are easily measurable with standard al - based sinis or sns thermometry techniques @xcite . vs. @xmath18 and @xmath293 calculated for @xmath294 , @xmath295 and @xmath296 , respectively . panels ( d ) , ( e ) and ( f ) show three selected profiles of @xmath249 vs @xmath293 for the same values of @xmath30 corresponding to the colored straight lines in ( a ) , ( b ) and ( c ) . notice that the scale is logarithmic above the break in the vertical axis . in addition , a dashed line indicates @xmath268 . all curves have been calculated for @xmath297 . ] we consider finally the case of a sis junction for which we define @xmath298 in this case , the forward and reverse total heat currents flowing through the jj read @xcite @xmath299 and @xmath300 $ ] , where the negative sign before brackets is set so that @xmath257 is positive by definition . by fixing the temperature of the second electrode to @xmath297 , we calculate @xmath249 as a function of @xmath18 and as a function of @xmath293 . the result is plotted in fig . [ fig43](a ) , ( b ) and ( c ) for three representative cases , corresponding to @xmath294 , @xmath295 and @xmath296 , respectively . three selected profiles of @xmath249 vs. @xmath293 for different @xmath18 values , i.e. , for different thermal gradients , are shown as well in fig . [ fig43](d ) , ( e ) and ( f ) . the inspection of these graphs reveals , on the first place , how phase biasing across the junction does make a substantial difference . in particular , the heat rectification coefficient does not only change by almost two orders of magnitude from @xmath294 to @xmath296 but it also switches its sign . it is worthwhile to emphasize that the sis junction rectifies heat only if @xmath301 . as for the case of the nis diode , heat rectification demands the combination of two different dos being ( at least one of them ) strongly temperature - dependent . let us analyze in more detail the case for which @xmath296 . in the forward configuration @xmath255 exhibits a local maximum when the temperature - dependent superconducting gaps of both superconducting electrodes coincide , which corresponds to the bright bend curve clearly visible in fig . [ fig43](c ) . on the other hand , for the reverse configuration , @xmath257 exhibit a local minimum when the @xmath15 electrode reaches its critical temperature , which corresponds to the straight line defined by @xmath302 . as a result , @xmath249 is maximized for @xmath303 and @xmath302 . in particular , @xmath304 is reached at @xmath305 and @xmath306 , which accounts well for the subgap leackage in realistic sis junctions . @xmath249 is reduced to @xmath307 when considering @xmath308 and down to @xmath309 for @xmath310 . ] . the experimental realization of the sis-based thermal rectifier is similar to that of the nis diode . in this case , we substitute n by a second superconducting electrode s@xmath21 made , for instance , of mn - doped al since the latter allows for fine tuning of the aluminum superconducting gap @xcite . optimum phase biasing can be achieved by using a rf squid configuration pierced by a control flux @xmath60 [ see fig . [ fig45](a ) ] . for such a purpose , the thermal diode can be enclosed through clean contacts within a superconducting ring s@xmath82 with energy gap @xmath311 so to suppress heat losses . neglecting the inductance of the loop , the phase - flux relation is given by @xmath312 @xcite enabling the phase drop across the junction to vary within the whole phase space , i.e. , @xmath313 . the experimental operation procedure will be that described in the previous section by @xmath314 as given in eq . [ eph ] and setting @xmath315 [ eq . ( [ qqp ] ) ] with @xmath126 and substituting @xmath119 for @xmath129 [ see fig . [ fig45](b ) ] . as representative parameters we set again @xmath316 m@xmath135 , @xmath317 k@xmath11 and @xmath318 k@xmath11 . we use cu for which @xmath319 wk@xmath132m@xmath133 @xcite and al and mn - doped al with @xmath290 wk@xmath132m@xmath133 @xcite . for these materials we set @xmath320 k and @xmath321 . ] . solving the thermal model for @xmath322 mk , the latter would give temperature difference approaching @xmath323 mk as it can be seen in fig . [ fig45](c ) . even more interesting , phase - coherence fingerprints are clearly observable as well . notably , @xmath324 shows the expected @xmath325-periodicity as shown in fig . [ fig45](d ) . these temperature differences are easily measurable using sns thermometry techniques @xcite based on , for instance , vanadium with @xmath326 k. as a function of @xmath291 and @xmath30 , respectively . black horizontal line indicates @xmath277 . ] along this manuscript we have proposed the experimental realization of different josephson - based devices that exploit phase - coherence of heat currents to provide with a robust magnetic flux - to - temperature control . the successful realization of these architectures relies on the recent experimental demonstration of heat current interference in a simpler device . the latter consist of a `` thermal '' version of a conventional electric squid that provided with an experimental temperature modulation of @xmath104 mk in amplitude upon the application of an external magnetic field as reviewed in sec . [ sec1 ] . in sec . [ sec2 ] we have theoretically demonstrated how this modulation can be tuned _ in - situ _ thanks to an improved design . considering realistic parameters , such tunability leads to a much more robust drain temperature oscillations of @xmath327 mk . this is possible by replacing one of the squid jj by an additional dc squid leading to a double - loop thermal interferometer that allows to maximize / minimize the strength of the phase - dependent component of the heat current . the existence of two control knobs , i.e. , two externally applied magnetic fluxes , may be exploited to perform non - trivial adiabatic cycles in its control space eventually leading to the realization of a heat pump @xcite . in sec . [ sec4 ] we have shown that the heat current through a temperature - biased extended jj under the influence of an in - plane magnetic field displays _ coherent diffraction _ , in full analogy with the josephson _ electric _ critical current @xcite . depending on the junction geometry , the latter will lead to a plethora of magnetic flux - to - temperature diffraction patterns with modulation amplitudes in the range of @xmath328 mk . the experimental realization of this device has been reported very recently @xcite setting a complementary and conclusive demonstration of the `` thermal '' josephson effect in weakly - coupled superconductors . we have finally proposed and analyzed in sec . [ sec5 ] the concept of a nis - based and a sis josephson thermal rectifier . under appropriate conditions , remarkably large rectification coefficients of @xmath329 and @xmath304 can be obtained for the nis and sis diodes , respectively . combining the diode within a realistic device , temperture differences between the forward and reverse configurations of the order of @xmath330 and @xmath107 mk could be reached for the nis and sis diodes , respectively . in addition , thanks to phase - coherence , the rectification character of the sis diode can be maximized or even inverted _ in - situ_. such a device might find a straightforward application of technological interest , e.g. , in the field of electronic refrigeration enabling magnetic - flux dependent heat management and thermal isolation at the nanoscale . the structures described here can be integrated within , not only superconducting elements , but also hybrid mesosocopic circuits composed of , e.g. , normal metals , two dimensional electron gases and semiconductor nanowires as well . in this way , many fields of research such as radiation detectors or quantum computing might benefit from our approach @xcite . on the other hand , these devices represent the first step towards the realization of coherent caloritronic circuits . the discipline usually referred to as _ ( from _ calor- _ , i.e. , `` heat '' in latin + _ electronics _ ) , deals with the generation and mastering of heat currents through an electronic drive . yet , our approach provides with the possibility of manipulating heat currents to flow through a series of superconducting components exploiting their coherent character . the core papers described in this review deal only with static parameter configurations where the superconducting phase and temperature are time - independent . the natural extension is to control and change the superconducting phase in time by applying a magnetic field . in principle this opens the possibility to transfer dynamically the heat between the superconductors , cool or heat one them and build a thermal mesoscopic engine . this new field requires both theoretical and experimental advance and will be the subject of future research . the fp7 program no . 228464 microkelvin , the italian ministry of defense through the pnrm project terasuper , the marie curie initial training action ( itn ) q - net 264034 and firb - futuro in ricerca 2012 under grant no . rbfr1236vv hybridnanodev . are acknowledged for partial financial support

spatial discretizations or fourier expansions of the solutions of partial differential equations ( pdes ) which depend on time lead to systems of ordinary differential equations ( odes ) . the most difficult case arises when the solution of a pde becomes singular in finite time . at such instants the solution of the pde develops activity down to the zero length scale . a brute force numerical simulation ( no matter how large ) of such a solution is bound to fail because the simulation has a finite resolution and thus will be unable to resolve all the length scales down to the zero scale . when the solution develops activity at a scale smaller than the smallest scale available to the simulation , the numerically computed solution becomes underresolved . this leads to a rapid deterioration of the accuracy of the simulation . the notion of propagation of activity to smaller and smaller scales depends on the physical context of the pde . in some cases , e.g. the 3d euler or navier - stokes equations @xcite , this could mean a cascade of energy to smaller and smaller scales . in other cases , e.g. the nonlinear focusing schrdinger equation @xcite , this could mean a cascade of mass to smaller and smaller scales . irrespective of the specific physical context , the problem facing the numerical analyst is how to use a _ finite _ simulation and yet prevent the computed solution from suffering a loss of accuracy . in other words , how to construct a numerical method which reproduces correctly the features of the solution of the original equation at the length scales that are available numerically . this is the motivation behind the construction of reduced models ( see e.g. @xcite ) . by construction , a reduced model must allow for energy ( mass ) to escape from the scales that are accessible to the simulation ( called resolved scales or modes ) to the inaccessible scales ( called unresolved ) . the main difficulty in constructing an accurate reduced model is the need to estimate the _ correct _ rate at which activity is propagated from the resolved to the unresolved scales . the mori - zwanzig ( mz ) formalism @xcite proceeds by dividing the available resolution into resolved and unresolved parts . then , it constructs a reduced model for the resolved scales and uses the unresolved scales to effect the drain of energy ( mass ) out of the resolved scales . although the mz formalism allows for the construction , in principle , of an exact reduced model it has two drawbacks ( which are also shared by _ any _ other reduction formalism ) . first , the reduced model can be , in general , prohibitively expensive to calculate . the reason is that one must obtain an accurate representation of the behavior of the unresolved scales before they can be safely eliminated . obtaining this representation can be rather costly . the second drawback is more subtle and has not been adequately appreciated by the scientific computing community . it is specific to the case of constructing reduced models for singular pdes or in general for systems of ordinary differential equations which are larger than any available numerical resolution . suppose that you have to construct a reduced model of a full system which is larger than any available numerical simulation . let us call this system s1 . exactly because s1 is larger than any available numerical simulation , if we want to construct a reduced model we have to use as a starting point a system , call it s2 , whose size is smaller than the size of s1 . suppose that you start with s2 and use the mz formalism ( or any other reduction formalism for that matter ) and construct an _ exact _ reduced model s3 for a subset of s2 . an exact reduced model means that if one evolves s2 and s3 separately , then the behavior of the scales resolved by s3 will be the same as the behavior of the scales in s3 predicted by the simulation of the system s2 . however , and this is the heart of the problem , since s2 itself will become eventually underresolved , the exact reduced model s3 will also become underresolved . in other words , the predictions of the exact reduced model s3 can only be trusted for as long as the predictions of the system s2 can be trusted . as a result , _ any reduced model that has any chance of being accurate for longer times can not be exact . _ there are examples of _ inexact _ reduced models , e.g. the @xmath0-model @xcite , coming from the mz formalism , which have been applied to singular pdes and shown numerically to be relatively accurate for long time intervals . however , the @xmath0-model s accuracy is difficult to assess beforehand and the reason for its relative success has remained a mystery ( see also @xcite for an application to the 3d navier - stokes equations which shows that the @xmath0-model , while not bad , is in need of some modification ) . in order to construct better reduced models we need to incorporate dynamic information from the full system which will help us decide which of the terms appearing in the exact reduced model are the ones that are most important . in this way , we can construct an inexact but accurate reduced model by keeping the important terms and disregarding the unimportant ones . the underlying principle of this approach is the same as the principle behind the construction of effective field theories in high energy physics @xcite . the way we propose to address the problem of constructing better reduced models is to embed the mz reduced models in a larger class of reduced models which share the same functional form as the mz reduced models but have different coefficients in front of the various terms that appear in the reduced models . then , one can estimate these coefficients on the fly while the original system of equations is still valid . the estimation of the coefficients is achieved by requiring that certain integral quantities ( e.g. @xmath1 norms ) involving only resolved scales , should acquire the same values when computed from the original system and the reduced model . the constraints used to obtain the coefficients are the analog of the matching conditions " used in effective field theory . also , the approach is the time - dependent analog of the process of renormalization used in high energy and condensed matter physics @xcite . before the original system ceases to be valid , one reverts to the reduced model with the various coefficients having their estimated values . we call the proposed approach the renormalized mori - zwanzig algorithm . in addition to computing the coefficients of the reduced model on the fly , the proposed approach allows us to gauge the importance of the different terms appearing in the reduced model . while at the moment a rigorous proof is lacking , the numerical results ( see section [ universality ] ) suggest that the values of the coefficients depend on the ratio of the smallest scale present in the initial condition to the smallest scale resolved by the reduced model . the larger is this ratio , the larger are the coefficients for more terms and thus , more terms are needed in the reduced model . on the other hand , for the case when the ratio is small only a few number of terms need to be kept without sacrificing the accuracy of the model . as we have already said , this situation is analogous to the principle underlying the effective field theory method in high energy physics @xcite . it is also interesting to investigate to what extent the values of the coefficients are determined by the scaling symmetries of the underlying pde . this could allow us to identify different universality classes of pdes based on their scaling symmetries ( see also section [ universality ] ) . finally , we note that the proposed method of computing the coefficients was first presented by the author in @xcite . the goal in @xcite was to construct a mesh refinement scheme to allow us to reach the singularity instant more efficiently . then , the computed coefficients were used in a fixed point analysis of the singularity in order to estimate the blow - up exponents . in the current work the goal is different . we not only want to reach the singularity instant or , more generally , the time that the full system runs out of resolution , but also follow the solution for later times . also , in the current work , we present a new way of expanding the memory term of the mz formalism . this allows us , at least for burgers and euler , to calculate recursively and efficiently ( and with minimal storage requirements ) the higher order terms in the memory expansion . this is a welcome feature since analytical computation of the higher order terms quickly becomes exhausting and , thus , prone to errors . the paper is organized as follows . section [ algo ] presents the main construction along with a brief presentation of the mz formalism and how to compute the coefficients of the reduced model on the fly . section [ numerics ] presents numerical results for the 1d burgers and the 3d incompressible euler equations . section [ conclusions ] contains a discussion of the results and presents some directions for future work . in this section we explain the main ideas behind the proposed approach . in section [ full_reduced ] we set up the notation for the original system and the reduced model in an abstract way which does not make reference to any specific method for obtaining the reduced model . in section [ coefficient_compute ] we show how to obtain the coefficients for the reduced model . in section [ mzformalism ] we give a brief presentation of the mz formalism which allows us to obtain the functional form of the terms appearing in the reduced model . in section [ combine ] we combine the ideas in section [ coefficient_compute ] with the mz formalism from section [ mzformalism ] to derive the proposed algorithm for computing renormalized mz reduced models . suppose that we want to construct a reduced model for the partial differential equation ( pde ) @xmath2 where @xmath3 is a , in general nonlinear , operator and @xmath4 ( the construction extends readily to the case of systems of partial differential equations ) . after spatial discretization or expansion of the solution in series , the pde transforms into a system of ordinary differential equations ( odes ) . for simplicity we restrict ourselves to the case of periodic boundary conditions , so that a fourier expansion of the solution leads to system of odes for the fourier coefficients . to simulate the system for the fourier coefficients we need to truncate at some point the fourier expansion . let @xmath5 denote the set of fourier modes retained in the series , where we have split the fourier modes in two sets , @xmath6 and @xmath7 we call the modes in @xmath6 resolved and the modes in @xmath8 unresolved . the reduced model involving only the resolved modes @xmath6 will be called the reduced system and the system involving both the resolved _ and _ unresolved modes @xmath5 will be called the full system . the main idea behind the algorithm is that the evolution of moments of the reduced set of modes , for example @xmath1 norms of the modes in @xmath6 , should be the same whether computed from the full or the reduced system . this requirement will eventually allow us to compute the coefficients appearing in the reduced model ( see section [ coefficient_compute ] ) . the full system of equations for the modes @xmath5 is given by @xmath9 where @xmath10 is the vector of fourier coefficients of @xmath11 and @xmath12 is the fourier transform of the operator @xmath13 the system should be supplemented with an initial condition @xmath14 the vector of fourier coefficients can be written as @xmath15 where @xmath16 are the resolved modes ( those in @xmath6 ) and @xmath17 the unresolved ones ( those in @xmath8 ) . similarly , for the right hand sides ( rhs ) we have @xmath18 note that the rhs of the resolved modes involves both resolved and unresolved modes . in anticipation of the construction of a reduced model we can rewrite the rhs as @xmath19 in general , when one constructs a reduced model , additional terms appear on the rhs of the equations of the reduced model ( see section [ mzformalism ] for more details ) . the role of these additional terms is to account for the interactions between the resolved and unresolved modes , since the unresolved modes no longer appear explicitly in the reduced model . as is standard in renormalization theory @xcite , one can augment the rhs of the equations in the full system by including such additional terms . that is accomplished by multiplying each of these additional terms by a zero coefficient . in this way , the reduced and full systems rhss have the same functional form . in particular , for each mode @xmath20 we can rewrite @xmath21 as @xmath22 where @xmath23 and @xmath24 are of the same functional form as the additional terms which appear in the reduced model . this is easy to do by taking @xmath25 and @xmath26 thus , the equation for the the mode @xmath27 is written as @xmath28 correspondingly , the reduced model for the mode @xmath29 is given by @xmath30 with initial condition @xmath31 we repeat that the functions @xmath32 have the same form as the functions @xmath33 but they depend _ only _ on the reduced set of modes @xmath34 dimensional reduction transforms the vector @xmath35 to @xmath36 this allows one to determine the relation of the full to the reduced system by focusing on the change of the vector @xmath37 to @xmath38 also , the vectors @xmath37 and @xmath39 do not have to be constant in time . this does not change the analysis that follows . define @xmath40 quantities @xmath41 involving only modes in @xmath34 for example , these could be @xmath1 norms of the reduced set of modes . to proceed we require that these quantities rates of change are the same when computed from ( [ full ] ) and ( [ reduced ] ) , i.e. @xmath42 note that similar conditions , albeit time - independent , lie at the heart of the renormalization group theory for equilibrium systems ( @xcite p. 154 ) . also , the conditions are the analog of the matching conditions " underlying the construction of effective field theories @xcite . when we only know the functional form of the terms appearing in the reduced model but not their coefficients it is not possible to evolve a reduced system . we present a way of actually computing the coefficients of the reduced model as needed . if the quantities @xmath41 are e.g. @xmath1 norms of the fourier modes , then we can multiply equations with appropriate quantities and combine with equations to get @xmath43 where @xmath44 are the new rhs functions that appear . note that the rhs of the equations above does not involve primed quantities . the reason is that here the reduced quantities are computed by using the values of the resolved modes from the full system . the above system of equations is a linear system of equations for the vector of coefficients @xmath38 the linear system can be written as @xmath45 where @xmath46 this system of equations can provide us with the time evolution of the vector @xmath38 the determination of coefficients for the reduced model through the system is a time - dependent version of the method of moments . we specify the coefficients of the reduced model so that the reduced model reproduces the rates of change of a finite number of moments of the solution of the original system . this ensures that each term in the model is properly weighted so that the resulting reduced model reproduces , at the scales accessible to the reduced model , the dynamics ( see ) of the original system . by construction , the entry @xmath47 of the matrix @xmath48 measures the contribution of the @xmath49-th term of the reduced model to the rate of change of @xmath50 in fact , the @xmath49-th column of the matrix @xmath48 is comprised of all the contributions of the @xmath49-th term in the reduced model to the rates of change of the different @xmath50 while the reduced system has no need to transfer activity from the resolved to the unresolved scales , the columns of @xmath48 corresponding to the activity - transferring terms will be zero ( to the numerical precision used ) . thus , the matrix @xmath48 will be singular . this can be monitored by estimating the rank of the matrix through the singular value decomposition ( svd ) @xcite . when the smallest singular value becomes non - zero for the numerical precision used the reduced system starts transferring activity to the unresolved scales . after that instant we can use the system to estimate the coefficient vector @xmath39 ( see @xcite for more details ) . the accurate determination of the coefficients guarantees that the rate at which the reduced model transfers activity to the unresolved scales is in agreement with the rate of transfer dictated by the full system . as we will see in section [ numerics ] , the accurate determination of the coefficients is crucial and moreover it can be hard . this is because the matrix @xmath48 can turn out to be extremely ill - conditioned . such ill - conditioning is to be expected because the contributions of the different terms in the reduced model to the rate of change of the quantities @xmath51 can vary dramatically from term to term . in fact , it is this significant variation of the contributions of the various terms which gauges the importance of the different terms in the reduced model . eventually , the difference of the importance of the different terms is reflected on the values of the coefficients of the terms . this will become apparent when we present the numerical results in section [ numerics ] . we have presented in the previous section an abstract way of writing the reduced system which does not make any reference to a specific method for obtaining the functions @xmath52 appearing on the rhs of . in order to proceed we need to specify the functions @xmath53 we will do that through the mori - zwanzig formalism @xcite . suppose we are given the full system @xmath54 where @xmath10 with initial condition @xmath14 the system of ordinary differential equations we are asked to solve can be transformed into a system of linear partial differential equations @xmath55 where @xmath56 the solution of is given by @xmath57 . using semigroup notation we can rewrite ( [ pde ] ) as @xmath58 suppose that the vector of initial conditions can be divided as @xmath59 where @xmath60 is the vector of the resolved variables and @xmath61 is the vector of the unresolved variables . let @xmath62 be an orthogonal projection on the space of functions of @xmath60 and @xmath63 we delay the specification of the projection operator until section [ numerics ] where we present applications to specific pdes . equation can be rewritten as @xmath64 where we have used dyson s formula @xmath65 equation ( [ mz ] ) is the mori - zwanzig identity . note that this relation is exact and is an alternative way of writing the original pde . it is the starting point of our approximations . of course , we have one such equation for each of the resolved variables @xmath66 . the first term in ( [ mz ] ) is usually called markovian since it depends only on the values of the variables at the current instant , the second is called `` noise '' and the third `` memory '' . if we write @xmath67 @xmath68 satisfies the equation @xmath69 if we project ( [ ortho ] ) we get @xmath70 since @xmath71 . also for the initial condition @xmath72 by the same argument . thus , the solution of ( [ ortho ] ) is at all times orthogonal to the range of @xmath73 we call ( [ ortho ] ) the orthogonal dynamics equation . since the solutions of the orthogonal dynamics equation remain orthogonal to the range of @xmath62 , we can project the mori - zwanzig equation ( [ mz ] ) and find @xmath74 in order to proceed with the computation of the reduced model we need to compute the markovian term and the memory term . while the markovian term is usually rather straightforward to compute ( see section [ numerics ] ) , the memory term computation is rather involved due to the presence of the orthogonal dynamics evolution operator @xmath75 in fact , it is the presence of this operator which makes , in general , the computation of mz reduced models prohibitively expensive ( see @xcite for a thorough discussion ) . one can start from and based on assumptions derive simplified reduced models that are easier to calculate @xcite . however , it has proven difficult to justify the assumptions underlying the derivation of the simplified models . in addition , it is hard to work with the expression for the memory term given in . we offer here an alternative presentation of the memory term which allows us to fit the mz formalism in the framework described in section [ full_reduced ] . note that the memory term @xmath76 can be written , through dyson s formula , as @xmath77 which can be rewritten ( using the linearity of the operator @xmath78 ) as @xmath79 since @xmath80 we find @xmath81 where @xmath82 + \ldots $ ] is given by the bch series and @xmath83=-tltql - tql(-tl)$ ] ( see e.g @xcite ) . we have @xmath83=[tl , tpl]=[tql , tpl].$ ] in general , @xmath84 \neq 0.$ ] however , depending on the initial conditions , @xmath84 $ ] may be small and thus allow the simplification of the memory term expression . the magnitude of @xmath84 $ ] will be examined in detail in a future publication . if we assume that @xmath85 expansion of the operator @xmath86 in taylor series around @xmath87 gives @xmath88 one can obtain different simplified models by truncating the series in after different values of @xmath89 in particular , if we omit all the terms after the first one we obtain the @xmath0-model which has been studied thoroughly @xcite . the series representation of the memory term in is based on the assumption of analyticity in time of the operator @xmath90 this assumption may be true for small @xmath0 but it does not have to hold for larger @xmath91 in other words , the taylor expansion of the operator @xmath86 has , in general , only a _ finite _ radius of convergence . insisting on using the taylor expansion of the operator @xmath86 as is for later times is dangerous and can lead to the instability of the reduced model ( see also section [ renovsnoreno ] ) . in fact , when dealing with full systems coming from discretizations of singular pdes , the breakdown of the taylor expansion of the operator @xmath86 is related to the onset of underresolution on the part of the full system . to proceed we need to put the mz model given by and in the framework of section [ full_reduced ] . to do that we set @xmath92 with this identification we have , in essence , embedded the reduced models derived through the mz formalism in a larger class of reduced models which share the same functional form with the mz models but who are allowed to have different coefficients . in the notation of section [ full_reduced ] , the original mz models correspond to the coefficient vector @xmath93 while the original mz models may suffer from instabilities , the new models can be made stable by _ assigning to each term in the reduced model the appropriate coefficient_. the magnitude of the coefficient of a term reflects the importance of the term in the reduced model . the values of the coefficients can now be determined by solving the linear algebraic system . this ensures that the coefficient of each term in the model is properly redefined ( renormalized ) so that the resulting reduced model reproduces , at the scales accessible to the reduced model , the dynamics ( see ) of the original system . we are now in a position to state the renormalized mori - zwanzig algorithm which constructs a reduced model with the necessary coefficients computed on the fly . 1 . choose a number of terms , say @xmath40 , to keep at the taylor expansion of the memory term . 2 . evolve the full system and compute , at every step , using the svd , the rank of the @xmath94 matrix @xmath95 3 . when the smallest singular value @xmath96 ( we assume that the singular values are indexed from largest to smallest ) where @xmath97 a prescribed tolerance , solve the system for the coefficients . 4 . for the remaining simulation time , switch from the full system to the reduced model with the estimated values of the coefficients . in order to apply the algorithm above , we need to specify the quantities @xmath98 also , we need to compute the expression for the markovian term , as well as the expressions for the terms in the taylor expansion of the memory term . this can be a formidable task . however , as we will show in the next section , for the case of burgers and euler , these computations can be automated and thus we can compute arbitrarily high order terms in the memory term expansion with minimal effort and storage . in this section we present results of the rmz algorithm for the 1d burgers and the 3d euler equations . we use the 1d inviscid burgers equation as an instructive example for the constructions presented in this section . the equation is given by @xmath99 equation ( [ burgersequation ] ) should be supplemented with an initial condition @xmath100 and boundary conditions . we solve ( [ burgersequation ] ) in the interval @xmath101 $ ] with periodic boundary conditions . this allows us to expand the solution in fourier series @xmath102 where @xmath103.$ ] we have written the set of fourier modes as the union of two sets in anticipation of the construction of the reduced model comprising only of the modes in @xmath104,$ ] where @xmath105 the equation of motion for the fourier mode @xmath106 becomes @xmath107 to conform with the mori - zwanzig formalism we set @xmath108 and we have @xmath109 for @xmath110 the system ( [ burgersodemz ] ) is supplemented by the initial condition @xmath111 we focus on initial conditions where the unresolved fourier modes are set to zero . we also define @xmath112 by @xmath113 note that @xmath114 we also need to define a projection operator @xmath73 for a function @xmath115 of all the variables , the projection operator we will use is defined by @xmath116 i.e. it replaces the value of the unresolved variables @xmath61 in any function @xmath115 by zero . note that this choice of projection is consistent with the initial conditions we have chosen . also , we define the markovian term @xmath117 the markovian term has the same functional form as the rhs of the full system but is restricted to a sum over only the resolved modes in @xmath34 the full system conserves the energy @xmath118 contained in all the modes . similarly , the markovian term of the reduced model does _ not _ alter the energy content of the resolved modes . the necessary energy transfer out of the resolved modes rests on the memory terms . based on our choice of projection operator and the scaling symmetries of the burgers equation we set @xmath119 with the definition of @xmath62 given above , we find for @xmath120 @xmath121 the expression for @xmath120 contains three terms which involve at least one wavenumber in the unresolved range @xmath7 the terms in the taylor expansion of the memory term are given by @xmath122 for the @xmath49-th term we have @xmath123 the expression for @xmath124 in for the @xmath49th term ( @xmath125 ) can be computed recursively using a simple construction based on a pascal triangle . note that for our choice of projection operator @xmath126 we have @xmath127 we begin with the ( first - order ) term for @xmath128 i.e. , @xmath129 we find @xmath130 the first order term can be computed by convolving the resolved part of @xmath131 with the unresolved part of @xmath132 in practice , all the convolutions sums can be computed using fast fourier transforms @xcite . note that the expression @xmath131 is linear in the fourier modes while @xmath133 is quadratic . thus , the convolution sum in @xmath134 ( including the factor @xmath135)can be denoted as @xmath136 where @xmath137 stands for convolution while @xmath138 and @xmath11 stand for the resolved and unresolved parts . this notation facilitates the recognition of the pattern for the higher order terms . with this notation , the first - order term can be written as @xmath139 where we have used bold face to denote the coefficient . we continue with the second order term @xmath140 we find @xmath141 the convolution sums in this term can be denoted as @xmath142 and @xmath143 the second order term can be written as @xmath144 to see the pattern more clearly we need one more term , @xmath145 we find @xmath146 the terms in the parenthesis can be denoted as @xmath147 @xmath148 and @xmath149 the third order term can be written as @xmath150 by examining the expressions in - we see that the memory terms can be computed as weighted sums of convolution sums where the weights are given by appropriate pascal triangle coefficients ( the bold face numbers ) . this was to be expected since we started with a convolution sum ( involving products ) of two functions and each new term in the taylor series involves a differentiation . moreover , the number of convolution sums that need to be added is equal to the order of the memory term in the taylor expansion . also , for each term , the convolution sums involve expressions whose degree ( in fourier modes ) follows an easily discernible pattern . for the @xmath151th order term in the taylor series we need the convolution sums @xmath152 finally , the expressions entering the convolution sums can also be computed by a pascal triangle construction . for example , in order to calculate the third order term , as can be seen from , one needs to compute and store only the quantities @xmath153 for @xmath154 note that the quantities @xmath155 and @xmath156 which are also needed are the same as @xmath157 for the _ resolved _ modes and zero for the _ unresolved _ modes . so , they do not need to be stored . they can be quickly constructed when needed . we see that the storage requirements for the calculation of the memory terms grows only linearly in the order of the taylor expansion . also , the ability to calculate the needed expressions through ffts speeds up significantly the calculation of the various memory terms . the recursive estimation of the memory terms allows us to calculate memory terms of very high order efficiently , without having to write down explicitly the analytical expressions which become very complicated after the first few orders in the expansion . figure [ plot_burgers_collective ] shows the evolution of the energy @xmath158 of the resolved modes computed by reduced models of different orders and the random choice method @xcite . the initial condition is @xmath159 this leads to the formation of a standing shock at @xmath160 all the reduced models use @xmath161 fourier modes while the full system has @xmath162 modes . the results of the reduced models are compared to a converged solution of the random choice method with @xmath163 points . the energy of the random choice method solution was computed using only @xmath161 modes . however , note that practically all the energy of the random choice method solution is concentrated in the first few fourier modes , so even if we had computed the energy for all @xmath163 fourier methods the results would not have changed . this is to be expected , since for the initial condition we are using , a standing shock forms at time @xmath164 and , thus , by time @xmath165 the only fourier modes having some energy left in them are the first few . the quantities @xmath51 used to set up the linear algebraic system needed to compute the coefficients of the reduced system are @xmath1 norms of the solution . in particular , for the first order model we use @xmath166 for the second order model @xmath167 and for the third order model @xmath168 in general , for the reduced model of order @xmath169 we need @xmath170 quantities because we also have to compute the coefficient of the markovian term . all the calculations are done in double precision . the tolerance @xmath97 which is used to decide when it is time to switch to the reduced model is set to @xmath171 the systems of ordinary differential equations for the different reduced models were solved using the runge - kutta - fehlberg method with the stepsize control tolerance set to @xmath172 @xcite . the numerical problem of solving the linear system for the coefficients is hard because the resulting system has very large condition number and very small determinant . this happens for three reasons . first , the dominant contribution to the linear system matrix comes from the markovian term ( except for the contribution to the rate of change of the @xmath173 which is zero ) . this means that the coefficient of the markovian term is practically 1 . second , the contributions of each memory term to the rates of change of the different @xmath51 vary dramatically . third , the contributions to each @xmath51 by the different memory terms also varies substantially . of course , this situation is exacerbated if we use more terms we use in the expansion . for the case when we retain up to the third order term in the memory expansion , we have to deal with condition numbers of the order @xmath174 and determinant values of order @xmath175 inevitably , even the use of double precision can not provide us with an accurate estimate of the coefficients . since the linear system matrix is practically singular ( for the numerical precision used ) we have chosen to solve the linear system using the svd algorithm @xcite . a partial remedy to the problem comes from a slight modification in the way of estimating the coefficients . since we know that the markovian term coefficient is practically 1 , we can set it to 1 , and subtract the column of contributions of the markovian term from the rhs of the linear system . this allows us to reduce the dimensionality of the linear system to be solved from @xmath176 to @xmath177 this practice of subtracting almost equal numbers is not advisable in general because it leads to loss of significant digits @xcite . however , in our case it helps to improve the results by lowering the condition number of the matrix from about @xmath174 to about @xmath178 in figure 1 , the estimation of the coefficients for the third order model using the reduced dimension matrix is denoted as `` rmz - reduced 3rd '' . as shown in figure [ plot_burgers_collective ] , the rmz models of first and second order give practically the same results as the @xmath0-model . the third order model gives a slight improvement . however , when the reduced dimension matrix is used , the energy evolution predicted by the third order model is practically identical to the correct energy evolution of the resolved modes predicted by the random choice method . if we increase the resolution of the reduced model , the numerical problems for the calculation of higher order coefficients become even more pronounced . this is to be expected , since a larger resolution means that the renormalized coefficients of the reduced model will be smaller . thus , computing them with accuracy is more difficult . we have to note , that the computation of the higher order coefficients is more difficult for our choice of initial condition since it only involves one active fourier mode . initial conditions with more active fourier modes will transfer activity to the unresolved scales at a higher rate and thus the corresponding renormalized coefficients will be larger . a detailed study of the behavior of the coefficients for different initial conditions will be presented elsewhere . for the first order model , we have already presented in @xcite a detailed study about the change of the value of the renormalized coefficient with resolution up to the order of @xmath179 fourier modes . in that work , the renormalized coefficient calculation was used to determine , in a fixed point analysis , the blow - up exponent . consider the 3d incompressible euler equations with periodic boundary conditions in the cube @xmath101 ^ 3 $ ] : @xmath180 where @xmath181 is the velocity , @xmath182 is the pressure and @xmath183 the system in ( [ euler ] ) is supplemented with the initial condition @xmath100 which is also periodic and incompressible and @xmath184 since we are working with periodic boundary conditions , we expand the solution in fourier series keeping @xmath185 modes in each spatial direction , @xmath102 where @xmath103\times[-\frac{m}{2},\frac{m}{2}-1]\times[-\frac{m}{2},\frac{m}{2}-1].$ ] also @xmath186 and @xmath187 the equation of motion for the fourier mode @xmath106 becomes @xmath188 where @xmath189 is the incompressibility projection matrix and @xmath190 is the @xmath191 identity matrix . the symbol @xmath192 denotes inner product in @xmath193 the system ( [ eulerode ] ) is supplemented by the initial condition @xmath194 where @xmath195 are the fourier coefficients of the initial condition @xmath196 to conform with the mz formalism we set @xmath197 and we have @xmath198 for @xmath110 the system ( [ eulerodemz ] ) is supplemented by the initial condition @xmath111 note that we focus on initial conditions where the unresolved fourier modes are set to zero . we also define @xmath199 by @xmath113 note that @xmath114 consider the subset @xmath200\times[-\frac{n}{2},\frac{n}{2}-1]\times[-\frac{n}{2},\frac{n}{2}-1]$ ] for @xmath201 we will construct the reduced models for the fourier modes @xmath106 with @xmath202 we need to define a projection operator @xmath73 for a function @xmath115 of all the variables , the projection operator we will use is defined by @xmath116 i.e. it replaces the value of the unresolved variables @xmath61 in any function @xmath115 by zero . note , that this choice of projection is consistent with the initial conditions we have chosen . based on our choice of projection operator and the scaling symmetries of the euler equations we set @xmath119 define @xmath203 the markovian term has the same functional form as the rhs of the full system but is restricted to a sum over only the resolved modes in @xmath34 the full system conserves the energy @xmath118 contained in all the modes . similarly , the markovian term of the reduced model does _ not _ alter the energy content of the resolved modes . the necessary energy transfer out of the resolved modes rests on the memory terms . with the definition of @xmath62 given above , we find for @xmath120 @xmath204 the expression for @xmath120 contains three terms which involve at least one wavenumber in the unresolved range @xmath7 the terms in the taylor expansion of the memory term are given by @xmath122 for the @xmath49-th term we have @xmath205 the different terms in the memory expansion can be computed recursively as in the case of burgers . however , there is a slight complication because the presence of the incompressibility operator and of the inner product on the rhs destroys the commutativity which allowed us in burgers to group terms ( the factor 2 which appears outside every parenthesis there ) . this problem can be addressed by a construction which uses 2 pascal triangles instead of 1 used in the case of burgers . for each order , one adds up the corresponding terms from the 2 pascal triangles and obtains the desired memory term . other than that , the recursive algorithm remains the same and we omit the details . also , note that all the higher order terms are divergence - free by construction . we have used the same quantities @xmath51 as in the case of burgers , with the obvious generalizations , since for 3d euler we have a 3-dimensional velocity vector instead of the scalar velocity in burgers . also , even though for 3d euler we have a 3-dimensional vector , we have assumed that the reduced model renormalized coefficients are the same for all 3 velocities . this is a simplifying assumption . of course , one can use different renormalized coefficients for the different velocities at the expense of having to solve a larger linear system for the renormalized coefficients . a detailed study of that case will be presented in future work . we have used the taylor - green initial condition ( see e.g. @xcite and references therein ) which is given by @xmath206 figure [ plot_euler_collective ] shows the evolution of the energy @xmath158 of the resolved modes for different resolutions computed by rmz reduced models of different orders and the @xmath0-model . we have presented results for the rmz reduced models using the reduced linear system matrix approach discussed above to tame the condition number of the matrix . based on these results , we make two observations . first , for @xmath207 resolved modes , the rmz third order model dissipates energy at a slower rate than the @xmath0-model with @xmath207 resolved modes . this is true not only for the third order model but also for the first and second order models ( we have omitted those results to avoid cluttering the figure ) . this slower rate of energy dissipation compared to the @xmath0-model holds also for the case of @xmath208 resolved modes . the second observation is that the rate of energy dissipation of the rmz models is consistent with the rate predicted by the @xmath0-model with _ higher _ resolution . this is to be expected since a higher order model should result in a more accurate prediction of the energy dissipation rate . the reader may be concerned about the small resolutions used in the numerical experiments . there are two reasons for that . first , if one keeps several terms in the memory expansion , then , for a very smooth initial condition like the one we use , the matrix @xmath48 becomes even more ill - conditioned for large resolutions . however , this is not a severe problem . on the contrary , it signifies that most of the higher order terms should have small coefficients and thus can be safely removed from the model . the second reason we have used small resolutions both for burgers and euler is because an accurate reduced model should be able to reproduce the correct energy content for its resolved scales no matter how small the resolution . for example , for burgers , where we know what the energy content should be after the singularity , we see that the rmz model with a small resolution ( @xmath208 ) indeed reproduces the correct energy content for this resolution . we have hinted in the introduction at the possibility that the renormalized coefficients may be determined by two factors : i ) the ratio of the smallest active scale in the initial condition to the smallest resolvable scale and ii ) the scaling symmetries of the equation under investigation . even though the numerical difficulties with the ill - conditioned linear system matrix do not allow us at present to study accurately the higher order renormalized coefficients , we have enough accuracy to study the first order renormalized coefficient both for 1d burgers and 3d euler . note that the two equations share the same scaling symmetries . also , we have chosen for burgers the initial condition @xmath209 which has only one active fourier mode ( for @xmath210 ) and for 3d euler , the taylor - green initial condition which also has only active fourier modes for @xmath211 figure [ plot_burgers_euler_coefficient ] shows the comparison of the value of the renormalized first - order coefficient for burgers and 3d euler as a function of the ratio of the smallest scale active in the the initial condition to the smallest scale of the reduced model . we make two observations . first , the values of the renormalized coefficient for the two equations are in remarkable agreement . second , from the slope of the linear fit , we see that the value of the coefficient is practically equal to the ratio of the smallest scale active in the the initial condition to the smallest scale of the reduced model . needless to say that one example is not enough to infer the generality of the result for arbitrary initial conditions . a theoretical explanation of this result is lacking at the moment . note that due to the way we have defined the terms in the expansion of the memory , all the terms have the same dimensions as the markovian term and the left - hand side of the equation for each fourier mode . so , the corresponding coefficients have to be dimensionless . thus , we expect the coefficients to depend on ratios of quantities with the same dimensions . here we have investigated the possibility that this ratio is that of the smallest active scale in the initial condition to the smallest active scale of the reduced model . we should comment here on the behavior of the rmz algorithm for the 2d euler equations for which the 2d version of the taylor - green initial condition is an exact solution , i.e. a steady state . exactly because it is a steady state there is no need for a reduced model . application of the rmz algorithm agrees with this . there is never any need to transfer energy to the unresolved scales and thus , no need to switch to a reduced model . the contributions of the different memory terms to the matrix @xmath48 are all well below the double precision threshold . this allows the freedom to assign to the renormalized coefficient the values shown in figure [ plot_burgers_euler_coefficient ] without incurring any trouble . in other words , the behavior of the solution of the 2d euler for the taylor - green initial condition does not contradict the agreement for the renormalized coefficient of the 1d burgers and 3d euler equations shown in figure [ plot_burgers_euler_coefficient ] . in this section we present results which show that the renormalized version of the mz formalism is advantageous with respect to the original mz formalism . in particular , we show that for the same order in the taylor expansion of the memory term , the renormalized algorithm leads to the stabilization of the reduced model . figure [ plot_burgers_renovsnoreno ] compares , for the 1d burgers equation , the energy @xmath158 for 16 resolved modes for the renormalized and unrenormalized third order models . figure [ plot_euler_renovsnoreno ] compares , for the 3d euler equations , the energy @xmath158 for @xmath207 resolved modes for the renormalized and unrenormalized third order models . the unrenormalized models quickly become unstable and lose all predictive ability . the unrenormalized expansion leads to divergence of the predicted energy content of the resolved modes . this is analogous to the divergences that plagued perturbative calculations in quantum field theory ( qft ) before the advent of renormalization @xcite . in qft , the reason for the divergences was that the perturbation expansion was performed in a quantity ( the bare mass , charge etc . ) which turns out to be ill - defined . the process of renormalization replaces the perturbation expansion in powers of the ill - defined quantity with a perturbation expansion in powers of the experimentally determined values of this quantity . this allows the subtraction of the terms that cause divergences and leads to finite results . in our case , the expansion of the memory of the mz formalism is ill - defined because the taylor expansion at @xmath87 breaks down after some time . on the other hand , the renormalized mz formalism takes into account dynamic information from the evolution of the full system ( while this system is still valid ) and prescribes to each term in the memory expansion an appropriate coefficient . the coefficient measures how important this term is . in this way , the divergences are averted and the results become finite . we have presented a new way of computing reduced models for systems of ordinary differential equations . the approach combines renormalization and effective field theory techniques with the mori - zwanzig formalism . the constructed reduced models are stable because they transfer activity out of the resolved scales at a rate which is dictated by the full system . the consistency between the rate of transfer activity of the reduced model and the rate of transfer activity dictated by the full system is the analog of the matching conditions employed in effective field theory . the matching conditions lead to a redefinition ( renormalization ) of the coefficients of a reduced model originally constructed through the mori - zwanzig formalism . the results we have obtained for the 1d burgers and 3d euler equations are rather encouraging . the proposed approach of calculating the renormalized coefficients leads to a linear algebraic system for the coefficients . the matrix of this linear system is ill - conditioned by construction . the reason is that the reduced model comprises of two parts . the markovian part which is of the same functional form as the full system but defined only on the resolved scales , and the memory part which has no analog in the full system . the memory part is there to compensate for whatever activity the markovian term can not reproduce . thus , the determination of the coefficients appearing in the memory part are , in essence , computed by subtracting the contribution of the markovian term from the full system . this difference may be small and also the different terms in the memory part can have varying contributions to the difference . these factors result in the ill - conditioning of the linear system that needs to be solved . in the future , we plan to address the problem through various techniques designed to deal with ill - conditioned matrices . the results for 1d burgers and 3d euler equations also point to the possibility that the renormalized coefficients may be completely determined by the structure of the initial condition and the scaling symmetries of the equation . this needs to be investigated further and a detailed study involving random initial conditions with various spectra will be presented elsewhere . the proposed approach can also be applied to the navier - stokes equations @xcite . it is straightforward to incorporate viscosity in the new expansion of the memory term proposed in the current work . we note that the viscosity starts contributing from the second order memory term . also , the inclusion of viscosity does not complicate considerably the recursive algorithm for the calculation of the higher order terms . the expressions needed to compute the viscosity contributions can be estimated through terms already computed in the pascal triangle construction for the inviscid terms . the approach presented in the current work opens new possibilities for the construction of accurate and stable reduced models for ( large ) systems of ordinary differential equations . it also highlights the affinity between problems of model reduction in scientific computing and the construction of effective field theories in high energy physics . we hope that this connection will benefit the problem of constructing reduced models and will be of use in tackling real world problems which are impossible to address through brute force calculations . in conclusion , as steven weinberg once put it @xcite , renormalization is indeed a good thing . i am grateful to profs . g.i . barenblatt , a.j . chorin and o.h . hald for their ongoing guidance and support .

the effects of the ionosphere have always been an obstacle for ground - based radio frequency observations of astronomical sources . this is especially true for interferometric observations used to make images of objects with relatively small angular sizes . vhf and uhf interferometers such as the very large array ( vla ) in new mexico , the westerbork synthesis radio telescope ( wsrt ) in the netherlands , the giant metrewave radio telescope ( gmrt ) in india , and the australia telescope compact array ( atca ) , among others , are all affected by the ionosphere in the same way . the fundamental principles for the operation of an interferometer are well described in the literature ( e.g. , * ? ? ? briefly , an interferometer measures the time - averaged correlation of the complex electric fields measured at pairs of antennas pointed at a particular object . these correlations , or `` visibilities '' provide a measure of the spectrum of the sky brightness distribution at different spatial frequencies . these frequencies are essentially the difference between the position vectors of the two antennas normalized by the wavelength of the observation in a coordinate system based on the position of the object in the sky . these spatial frequencies are commonly referred to as @xmath2 , @xmath3 , and @xmath4 , and the coordinate system used is defined such that @xmath2 is the spatial frequency in the east - west direction , @xmath3 corresponds to the north - south direction , and @xmath4 gives the spatial frequency along the line of sight to the object . thus , as the object moves through the sky , the @xmath2,@xmath3,@xmath4 coordinates of a pair of antennas , or `` baseline , '' changes . the visibility measured for a specific baseline at a particular time is given by @xmath5 where @xmath6 denotes solid angle , @xmath7 the frequency of the observed signal , and @xmath8 is the intensity on the sky at a position given by the direction cosines @xmath9 , @xmath10 , and @xmath11 which are measured relative to the position of the observed object on the sky . generally , interferometers are `` fringe - stopped , '' that is , the measured visibilities are multiplied by a factor of @xmath12 so that visibilities from a source at the center of the field of view will have a phase of zero ( i.e. , the source will not produce fringes ) . this is mainly done because with fringe - stopping and for small fields of view ( i.e. , @xmath13 ) , equation ( 1 ) becomes a simple two - dimensional fourier transform such that the observed visibilities , measured as a function of @xmath2 and @xmath3 , may be converted to maps of intensity on the sky using standard numerical methods . as signals from astronomical sources pass through the ionosphere , a phase term is added given by @xmath14 where @xmath15 is the electron density and @xmath16 denotes the path - length through the ionosphere . thus , the phase of the observed visibility for a baseline is altered by the difference between the ionospheric phase terms observed by the two antennas , which is proportional to the difference in the total electron content ( tec ) observed along the lines of sight from the two antennas to the observed object . to first order , if these phase terms are not removed , then during the fourier inversion process involved in making an image of the sky , the ionospheric phase terms have the effect of changing the apparent positions of objects in the image plane . when higher order ionospheric effects begin to dominate , objects begin to appear distorted in the image plane , and in extreme cases , almost disappear . therefore , to make an image , one must remove these phase terms through a calibration process which estimates the required phase corrections . typically , after initial calibrations for instrumental effects are performed , one usually uses some form of a procedure referred to as `` self - calibration '' @xcite . this involves dividing the visibilities by an assumed sky model for the observed field of view which removes any contribution by the sky brightness distribution to the observed visibility phases . following this , a linear fit is used to determine the complex gain for each antenna . since the interferometer only provides phase differences between antenna pairs , the absolute phase of the complex gain for each antenna can not be determined by this fitting process . the general practice is to choose one reference antenna for the interferometer and to set the phase for its complex gain to zero . since for @xmath17 antennas , there are @xmath18 baselines , this is generally an over - determined problem and can be done over relatively short time periods , depending on the brightness of the source . one may also use this calibration to make an image , deconvolve the image to produce a better sky model , and then repeat the process until it converges . this has been shown to be a rather robust procedure for determining the complex antenna gains ( see * ? ? ? * and references therein ) . according to equations ( 2 ) and ( 3 ) , the phase corrections obtained from the determined antenna gains are essentially measurements of the difference between the tec along the lines of sight of a particular antenna and that of the reference antenna . these equations also demonstrate that the effect of the ionosphere will be more substantial for vhf observations . given the size of available astronomical vhf interferometers ( baselines ranging from @xmath19 km to as large as @xmath20 km ) , the robustness of the self - calibration procedure on relatively small time scales ( typically @xmath21 minute , but as small as a few seconds for extremely bright sources ) , and the sensitivity of such interferometers to relatively small tec variations ( fluctuations in differential tec of as small as 0.001 tecu ) , these instruments are capable of studying tec fluctuations on substantially finer scales than many others available . subsequently , previous work has been done using astronomically motivated observations ( e.g. , * ? ? ? * ; * ? ? ? * ) as well as observations geared toward studying the ionosphere ( e.g. , * ? ? ? * ; * ? ? ? * ) to explore the phenomenology of the ionosphere on these fine scales . much of this work has been performed with the vla ( latitude@xmath22 n and longitude@xmath23 w ) . this is in part due to the fact that the vla is relatively well suited to the study of the ionosphere because its 27 antennas are distributed in a `` y''-shape which allows it to probe structures along three different directions . arrays such as the wsrt and atca are so - called `` east - west '' arrays because the antennas are aligned along a single east - west axis and can therefore only observe ionospheric fluctuations along one dimension . the vla is also unique in that the antennas are moved from time to time among four different configurations , referred to a , b , c , and d. the configurations go from larger and more spread out to smaller and more compact . at its largest in the a configuration , each of the array s three arms is about 20 km long with the shortest antenna separations being about 1.5 km . in the most compact configuration , d , the arms are at most 0.6 km long and the shortest spacings are about 0.04 km . this allows the vla the ability to study the ionosphere over a wider range of physical scales than other similar interferometers . finally , the vla had a somewhat unique vhf system in place that allowed observations to be made simultaneously at 74 mhz and 327 mhz using a pair of dipole antennas ( one for each band ) mounted near the prime focus of each antenna . recently , the vla electronics and receivers were upgraded to establish the new expanded vla ( evla ) which does not include the old vla vhf system . however , a new and improved vhf system is now being developed and will be available in the near future . past work using the vla and other similar instruments has led to interesting results . these include discoveries such as the new class of magnetic - eastward - directed ( med ) waves , predominantly found at night , discovered by @xcite as well as larger statistical studies such as the measurements by @xcite of the dependence of differential ionospheric refraction on relatively large angular scales ( @xmath24 ) using data from a 74 mhz all - sky survey which showed large dependences on time of day . however , there is much more to be learned from these types of data , especially at smaller time and amplitude scales . therefore , we are embarking on a program utilizing the vla data archives ( https://archive.nrao.edu ) that seeks to push this type of analysis to even finer scales . we will use previously unexplored data sets of the brightest vhf objects . we will apply new techniques for calibration and the mitigation of radio frequency interference ( rfi ) to other data sets to significantly improve their sensitivity to small amplitude tec fluctuations as well as fluctuations occurring on smaller time and spatial scales than could be explored previously . here , we describe the first step in this program , a thorough evaluation of the ionospheric information contained within a single , relatively long vhf observation of one of the brightest radio sources in the sky with the vla . this data set provides the opportunity to develop and establish techniques for processing , analyzing , and interpreting similar data in future segments of our program . this paper focusses on the data selection and calibration as well as the post - processing done on the phase information extracted from this exemplar data set to measure tec gradients with the array . in a companion paper @xcite , we detail new techniques for spectral analysis of these data . in the study presented in this paper , we have sought to explore the kinds of ionospheric phenomena capable of being observed with the vla at the lowest fluctuation levels and the smallest times scales possible . to ensure that we were able to determine ionospheric phase terms to the the level of accuracy needed at the smallest available time intervals , including the resolution of any @xmath25 ambiguities , we chose to use a data set focused on cygnus a ( or `` cyg a '' ; also known as 3c405 ) . with a total intensity of more than 17000 jy ( @xmath26 w m@xmath27 hz@xmath28 ) at 74 mhz , cyg a is one of the two brightest sources in the sky at frequencies below 100 mhz . its brightness distribution on the sky is also known relatively well , implying that it can be used within the self - calibration procedure described briefly in 1 to solve for the phase corrections ( both instrumental and ionospheric ) for each antenna over relatively short time intervals . the data set we have selected consists of simultaneous 74 mhz and 327 mhz dual polarization observations of cyg a with the vla over a period of more than 12 hours on 12 - 13 august , 2003 ( vla program number ak570 ) . during this period , there was a moderate amount of geomagnetic activity ( @xmath29 index @xmath304 ) and low solar activity ( @xmath31 sfu ; 1 sfu@xmath32 w m@xmath33 hz@xmath28 ) . for this data , the vla was in the a configuration with the vhf dipole system available on all but three antennas ( antenna numbers 11 , 13 , and 15 ; see 1 and fig . [ layout ] ) . the exact layout of the antennas is shown in fig . [ layout ] . the data set contains a 35 minute scan ( i.e. , a contiguous block of observing time ) , and a longer , nearly 13 hour scan which , taken together , cover a range of local time from roughly @xmath34 on 12 august to @xmath35 the next morning . for the 74 mhz band , the total bandwidth was 1.5 mhz ; it was 3 mhz for the 327 mhz band . for both bands , the time - averaged visibilities were measured over intervals of 6.67 seconds simultaneously in both rr and ll polarizations . the processing of vhf interferometric observations from the vla or other similar interferometers is , in the average case , a lengthy and difficult process . one usually must deal with significant sources of rfi , which typically appear much stronger on shorter baselines . for the vla , the field of view at 74 mhz is more than ten degrees in diameter , and the ionospheric conditions may vary significantly from one part of the field of view to another , which requires special calibration techniques . one must also typically produce an image for the entire field of view , regardless of which object(s ) one is interested in because the removal of sidelobes ( i.e. , secondary peaks of the impulse response in the image plane ) produced by other objects is crucial to minimizing noise in the final image . fortunately , the unusually high intensity of cyg a makes it substantially brighter than any source of rfi or any other object in the field of view . this makes the calibration process much simpler and more straight forward . the calibration of the data was performed using standard tasks within the astronomical image processing system ( aips ; * ? ? ? * http://www.aips.nrao.edu ) . the first step in the calibration was to determine the bandpass response of each band ( aips task ` bpass ` ) . this was done within one minute time intervals by dividing the data by the visibilities measure within a few central channels where the response typically peaks . the relative amplitude and phase responses across each bandpass were then measured and interpolated onto the full set of time steps . after trimming the first and last few channels from each band where the response drops substantially , the data were corrected for the bandpass responses . following this , at 6.67 second intervals ( i.e. , the shortest possible for these data ) , the data for each band and polarization were used with model visibilities computed using images of cyg a presented by @xcite ( currently publicly available at http://lwa.nrl.navy.mil/tutorial/ ) to compute the phase corrections for each antenna at each time step . this was done with the aips task ` calib ` which does a series of consistency and sanity checks as it determines the solutions needed to fit the measured visibilities to the model and flags antennas and intervals that appear spurious or of poor quality . we relaxed some of the criteria for these checks , namely the minimum number of antennas ( we used four ) required and the minimum signal to noise ratio ( we used three ) because of the relatively short time intervals used . as noted above , an interferometer like the vla can only measure relative phases , implying that for this calibration to work , a reference antenna must be chosen whose phase is arbitrary and is subsequently set to zero . typically , the antenna closest to the center of the vla is not used because the effects of rfi tend to be the worst for this antenna . however , as noted above , cyg a is bright enough that this is not a consideration for our data set , and we therefore used this antenna as our reference ( antenna 9 ; see fig . [ layout ] ) . following the determination of phase corrections using standard aips routines , several steps were taken to extract ionospheric information from the phase data . this was done using ad hoc , python - based software . the phase corrections measured by the calibration process contain contributions from several effects , the ionosphere being the largest , especially at 74 mhz . in particular , the phase difference between two antennas within a single baseline is given by @xmath36 where @xmath37 denotes the difference in the ionospheric phases along the lines - of - sight of the two antennas given by equation ( 3 ) , @xmath38 represents difference in the instrumental effects of the two antennas , @xmath39 is the contribution to the phase difference from the structure of the observed source , and @xmath40 is the contribution from @xmath25 ambiguities . we can remove @xmath39 by dividing the observed visibilities by a model of cyg a and @xmath40 by having short enough time sampling to `` unwrap '' the phases ( see below ) . the dual 74 and 327 mhz observing mode is very useful for removing instrumental effects since the ionospheric phase simply scales with wavelength [ see equations ( 2 ) and ( 3 ) ] , whereas instrumental effects do not necessarily . the instrumental components of the phase corrections include errors in the delays added to individual antenna signals used to fringe - stop the visibilities ( see 1 ) and offsets between the antenna pointing and the actual source position . these two effects and @xmath25 ambiguities were dealt with in three separate steps . first , the aips task ` calib ` flags antennas that are spurious at each time step so that the phase corrections are `` missing '' for some antennas at a relatively small fraction ( @xmath41 ) of time steps . generally , a small amount of missing data is not problematic . however , to facilitate the use of fast fourier transforms ( ffts ) within our spectral analysis of the data presented in a companion paper which work best with evenly sampled data , these missing time steps were filled in . we have done this by linearly interpolating the real and imaginary parts of the complex antenna gains , @xmath42 , computed by ` calib ` for all antennas onto a common time - step grid consisting of 7299 steps spaced by 6.67 seconds , and then recomputing the phase corrections \{@xmath43$]}. next , the time sampling of the data ( 6.67 seconds ) was sufficiently short that the phase , @xmath44 , as a function of time for each antenna could be `` unwrapped '' in the conventional way , i.e. , by correcting phase jumps of more than @xmath45 radians by adding or subtracting @xmath46 . there was one caveat to this process , however . for each antenna , there were a few ( ranging from zero to five ) times steps where , for one reason or another , the phase correction was either spurious or represented a real and very short jump in the instrumental phase , appearing as sharp spikes in the unwrapped phases . since there were a total of 7299 time steps , a few short jumps in phase would not be an issue except for their effect on the unwrapping process . any of these spikes can cause an artificial large phase jump if it is included in the unwrapping process , which we have illustrated in the upper panel of fig . [ despike ] . to combat this , we wrote a simple algorithm that computes the difference between @xmath47 at each time step and the value for the next time step , where @xmath44 is the wrapped phase . any time step where the absolute value of this difference was more than ten times the standard deviation among all time steps for a give antenna was flagged and not included in the unwrapping process . these empty time steps were then filled by linearly interpolating the unwrapped phase data for the un - flagged time steps . it should be noted that these spikes only occupy 12 time steps ( 6.6713.34 seconds ) and that the surrounding data are otherwise well - behaved , making interpolation a reasonable and straightforward solution to eliminating instances of such spikes . an example of how the data were flagged is illustrated in the middle panel of fig . [ despike ] while the resulting unwrapped phases are shown in the lower panel . from this result , one can see that the spikes are not always completely removed from the data , but the goal of eliminating their effect on the unwrapping process has been achieved . a number of instrumental effects can contribute to the phase corrections , including errors in the fringe - stopping process ( see 1 ) and offsets between the position of the source and that of the observed field center ( i.e. , pointing errors ) . these effects are generally stable in time , changing insignificantly over periods of days ( see , e.g. , * ? ? ? . however , for the vla , the instrumental phase is known to occasionally have short jumps that one must be wary of . we have found such a jump in our data occurring at a local time of about @xmath48 on 12 august . the jump is most obvious when one plots the difference between the 74 mhz phase and the 327 mhz phase scaled by a factor of ( 327/74 ) so that the ionospheric phases cancel out ( note , this only gives us the scaled difference between the 327 and 74 mhz instrumental phases , not the instrumental phases themselves ) . we have plotted this difference as a function of time for antenna 14 for both polarizations in fig . [ pjump ] to show the location of what we will refer to as the `` phase jump region . '' while being fairly subtle in the ll polarization , it is quite obvious in the rr polarization . the jump lasted about 6 minutes and can be seen in the data for several antennas . to deal with this , we have treated the phase jump region , as well as the time periods before and after it , as separate scans , assuming that each scan has its own instrumental phase . this basically resulted in us treating the data as if it contained four scans instead of two . to remove instrumental effects , we have used a kind of continuum subtraction process . within this process , we have treated ionospheric fluctuations as features superimposed on a smooth continuum consisting of the instrumental phases , which vary relatively slowly with time as well as any slowly varying component of the ionospheric phase . with the current data , we unfortunately do not have the means to separate the slowly varying component of the ionospheric phase from the instrumental effects and can therefore only measure fluctuations in tec on relatively short ( @xmath19 hour ; see below ) time scales . in the future , with instruments with larger bandwidths , it will be possible to use the wavelength dependence of the phase corrections to separate these effects since the instrumental phases are @xmath49 and the ionospheric phase is @xmath50 . we have chosen to perform our continuum subtraction process for each antenna , band , and polarization by smoothing the unwrapped phases with a one - hour - wide boxcar which appeared to preserve any apparent fluctuations while giving a good representation of the continuum . for the first scan and the phase jump region , we simply subtracted a single mean value from all the phases since they are each shorter than an hour . we did the same for the first and last hour of each of the remaining two scans so that the same filter width would be used for all times steps . following this , we found that the position offset component of the instrumental phases presented a problem for this process near the edges of each scan . this is because , depending on the antenna , these phases can vary significantly over one hour , especially at 327 mhz . from the fringe - stopped ( see 1 ) version of equation ( 1 ) , one can see that a position offset in the direction cosines @xmath9 and @xmath10 of @xmath51 and @xmath52 will produce an additional phase of @xmath53 . since @xmath2 and @xmath3 are normalized by the observed wavelength , any such phase will be 4.4 times larger for the 327 mhz band . in addition , the offsets can be different for each band and this difference can vary with baseline . this is due to a number of factors including the fact that different model images were used for the bands which may not be exactly aligned and that cyg a has a significant amount of resolved structure which larger baselines are more sensitive to , especially at 327 mhz where the angular resolution in the image plane is 4.4 times better than that at 74 mhz . we have demonstrated this in the upper panel of fig . [ edge ] where we have plotted @xmath54 for antenna 3 , ll polarization as a function of time . with @xmath55 scaled by ( 327/74 ) , the ionospheric phases are removed and all that is left is the difference between the instrumental phases for the two bands . the data follow a smooth curve which is inconsistent with the known behavior of vla instrumental phases , especially at 74 mhz . furthermore , the curve that the data follows is easily fit by a linear combination of the un - normalized versions of the @xmath2 and @xmath3 coordinates ( see the red curve in the upper panel of fig . [ edge ] ) . a single baseline ( antenna 3 with the reference antenna ) will sweep out an ellipse in the @xmath2,@xmath3-plane because of the rotation of the earth ( see , e.g. , * ? ? ? therefore , this is exactly what one would expect for a scenario where there is a single position offset for each of the two bands during each scan . to show the effect of the time dependence of the position offset phase on our continuum determination process , we have plotted the continuum - subtracted version of @xmath54 versus time in the middle panel of fig . one can see from this plot that within the first scan and within the last hour of the second and fourth scans where a single mean continuum value was subtracted from each , the gradient of the position offset phases has introduced an artificial difference which increases / decreases with time . since the position offset phase is much larger for the 327 mhz band , we have introduced the following additional step for the continuum subtraction of the 327 mhz data . within the first scan , the phase jump region , and the first and last hours of the other two scans , we have fit a linear combination of the un - normalized versions of @xmath2 and @xmath3 to @xmath54 separately for each time range . within each of these time periods , we used the mean values for @xmath56 , @xmath57 , and the @xmath2,@xmath3 fit to construct a time - variable continuum for the 327 mhz data . the benefits of this are illustrated in the bottom panel of fig . [ edge ] where we have plotted the continuum - subtracted version of @xmath54 , this time , including the additional computation for the time - variable 327 mhz continuum with the first / last hour of each scan . one can see that the roughly linear features seen in the middle panel of fig . [ edge ] have been removed and that the remaining difference between the 74 mhz and 327 mhz continuum - subtracted phases is essentially noise . following the application of the corrections detailed above , we used equation ( 3 ) to convert the continuum subtracted phases for each antenna , band , and polarization to values of differential tec , or @xmath0 . then , at each time step and antenna , we computed the median @xmath0 among the four values ( i.e. , two bands and two polarizations ) as well as the median absolute deviation ( mad ) as an estimate of the uncertainty in the median . to increase the reliability of the mad computations , we included with each time step the four nearest time steps ( using their own individual median values ) for a total of 20 data points per time step . in both computations ( median and mad ) , the median was used to minimize the effects of any spurious data which remained . the resulting @xmath0 values are plotted as functions of time for each antenna in the northern arm in fig . [ north ] , the southeastern arm in fig . [ east ] , and the southwestern arm in fig . [ west ] along with the mad values to illustrate the relatively high precision to which @xmath0 has been measured . the typical @xmath0 uncertainty , represented by the mad computations , is about @xmath1 tecu , demonstrating the remarkable ability of the vla to detect extremely small tec fluctuations , even on time scales @xmath58 seconds when an object as bright as cyg a is used . with the fully reduced @xmath0 data , including robust estimates of the uncertainties , we are in a position to explore a wide range of ionospheric phenomena . first , we note that since the vla measures differential tec values between antenna pairs , it is essentially only sensitive to changes in the tec gradient . given the geometry of the array ( see fig . [ layout ] ) , we can not numerically compute the tec gradient at each antenna location from our data , and measuring the full tec gradient requires a somewhat ad hoc approach . such measurements are crucial for any analysis of observed tec fluctuations because without modification , the set of @xmath0 time series can only be spectrally analyzed for specific assumed pattern models ( e.g. , a single plane wave , see * ? ? ? * ) . this is different from the normal mode of operation for radio interferometers in which standard techniques are used to invert and de - convolve sparsely sampled visibility data to make an image . as equation ( 1 ) demonstrates , the observed visibilities are functions of the _ differential _ antenna positions , @xmath2 , @xmath3 , and @xmath4 . for small fields of view , the contribution of the @xmath4 term is negligible if fringe - stopping is applied ( see 1 ) . thus , even for a y - shaped array , reasonably good @xmath2,@xmath3-coverage can be obtained . this is improved further by the rotation of the earth which causes each baseline to sweep out an ellipse in the @xmath2,@xmath3-plane ( see 2.2 and * ? ? ? using this fact to obtain better @xmath2,@xmath3 coverage is sometimes referred to as `` earth rotation synthesis . '' in contrast , the gradient of an arbitrary set of tec fluctuations varies over the array as a function of the _ actual _ antenna positions projected onto the ionosphere . improved spatial coverage can be obtained by exploiting the change in the apparent position of the observed source ( i.e. , rotation of the earth ) and the movement of the fluctuations themselves , effectively converting temporal baselines into spatial ones . however , since the tec fluctuations presumably have a distribution of speeds and directions , this is not as straightforward as in earth rotation synthesis . one must decompose the time series into temporal spectral modes and then analyze how the properties of each mode vary across the array to extract the size , speed , and direction of the dominant pattern(s ) for that mode ( such spectral techniques are detailed in a subsequent paper ) . therefore , we must still contend with data that has been sampled in a y - pattern which can not be inverted in a straightforward manner . we have consequently developed two ad hoc techniques designed to provide measurements of the tec gradient time series over the full array and along each of the vla arms . before implementing either technique , we first had to perform two basic geometric corrections to the data so that the measured tec gradients would correspond to vertical tec gradients as closely as possible . first , to ensure that our characterization of the shape of the observed tec surface is physically meaningful , we needed to project the antenna pattern displayed in fig . [ layout ] onto the locations where the lines of sight of the antennas pass through the ionosphere , or `` pierce - points . '' second , we needed to compute the slant - to - vertical tec corrections for the line of sight to cyg a as its apparent position on the sky changed throughout the observation . for a plane parallel approximation , both of these corrections are relatively straightforward . however , since cyg a was as low as @xmath59 above the horizon during the observing run , a plane parallel approximation was far from valid at all time steps . we have therefore computed the two required geometric corrections using a spherical model detailed in appendix a. within this model , the ionosphere was approximated with a thin shell located at the height of the maximum electron density , or `` peak height '' ( see * ? ? ? * ; * ? ? ? we obtained estimates of the peak height as a function of time by using the international reference ionosphere ( iri ; * ? ? ? * ) software , inputting the date and time of our observations and the latitude and longitude of the vla . we then re - determined the peak heights using the latitudes and longitudes of the pierce - points . we found that additional iterations of this process only marginally changed the results and chose to use one iteration only . the final peak heights used are plotted in the upper panel of fig . [ zion ] along with the projected separations from the array center for the farthest antennas of each arm ( antennas 1 , 7 , and 22 ; see fig . [ layout ] ) and the corresponding slant - tec corrections . after applying the geometric corrections to the antenna positions and the @xmath0 measurements , we sought to characterize the full two - dimensional tec gradient observed by each antenna at each time step . rather than assume a particular dominant structure ( e.g. , a plane wave ) , we simply assumed that since the array is smaller than many transient ionospheric phenomena , the observed tec surface at any time step could be approximated with a low - order taylor series . we examined many time steps and found that a second order , two - dimensional taylor series adequately approximated the amount of curvature in the tec surface detected by the vla . this taylor series has the following form @xmath60 where @xmath16 and @xmath61 are the north - south and east - west antenna positions , respectively , projected onto the surface of the ionosphere at the estimated peak height . to maximize the amount of data used to constrain the parameters of each fit , we used the difference between @xmath0 for each of the 300 unique antenna pairs at each time step . thus , the form of equation ( 5 ) actually fit to the data was @xmath62 where the @xmath63 and @xmath64 subscripts denote the values for the @xmath65 and @xmath66 antennas , respectively . we also utilized some standard sigma - clipping during the fitting process for each time step by computing the rms of the fit residuals , rejecting all antenna pairs with absolute residuals @xmath67 , and repeating 50 times . as many as about 10 and as few as zero were rejected for any given time step . we note that each time step was fit independently to preserve the presence of any small - scale spatial / temporal fluctuations . the fitted coefficients as a function of time are plotted in fig . [ pfit ] . from these , one can see the same large amplitude and period fluctuations at the beginning of the observing run that are visible in the individual antenna data plotted in fig . [ north][west ] . note that they are not quite as large here because of the applied slant - tec correction discussed above . here , we can see that they are most visible in the @xmath68 coefficient which is the partial derivative of the tec surface at the center of the array along the north - south direction . with the plots in fig . [ pfit ] , one can also see the same thing beginning to happen near the end of the run toward dawn . this is qualitatively consistent with the known behavior of medium - scale traveling ionospheric disturbances ( mstids ) which are prevalent near sunrise and sunset ( e.g. , * ? ? ? during the middle of the night , the second - order terms become more significant . while the polynomial - based measurements provide useful information about the variation of the full two - dimensional tec gradient , they neglect the ability of the vla in its a configuration to detect fluctuations on scales as small as a few kilometers . in principle , one could do this by simply increasing the order of the polynomials used . however , it is likely that the small - scale structures observed do not span the array . this implies that such fits would not yield accurate representations of the full tec gradient at each antenna , especially those near the ends of the arms ( see fig . [ layout ] ) . therefore , instead of using higher order polynomial fits , we have opted for an alternative approach to make full use of the data . this complementary method computes the projection of the tec gradient ( or , the spatial derivative of @xmath0 ) along each vla arm . the projected gradient was computed at each time step separately for the antennas of each arm using simple three - point lagrangian interpolation . given the typical @xmath0 precision of @xmath1 tecu and the mean separation between antennas of 2.5 km , the precision of these projected tec gradient measurements is typically about @xmath69 tecu km@xmath28 . the time series of the projected gradient at each antenna is plotted in fig . [ tecgrad ] . we have also plotted in red the projected gradient computed using the polynomial coefficients plotted in fig . one can see that for the larger amplitude , longer period disturbances , the polynomial fits largely recover the structure observed using the data for individual antennas . however , during the middle of the night , there appears to be a significant amount of smaller - scale structure missed by the polynomial fits that can only be observed using the individual antenna gradients , especially for the shortest baselines near the center of the array . our exploration of a long , vhf observation of cyg a with the vla has successfully demonstrated the power of this instrument to characterize a variety of transient ionospheric phenomena . for this observation , the typical @xmath70 uncertainty in the @xmath0 measurements was @xmath1 tecu , yielding more than an order of magnitude better sensitivity to tec fluctuations than can be achieved with gps - based relative tec measurements ( see , e.g. , * ? ? ? * ) . large amplitude , long period waves reminiscent of mstids are visible within the @xmath0 data near dusk and dawn as well as other times intermittently throughout the night . the polynomial - based approach we have detailed in 3.2 appears to be able to recover the properties of the full two - dimensional tec gradients associated with these relatively large disturbances as they passed over the array . this information can be used to estimate the size , speed and directions of such patterns down to scales of roughly half the size of the array ( @xmath71 km ) . this is demonstrated in more detail in a subsequent paper describing the spectral analysis of these data . in addition , our approach of measuring the projected gradient at each antenna along each arm has shown that there are smaller - scale tec fluctuations observed throughout the night , most prominently after midnight local time . thus , the vla can be used to simultaneously study fine - scale ionospheric dynamics . this may include a host of phenomena such as the small - scale distortions / structure within mstid wavefronts , turbulent fluctuations from ion - neutral coupling within the lower ionosphere / thermosphere , and sporadic - e ( @xmath72 ) layers . in the case of @xmath72 , @xcite demonstrated with a combination of vla data and optical observations that many of the small - scale fluctuations seen by the vla during summer nighttime are likely associated with @xmath72 layers . @xcite showed that the tec gradients caused by these layers are typically @xmath73 tecu km@xmath28 which is easily detectable using the arm - based gradient method . a specialized spectral analysis technique has also been developed for these measurements and will likewise be detailed in the companion manuscript to this paper . two basic geometric corrections must be applied to the antenna positions and @xmath0 measurements so that they more accurately represent the actual conditions within the ionosphere . since we have used observations of cyg a that include times when it is relatively close to the horizon , we can not use a plane - parallel approximation . instead , we have used a thin shell approximation for the ionosphere where the shell is located at the height of maximum electron density , @xmath74 , as computed by the iri software for the dates and times of the observations ( see 3.1 and fig . [ zion ] ) . the full spherical corrections used are detailed below . first , the positions of the antennas on the ground must be converted to projected positions within the ionosphere which , for a non - plane - parallel atmosphere , change with the elevation of the observed source . for a spherical shell , we may define a `` pierce - point '' for each antenna where its line of sight to the source passes through the ionosphere . the positions of these pierce - points relative to that for the center of the array can then be used as their projected ionosphere positions . [ geo ] provides a schematic illustration ( not to scale ) of how these positions are determined . we first define a set three position vectors , @xmath75 , @xmath76 , and @xmath77 , which define the positions of the array center / antenna , the ionosphere pierce - point , and the observed source , respectively , with the center of the earth as the origin of the coordinate system . next , we note that the vast majority of astronomical sources , including cyg a , are essentially infinitely far away , which implies that the line of sight from the array center / antenna location to the source is essentially parallel to that from the center of the earth to the sources , or @xmath78 . if we define a `` left - handed '' coordinate system such that @xmath75 points along the @xmath79-axis , then the source position is given by @xmath80 where @xmath81 is the angular elevation of the source , @xmath82 is the azimuthal angle measured from north though east , and the @xmath83 and @xmath84 unit vectors point toward the north and east , respectively , as viewed from the array / antenna . combining this with the assumption of parallel lines of sight to the source and the fact that the length of @xmath76 is set to @xmath85 yields the following expression @xmath86^{2 } \nonumber + \left [ r \ ; \mbox{cos}(h)\mbox{sin}(a ) + y_{a } \right ] ^{2 } \nonumber + \left [ r \ ; \mbox{sin}(h ) + z_{a } + \mbox{r}_{\mbox \scriptsize earth } \right ] ^{2}\ ] ] where @xmath87 , @xmath88 and @xmath89 are the coordinates of the antenna relative to the array center and @xmath90 . since the antenna positions are known , @xmath91 is the only unknown variable . equation ( a2 ) can then be rewritten as a quadratic equation and solved for @xmath91 keeping in mind that @xmath92 which allows one to compute the @xmath16 , @xmath61 , and @xmath79 coordinates of @xmath76 for the array center and each antenna in the current coordinate system . following this , a coordinate rotation was performed such that @xmath76 for the array center pointed along the @xmath79-axis and the @xmath16 and @xmath61 axes pointed toward north and east , respectively , as viewed from the location on the earth directly below the array center pierce - point . these rotated coordinates were then taken to be the @xmath16 and @xmath61 antenna positions projected onto the ionosphere thin shell for each time step . [ geo ] shows a graphical representation of these computations for the array center ( in black ) and for an exemplar antenna ( in gray ) . the second correction deals with the fact that the path length through the ionosphere is increased when the observed source is closer to the horizon . for a thin spherical shell , it is increased by a factor of @xmath93 where @xmath94 is the angle between the line of sight from the vla to the source and a line from the ionosphere pierce - point to the location on the earth directly below it . in the schematic in fig . [ geo ] , @xmath94 is the angle between the position vectors @xmath76 and @xmath95 . therefore , to compute the factor needed to correct our @xmath0 measurements , @xmath96 , we simply computed the dot product between these two vectors and divided by the product of their lengths , @xmath97 . finally , while computing the above geometric corrections , we also computed estimates of the apparent motion of cyg a within the coordinate system of each time step . this was done to estimate the degree of doppler shifting of the temporal / spatial frequencies of any detected wave phenomena . we did this for each time step by recomputing the position of the array center pierce - point for the two nearest time steps within the coordinate system of the current time step . these positions were then used to numerically compute the time derivatives of the @xmath16 and @xmath61 coordinates of the array center pierce - point to obtain the north - south and east - west components of the sidereal velocity . these are plotted in fig . [ svel ] as functions of time along with a histogram for the azimuth angle ( measured north through east ) of the sidereal velocity vector for the entire observing run . one can see from this figure that the velocities were sometimes significant , especially when the source was at lower elevations . in addition , while the motion is generally from east to west , as one would naively assume , there is a significant spread in position angle of more than @xmath98 . the authors would like to thank the referees for useful comments and suggestions . basic research in astronomy at the naval research laboratory is supported by 6.1 base funding . the vla was operated by the national radio astronomy observatory which is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . part of this research was carried out at the jet propulsion laboratory , california institute of technology , under a contract with the national aeronautics and space administration . coker , c. , s. e. thonnard , k. f. dymond , t. j. w. lazio , j. j. makela , and p. j. loughmiller ( 2009 ) , simultaneous radio interferometer and optical observations of ionospheric structure at the very large array _ rad . , rs0a11 cornwell , t. and e. b. fomalont ( 1999 ) , self - calibration , in _ synthesis imaging in radio astronomy ii , asp conference ser . 180 , edited by g. b. taylor , c. l. carilli , and r. a. perley , pp . 187199 , asp , san francisco , calif . helmboldt , j. f. , t. j. w. lazio , h. intema , and k. dymond ( 2011 ) , a new technique for spectral analysis of ionospheric tec fluctuations observed with the very large array vhf system : from qp echoes to mstids _ rad . , _ in press thompson , a. r. ( 1999 ) , fundamentals of radio interferometry , in _ synthesis imaging in radio astronomy ii , asp conference ser . _ , vol . 180 , edited by g. b. taylor , c. l. carilli , and r. a. perley , pp . 187199 , asp , san francisco , calif . for antenna 4 , rr polarization , as a function of time ( relative to midnight , 13 august ) when no attempt was made to `` de - spike '' the phase data ( see 2.2 ) middle panel : the difference between the cosine of the wrapped phase at a time step i , @xmath99 , and the next time step as a function of time used to find spikes in the phase data . flagged spikes are highlighted with circles . lower panel : the phase as a function of time , unwrapped with the flagged spikes excluded.,width=576 ] , and the 327 mhz phase , @xmath55 , with the 327 mhz phase scaled such that the ionospheric components of the two phases canceled out for the rr ( upper ) and ll ( lower ) polarizations for antenna 14 . the region where the instrumental phases have changed relatively abruptly ( i.e , the `` phase jump '' region first referred to in 2.2 ) is shaded in gray.,width=576 ] , and the 327 mhz phase , @xmath55 , with the 327 mhz phase scaled such that the ionospheric components of the two phases canceled out for antenna 3 , ll polarization . a fit to this data of a linear combination of the spatial frequencies @xmath2 and @xmath3 [ see equation ( 1 ) ] which assumes a single position offset ( see 2.2 ) is also plotted in red . middle : the same as in the upper panel , but for the phases with the smoothed ( with a one - hour wide box ) phases subtracted from each band . lower : the same as the middle panel , but with a fit of a linear combination of @xmath2 and @xmath3 used in the continuum subtraction of the 327 mhz phases at the edges of each scan ( see the discussion in 2.2).,width=576 ] ) of the vla , the the difference between the tec fluctuation ( i.e. , above or below the mean tec ) along its line of sight and that measured along the reference antenna s line of sight , @xmath100 . the estimated uncertainty ( see 2.2 ) is plotted in red in each panel.,width=576 ] ) of the vla , the the difference between the tec fluctuation ( i.e. , above or below the mean tec ) along its line of sight and that measured along the reference antenna s line of sight , @xmath100 . the estimated uncertainty ( see 2.2 ) is plotted in red in each panel.,width=576 ] ) of the vla , the the difference between the tec fluctuation ( i.e. , above or below the mean tec ) along its line of sight and that measured along the reference antenna s line of sight , @xmath100 . the estimated uncertainty ( see 2.2 ) is plotted in red in each panel.,width=576 ] ) at each antenna as a function of local time ( black curves ) . plotted in red are the projected tec gradients computed using the fitted polynomial coefficients plotted in fig . [ pfit ] . in each panel , the antenna number and the mean separation among the nearest antennas ( @xmath101 ) are printed.,width=576 ]

the theory of gravitational lensing has by now reached a considerable degree of sophistication @xcite . theorems have been found predicting ( or bounding ) the number of images in very general systems @xcite . shapes of stable caustics have been exhaustively classified @xcite , a non - perturbative notion of the lens map has been obtained @xcite , and various universal behaviors of images have been found for sources lying near caustics @xcite . these general results have been complemented by a number of detailed calculations for specific types of lenses . the majority of such calculations have been performed within the quasi - newtonian viewpoint of gravitational lensing commonly used in astrophysics @xcite . while various flavors of this formalism exist , most require that bending angles be small and that all lenses be nearly - newtonian mass distributions . this is to be contrasted with the more fundamental picture of gravitational lensing where light rays are modelled as null geodesics in a lorentzian spacetime . within this context , specific lensing calculations have been performed in kerr , reissner - nordstrm , and a handful of other geometries @xcite . while curvatures and bending angles may be large in these examples , they all involve ( at least conformally ) stationary spacetimes . it is of interest to understand if qualitatively new effects appear in dynamical cases . this paper considers the bending of light by ( decidedly non - stationary ) plane - symmetric gravitational waves . gravitational lensing by gravitational waves has previously been considered by a number of authors , although almost all of this work has been carried out within the weak - field regime @xcite . one exception is @xcite , where redshifts were computed in an exact solution to the vacuum einstein equation representing a plane gravitational wave . this work considered only very specific waveforms , and was confined to a coordinate patch too small to include caustics and other effects associated with the formation of multiple images . separately , extensive work has been devoted to non - perturbatively understanding the geodesic structure of generic plane wave spacetimes @xcite . this is clearly a subject closely related to gravitational lensing , although few explicit relations between the two subjects appear to have been made ( see , however , remarks in @xcite ) . it is the purpose of this work to provide a comprehensive and non - perturbative discussion of lensing in plane wave spacetimes . these geometries are a well - known subclass of _ pp_-waves ; plane - fronted waves with parallel rays . many plane wave spacetimes are exact solutions to the vacuum einstein equation . others may be interpreted as , e.g. , exact solutions in einstein - maxwell theory . while rather idealized from the astrophysical perspective , plane wave spacetimes admit a wide variety of interesting phenomena . depending on the waveform , any number of astigmatic and anastigmatic caustics may exist . examples admitting any specified number of images even an infinite number are easily constructed . in particular , even numbers of images can exist @xcite . the image count can also change in time even when a source does not cross an observer s caustic . despite all of these properties , plane wave spacetimes are geometrically very mild . they are topologically equivalent to @xmath0 and admit coordinate systems which cover the entire manifold . aside from their value as models of gravitational radiation , the plane wave spacetimes considered here have also found numerous applications via the penrose limit . this limit provides a sense in which the metric near any null geodesic in any spacetime is equivalent to the metric of an appropriate plane wave spacetime @xcite . it allows problems in relatively complicated spacetimes to be reduced to equivalent problems in plane wave spacetimes ( which are often much simpler ) . this has been particularly valuable within string theory and related fields @xcite . penrose limits have also been applied to ordinary quantum field theory in order to investigate causality and effective indices of refraction for photons and gravitons propagating in curved spacetimes @xcite . more recently , penrose limits were used to deduce the effect of caustics on green functions associated with the propagation of classical fields in curved spacetimes @xcite . given the content of the penrose limit , lensing in plane wave spacetimes might be related to lensing in generic spacetimes as seen by ultrarelativistic observers . we make no attempt to justify this conjecture , however . this paper starts by providing a self - contained review of plane wave spacetimes in sect . [ sect : planereview ] . although most of this material is not new @xcite , it is not widely known . [ sect : numimages ] then derives the number of images of a point source that may be viewed in plane wave spacetimes . under generic conditions , this is found to depend only on the waveform and a certain time parameter associated with the observation event . the number of images does not depend on any properties of the source . once this is established , sect . [ sect : imageproperties ] computes image positions , frequency shifts , angles , and image distortion parameters in general plane wave spacetimes . sect . [ sect : symwaves ] applies these results to symmetric plane waves , which have constant waveforms . these geometries produce an infinite number of images of almost every source . their lensing properties are found to change significantly if the ricci tensor is increased beyond a certain threshold . lastly , sect . [ sect : sandwaves ] discusses `` sandwich waves ; '' wavepackets with finite width . these spacetimes generically admit images which appear at discrete times and then persist indefinitely . such images initially provide infinitely blueshifted , infinitely bright pictures into the infinitely distant past . very quickly , however , such images become highly redshifted and effectively fade away . one of these `` transient flashes '' is produced by almost every sufficiently weak vacuum ( ricci - flat ) wave . for weak waves , a second flash appears only if the ricci tensor of the wavepacket exceeds a certain threshold . throughout this work , the spacetime is assumed to be everywhere transparent . the language used also assumes that the geometric optics approximation @xcite holds even in situations where it would be severely strained ( such as when light is emitted near an observer s caustic ) . this paper restricts attention only to plane wave spacetimes in four spacetime dimensions . our sign conventions follow those of wald @xcite . the signature is @xmath1 . latin letters @xmath2 ( and occasionally @xmath3 ) from the beginning of the alphabet are used to denote abstract indices . greek letters @xmath4 are used to denote four - dimensional coordinate indices . the latin letters @xmath5 are instead coordinate indices associated with the two directions transverse to the direction of wave propagation . objects involving the latter type of index are often written in boldface with all indices suppressed . they are then manipulated using the standard notation of linear algebra [ e.g. , @xmath6 and @xmath7 . overall , notation related to plane wave spacetimes closely follows the conventions of @xcite . plane wave geometries may be interpreted as simple models for gravitational waves emitted from distant sources . alternatively , they arise as universal limits for the geometries near null geodesics in any spacetime @xcite . the typical definition of a plane wave spacetime @xmath8 requires that @xmath9 and that there exist global coordinates @xmath10 such that the line element takes the form @xmath11 @xmath12 is any symmetric @xmath13 matrix . its components describe the waveforms associated with a wave s three polarization states . this leaves the usual two polarizations associated with vacuum general relativity . note that the six polarization states typically stated to exist for linearized `` plane '' gravitational waves in generic theories of gravity @xcite can not all be represented by the metric ( [ planewavemetric ] ) . three of these polarizations may be realized only as geometries which are rather less plane - symmetric than those considered here . ] . the @xmath14 coordinate is interpreted as a phase parameter for the wave , while @xmath15 affinely parametrizes its rays . the remaining two coordinates @xmath16 span spacelike wavefronts transverse to the wave s direction of propagation . note that if @xmath17 in some region , the spacetime is locally flat there . in terms of a minkowski coordinate system @xmath18 , @xmath14 and @xmath15 satisfy @xmath19 and @xmath20 in such a region . we consider only nontrivial plane waves , so @xmath21 can not vanish everywhere . the physical interpretation of the so - called brinkmann metric ( [ planewavemetric ] ) as a plane - symmetric gravitational wave follows from considering the integral curves of the vector field @xmath22 . these curves form a null geodesic congruence which may be interpreted as the rays of the gravitational wave . @xmath23 , so these rays have vanishing expansion , shear , and twist . there is therefore a sense in which they are everywhere parallel to one another . all rays are also orthogonal to the family of spacelike 2-surfaces generated by the two commuting spacelike vector fields @xmath24 . the induced metric on each such surface is flat : the wavefronts are 2-planes . the curvature is constant on these planes in the sense that the @xmath25 are curvature collineations : @xmath26 despite this , the @xmath25 are not everywhere killing . there do , however , exist linear combinations of @xmath27 and @xmath28 which _ are _ killing . @xmath28 itself is also killing , which may be interpreted as a statement that plane waves do not deform along their characteristics . plane wave spacetimes admit a minimum of five linearly independent killing fields . note that in flat spacetime , five ( out of the total of ten ) killing fields are symmetries of all electromagnetic plane waves @xcite . killing fields of plane wave spacetimes are discussed more fully in sect . [ sect : geodesic ] . all non - vanishing coordinate components of the riemann tensor may be determined from @xmath29 it follows from this that the ricci tensor is @xmath30 where @xmath31 denotes the ordinary ( euclidean ) trace of the @xmath13 matrix @xmath32 . the ricci scalar always vanishes in plane wave spacetimes . more generally , there are no nonzero scalars formed by local contractions of the metric , the curvature , and its derivatives : @xmath33 . this is analogous to the fact that plane electromagnetic waves satisfy , e.g. , @xmath34 . note , however , that electromagnetic plane waves are not the only electromagnetic fields with vanishing field scalars . similarly , plane wave spacetimes are not the only curved geometries with vanishing curvature scalars @xcite . it follows from ( [ ricci ] ) that plane wave spacetimes satisfying the vacuum einstein equation ( and the vacuum equations of many alternative theories of gravity @xcite ) are characterized by the simple algebraic constraint @xmath35 . for vacuum waves , there exist two scalar functions @xmath36 and @xmath37 such that @xmath38 @xmath39 and @xmath40 describe the waveforms for the two polarization states of a gravitational plane wave propagating in vacuum . a plane wave is said to be linearly polarized if @xmath39 and @xmath40 are linearly dependent ( in which case one of these functions can be eliminated by a suitable rotation of the transverse coordinates @xmath41 ) . if @xmath39 and @xmath40 have compact support , the geometry is said to be a sandwich wave . this name evinces the image of a curved region of spacetime `` sandwiched '' between null hyperplanes in a geometry that is otherwise minkowski . physically , it corresponds to a wavepacket of finite length . note that the planar symmetry considered here is very special in the sense that passing waves do not necessarily leave any `` tail '' behind them . after interacting with a sandwich wave , all observers enter a region of spacetime which is perfectly flat . there is a sense in which test fields propagating on plane wave spacetimes also have no tails @xcite . a general ( not necessarily vacuum ) wave profile @xmath32 may be built by adding to ( [ hvac ] ) a term proportional to the identity matrix @xmath42 . there then exists a third polarization function @xmath43 such that @xmath44 if the ricci tensor of such a wave is associated with a stress - energy tensor via einstein s equation , that stress - energy tensor obeys the weak energy condition if and only if @xmath45 . assuming this , the stress - energy tensors associated with ( [ hgen ] ) are very simple . they could be generated by , e.g. , electromagnetic plane waves with the form @xmath46 } x^1 . \label{emfield}\ ] ] alternatively , ( [ hgen ] ) could be associated with the stress - energy tensor of the massless klein - gordon plane wave @xmath47 besides the vacuum case @xmath48 , another interesting class of wave profiles are those that are conformally flat . these satisfy @xmath49 , so @xmath50 . as gravitational lenses , all caustics of conformally - flat plane waves are associated with `` perfect '' anastigmatic focusing . for more general plane waves , caustics are typically ( but not necessarily ) associated with astigmatic focusing . the geometry of plane wave spacetimes has been analyzed in detail by a number of authors @xcite . one essential conclusion of this work is that nearly all interesting properties of plane wave spacetimes may be deduced from the properties of @xmath13 matrices @xmath53 satisfying the differential equation @xmath54 this is a `` generalized oscillator equation '' with @xmath55 acting like a matrix of squared frequencies . ( [ edef ] ) arises when solving for geodesics or jacobi fields in plane wave spacetimes . bitensors such as synge s function and the parallel propagator may be written explicitly in terms of its solutions . the same is also true for a plane wave s killing vectors . it is convenient to write all possible matrices @xmath56 in terms of two particular solutions . fix any is used here because this will later be interpreted as a @xmath14 coordinate associated with some observer . similarly , @xmath57 is often interpreted below as a @xmath14 coordinate associated with a source ( not necessarily when that source is observable at any observation point with @xmath58 ) . later , the notation @xmath59 is used to denote a @xmath14 coordinate associated with the emission of light from a source . ] @xmath60 and define @xmath61 and @xmath62 to be solutions to ( [ edef ] ) ( where derivatives are applied to the first arguments of @xmath51 and @xmath52 ) with the initial conditions @xmath63 here , the notation @xmath64 indicates a partial derivative with respect to the first argument of @xmath51 . we assume for simplicity that @xmath65 is a matrix of piecewise - continuous functions and that @xmath51 and @xmath66 are at least @xmath67 ( and piecewise-@xmath68 ) in both of their arguments . example expressions for @xmath51 and @xmath52 are discussed in sects . [ sect : symwaves ] and [ sect : sandwaves ] . given any two solutions @xmath69 and @xmath70 to ( [ edef ] ) , it is easily verified that their wronskian is conserved : @xmath71 here , @xmath72 denotes a matrix transpose . applying this formula with @xmath73 and @xmath74 shows that @xmath75 neither @xmath76 nor @xmath66 are necessarily symmetric matrices . nevertheless , ( [ genwronskian ] ) and ( [ wronskian ] ) may be used to show that the products @xmath77 _ are _ symmetric wherever they exist @xcite . letting @xmath78 and @xmath79 in ( [ genwronskian ] ) shows that @xmath80 for all @xmath81 . this is essentially etherington s reciprocity law @xcite . a similar calculation may be used to show that @xmath82 as well . it is sometimes useful to consider partial derivatives @xmath83 with respect to the second arguments of @xmath51 and @xmath52 . the resulting matrices remain solutions to ( [ edef ] ) . comparing initial conditions shows that @xmath84 @xmath51 may therefore be derived from @xmath52 . the opposite is also true wherever @xmath85 . geometrically , @xmath66 corresponds to the transverse coordinate components of a jacobi propagator describing the evolution of deviation vectors along geodesics passing between different pairs of points @xcite . @xmath52 is also related to image distortion . up to an overall time dilation factor , it translates small differences in image position on an observer s sky to spatial deviations from a fiducial source point . in the language of @xcite , @xmath52 is proportional to the jacobi map . the symmetric matrix @xmath86 plays a similar role , but translates source separations to _ emission _ ( rather than observation ) angles . it is proportional to an object typically referred to as the optical deformation matrix . these statements are explained more fully in sect . [ sect : imagedist ] . it is often useful when working with plane wave spacetimes to consider hypersurfaces of `` constant phase . '' recalling the interpretation of the @xmath14 coordinate as a phase , let @xmath87 denote such a hypersurface . given two points @xmath88 and @xmath89 ( with @xmath90 ) , define the multiplicity or `` index '' of these points to be @xcite @xmath91 we also set @xmath92 whenever @xmath93 . it follows from ( [ bswitch ] ) that @xmath94 . for most pairs of points , @xmath95 . such pairs are said to be `` disconjugate . '' pairs @xmath96 satisfying @xmath97 are instead said to be conjugate with multiplicity @xmath98 . similarly , we call the pairs @xmath99 `` conjugate hyperplanes '' and the pairs of real numbers @xmath100 `` conjugate phases '' when @xmath101 . despite the appearance of ( [ idef ] ) , the index map @xmath102 describes phenomena which do not depend on any choice of coordinate system . conjugate pairs as described here are closely related to the conjugate points commonly considered in differential geometry and optics . in general , distinct points @xmath103 and @xmath104 on a given geodesic are said to be conjugate if and only if there exist nontrivial deviation vectors along that geodesic which vanish at both @xmath104 and @xmath103 . in plane wave spacetimes , this condition reduces to @xmath105 . defining the multiplicity of a pair of conjugate points to be the number of linearly independent deviation vectors which vanish at those points , that multiplicity is equal to @xmath106 . the concept of conjugacy associated with @xmath107 does not , however , require the specification of any particular geodesic . it is uniquely defined even for pairs of points connected by multiple geodesics or by none . indeed , these are the only cases where @xmath108 . all strong lensing effects associated with plane wave spacetimes are related to the existence of conjugate hyperplanes . it follows from ( [ idef ] ) that every pair of conjugate phases @xmath100 satisfies @xmath109 finding conjugate pairs and their multiplicities may be viewed as a matter of direct computation once @xmath65 is specified . alternatively , various sturm - type comparison theorems can be used to make general statements regarding the existence and separations of conjugate pairs for various classes of plane wave . see , e.g. , chapt . xi of @xcite for results relating to mathematical problems of this type and @xcite for an application to `` tame '' plane wave spacetimes . more specific examples are discussed in sects . [ sect : symwaves ] and [ sect : sandwaves ] below . the qualitative structure of geodesics in plane wave spacetimes is closely related to the index @xmath107 . first note that every disconjugate pair of points @xmath110 is connected by exactly one geodesic . if two points are conjugate , the number of connecting geodesics is either zero or infinity . sizes of geodesically connected regions may be summarized by @xmath111 + i ( p_s , p_o ) = 3 . \label{indextheorem}\ ] ] a similar relation exists for null cones when @xmath112 : @xmath113 + i ( p_s , p_o ) = 2 . \label{indextheorem2}\ ] ] the latter result has been referred to as an `` index theorem '' in @xcite . ( [ indextheorem ] ) implies that geodesics emanating from @xmath104 and intersecting a hyperplane @xmath114 that is disconjugate to @xmath115 form a 3-dimensional region . indeed , these geodesics fill the entire hyperplane . more interestingly , geodesics intersecting a conjugate hyperplane with multiplicity 1 fill only a 2-dimensional region on that 3-dimensional surface . geodesics intersecting a hyperplane with multiplicity 2 form a line . similarly , the null cone of a point reduces to a 1-dimensional curve on every hyperplane with multiplicity 1 . a null cone intersecting a hyperplane with multiplicity 2 is focused to a single point on that hyperplane . these two cases correspond to astigmatic and anastigmatic focusing , respectively . anastigmatic focusing tends to be unstable in the sense that perturbations tend to split a single multiplicity 2 phase into two closely - spaced phases each with multiplicity 1 . or a preferred @xmath116 hyperplane @xmath115 , plane wave spacetimes naturally divide into a number of open four - dimensional regions @xmath117 . these regions are separated from each other by the hyperplanes @xmath118 conjugate to @xmath115 . ] in many applications , there exists a preferred point @xmath104 , or perhaps a preferred hyperplane @xmath115 . @xmath104 may , for example , represent the position of an observer at a particular time . fixing this point , the set of hyperplanes conjugate to @xmath115 divides a plane wave spacetime into a ( possibly infinite ) number of open regions @xmath117 . let @xmath119 denote the largest connected region containing @xmath115 and excluding any portion of a hyperplane conjugate to @xmath115 . if there exists a smallest @xmath120 conjugate to @xmath121 , the surface @xmath122 is clearly contained in the boundary of @xmath119 . @xmath123 may then be defined as the largest connected region which includes @xmath122 as a boundary and contains points @xmath103 satisfying @xmath124 and @xmath125 . this continues the spacetime `` above '' @xmath122 . similar constructions may be used to define @xmath126 and @xmath117 for values of @xmath127 other than 1 . [ fig : nn ] . fixing a particular nonzero integer @xmath127 and real number @xmath121 , it is not necessary that @xmath126 exist at all . in general , the domain of @xmath128 is an open subset of @xmath129 . this domain can be empty for some @xmath127 . the geodesic uniqueness result described above can now be reduced to the statement that a point @xmath104 is connected to another point @xmath130 by exactly one geodesic if and only if there exists some @xmath127 such that @xmath131 . two - point tensors like synge s world function , the parallel propagator , and the van vleck determinant may be defined unambiguously throughout `` @xmath132 . '' this excludes from @xmath133 only a set of measure zero . beyond the qualitative geodesic structure of plane wave spacetimes discussed above , it is not difficult to obtain explicit coordinate expressions for all geodesics . let @xmath134 denote some geodesic and @xmath135 an affine parametrization of it . the vector field @xmath136 generating the characteristics of the gravitational wave is killing , so @xmath137 must be constant on @xmath138 . if this constant vanishes , @xmath138 is confined to a hypersurface of constant phase . such a geodesic has the form of a ( euclidean ) straight line in the coordinates @xmath139 . geodesics satisfying @xmath140 are more interesting . in these cases , the affine parameter can always be rescaled such that @xmath141 . it is then possible to identify that parameter with the phase coordinate @xmath14 . doing , so @xmath142 for all @xmath143 . the spatial components @xmath144 of any geodesic are fixed everywhere once @xmath145 and @xmath146 have been specified at some fiducial phase @xmath121 . in terms of the matrices @xmath51 and @xmath52 defined in sect . [ sect : ab ] , @xmath147 the associated @xmath15 coordinate of @xmath138 may be efficiently derived using that fact that the vector field @xmath148 is a homothety @xcite . this implies that @xmath149 , \label{vgeo}\ ] ] where @xmath150 is a constant . @xmath151 is closely related to the conserved quantity on @xmath138 associated with the killing field @xmath28 . the ( unit ) 4-velocity @xmath152 tangent to @xmath138 is related to @xmath153 via @xmath154 it follows that @xmath155 if the spacetime is nearly flat ( so @xmath156 ) , @xmath151 reduces to a particle s specific energy minus its specific momentum in the direction of the gravitational wave . the limit @xmath157 ( or @xmath158 ) may therefore be interpreted as ultrarelativistic motion in the direction of the gravitational wave . by contrast , the limit @xmath159 ( or @xmath160 ) corresponds to ultrarelativistic motion _ against _ the background wave . other killing fields present in essentially all plane wave spacetimes may be written as @xmath161 where @xmath162 is any 2-vector with the form @xmath163 for each choice of @xmath121 , there exists a four - parameter family of such vector fields . each of these is associated with a conservation law . two such conserved quantities may be summarized by @xmath164 . \label{pdef}\ ] ] fixing any @xmath121 , this 2-vector is conserved in the sense that it is independent of @xmath57 . another conserved 2-vector may be defined by @xmath165 . \label{cdef}\ ] ] in the weak - field limit , @xmath166 corresponds to the specific momentum transverse to the gravitational wave . in this same context , @xmath167 may be interpreted as the conserved quantity associated with boosts transverse to the gravitational wave . it constrains transverse displacements . note that both @xmath166 and @xmath167 depend on a choice of @xmath121 . this is analogous to the choice of origin necessary to define angular momentum in elementary mechanics . in stationary spacetimes , it is common to discuss various quantities related to gravitational lensing in terms of stationary observers ( and often stationary sources ) . while plane wave spacetimes are not stationary , there does exist sufficient symmetry to define similarly preferred sources and observers . two geodesics @xmath138 and @xmath168 can be said to be `` instantaneously comoving '' at @xmath58 when @xmath169 these conditions imply that the 4-velocities of both geodesics are parallel - transported versions of each other on the constant - phase hyperplane @xmath115 . note , however , that geodesics which are comoving at one phase are not necessarily comoving at any other phase . as implied by ( [ indextheorem ] ) , bundles of geodesics are strongly focused on conjugate hyperplanes . consider such a hyperplane associated with a phase @xmath126 conjugate to @xmath121 with multiplicity 1 . it is clear from ( [ idef ] ) that @xmath170 is a matrix with rank 1 . there therefore exists a unit 2-vector @xmath171 such that @xmath172 @xmath171 is unique up to sign . choosing any @xmath173 orthogonal to @xmath171 , the transverse spatial coordinates of all geodesics starting at a given point @xmath104 focus to the line @xmath174 as they pass through @xmath175 . here , @xmath176 is any real number . all @xmath15 coordinates may be reached on @xmath175 by appropriate geodesics . geodesics starting at @xmath104 and intersecting a hyperplane @xmath118 with multiplicity 2 all focus to the single transverse position @xmath177 as they pass through @xmath118 . as in the multiplicity 1 case , all values of @xmath15 may be reached by appropriate geodesics . ( [ focusnondeg ] ) and ( [ focusdeg ] ) illustrate explicitly how conjugate hyperplanes with multiplicities 1 and 2 are associated with astigmatic and anastigmatic focusing , respectively . the former case involves focusing in only one transverse direction , while the latter case involves simultaneous focusing in both directions transverse to the gravitational wave . as noted above , all pairs of points not lying on conjugate hyperplanes are connected by exactly one geodesic . there is therefore no ambiguity in ascribing geodesic distances to these pairs . in particular , synge s world function @xmath178 is well - defined whenever its arguments do not lie on conjugate hyperplanes . plane wave spacetimes constitute one of the few examples where @xmath179 is known essentially in closed form : @xmath180 . \label{sigdef}\end{aligned}\ ] ] this is symmetric in its arguments : @xmath181 . the appearance of @xmath182 in ( [ sigdef ] ) indicates that @xmath179 tends to diverge when its arguments approach conjugate hyperplanes . more specifically , suppose that @xmath183 and @xmath184 . then , @xmath185 ^ 2 \label{signondeg}\ ] ] is an asymptotic approximation for @xmath179 if the bracketed term on the right - hand side of this equation is nonzero @xcite . it follows from ( [ focusnondeg ] ) that this expression is valid only when there does not exist any geodesic passing from @xmath104 to a point on @xmath118 with the same transverse coordinates as @xmath103 . the equivalent result if @xmath103 is near a hyperplane @xmath118 with multiplicity 2 is @xmath186 again , this is valid only when there does not exist any geodesic passing from @xmath104 to a point on @xmath118 with the same transverse coordinates as @xmath103 . one of the most basic questions that can be asked regarding a gravitational lens is the number of images that it produces of a particular source . stated somewhat differently , how many future - directed null geodesics connect a given timelike curve ( the source ) to a particular spacetime event ( the observer at a particular time ) ? this may be answered using the geodesic structure of plane wave spacetimes summarized above . to an observation event @xmath104 . the arguments of sect . [ sect : numimages ] show that under generic conditions , exactly one image is emitted from each region @xmath117 lying before the observation event . the dashed lines correspond to constant-@xmath14 hyperplanes conjugate to @xmath115 . ] to fix the notation , let @xmath104 denote a fixed observation event and @xmath138 the timelike worldline of a point source . assume that @xmath138 may be parametrized by an everywhere-@xmath67 function @xmath135 . the phase coordinate @xmath14 serves as a useful `` quasi - time '' @xcite for plane wave spacetimes is not a `` true '' time function because @xmath187 is null . generically , there do exist well - behaved functions @xmath188 where @xmath189 is everywhere timelike @xcite . these are time functions in the usual sense , although they do not appear to simplify any computations performed here . incidentally , the existence of such time functions implies that plane wave spacetimes are stably causal . plane waves are not , however , globally hyperbolic @xcite . hypersurfaces of constant @xmath190 are not cauchy surfaces . ] , so let @xmath191 satisfy @xmath192 for all @xmath143 . the source s worldline is not required to be geodesic . this section establishes that under generic conditions , an ideal observer at @xmath104 may see exactly one image of @xmath138 from each of the `` epochs '' @xmath117 described in sect . [ sect : conjugate ] . somewhat more precisely , there typically exists exactly one future - directed null geodesic from @xmath193 to @xmath104 for each @xmath194 such that @xmath117 exists . [ fig : multimage ] . recall that the boundaries of @xmath117 depend only on the spacetime under consideration , and not at all on the behavior of any particular source . this allows generic bounds to be placed on time delays associated with the various images that may be observed . recall from sect . [ sect : conjugate ] that each point in @xmath195 is connected to @xmath104 via exactly one ( not necessarily null ) geodesic . synge s function @xmath196 is therefore well - defined and explicitly given by ( [ sigdef ] ) for all @xmath57 such that @xmath197 . consider instead the rescaled function @xmath198 if @xmath194 , the domain of @xmath199 is equal to all @xmath57 such that @xmath197 . if @xmath200 , we additionally suppose that @xmath201 for reasons of causality . images of @xmath195 produced by a plane gravitational wave correspond to the zeros of @xmath202 . recalling the form ( [ sigdef ] ) for @xmath179 , it is clear that @xmath202 depends on @xmath138 as well as the matrices @xmath51 , @xmath52 , @xmath182 , and @xmath203 . we have assumed in sect . [ sect : ab ] that @xmath76 and @xmath52 are at least @xmath67 . so is @xmath191 . the definition of @xmath117 ensures that @xmath204 everywhere @xmath199 is defined . @xmath182 is therefore @xmath67 and @xmath202 is continuous . @xmath202 is also monotonic . to see this , note that ( [ edef ] ) , ( [ wronskian ] ) , ( [ symmatrices ] ) , ( [ sigdef ] ) , and ( [ sigcapdef ] ) may be used to show that @xmath205 > 0 . \label{sigdot}\end{aligned}\ ] ] @xmath202 is both continuous and monotonic , so at most one zero can exist for each @xmath127 . this means that at most one image of a source may reach an observer from each epoch @xmath117 . _ exactly _ one such image exists if some @xmath206 are known to satisfy @xmath207 such bounds are easily established . the simplest case to consider is one where @xmath117 lies `` in between '' hyperplanes conjugate to @xmath115 . suppose that both @xmath126 and @xmath208 exist for some @xmath209 . this is true in fig . [ fig : multimage ] for @xmath210 . more generally , ( [ signondeg ] ) and ( [ sigdeg ] ) imply that if there are no geodesics connecting @xmath104 to either @xmath211 or @xmath212 , @xmath213 it follows from these limits together with continuity that @xmath202 is surjective on @xmath129 . since this function is also monotonic , there must exist exactly one emission phase @xmath214 such that @xmath104 and @xmath215 are connected by a null geodesic . under the same assumptions , projectiles moving on timelike geodesics may be thrown from @xmath197 to @xmath104 only if @xmath216 . choosing @xmath217 to be sufficiently small ( but positive ) , these projectiles can require an arbitrarily large amount of proper time to intersect @xmath104 . it is somewhat curious that points @xmath197 satisfying @xmath218 can not be connected to @xmath104 by any causal geodesic . such points may , however , be reached by suitably accelerated curves which are everywhere causal . next , consider the case @xmath200 when there exists at least one conjugate hyperplane in the observer s past ( as occurs in the example illustrated by fig . [ fig : multimage ] ) . the arguments given above imply that if @xmath219 is geodesically disconnected from @xmath104 , @xmath220 the other boundary of the domain of @xmath221 occurs at @xmath222 . here , @xmath223 this limit clearly tends to @xmath224 if @xmath225 . physically , @xmath225 implies that @xmath104 can not be connected to @xmath226 by any null geodesic . assuming that this is true , there must exist exactly one @xmath227 such that @xmath104 and @xmath228 are connected by a null geodesic . the results of sects . [ sect : inbetween ] and [ sect : young ] typically suffice to describe the images formed in plane wave spacetimes containing an infinite number of conjugate hyperplanes in an observer s past . it is , however , important to consider cases where only a finite number of conjugate points exist . this occurs , e.g. , for finite wavepackets where @xmath32 has compact support . it is also true of linearly polarized vacuum waves that are tame in the sense described in @xcite . if there is at least one conjugate hyperplane in the observer s past , let @xmath229 denote the smallest negative integer such that @xmath230 exists . if there are no conjugate hyperplanes in the observer s past , set @xmath231 . the case illustrated in fig . [ fig : multimage ] corresponds to @xmath232 if @xmath233 extends into the infinite past . regardless , we ask whether there exist any future - directed light rays from @xmath234 to @xmath104 . unlike in the cases considered above , sources spend an infinite amount of proper time in @xmath235 . even in flat spacetime , a source that accelerates for an infinitely long time may be causally - disconnected from certain observers via a rindler horizon . such phenomena can be ruled out here by supposing that there exists some finite constant @xmath236 such that @xmath237 > \kappa_{\mathrm{min } } \label{vdotbound}\ ] ] for all @xmath57 less than some cutoff . as is clear from ( [ kappadef ] ) and ( [ 4vel ] ) , the left - hand side of this inequality acts like the square root of a time dilation factor between the coordinate @xmath14 and the source s proper time . eq . ( [ vdotbound ] ) implies that the source s 4-velocity @xmath238 satisfies @xmath239 sufficiently far in the past . it therefore excludes sources which experience arbitrarily large boosts against the background gravitational wave in the distant past . it is satisfied by , e.g. , sources whose motion is geodesic sufficiently far into the past . assuming that @xmath240 exists , it is clear from ( [ sigdot ] ) that @xmath241 may be chosen sufficiently small that @xmath242 further assuming that @xmath243 is geodesically disconnected from @xmath104 ( if @xmath244 ) or that @xmath226 is not null - separated from @xmath104 ( if @xmath231 ) , it follows that there exists exactly one @xmath59 smaller than @xmath230 ( if @xmath245 ) or @xmath121 ( if @xmath231 ) such that @xmath246 is connected to @xmath104 via a future - directed null geodesic . the results just described may be summarized as follows : suppose that there does not exist any past - directed geodesic segment from @xmath104 to @xmath138 whose endpoints are conjugate ( in the usual sense ) . also assume that the source and observer are not instantaneously aligned with the background wave : @xmath247 . if there exists an `` oldest '' phase conjugate to @xmath121 , further require that there be some @xmath248 such that the source s motion is bounded by ( [ sourcebound ] ) sufficiently far in the past . for each @xmath194 , these assumptions imply that an observer at @xmath104 sees exactly one image of @xmath138 as it appeared in @xmath117 . this provides a strong bound on the possible emission times of different images . if @xmath235 exists for every negative integer @xmath229 , an infinite number of images are formed . if , however , there is some smallest @xmath249 such that @xmath235 exists , @xmath250 images appear at @xmath104 . note that these results depend only on the waveform @xmath65 and the phase coordinate @xmath121 associated with the observer . the total number of images is the same for all sources satisfying the hypotheses outlined above . these hypotheses are generic . recalling ( [ indextheorem ] ) , the requirement that @xmath138 exclude any points conjugate to @xmath104 along a connecting geodesic is equivalent to demanding that the source s worldline avoid certain well - behaved one- or two - dimensional subsets of the three - dimensional hyperplanes conjugate to @xmath115 . similarly , the assumption that @xmath247 demands only that @xmath138 avoid a certain line on @xmath115 . the bound ( [ sourcebound ] ) on a source s asymptotic 4-velocity can fail to hold only for sources which accelerate for an infinitely long time . violating any of these conditions for a fixed observer would require that a source s worldline be quite exceptional . moreover , we now show that @xmath250 images appear under even broader ( but more difficult to state ) conditions than those just discussed . despite the comments made above , our assumptions on the behavior of a source s worldline can be violated in certain cases . suppose , contrary to these assumptions , that there exists at least one point @xmath251 which is conjugate to @xmath104 along a geodesic connecting these two points . an infinite number of geodesics then pass between @xmath252 and @xmath104 . if the connecting geodesics are null , @xmath252 lies on a caustic of the observer s past light cone . continuous images of point sources `` einstein rings '' are then formed at @xmath104 ( ignoring the associated breakdown of geometric optics ) . such cases are not considered any further here . if conjugate points between the source and observer are associated with non - null geodesics , discrete images of @xmath138 appear at @xmath104 . in these cases , the methods used above are easily adapted to find how many images of @xmath195 arrive at @xmath104 . as already mentioned , there can be no more than one root for each @xmath253 . depending on the details of the system , the associated image from @xmath117 may or may not exist . the case considered in sect . [ sect : inbetween ] where @xmath195 lies in between successive conjugate hyperplanes is the simplest to analyze . assuming that @xmath138 does not intersect the caustic of @xmath104 , no images are formed of @xmath195 if either @xmath211 is connected to @xmath104 via a timelike geodesic or @xmath212 is connected to @xmath104 via a spacelike geodesic . otherwise , exactly one image exists from this region . if there exists an oldest conjugate hyperplane @xmath254 as described in sect . [ sect : oldest ] , suppose that the source satisfies ( [ vdotbound ] ) for some @xmath255 . there are then zero images of @xmath234 if @xmath243 is connected to @xmath104 via a timelike geodesic . there is exactly one image if these points are either geodesically disconnected or are connected by a spacelike geodesic . the last cases to consider concern images of @xmath256 . first suppose that @xmath247 . if there exists at least one conjugate hyperplane in the observer s past , one image is formed of @xmath256 if either @xmath219 is geodesically disconnected from @xmath104 or it is connected by a timelike geodesic . no images are formed if @xmath219 and @xmath104 are connected by a spacelike geodesic . if @xmath247 and there are no conjugate points in the observer s past , condition ( [ vdotbound ] ) implies that there exists exactly one image of @xmath256 . cases where the source and observer are instantaneously aligned are more interesting . suppose that @xmath257 . there then exists one image of @xmath256 with @xmath258 . recalling that @xmath121 is not in the domain of @xmath221 , it is possible for a second image to be emitted from @xmath256 if @xmath259 somewhere . this may be seen by noting that @xmath260 two images of @xmath256 can therefore exist when @xmath261 and @xmath257 . if a source includes points which are conjugate to the observer ( in the ordinary sense ) , there is no simple result for the total number of images formed . nevertheless , it is always possible to say that the total number of images is less than or equal to @xmath262 if a source does not intersect a caustic of the observer s light cone . plane wave spacetimes typically produce multiple images of each source . even for sources whose intrinsic properties remain constant , these images can appear with different spectra , brightnesses , etc . we now compute these properties for generic configurations satisfying the hypotheses summarized in sect . [ sect : totimages ] . for each image of a timelike worldline @xmath138 seen at @xmath104 , there is an associated null geodesic segment connecting @xmath104 to an appropriate emission point @xmath263 . in terms of synge s function ( [ sigdef0 ] ) , these points satisfy @xmath264 first derivatives of @xmath179 are always tangent to the connecting light ray . in particular , the vector @xmath265 at @xmath104 points along the geodesic which eventually intersects @xmath252 ( and is therefore past - directed ) . the derivative operator here is understood to act on the second argument of @xmath179 . also note that @xmath266 is normalized such that @xmath267 . parallel - transporting @xmath266 to the observation point yields @xmath268 the derivative operator in this equation is understood to act on the first argument of @xmath179 . both @xmath266 and @xmath269 may be viewed as ( dimensionless ) separation vectors between @xmath104 and @xmath252 . ( [ sigdef ] ) and the various identities of sect . [ sect : ab ] may be used to compute the explicit coordinate components of @xmath269 and @xmath266 . components transverse to the direction of wave propagation are @xmath270 , \label{tangent1}\ ] ] where @xmath271 and @xmath272 denote the transverse coordinates of the observer and source . a similar calculation shows that @xmath273 . \label{tangent2}\ ] ] @xmath274 and @xmath275 are related via @xmath276 much of the discussion below considers sources moving on geodesics . in these cases , use of ( [ spatgeo ] ) shows that @xmath277 here , @xmath278 . the various identities involving @xmath51 and @xmath52 discussed in sect . [ sect : ab ] may also be used to reduce the imaging condition ( [ sigzero ] ) to @xmath279 here , @xmath280 . ( [ siggeo ] ) is a nonlinear relation for the emission `` time '' @xmath59 in terms of the observer s position @xmath104 and the parameters @xmath281 , @xmath282 , @xmath283 , @xmath151 describing the source s worldline . as discussed in sect . [ sect : numimages ] , there can be many solutions to ( [ siggeo ] ) . these correspond to different images . neither @xmath59 nor @xmath266 depends on the observer s motion . nevertheless , redshifts and angles on the observer s sky do depend on that motion ( as is true even in flat spacetime ) . it is often useful to fix this effect by supposing that the observer is instantaneously comoving with the source . following ( [ comoving ] ) , this is taken to mean that the unit 4-velocities @xmath284 of the source and observer on @xmath115 satisfy @xmath285 recall from sect . [ sect : geodesic ] that the `` transverse momentum '' @xmath286 is generated by contracting @xmath287 with the two killing fields equal to @xmath25 at @xmath104 and having vanishing first derivative at that point . also note that ( [ comoving3 ] ) implicitly defines an instantaneous observer velocity @xmath288 normalized ( like @xmath153 ) such that @xmath289 . gravitational lenses typically discussed in astrophysics involve nearly - newtonian mass distributions which may be regarded as approximately stationary ( at least on sub - cosmological timescales ) . if both a source and an observer are sufficiently far from such a lens , there can be no significant redshift or blueshift from the gravitational field of that lens . roughly speaking , a light ray falling into any stationary gravitational potential must climb out of that same potential . this result breaks down if light passes through non - stationary regions of spacetime . indeed , plane wave spacetimes may produce images with significant frequency shifts @xcite . consider an approximately monochromatic beam of light emitted from @xmath252 and received at @xmath104 . a future - directed tangent vector @xmath290 to the emitted light ray may always be chosen such that @xmath291 is the angular frequency of the light as seen by its source . the frequency @xmath292 of this same light ray as measured by an observer at @xmath104 is @xmath293 , where @xmath294 is equal to @xmath295 parallel transported from the source to the observer . the observed and emitted frequencies are therefore related by @xmath296 here , @xmath297 and @xmath298 . the 2-vectors @xmath299 and @xmath300 appearing here are determined by the source and observer positions via ( [ sigdef ] ) , ( [ sigzero ] ) , ( [ tangent1 ] ) , and ( [ tangent2 ] ) . the resulting frequency shift is valid for all emission points not contained in a caustic of @xmath104 . now suppose that a source moves on a geodesic and that the observer is instantaneously comoving with this geodesic in the sense of ( [ comoving2 ] ) and ( [ comoving3 ] ) . then @xmath301 does nt depend on which image is chosen . ( [ tangentdiff ] ) , ( [ lambdadotgeo ] ) , and the symmetry of @xmath302 may be used to rewrite ( [ redshiftgen ] ) as @xmath303 the matrix in parentheses on the right hand side of this equation acts like a metric for the `` separation '' 2-vector @xmath304 $ ] . if both eigenvalues of @xmath305 are negative , the source is necessarily redshifted . conversely , sources are always blueshifted when this matrix is positive definite . if @xmath306 has both positive and negative eigenvalues , the sign of the frequency difference depends on the direction of @xmath307 . for special configurations , there is no frequency shift at all . various images formed from a single source appear at different points on an observer s sky . like redshifts , the relative angles between images change depending on an observer s 4-velocity @xmath308 . the angle @xmath309 between two images arriving at @xmath104 with tangents @xmath266 and @xmath310 is @xmath311 simplifying , @xmath312 this expression is valid for arbitrary source and observer configurations . specializing to geodesic sources and comoving observers , @xmath313 here , @xmath314 and @xmath315 . it is evident that angles are largely controlled by the difference between @xmath316 at the two emission times . another interesting angle to consider is the observed separation @xmath317 between a single image ( emitted at @xmath252 ) and a generator @xmath28 of the background gravitational wave . for arbitrarily moving source and observer configurations , @xmath318 for observers comoving with geodesic sources , @xmath319 this may be used to rewrite the angle @xmath309 between two different images partially in terms of the angles @xmath317 and @xmath320 those images make with @xmath28 . using ( [ costheta ] ) , @xmath321 similarly , the frequency shift ( [ redshiftgeodesic ] ) of an individual image may be rewritten as @xmath322 it is evident from this equation that images which appear highly blueshifted to comoving observers must satisfy @xmath323 . thus far , all sources here have been modelled as though they were confined to timelike worldlines . real objects are not pointlike , however . they form extended worldtubes in spacetime . images of such worldtubes form null geodesic congruences which converge on @xmath104 . these images can be significantly distorted by the curvature of spacetime . it is simplest to quantify such distortions by first fixing a particular null geodesic @xmath324 passing between some part of the source and @xmath104 . precisely which geodesic is chosen is not important . @xmath324 serves only as an origin from which to discuss nearby light rays connecting @xmath104 to other points in the source . once this origin has been fixed , the image of an extended source may be described entirely using deviation vectors on @xmath324 ( at least for sufficiently small sources ) . [ fig : distortion ] . is indicated together with another light ray separated from it by a deviation vector @xmath325 . the vector @xmath266 is also drawn . this is tangent to @xmath324 at the observation point @xmath104 . ] deviation vectors ( or jacobi fields ) satisfy the geodesic deviation ( or jacobi ) equation along @xmath324 . letting @xmath326 denote the past - directed null vector tangent to @xmath324 and obtained by parallel - transporting @xmath266 from @xmath104 , every deviation vector @xmath325 is a solution to @xmath327 this equation is linear , so @xmath325 must depend linearly on initial data . in particular , all deviation vectors can be written as linear combinations of appropriate bitensors contracted into the initial data . this is done to avoid confusion when writing down two - point tensors such as @xmath328 [ see ( [ bdef ] ) ] . ] @xmath329 and @xmath330 . all light rays observed at @xmath104 must necessarily intersect that point , so it suffices to set @xmath331 . the first derivative of a deviation vector at @xmath104 describes an angular deviation between one point of an image and the center associated with @xmath324 . we therefore consider deviation vectors @xmath325 with the form @xmath332 @xmath328 is known as a jacobi propagator . it satisfies the jacobi equation @xmath333 along @xmath324 together with the initial conditions @xmath334 note that @xmath328 is a bitensor . it maps vectors at @xmath104 into vectors at others points on @xmath324 . the transverse components of @xmath328 are @xmath335 where @xmath52 is the matrix defined in sect . [ sect : ab ] . other components of @xmath328 may be deduced from the eigenvector relations @xcite @xmath336 all parts of an image must arrive at an observer along null geodesics . additionally , an observer with 4-velocity @xmath308 can only measure angles of vectors orthogonal to @xmath308 . it therefore suffices to restrict attention to deviation vectors satisfying @xmath337 at @xmath104 . these constraints restrict all interesting initial data to a two dimensional space . the orthonormal vectors @xmath338 form a basis for this space at @xmath104 . they satisfy @xmath339 parallel - transporting @xmath340 to another point on @xmath324 yields @xmath341 which forms a sachs basis @xcite on @xmath324 . initial data appearing in ( [ bdef ] ) must be of the form @xmath342 where @xmath343 is an unconstrained 2-vector . the factor @xmath344 is included here so that @xmath343 is directly related to angles on an observer s sky . a sufficiently small image may be described by a suitable set of 2-vectors @xmath343 representing the angular locations of each portion of the image with respect to the center defined by @xmath324 . each @xmath343 may be translated into a physical displacement at the source using ( [ bdef ] ) and ( [ initdata ] ) . first note that for every particular @xmath343 , ( [ rdotb ] ) implies that @xmath345 throughout @xmath324 . indeed , @xmath325 is always a linear combination of the @xmath346 together with @xmath326 . components of @xmath325 proportional to @xmath326 are physically irrelevant , so we consider only the sachs components @xmath347 \alpha^{(j)}.\ ] ] defining the @xmath13 matrix @xmath348 it is then clear that @xmath349 for any @xmath343 . @xmath350 is referred to as the jacobi matrix or jacobi map @xcite . using ( [ btrans])-([elldotb ] ) , ( [ sachso ] ) , ( [ sachsgen ] ) , and ( [ ddef ] ) , @xmath351 if a source moves on a geodesic which is instantaneously comoving with the observer , @xmath352 this discussion implies that a portion of an image with angular separation @xmath353 from the fiducial direction associated with @xmath354 is spatially separated from the fiducial emission point @xmath355 by @xmath356 the factors of @xmath357 have been introduced here to emphasize that this description is valid only for infinitesimal deviations . regardless , ( [ binterpret ] ) shows that up to the time dilation factor @xmath358 , the matrix @xmath52 central to all aspects of plane wave geometry may be physically interpreted as a transformation converting infinitesimal angles on the vertex of a light cone into infinitesimal separations elsewhere on that light cone . @xmath52 depends only on the @xmath14 coordinates of the source and emission points , and not on any other aspects of the physical configuration . it may be computed for all possible observer - source pairs directly from the wave profile @xmath65 . angles of emission ( as opposed to observation ) of the various light rays travelling from the source to the observer may be found by differentiating ( [ binterpret ] ) and applying the appropriate time dilation factor : @xmath359 the last equality here makes use of ( [ redshiftgen ] ) . applying ( [ binterpret ] ) shows that @xmath360 the symmetric matrix @xmath361 therefore converts spatial locations to emission angles within the source ( with the constraint that all light rays intersect @xmath104 ) . it is referred to as the optical deformation matrix @xcite . ( [ binterpret ] ) implies that there is a sense in which circles on the observer s sky correspond to ellipses near @xmath252 . this deformation may be parametrized by performing a polar decomposition on @xmath362 : @xmath363 here , @xmath364 and @xmath365 represent rotation matrices through some angles @xmath366 and @xmath367 . the ratio @xmath368 is related to the ellipticity of the aforementioned ellipse . @xmath367 represents the angle between the principal axes of that ellipse and the sachs basis . @xmath369 and @xmath367 are referred as shape parameters @xcite . recalling that polarization vectors are parallel - transported in the geometric optics approximation @xcite , any polarization vector must have sachs components which are constant along @xmath324 . in principle , the angle @xmath367 might therefore be measured by comparing the relative `` rotation '' between an object s observed shape and an appropriate polarization angle @xcite . for linearly polarized waves where @xmath65 can be made diagonal by an appropriate coordinate choice , @xmath370 with respect to this coordinate system and the basis ( [ sachsgen ] ) . it is shown in sect . [ sect : weakwave ] that @xmath367 also vanishes in a natural way for all sufficiently weak wavepackets which are nonzero only for short times . ( [ binterpret ] ) implies that @xmath371 converts angles at the observer to separations within the source . the determinant of @xmath350 must therefore relate solid angles at @xmath104 to physical areas near @xmath252 : @xmath372 it follows that @xmath373 may be interpreted as an `` angular diameter distance . '' absolute value signs are necessary here because @xmath374 changes sign after each pass through a conjugate hyperplane with multiplicity 1 . physically , such sign changes represent parity inversions of the resulting image . note that @xmath375 does not necessarily increase monotonically with the age of an image ( as computed using the source s proper time ) . closely related to the angular diameter distance is the luminosity distance @xmath376 one factor of @xmath377 arises here from considering light cones emanating from the source instead of the observer . the other factor of @xmath378 is related to the energy change associated with frequency shifts . now that various optical quantities have been computed for general plane wave spacetimes , we consider their application to various special cases . the simplest nontrivial plane waves are the symmetric waves . these are locally symmetric in the sense that @xmath379 . it follows from ( [ riemann ] ) that symmetric plane waves must have constant waveforms . also note that @xmath380 is killing in these examples [ as well as @xmath136 , which is killing in all plane wave spacetimes ] . particular symmetric plane waves may be specified entirely by the ( constant ) eigenvalues of @xmath65 . recalling the decomposition ( [ hgen ] ) of @xmath65 into @xmath39 , @xmath40 and @xmath43 , a coordinate rotation may always be used to set @xmath381 for symmetric waves . it is then evident that the two eigenvalues of @xmath65 are given by @xmath382 . it is always possible to set @xmath383 where @xmath384 the weak energy condition implies that @xmath385 , so at least one eigenvalue of @xmath65 must be negative ( implying that at least one of the @xmath386 must be positive ) . we assume for definiteness that @xmath387 . then , @xmath388 if the vacuum einstein equation is imposed , @xmath389 and @xmath390 . for conformally - flat geometries representing spacetimes associated with , e.g. , pure electromagnetic plane waves , @xmath391 and @xmath392 . other cases may be viewed as superpositions of gravitational and ( `` gravito''-)electromagnetic waves . all symmetric waves produce an infinite number of images of almost every source . it is clear from ( [ symwave ] ) that these waves are also linearly polarized . the angles @xmath367 and @xmath366 appearing in ( [ distortion ] ) therefore vanish when considering image deformations with respect to the sachs basis ( [ sachsgen ] ) . other lensing properties depend on the sign of @xmath393 . we call the case @xmath394 `` gravity - dominated '' and the case @xmath395 `` matter - dominated . '' consider symmetric plane wave spacetimes where @xmath396 . gravity - dominated waves such as these generalize the vacuum waves satisfying @xmath397 . symmetric vacuum waves arise from , e.g. , the penrose limit of a null geodesic orbiting a schwarzschild black hole on the light ring . for any gravity - dominated symmetric wave , the matrices @xmath51 and @xmath66 defined in sect . [ sect : ab ] are @xmath398 it is clear that @xmath399 has an infinite number of zeros for any choice of @xmath121 . each of these zeros represents a phase conjugate to @xmath121 . there are an infinite number of such phases in both the past and future of every observer . the discussion in sect . [ sect : numimages ] therefore implies that under generic conditions , an infinite number of images appear for almost every source . explicitly , all conjugate phases are given by @xmath400 where @xmath127 is any nonzero integer . it is evident from ( [ idef ] ) that all of these phases have multiplicity 1 . for any @xmath209 and any observation point @xmath104 with @xmath401 , exactly one image of each source is visible as that source appeared in @xmath117 . this corresponds to the region between @xmath402 and @xmath403 . note that @xmath399 switches sign on each pass through a conjugate phase . the parity of an image emitted from @xmath117 is therefore opposite to the parity of an image emitted from @xmath404 . specializing to cases where the source is a geodesic and the observer is instantaneously comoving with that source on @xmath115 , some configurations lead to redshifts and others to blueshifts . using ( [ redshiftgeodesic ] ) , @xmath405 an image is therefore redshifted if and only if @xmath406 it is blueshifted when @xmath407 . there is no frequency shift at all in cases where @xmath408 the direction of the frequency shift clearly depends only on the instantaneous orientation @xmath409 of the source and the observer on @xmath115 . in particular , it does not depend on which image is considered . all images of a particular source experience the same type of frequency shift . emission times @xmath59 for an observer comoving with a geodesic source may be found by solving ( [ siggeo ] ) . for gravity - dominated symmetric waves , this equation reduces to @xmath410 if @xmath411 , it is evident that the image from @xmath412 must satisfy @xmath413 . images from the distant past are therefore emitted at phases @xmath59 very nearly conjugate to @xmath121 : @xmath414 substituting this relation into ( [ psidef ] ) and ( [ redshiftgravsym ] ) shows that very old images cluster near @xmath28 on the observer s sky and experience increasingly - negligible frequency shifts : @xmath415 old images of slightly extended sources are also highly distorted and demagnified . their angular diameter and luminosity distances both scale like @xmath416 gravity - dominated symmetric waves therefore produce an infinite number of exponentially dimming images for almost every source . matter - dominated symmetric waves satisfying @xmath395 act somewhat differently than gravity - dominated waves . in these cases , @xmath417 phases conjugate to @xmath121 occur at @xmath418 and at @xmath419 , where @xmath420 are any nonzero integers . if @xmath421 is an irrational number , these two families of phases are distinct . each conjugate pair then has multiplicity 1 . if @xmath421 is rational , some conjugate pairs have multiplicity 2 . in the conformally - flat case where @xmath422 , all conjugate phases have multiplicity 2 . in every other case where @xmath421 is rational , an infinite number of conjugate phases occur with each multiplicity . regardless of @xmath393 , an infinite number of images are formed for almost every source . now consider a luminous source moving on a geodesic . if the source and observer are instantaneously comoving in the sense of ( [ comoving2 ] ) and ( [ comoving3 ] ) , frequency shifts associated with each image are given by ( [ redshiftgeodesic ] ) . since @xmath423 is negative - definite in this case , all images are redshifted . if @xmath411 , an image originating from @xmath412 must be emitted just before the source intersects @xmath118 . all such images cluster towards @xmath28 in the observer s sky and have negligible frequency shifts . images emitted near conjugate hyperplanes with multiplicity 1 are highly distorted . images emitted near conjugate hyperplanes with multiplicity 2 are not significantly distorted at all . in both cases , however , older images are dimmer ( although the rate at which this occurs is much slower than for gravity - dominated symmetric waves ) . regardless of the sign of @xmath393 , the oldest images formed by symmetric plane wave spacetimes depend on the spacetime structure at arbitrarily large transverse distances . if the metric is modified so that the wave decays at large distances , only a finite number of images discussed here would be unaffected . the oldest images found in pure symmetric waves likely do not appear at all in perturbed symmetric waves . symmetric plane waves are mathematically simple , but are not reasonable models for gravitational radiation emitted from compact sources . more interesting are waves where @xmath21 is nonzero only for a finite time : sandwich waves . suppose , in particular , that there exists some @xmath424 such that @xmath425 for all @xmath426 $ ] . it follows from ( [ riemann ] ) that such spacetimes are locally flat whenever @xmath427 or @xmath428 . the curved region containing the gravitational wave is effectively sandwiched between the two null hyperplanes @xmath429 and @xmath430 . every timelike curve eventually passes entirely through such a wave . before an observer interacts with the wave , spacetime is flat and optics is trivial . the case @xmath431 where an observer has already passed through the wave is more interesting . in this case , @xmath51 and @xmath52 reduce to their flat space forms @xmath432 when @xmath433 . the forms of @xmath51 and @xmath52 inside the wave depend on the details of @xmath65 , and will not be discussed here . if @xmath434 , however , there always exist four constant @xmath13 matrices @xmath343 , @xmath435 , @xmath436 and @xmath437 such that @xmath438 note that the dots on @xmath435 and @xmath437 do not refer to derivatives in this case . they are only used as a labelling device . it follows from ( [ primedd ] ) that @xmath439 is independent of @xmath121 . moreover , @xmath440 if there were no wave at all , @xmath441 and @xmath442 . in general , @xmath435 and @xmath437 have a simple physical interpretation . if two geodesics are comoving and have a transverse separation @xmath443 when @xmath428 , it follows from ( [ spatgeo ] ) that the transverse separation between these geodesics is @xmath444 immediately before they interact with the wave at @xmath445 . similarly , the relative transverse velocity of these geodesics is @xmath446 when @xmath427 . interpretations for @xmath343 and @xmath436 are somewhat less direct . consider two geodesics which intersect at some time @xmath431 after the wave has passed , but which have a relative transverse velocity @xmath447 at @xmath58 . the difference in transverse velocities between these two geodesics is then @xmath448 when @xmath427 . similarly , the difference in the transverse positions of these geodesics is @xmath449 at @xmath445 . it follows that @xmath343 controls shifts in position that are independent of the time @xmath121 at which the two geodesics cross each other . in principle , @xmath343 , @xmath435 , @xmath436 , @xmath437 may all be found by solving ( [ edef ] ) if @xmath65 is known . at first glance , this would appear to imply that @xmath450 numbers are required to describe observations through a sandwich wave . the actual number of required parameters is somewhat less than this . first note that the wronskian identity ( [ wronskian ] ) implies that @xmath451 is a symmetric matrix . it also implies that @xmath452 further simplifications arise by recalling from ( [ symmatrices ] ) that @xmath453 and @xmath454 are symmetric . it follows that @xmath455 are symmetric as well . these expressions are completely general , and hold for any sandwich wave . they significantly constrain the number of independent parameters needed to specify @xmath51 and @xmath52 . equivalently , they limit the number of parameters that must be extracted from @xmath65 . it follows from the arguments of sect . [ sect : numimages ] that the number of images of a generic source observable in any plane wave spacetime is governed by the number of hypersurfaces conjugate to the @xmath456 hypersurface @xmath115 containing the observation event @xmath104 . continuing to assume that @xmath431 , all phases conjugate to @xmath121 must be smaller than @xmath457 . it follows from ( [ zerodet ] ) and ( [ bsandwich ] ) that conjugate phases occurring before the wave may be found by solving @xmath458 = 0 \label{conjsand}\ ] ] for all @xmath459 . this equation is quadratic , so at most two solutions exist . an observer ahead of the wave may therefore see at most three images of a source as it appeared behind the wave . there may also be at most one image of a source as it appeared ahead of the wave . in principle , any number of images may arise from inside the wave [ where ( [ bsandwich ] ) is not valid ] if @xmath21 is sufficiently large . conjugate phases found by solving ( [ conjsand ] ) clearly depend on the observation time @xmath121 . less obviously , the _ number _ of conjugate phases can also depend on @xmath121 . for an observer moving on a timelike worldline ( where @xmath121 increases monotonically ) , new conjugate phases and therefore new images sometimes appear at discrete times . these images correspond to observation times where ( [ conjsand ] ) momentarily degenerates to a linear equation . new images can therefore arise when @xmath460 and @xmath461 if @xmath462 , the two solutions to this equation are @xmath463 \pm \sqrt { [ \tr \bm{\beta } \tr \dot{\bm{\beta } } - \tr ( \bm{\beta } \dot{\bm{\beta}})]^2 - 4 \det \bm{\beta } \det \dot{\bm{\beta } } } } { 2 \det \dot{\bm{\beta } } } .\ ] ] only solutions satisfying @xmath464 are physically relevant . when a conjugate phase of this type first appears , it satisfies @xmath465 the associated image therefore provides a picture of the infinitely distant past . furthermore , an infinite amount of the source s history appears to the observer within a finite amount of proper time . this implies that new images are highly blueshifted . no matter how long an observer waits , no conjugate phase can exceed @xmath457 . the emission time for an associated image might therefore be expected to tend towards a constant value as @xmath466 . this means that a very large amount of proper time at the observer corresponds to only a small amount of proper time at a source . images which appear suddenly and are initially highly blueshifted become highly redshifted at late times . it is unclear precisely what these types of images imply . to the extent that geometric optics remains valid , all observers passing through @xmath467 momentarily see almost the entire universe appear infinitely blueshifted as it was in the infinitely distant past . furthermore , ( [ angleredshift ] ) implies that all of the universe is briefly compressed into a single point on each observer s sky . of course , such phenomena lie outside the domain of geometric optics . they may even lie outside of the realm of test fields propagating on a fixed background spacetime . extreme focusing events like these might indicate instabilities inherent in the plane wave geometry itself . it should , however , be noted that all of the infinities just alluded to are likely to have finite cutoffs in `` realistic '' plane waves which decay at large transverse distances . one important class of sandwich waves are those that are very weak and last only for a short time . in these cases , @xmath343 , @xmath435 , @xmath436 , and @xmath437 may be expanded as integrals involving successively higher powers of @xmath65 . to lowest order in such a scheme , @xmath51 and @xmath52 are approximately unaffected by the wave while inside of it . assuming that @xmath468 , the first corrections to this assumption are @xmath469 this approximation is consistent with ( [ wronskian ] ) [ and therefore ( [ wronsksand ] ) as well ] . in terms of the matrices appearing in ( [ bsandwich ] ) , @xmath470 , @xmath471 , and @xmath472 note that eqs . ( [ asand ] ) and ( [ bsand ] ) should be applied with care if @xmath65 involves many oscillations of an approximately periodic function . in these cases , the integral of @xmath65 can be very nearly zero . terms nonlinear in @xmath65 might then be significant . assuming that ( [ asand ] ) and ( [ bsand ] ) are indeed adequate approximations for @xmath51 and @xmath52 , a coordinate rotation may always be used to diagonalize @xmath437 . there then exist two constants @xmath473 and @xmath474 such that @xmath475 in this sense , all sufficiently short gravitational plane waves act as though they are linearly polarized [ so @xmath476 in ( [ distortion ] ) ] . in terms of the individual wavefunctions appearing in ( [ hgen ] ) , @xmath477 , \\ \mathcal{h}_2 & = & \int_0^{u_+ } \rmd w [ h_\|(w ) - h_+(w)].\end{aligned}\ ] ] the transverse coordinates @xmath16 have also been chosen such that @xmath478 if a wave satisfies the vacuum einstein equation , @xmath48 and @xmath479 . more generally , it follows from the weak energy condition that @xmath480 now assume that the integral of @xmath39 is non - negative , which entails only a minimal loss of generality . then , @xmath481 we say that a wave is gravity - dominated if @xmath482 and matter - dominated if @xmath483 . these definitions are closely analogous to those used to classify symmetric plane waves in sect . [ sect : symwaves ] . there , a wave was said to be gravity- or matter - dominated depending on the sign of the constant @xmath484 appearing in ( [ symwave ] ) . for weak gravity - dominated wavepackets , there can be at most one phase conjugate to an observer satisfying @xmath485 . if this exists , it evident from ( [ bsand ] ) that @xmath486 is conjugate to @xmath121 with multiplicity 1 . this equation is valid only if @xmath487 . a conjugate phase therefore exists only for observers satisfying @xmath488 note that @xmath489 is the unique physically - relevant solution to ( [ newtau ] ) in the gravity - dominated case . waves that are matter - dominated ( so @xmath483 ) also admit the conjugate phase ( [ tau1 ] ) when @xmath121 satisfies ( [ observer1 ] ) . in the conformally - flat case where @xmath490 , this is the only conjugate phase . unlike in the gravity - dominated case , the multiplicity of @xmath491 is equal to 2 for conformally - flat waves . in all other matter - dominated cases , @xmath491 has multiplicity 1 and a second conjugate phase is admitted ( also with multiplicity 1 ) for all observers satisfying @xmath492 this occurs at @xmath493 note that ( [ observer2 ] ) is a more stringent condition than ( [ observer1 ] ) . as implied by the notation , @xmath494 . .46 .46 .47 consider a point source moving on a timelike worldline @xmath138 in a weak sandwich wave spacetime . such a source appears differently when observed at different times . it is clear that when @xmath495 , exactly one image of @xmath138 is viewable under generic conditions . as time passes , the wave eventually passes through the observer . a second image then appears when @xmath496 . this image is always emitted before the first . if the wave involves a sufficient amount of ricci curvature ) and ( [ bsand ] ) be valid . it is possible for , e.g. , sufficiently strong vacuum waves to admit two conjugate phases in the region @xmath427 . ] ( from e.g. , electromagnetic plane waves ) and is not conformally - flat , a third image appears when @xmath497 . this is emitted before the first two images . all images persist indefinitely once they appear . sufficiently far in the future , one image is observed of the source as it appeared after interacting with the gravitational wave . all other images predate this interaction . [ fig : emissionsand ] . when the second image first appears at @xmath498 , the new conjugate phase @xmath499 is divergent . at all later times , it is finite . the same is also true for the emission times associated with the second image . almost the entire past history of the source is therefore observable within a finite proper time . this implies an infinite blueshift . at late times , @xmath500 . the observed evolution of the source via the second image effectively freezes as @xmath59 asymptotes to @xmath501 ( which predates the source s interaction with the wave ) . images such as these are highly redshifted , as indicated in fig . [ fig : redshift ] . note that a similar transition from infinite blueshift to infinite redshift also applies to the third image if it exists . except that three different choices are made for the value of @xmath502 . curves for both images are shifted to the right as @xmath502 is decreased . the younger ( solid ) curves are almost unaffected by the wave if @xmath502 is sufficiently large . for smaller values of @xmath502 , both curves are strongly bent by the constraint that they ca nt pass through the dotted curve representing @xmath491 . ] .46 .46 another qualitative feature of the emission times plotted in fig . [ fig : emissionsand ] is that there is a sense in which pairs of images can `` switch roles . '' consider , e.g. , the left panel of that figure . at late times , the solid curve ( corresponding to the younger image ) is perfectly linear . indeed , it remains very nearly linear until @xmath503 . a rapid transition then occurs where the dashed curve effectively takes over this linear behavior while the solid curve strongly deviates from it . in a sense , the two images reverse their roles . this phenomenon occurs one more time ( somewhat less sharply ) around @xmath504 when the second image first appears . it arises essentially because @xmath505 for the younger image and @xmath506 for the older image . these constraints can cause emission curves to bend sharply with large accompanying frequency shifts in order to avoid intersecting @xmath491 . whether or not this occurs depends on whether the `` average '' linear increase of @xmath59 ever comes near @xmath491 . if it does , this role switching occurs . if not , the younger image is barely affected by the gravitational wave at all . this is illustrated in fig . [ fig : changingv ] , where emission curves for several sources are plotted simultaneously . to summarize , images which appear at discrete times briefly appear as bright , highly blueshifted `` flashes . '' indeed , fig . [ fig : luminosity ] shows that their luminosity distances go to zero . simultaneously , the angular diameter distance of each new image tends to infinity . it is implied by ( [ angleredshift ] ) that all highly blueshifted images make must a very small angle @xmath317 with @xmath28 on the observer s sky . the second ( and third ) images therefore appear aligned with the direction of propagation of the gravitational wave when they first appear . this direction could be quite different from the location of the other image(s ) . as time progresses , all images migrate across the observer s sky as illustrated in fig . [ fig : angles ] . different images may remain separated from each other by large angles at all times . .46 .46 despite their simplicity , plane gravitational waves behave in qualitatively different ways from lenses associated with quasi - newtonian mass distributions . as expected from their dynamic nature , plane waves generically shift the observed frequencies of various images . they may also admit images which appear to move , deform , change brightness , and shift color as time progresses . more subtle differences relate to the number of images that are produced of a given source . for example , even numbers of images can appear generically ( which has led plane wave spacetimes to be cited @xcite as well - behaved examples where the odd number theorem @xcite does not apply ) . some plane waves can even produce an infinite number of discrete images . perhaps most striking of all are the bright flashes shown to be produced by generic sandwich waves in sect . [ sect : sandwaves ] . these correspond to individual images which appear at discrete times . more typical gravitational lenses can produce new images if a source crosses an observer s caustic . individual images then split into two ( or vice versa ) . the flashes produced by sandwich waves are quite different . their appearance does not require that a source pass through an observer s caustic . such images appear individually from the infinitely distant past . initially , they are infinitely bright and infinitely blueshifted points of light appearing in the direction of propagation associated with the gravitational wave . many of these effects depend at least partially on the idealization that a plane wave extends undiminished to infinitely - large transverse distances . plane wave spacetimes are not asymptotically flat . despite being topologically trivial and locally well - behaved , they are not even globally hyperbolic : null geodesics passing between appropriately - chosen pairs of points can extend to arbitrarily large transverse distances in between those points . it is this property which permits the infinite number of images described in sect . [ sect : symwaves ] to be produced by symmetric waves . the flashes described in sect . [ sect : sandwaves ] also depend on the spacetime structure at arbitrarily large distances . this structure likely affects the formation of even numbers of images as well . indeed , the usual proofs of the odd number theorem require global hyperbolicity , among other assumptions @xcite ( see , however , @xcite for a more general formulation ) . if a spacetime has the geometry of a plane wave only out to some finite transverse distance , all results derived here remain valid if the associated images involve light rays which never extend sufficiently far to interact with any large - distance modifications . the infinite sequence of images formed by a symmetric plane wave would then be expected to become finite for spacetimes which are only approximately plane waves . calculations involving the oldest images could no longer be trusted in these cases . similarly , the bright flashes associated with ideal sandwich waves are likely to be somewhat less extreme for waves which decay at infinity . large brightnesses and large blueshifts can still exist , but these will be cut off at some finite maximum . such maxima may , however , remain quite large . it is reasonably clear that modifications of the geometry at large distances can remove some images . less obviously , these modification can also introduce new images . consider , for example , the @xmath507-wave spacetimes obtained by substituting @xmath508 in the metric ( [ planewavemetric ] ) . these generalize the plane wave spacetimes . it has been shown that if @xmath509 grows subquadratically as @xmath510 , the resulting geometries are globally hyperbolic @xcite . moreover , every pair of points is connected by at least one geodesic in these cases ( unlike in pure plane wave spacetimes where the growth of @xmath509 is precisely quadratic ) . this implies that modifications of the geometry at large transverse distances can introduce new null geodesics even between points at small transverse distances . it would be interesting to explore these effects in more depth to understand precisely how modifications of this sort ( or more general ones ) alter the lensing properties described here for ideal plane waves . it would also be interesting to better understand what the transient flashes of sect . [ sect : sandwaves ] imply for waves propagating on plane ( or almost - plane ) wave spacetimes . this can likely be facilitated by the green functions derived in @xcite . 1 perlick v 2004 _ living rev . relativity _ * 7 * 9 petters a o , levine h and wambsganss j 2001 _ singularity theory and gravitational lensing _ ( boston , ma : birkhuser ) schneider p , ehlers j and falco e e 1992 _ gravitational lenses _ ( new york , ny : springer - verlag ) petters a o and werner m c 2010 _ gen . * 42 * 2011 mckenzie r h 1985 * 26 * 1592 perlick v 1996 * 13 * 529 perlick v 2001 _ comm . math . phys . _ * 220 * 403 ehlers j and newman e t 2000 * 41 * 3344 friedrich h and stewart j m 1983 _ london _ a * 385 * 345 frittelli s and newman e t 1999 d * 59 * 124001 seitz s , schneider p and ehlers j 1994 * 11 * 2345 werner m c 2009 * 50 * 082504 blandford r and narayan r 1986 _ astrophys . j. _ * 310 * 568 damour t and esposito - farse g 1998 d * 58 * 044003 kovner i 1990 _ astrophys . j. _ * 351 * 114 frieman j a , harari d d and surpi g c 1994 d * 50 * 4895 faraoni v 2008 _ new astron . _ * 13 * 178 faraoni v 1998 _ int . _ d * 7 * 409 faraoni v 1991 _ gen . * 23 * 583 ehrlich p e and emch g g 1991 the conjugacy index and simple astigmatic focusing _ geometry and nonlinear partial differential equations _ eds v oliker and a treibergs ( providence , ri : american mathematical society ) ehrlich p e and emch g g 1993 geodesic and causal behavior of gravitational plane waves : astigmatic conjugacy _ differential geometry : geometry in mathematical physics and related topics _ eds r e greene r e and s - t yau ( providence , ri : american mathematical society ) beem j , ehrlich p and easley k 1996 _ global lorentzian geometry _ ( new york , ny : dekker ) harte a i and drivas t d 2012 d * 85 * 124039 hollowood t j and shore g m 2008 _ j. high energy phys . _ * 12 * 091 hollowood t j , shore g m and stanley r j 2009 _ j. high energy phys . _ * 08 * 089 bondi h and pirani f a e 1989 _ london _ a * 421 * 395 bondi h , pirani f a e and robinson i 1959 _ london _ a * 251 * 519 penrose r 1976 any spacetime has a plane wave as a limit _ differential geometry and relativity _ eds . m cahen and m flato ( dordrecht : reidel ) blau m , frank d and weiss s 2006 * 23 * 3993 berenstein d , maldacena j m and nastase h 2002 _ j. high energy phys . _ * 04 * 013 grignani g , harmark t , marini a and orselli m 2010 _ j. high energy phys . _ * 06 * 034 stanley r and hollowood t j 2011 _ j. high energy phys . _ * 10 * 096 wald r m 1984 _ general relativity _ ( chicago , il : university of chicago press ) eardley d m , lee d l and lightman a p 1973 d * 8 * 3308 pravda v , pravdov a , coley a and milson r 2002 * 19 * 6213 torre c g 2006 _ gen . rel . grav . _ * 38 * 653 gnther p 1988 _ huygens principle and hyperbolic equations _ ( london : academic press ) etherington i m h 1933 _ phil . mag . _ * 15 * 761 ellis g f r 1971 relativistic cosmology _ general relativity and cosmology : proceedings of the 47th international school of physics `` enrico fermi '' _ ed r k sachs ( new york , ny : academic press ) hartman p 1982 _ ordinary differential equations _ ( boston , ma : birkhuser ) hubeny v , rangamani m and ross s 2004 d * 69 * 024007 penrose r 1965 * 37 * 215 frittelli s , kling t p and newman e t 2000 d * 63 * 023006 surpi g c and harari d d 1999 _ astrophys . j. _ * 515 * 455 flores j l and snchez m 2006 _ lect . notes phys . _

identifying members of known moving groups ( mg ) or open clusters can provide an important constraint on their age and composition . low - mass stars and brown dwarfs ( bds ) members of known mgs or open clusters can provide important feedback to atmospheric and evolutionary models of such objects at young ages , which currently await good calibration from constraints on age and composition . therefore , there exists an imperative to compile a sample of mg members in this mass range that can be used as anchor points or test - beds for these models . in this paper we present the last in a series of results of a major survey of ultra - cool dwarfs ( ucds ; objects with a spectral classification of m7 or later corresponding to teff@xmath12500 k ) , in young moving groups which began with the results presented in clarke et al . ( 2010 ) , hereafter paper i , and glvez - ortiz et al . ( 2010 ) , hereafter paper ii . we focus here on the study of age - sensitive spectroscopic signatures such as gravity sensitive features ( e.g. gorlova et al . 2003 ; mcgovern et al . 2004 ) , h@xmath0 emission ( e.g. west et al . 2004 , 2008 , 2011 ) and the presence of li i 6708 in a sample of objects previously classified as members of a mg based on kinematic or astrometric criteria . despite the recent disagreement about the origin of moving groups ( famaey et al . 2007 , 2008 ; antoja et al . 2008 ; klement et al . 2008 ; francis & anderson 2009 ; zhao et al . 2009 ; lpez - santiago et al . 2009 ; bovy & hogg 2010 ; murgas et al . 2013 ) , and as we discussed in paper ii , we are assuming the classical concept : a moving group is a young stellar population that shares a common space motion ( e.g. pinfield et al . 2006 ) whose members have a common origin , and therefore , age and composition . we focused our studies on well - documented groups : the hyades supercluster ( hy ; 600 myr ) , the ursa major group ( sirius supercluster ) ( si ; 300 myr ) , ic 2391 supercluster ( ic ; 35 - 55 myr ) , the castor moving group ( ca ; 200 myr ) , and the local association ( 20 - 150 myr ) or pleiades ( pl ) moving group ( see paper i and ii for details of mg properties and references thereof ) . many efforts have been made to obtain accurate models to understand the cool and complex atmospheres of low mass stellar and substellar objects ( e.g. kirkpatrick et al . 1993 ; rajpurohit et al . 2011 ; reyle et al . 2011 ; rajpurohit et al . but the models still show discrepancies , for example in the strength of some absorption bands : discrepancies likely due to inaccurate atomic parameters and/or missing molecular opacities . also , dust formation and its behaviour with temperature changes spectral characteristics in many ways . models that describe very low - mass stars ( vlms ) and bd atmospheres can not hope to be in agreement with observations without the inclusion of dust . the onset of iron and silicates dust grain formation suspended in the photospheres of late - type m dwarfs through the l dwarf spectral sub - types is accompanied by spectral reddening , especially in near - ir ( based upon the size , quantity and distribution of this dust ) . the t dwarf spectral sub - type sequence demonstrates a reversal to bluer colours due to the dust settling below the photosphere , thus reducing the reddening . consequently , atmospheric models that describe dwarf objects with t@xmath2 @xmath3 2800 k should include treatment for the effects of atmospheric dust evolution . up to date , there are several families of stellar model atmospheres of very low mass stars . these models are still incomplete or approximate in some physical properties ( opacities , oscillator strengths for some lines and molecular bands , etc ) , showing few or considerable discrepancies depending on the atmospheric regions , temperatures , etc , and there are also non covered areas or `` gaps '' . a comprehensive study of models tested by observations is needed to fully understand this cool atmospheres . since bds are objects occupying an intermediate position between stars and giant planets , studying their atmospheres can help us to better understand the processes in giant exo - planet atmospheres . the determination of the physical properties of vlms and bds is also important for understanding a broad range of topics including stellar and planetary formation , circumstellar disks , dust formation in cool atmospheres , and the initial mass function . by comparing high and low resolution optical spectra of our 25 low - mass stars and bd kinematic mg candidates with a grid of atmospheric models , we can obtain effective temperature and surface gravity estimates . in section 2 we describe the sample characteristics and selection criteria , while in section 3 we present details of our observations and data reduction techniques . in section 4 we explain the model atmosphere fitting process , and in section 5 we describe the age indicators and the analysis used to assess the sample memberships . finally , in section 6 we present a brief discussion and summary of our results . we present here the study of two different samples . we used photometric and astrometric criteria to combine an extended version of the liverpool - edinburgh high proper motion survey ( elehpm ; pokorny , jones & hambly 2003 ; pokorny et al . 2004 ) and southern infrared proper motion survey ( sips ; deacon et al . 2005 ; deacon & hambly 2007 ) with the two micron all sky survey ( 2mass ) , and applied colour cuts of @xmath4-@xmath5@xmath61.0 and @xmath7-@xmath5@xmath65.0 to select objects with spectral type predominantly later than m6v . to these we applied proper motion restriction , @xmath8@xmath9 ( @xmath10 is proper motion and @xmath11 is the proper motion one sigma uncertainty ) to ensure a 4-@xmath12 level of proper motion accuracy before searching for stars with proper motions consistent with any of the mgs . of the 817 red object catalogue we obtained by using both astrometric and photometric criteria , we concluded that 132 of these objects were possible members of one or more of the five mgs in the study ( see paper i for details ) . from this sample we then obtained high resolution spectra of 68 objects with spectral types ranging m4.5-l1 . using these high resolution spectra we derived galactic space - velocity components and , applied a simple kinematic criterion . we found that 49 targets have kinematics consistent with the young disk population ( yd ; see e.g. eggen 1984a , b , 1989 ) and that 36 of them possibly belong to one of the five moving groups in this study . we also measured ( where possible ) projected rotational velocities and used the rotation rate as a supporting criterion of youth in confirming the kinematic members . we found that 31 young disk candidates have moderately high rotational velocities which also suggest youth and that 12 others have significantly high rotations to consider for further investigation as possible young stars ( see paper ii for details ) . we present here high resolution follow - up observations for the 13 brightest stars in this sample , that we took in order to use other spectroscopic age criteria to confirm kinematic membership . in table [ tab : samplea ] we give the name , coordinates , 2mass @xmath4 magnitude and the mg candidature assigned from the kinematic study and the rotational velocities obtained in paper ii . our second sample consists of objects selected from the catalogue of folkes et al . ( 2012 ) , and the reader is refered to this publication for full details of the selection method used in the creation of the catalogue . these authors present an ucd catalogue containing 245 objects with spectral types of late - m through to the l t transition from low southern galactic latitudes , in an area covering 5042 sq.degs within 220@xmath13 @xmath3 l @xmath3 260@xmath13 and 0@xmath13 @xmath3 l @xmath3 60@xmath13 for @xmath14b@xmath14@xmath315@xmath13 ) , identified as part of a deep search using 2mass near - ir and supercosmos optical photometry selection criteria as well as reduced proper motion constraints . * photometry : using an optimal set of optical / near - ir selection criteria for the ( @xmath4-@xmath15)/(@xmath4-@xmath16 ) two - colour plane along with optical / near - ir colours of ( @xmath17-@xmath16 ) , ( @xmath18-@xmath16 ) , ( @xmath19-@xmath16 ) and the optical colour of @xmath20r@xmath21i@xmath22 , with @xmath4@xmath15@xmath16 from 2mass , we obtained a sample of objects in the spectral range of @xmath23m8v to @xmath23l9 . the selection method also included an @xmath24 and @xmath25 photometric surface gravity test with additional photometric constraints . see figures 1 , 2 and 3 , and section 2 of folkes et al . ( 2012 ) for details . * astrometry : to eliminate bright distant contaminants , which one would expect to display small proper motions , a reduced proper - motion diagram was used to segregate giant from dwarfs ( e.g. luyten 1978 ) . in this case we used a diagram of reduced proper motion in 2mass @xmath16-band plotted against ( @xmath26-@xmath16 ) colour . see fig . 6 and section 2.5 of folkes et al . ( 2012 ) for details . to find possible mg members from this ucd sample , folkes ( 2009 ) applied the astrometric and photometric test described in sections 3.2 and 3.3 of paper i ( used in the selection process of sample a. ) spectral types and photometric distances were derived from the ( @xmath27-@xmath4 ) colour using the @xmath28/(@xmath29-@xmath4 ) , and @xmath28/spt relations of dahn et al . ( 2002 ) . from these photometric and astrometric criteria folkes ( 2009 ) identified 23 objects , with spectral types ranging from m3 to l0 as possible moving group members , from which we could observe 12 that form sample b. here we obtain spectral classification and confirm ( or otherwise ) the youth of these candidates , through low resolution spectroscopy and by studying age - sensitive spectral features . in table [ tab : sampleb ] we give the name , coordinates , 2mass @xmath4 magnitude and the mg candidature assignation . to improve the previous photometric spectral type determination from folkes ( 2009 ) , we calculated a variety of different spectroscopic indices : pc3 , tio1+tio2 and vo1+vo2 from martn et al . ( 1999 ) ; the vo index from kirkpatrick et al . ( 1995 ) ; vo - a and tio5 from cruz & reid ( 2002 ) . since the tio1+tio2 and vo1+vo2 indices can imply two different spectral types ( see fig . 9 of martn et al . 1999 ) , we only considered them if they were consistent with the spectral types obtained by the other indices . from these a mean spectral type was calculated . some objects show considerable discrepancies between spectral type derived from photometry and that from spectroscopic indices . that is , with the exception of two objects ( 2mass1756 - 4518 and 2mass1909 - 1937 ) with very similar photometric and spectroscopic spectral classification . this difference is marked by the systematic assignment of a later spectral type by the photometric method . also the difference between the two classifications appears greater for redder objects . even with care during selection , the objects of sample b are situated in highly populated and reddened areas within the photometric selection planes , which could have produced these differences . also , the possible presence of dust in these atmospheres could account for redder colours , therefore suggesting later spectral types . we used the spectral types derived from spectroscopy for all calculations . for this paper we analysed optical high resolution echelle spectra of 13 objects ( sample a ; table [ tab : samplea ] ) and low resolution long slit spectra of 12 objects ( sample b ; table [ tab : sampleb ] ) . the data were obtained during four observing runs detailed in table [ tab : obs ] . both sets of high and low resolution spectra were extracted using the standard reduction procedures in the iraf twodspec and echelle packages respectively : bias subtraction , flat - field division , extraction of the spectra , telluric correction as well as wavelength and flux calibration . we obtained the solution for the wavelength calibration by taking spectra of a th - ar lamp . the average signal - to - noise ( s / n ) of the data , measured as the square root of the signal at @xmath30 8100 , is @xmath3030 for uves data ( except for 2mass2039 - 1126 that has s / n of @xmath3015 ) and @xmath30200 for both runs of imacs data . figure [ fig : todashl ] shows high resolution spectra of sample a on the left and low resolution spectra of sample b on the right , ordered by spectral type classification . well studied late - type objects with known spectral type were observed at the same time as the targets to be used as reference ( m9 lp944 - 20 ; m4 gl876 ; and m6 gl 406 ) and are also included in the figure for comparison . to determine spectral characteristics of the targets ( t@xmath2 , @xmath31 , etc ) , to aid us in age determination , we compared a grid of generated synthetic spectra with the observations . according to current concepts , dust plays an important role in the formation of late m dwarf spectra ( tsuji et al . 1996 ; hauschildt & allard 1992 ) . although there is no unified opinion about temperature range where dust should be considered , the temperatures of transition between dust - free and dusty models range from @xmath32@xmath12600 ( allard et al . 2011 ) to @xmath32@xmath13000 k ( jones & tsuji 1997 ) . therefore , synthetic spectra were computed for dust - free nextgen atmospheric models and for semi - empirical models ( i.e. modified nextgen models ) in the wavelength range 6500 - 9000 . the dust effects " are included in semi - empirical models for m3-m8 stars . synthetic spectra based on dusty , cond and corresponding semi - empirical models were calculated for m7-l0 stars . synthetic spectra were computed for nextgen atmospheric models ( hauschildt & allard 1992 ) by the wita6 program ( pavlenko et al . 1997 ; 2007a ) . the calculations were made under the assumption of local thermodynamic equilibrium ( lte ) , hydro - static equilibrium , in the absence of sources and sinks of energy , for a one - dimensional model atmosphere . we used the atomic line list from the vald database ( kupka et al . 1999 ) and the line list of titanium oxide ( tio ) from plez ( 1998 ) to produce the synthetic spectra . vanadium oxide ( vo ) band opacities were computed in the framework of the jola approximation ( see lyubchik & pavlenko 2000 ) . synthetic spectra were calculated for a grid of models having t@xmath2=2600@xmath213400 k with @xmath33 k step , surface gravities of @xmath31=4.0@xmath215.5 with 0.5 step and metallicities of @xmath34$]=0.0 , -0.5 and -1.0 dex . solar abundances were taken from anders & grevesse ( 1989 ) . the artificial rotational - broadening of spectral lines was implemented following the methodology of gray ( 2005 ) using the rotational velocity of the objects obtained in paper ii . we carried out a few numerical experiments to investigate the effect of microturbulence velocity on the spectra . the study shows that the differences in synthetic spectra associated with @xmath35 in the 1.0 - 5.0 km s@xmath36 range are negligible . thus all theoretical spectra were computed for a microturbulence velocity @xmath35 = 3.0 km s@xmath36 . we used the semi - empirical atmospheric models described in pavlenko et al . ( 2007a ) to include the effects of dust formation . the semi - empirical models were obtained by modifying nextgen , dusty and cond models . the dust opacity originates in the shell - like structures lying above the photosphere ( pavlenko et al . 2007a ) and we expect the influence of dust in a spectrum to be more prevalent in the shortest wavelength spectral region . we assumed that dust clouds are located in the upper most layers of stellar atmosphere and do not affect the distribution of temperatures and pressures . dusty effects were treated in two ways : ( 1 ) decreasing molecular abundance in the gas due to molecular condensation of dust particles and ( 2 ) radiation scattering in the dust clouds . the decrease in the concentration of tio and vo molecules was modelled by two parameters : ( 1 ) the number of layers in the above model for which there is no absorption of tio ( i.e. , all titanium oxide condenses on dust particles ) and ( 2 ) the coefficient of tio molecular density reduction for all other model layers . radiation scattering in the dust clouds was also modelled by two parameters : 1 ) the optical thickness of the dust cloud and 2 ) the location of maximum opacity as a result of these clouds . this method is described in pavlenko et al . ( 2007a ) and kuznetsov et al . ( 2012 ) . the theoretical spectra were computed using a semi - empirical atmospheric model for a grid with effective temperatures of t@xmath2=2600@xmath213000 k ( based on nexgen models ) and @xmath32=2000@xmath212600 k ( based on dusty and cond models ) with a step of 100 k , surface gravities of @xmath31=4.0@xmath215.5 with a step of 0.5 , and solar ( anders & grevesse 1989 ) metallicity . these models were calculated for the same rotational and microturbulence velocity as for the nextgen models . we are thus able to compare the observed high and low resolution spectra with synthetic spectra assuming both the presence and the absence of dust . the @xmath37-function analysis described in pavlenko et al . ( 2007a ) was used to determine the best fit for each grid of the model independently , @xmath38 where @xmath39 is a normalization parameter , @xmath40 is the synthetic flux and @xmath41 is the observed flux . the minimum value of @xmath37-function corresponds to the best fit . the @xmath37-function is integrated in the 6900 - 7200 , 8100 - 8400 and 7660.8 - 7734 , 8160 - 8220 wavelength ranges for low and high resolution spectra respectively . the use of these regions allowed us to analyze the tio band at @xmath237100 and na i ( 8183 , 8199 ) doublet in low resolution spectra , the k i ( 7665 and 7669 ) resonance doublet and na i ( 8183 , 8199 ) subordinate doublet in high resolution spectra . analysis of na i and k i lines is significant because their profile is sensitive to t@xmath2 and @xmath31 . the vo ( @xmath237400 , @xmath237800 ) bands were excluded from the fitting procedure since we do not have a satisfactory line list for this molecule . we carried out fits of the observed spectra for two model grids , original nextgen and semi - empirical , for m4-m8 objects . figures [ fig : fight ] and [ fig : figlt ] show example fits for the red part of the spectra in sample a , and the complete spectra in sample b. figures [ fig : fighk ] and [ fig : fighna ] show examples of the fitting of the atmospheric models , magnified around the na i@xmath238190 and k i@xmath237700 doublet area for sample a. the minimum value of @xmath37-function for each target with m4-m8 spectral type is shown in tables [ tab : resulta ] , [ tab : resultb ] , [ tab : resultab ] and [ tab : resultbb ] . the atmospheric structure of objects of m8.5-m9.5 spectral type differs greatly from early m dwarfs . the dust has a significant impact on the distribution of temperature vs. pressure in these atmospheres . thus , it is incorrect to use the nextgen model for objects with temperatures t@xmath42 2500 k. we used the dusty and cond ( allard et al . 2001 ) atmospheric models for the study of the late m dwarfs synthetic spectra . the dusty and cond models were modified in an identical way to the model improvements for early m dwarfs : we introduced additional scattering of radiation by dust clouds and reduced the abundance of the molecules tio and vo . as with earlier spectral types , we utilised fits to the tio and vo absorption bands and the k i and na i doublets , to determine the stellar fundamental parameters . in weakly ionized atmospheres at t@xmath2@xmath1 3000 k the natural broadening is several orders of magnitude weaker than pressure and stark broadening , and can thus be neglected . the broadening of k i and na i resonance lines is dominated by pressure effects , but the theory of this is not particularly advanced . pressure broadening is calculated using two methods : a collisional approximation ( van der waals theory ) and quasi - static theory ( see burrows & volobuyev 2003 ; allard et al . the use of a collision approximation allows us to describe the profiles of metal lines in early m dwarfs . the quasi - static theory is used in the study of l dwarfs , where energy levels of atomic sodium or potassium are immersed in a sea of molecular hydrogen and are subsequently perturbed by the potential field of the diatomic hydrogen . the modelling of atomic - line broadening for objects that are close to the transition between m and l spectral classes is complicated and this issue is beyond this paper aims . since in our case we could see only the core of the lines and do not see extended wings , our fits should be sufficient without including any broadening . physical properties were derived for objects with m8.5-m9.5 spectral types based on the fitting of high or low resolution spectra and synthetic spectra calculated for cond and dusty models . in addition @xmath37-function analysis provided the best fits . since in sample b there are some objects classified as m8 , we applied fits with the synthetic spectra ( nextgen , dusty and cond ) as well as with appropriate semi - empirical models , to provide the transition between the nextgen and the dusty / cond models . results of these fits are shown in tables [ tab : resultab ] , [ tab : resultbb ] , [ tab : resultac ] , and [ tab : resultbc ] . lp944 - 20 : lp944 - 20 is a known bd of spectral type m9 and we used it as a test for the atmospheric fits as well as to benchmark the age constraints described in following sections . several authors have studied this object in detail to determine its characteristics . tinney ( 1998 ) estimated an age between 475 and 650 myr by evolutionary tracks , while ribas ( 2003 ) determined lp944 - 20 as a kinematic member of the castor moving group and therefore estimated an age of 320@xmath4380 myr . pavlenko et al . ( 2007a ) used semi - empirical atmospheric models to determine the parameters of lp944 - 20 including lithium abundance , of which the best fit gave log n(li)=3.25 @xmath43 0.25 , and found a similar age to the one found by ribas ( 2003 ) . according to previous studies the effective temperature of lp944 - 20 is in the range 2040 @xmath44 t@xmath2 @xmath44 2400 k ( see basri et al . 2000 ; dahn et al . 2002 ; pavlenko et al . the best fit obtained here , with the dusty model , provides t@xmath2=2400 k and log @xmath45= 4.0 ( see table [ tab : resultac ] ) . the log @xmath45 obtained by pavlenko et al . ( 2007a ) was 4.5 , with temperatures of 2000 - 2200 k using cond models . they remarked that temperatures in the outermost layers of the dusty models were higher in comparison to the cond models . this difference is consistent within the uncertainties . all the targets in sample a , with the exception of 2mass0334 - 2130 , present a better fit when a semi - empirical model with dust is used instead of the dust - free nextgen model , as shown by @xmath46 of table [ tab : resulta ] . this would imply that the presence of dust should be taken into account for effective temperatures up to @xmath302900 ( jones & tsuji 1997 , kuznetsov et al . 2012 ) . similarly to lp944 - 20 , the two targets of sample a cooler than m8 spectral type are better fitted by the dusty model ( tables [ tab : resultab ] and [ tab : resultac ] , and figures [ fig : fight ] , [ fig : fighk ] and [ fig : fighna ] ) . for the analysis , we divided sample b into three different spectral type ranges for the fits : @xmath1m8 , @xmath47m8 and as we have many objects with an m8 spectral classification we treated them as a third set and applied both sets of models for them ( see section 4.4 . ) . we used gl 406 ( m6.0 ) as reference star to test spectral typing and model fits . however , we noticed that the flux calibration is very poor for @xmath48@xmath17500 . the cause is uncertain since all other spectra were calibrated correctly . the k i and na i lines were not present in the fault area , hence we centered our fit on @xmath48@xmath477500 . results of objects that were spectroscopically classified as earlier than m8 are shown in table [ tab : resultb ] , where @xmath46 is given in columns 8 and 12 . from these targets , four present a better fit with a semi - empirical model with dust , and two , 2mass1236 - 6536 and 2mass1736 - 0407 , with nextgen . these two , were included in our sample due to their cold photometric " classification ( section 3.2 ) but spectral classification and model fit suggest warmer temperatures . for objects classified as m8 , we applied all models . results are given in table [ tab : resultbb ] . in the case of 2mass0814 - 4020 , dusty and cond models have the same @xmath46 , but the dusty model is more appropriate for the target s temperature . also , despite the higher value of @xmath46 for the nextgen model , we could appreciate by visual inspection and by the @xmath37-function value that in parts of the spectrum nextgen was the best fit . for 2mass1557 - 4350 , the dusty and cond models also have the same @xmath46 . we chose the dusty model as being more appropriate for its temperature . for objects with spectral types later than m8 , parameters determined from the best fit to the observed seds are given in table [ tab : resultbc ] . in this case , it was not easy to find a good fit for the blue part of the spectrum ( @xmath48@xmath17500 ) . in the case of 2mass1909 - 1937 no cond model provided an acceptable result and the dusty model gave no reliable gravity values . it is likely that these objects are also suffering from relatively uncertain flux calibration , but the fact that these are the coolest objects and around the m - l transition , may indicate other issues like the use of an inapropriate set of models . since the k i and na i spectral regions seem to be correctly fitted we keep the results but consider them with caution for 2mass1734 - 1151 and 2mass1745 - 1640 , while we could not extract any acceptable conclusion from the fit for 2mass1909 - 1937 ( see table [ tab : resultbc ] ) . sample a is formed of objects that are already confirmed as kinematic candidates of mgs . since kinematics is not sufficent to confirm membership , determination of a candidate s age can further constrain its group membership to a point of robustness or dismiss it as an old field target that shares kinematics with a young mg . for our candidates , age in some cases can also discriminate between multiple mg candidatures . sample b is formed of astrometric candidates and so further study should be applied to confirm any membership . this can be achieved by : 1 ) using gravity sensitive spectroscopic features to distinguish between young and old ultracool dwarfs , 2 ) using the activity / age relation for late - type stars up to a spectral type of m7 ( mochnacki et al . 2002 ; silvestri et al . 2006 ; reiners & basri 2008 ; jenkins et al . 2009 ) , 3 ) measuring the lithium 6708 doublet ( e.g. rebolo et al . 1996 ; pavlenko et al . 2007a ) as the abundance of lithium is related to mass and age for substellar objects and 4 ) determining @xmath49 to differentiate between young and older m types . figure 9 of paper i shows a plot of evolution - time ( based upon potential group membership ) versus spectral type for the complete sample of candidates . the figure , reproduced and updated here as fig . [ fig : modelo ] , illustrates how an object would be selected for various methods of age constraining follow - up mentioned . candidates that appear to the right of the lithium edge can be followed up with a lithium test programme . objects that appear younger than 200 myr , are eligible for follow - up using spectroscopic gravity sensitive features . candidates that fall to the left of the spectral type m7 , would thus be suitable for age / activity relation follow - up , although candidates with a spectral type close to m7 may be subject to large uncertainties on their age . some candidates can not be tested by any of these methods but will be eligible for age testing using @xmath49 . in paper ii we presented a @xmath49 study ( for sample a ) where we applied this criterion of youth based on figures 9 and 10 of reiners & basri ( 2008 ) to support group membership ( see paper ii ) . we study here the rest of the age - constraining features . the radius of young ucds can be as much as 3 times greater than their eventual equilibrium state ( burrows et al . 2001 ) , and as a result young objects can exhibit significantly lower surface gravities ( 10 - 100 times ) than their evolved counterparts with the same spectral type . low - resolution ( e.g. r@xmath23350 - 2000 ) studies in the optical ( martn et al . 2010 ; martn et al . 1996 ; luhman et al . 1997 ) and infrared ( jones et al . 1996 : gorlova et al . 2004 ; mcgovern et al . 2004 ) have demonstrated that numerous features ( e.g. cah , k i , na i , vo , etc ) can be used as gravity - sensitive ( and thus age sensitive ) diagnostics for young objects . diagnostics should be sensitive to gravity and age for objects that are younger than @xmath23200 myrs ( e.g. barrado y navascus 2006 ) . schlieder et al . ( 2012 ) presented a study of the na i doublet equivalent width in giants , old dwarfs , young dwarfs , and candidate members of the @xmath50 pic moving group using medium resolution spectra . they concluded that the diagnostic is reliable for objects with spectral types later than m4 and younger than 100 myr , and that metallicity has an important role . therefore , this youth indicator is best used on samples with similar metallicity . thus , we aimed to apply this test by determining the surface gravity of the younger mg candidates , although we took into account all candidates up to 200 myr . as mentioned before , the main atomic gravity - sensitive features present in our spectra are : the k i resonance doublet ( 7665 and 7669 ) and the na i subordinate doublet ( 8183 and 8199 ) . we fitted the spectra of the candidates with a synthetic model ( see section 4 ) to determine the log @xmath45 of both samples a and b ( table [ tab : resulta ] , [ tab : resultb ] , [ tab : resultab ] , [ tab : resultbb ] , [ tab : resultac ] , and [ tab : resultbc ] ) . figures [ fig : fighk ] and [ fig : fighna ] show some examples of atmospheric model fitting magnified on the aforementioned features for sample a. we performed two approaches in the surface gravity studies : inferring age - limits from 1 ) the surface gravity obtained from the best fit synthetic models and 2 ) comparison of the equivalent width of gravity - sensitive na i doublet . as mentioned above , surface gravity features can be studied to discriminate if an object is young or old , but it is not easy to associate a log @xmath45 value to an age interval , that is , to have an absolute log @xmath45-age relation . many parameters affect the derived log @xmath45 : atmospheric model used , resolution and instrument . therefore , we observed lp944 - 20 , with known age and previously fitted by different models , to set the range of log @xmath45-age interval for our observations and as a test object for our method of late m dwarf investigation . taking into account the association to the castor mg and the age based on the presence of lithium , we used our log @xmath45 for lp944 - 20 as a reference and compared it to the log @xmath45 of the candidates . from this , we assume that an object with spectral type and log @xmath45 similar to lp944 - 20 is probably of similar age . models from the literature ( e.g. , nextgen , dusty , cond ) , associate similar or lower ages with warmer objects with the same log @xmath45 than to those that are cooler . for example , selecting log @xmath45 in the 4.0 - 4.5 range in dusty and cond models , an object with t@xmath51 @xmath52 2400 k intersects 8 - 30 myr age isochrones , whilst objects with t@xmath51 @xmath53 2400 k intersect 5 - 7 myr , and objects with t@xmath51 @xmath442400 k intersect with isochrones over 40 myrs . both the dusty and cond models present ambiguities for 30 myr isochrones . the cond model presents similar problems for @xmath5340 myr isochrones , but fortunately in this case with temperature ranges far from that of our targets . therefore , we can assume that objects warmer than lp944 - 20 would have , in general , ages @xmath44200 - 300 myr should they have log @xmath45 = 4.0 . since lp944 - 20 is only a single object , in order to constraint this log @xmath45-age relation , we also used the gl 876 field target , observed with the candidate a sample . this allows us to compare our target log @xmath45 values with those of old and intermediate age targets . we fit them in the same way as our candidates and results of the best fit are given in table [ tab : resulta ] and [ tab : resultac ] . as mentioned in section 4.4 , lp944 - 20 is a known member of the castor mg ( age@xmath23200 myrs ) and pavlenko et al . ( 2007a ) obtained an age of @xmath23300 myr through lithium depletion . applying the atmospheric model method described in section 4 , we obtained a log @xmath45 of 4.0@xmath430.5 . gl 876 is an m4 dwarf . it is known as an exoplanet host of four planets . literature provides few age calculations , such as 1 - 10 gyr from marcy et al ( 1998 ) , @xmath471 gyr from shankland et al . ( 2006 ) and 0.5 - 1 gyr from correia et al . ( 2010 ) . despite the variations it is clearly an old object that should show a high value of gravity . we find that the nextgen model best fit provides a log @xmath45 of 5.5@xmath430.5 . to set the age - gravity relation for our sample , we used these results as a basis . log @xmath45 obtained values of the candidates are 4.0 , 4.5 , 5.0 , and 5.5 , where we can state that objects with log @xmath45=4 are young and still in contraction and that objects with log @xmath45=5.5 are old . with a 0.5 uncertainly , an object with log @xmath45 of 4.5 ( 4.0 - 5.0 ) shows ambiguity when discriminating between young and old . accepting the limitations and the uncertainty in the constraints , spectral type and metallicity dependence , etc , for the atmospheric fits to our sample , objects with log @xmath45@xmath304.0 are probably young ( under @xmath23200 myrs ) , objects with log @xmath45@xmath304.5 are older but possibly still in contraction ( @xmath23300 myrs ) , and therefore objects with log @xmath45@xmath305.0 and log @xmath45@xmath305.5 are probably older than 300 myrs . the log @xmath45 obtained with model comparison are given in columns 5 and 9 of table [ tab : resulta ] and [ tab : resultac ] , columns 6 and 10 of table [ tab : resultb ] and [ tab : resultbc ] , and columns 5 , 8 and 11 of table [ tab : resultab ] and [ tab : resultbb ] . additionally , we checked if @xmath54(na i ) measurements can be used to determine youth , based on martn et al . ( 2010 ) and schlieder et al . ( 2012 ) . given the cool nature of the targets , the measurement of equivalent widths in the optical are generally measured relative to the observed local pseudo - continuum formed by molecular absorptions ( mainly tio ) , and therefore the @xmath54s are pseudo - equivalent widths " ( pavlenko 1997 ; zapatero osorio et al . however , we will call them @xmath54 for simplicity hereafter . although there is a considerable scatter in the data ( mainly from the use of different instruments ) martn et al . ( 2010 ) figure 3 plots na i @xmath54 versus spectral class for a few field objects and members or candidates of the upper scorpii ob association . to measure consistent equivalent widths in spectra of different resolutions , they established as a rule that the pseudo - continuum region was between 823 nm and 827 nm , and integrated the line from 817.5 nm to 821.0 nm . their figure shows a clear trend , and they inferred that objects with the weakest na i are likely to have low surface gravity . they established as a rule of thumb that any object with spectral class between m6 and l4 and with a na i doublet detectable but weaker than field counterparts observed with the same spectral resolution , is likely to have a low surface gravity and consequently a very young age ( i.e. , younger than 100 myr ) and substellar mass . sample b have similar resolution to the martn et al . ( 2010 ) data . as such we measured the combined equivalent widths of the two lines of the na i doublet in the same manner ( given in table [ tab : resultfinal ] ) allowing direct comparison . in the case of sample a , we convolved the spectra to the same resolution and then measured @xmath54(na i ) . the left panel of figure [ fig : ewnacomp ] presents equivalent widths of the na i doublet versus spectral type for all targets of sample a as blue crosses , and b as red filled circles , with the martn et al . ( 2010 ) data : 65 high - gravity field objects are plotted as six pointed asterisks , 6 low - gravity objects as open triangles , 12 reference field stars as open circles and 7 upper sco candidates as solid hexagons . while the metallicities of all mgs are similar to solar , the upper sco cluster does not yet have a robust metallicity determination , and other objects in martn et al . ( 2010 ) have different metallicities . we took the information from @xmath54(na i ) studies in schlieder et al . ( 2012 ) into account when compare samples a and b with martn et al . ( 2010 ) . for sample a , if we look into moving group candidatures ( thus age ) and @xmath54s(na i ) , they are inconsistent ( table [ tab : resultfinal ] column 2 ) . that is to say that some young mg candidates present a higher @xmath54(na i ) than others which have candidature to older mgs . but if we take into account the final mg membership given in column 11 of table [ tab : resultfinal ] , most of sample a targets are finally classified as candidates to mgs with ages @xmath55200 myr ( except one , sips2039 - 1126 ) , and therefore the @xmath54(na i ) become consistent ( see figure [ fig : ewnacomp ] right panel ) . sips0007 - 2458 and 2mass0334 - 2130 , both candidates to ic2391 present values of @xmath54(na i ) too high for ic2391 age . since we obtained @xmath54(na i)=6.5 for lp944 - 20 , the value of @xmath54(na i)=7.3 for sips2039 - 1126 is comparatively high for its pleiades membership . convolution might have introduced additional noise affecting the comparison and we discuss further this target in section 5.3 . for sample b , if we look into moving group candidatures and @xmath54s(na i ) ( table [ tab : resultfinal ] column 2 ) , they are consistent . that is in general , target candidates to ic2391 and the pleiades show the lowest @xmath54s , candidates to castor and sirius show higher , intermediate @xmath54 values and candidates to hyades show the highest values of @xmath54 ( figure [ fig : ewnacomp ] right panel ) . the two targets showing low values of @xmath54s(na i ) are 2mass1326 - 5022 , candidate to ic 2391 and 2mass1557 - 4350 , candidate to pleiades and ic 2391 . the rest of the candidates show higher values in agreement with their mg candidatures except for 2mass1734 - 1151 , m9 candidate to pleiades but showing larger values of @xmath54s ( na i ) than lp944 - 20 . subsequent studies by skumanich ( 1972 ) show that activity decreases over time for late - type stars . however , after a spectral type of @xmath30 m3 - 4 , stars become fully convective and the activity is driven by a turbulent dynamo in a mechanism that is still not clear . many studies have found evidence that the age - activity relation extends into the m dwarf regime ( see e.g , silvestri et al . 2006 ; mochnacki et al . 2002 ; mohanty & basri ( 2003 ) ; reiners & basri 2008 ) . recently , some authors ( berger et al . 2008 ; west et al . 2008 ) have shown that the fraction of active m0-m7 stars decreases with the vertical distance from the galactic plane , further evidence of an age - activity relation . comparing activity data to dynamical simulations , west et al . ( 2008 ) derive an `` activity life - time '' relation for m dwarfs of spectral type m0-m7 in the range 0.4 - 8 gyr . following this , we can identify older non - mg ucds in our sample through lack of chromospheric emission . comparing activity levels of our sample with activity levels of objects of the same spectral type and with known age , we can infer age ranges for our targets that would help us to confirm ( or otherwise ) their mg membership in the same way of other literature examples such us stauffer et al . ( 1995 ; fig . 13 ) , stauffer et al . ( 1997 ; fig . 8) , terndrup et al . ( 2000 ; fig . 9 ) , barrado y navascus et al . ( 2004 ; fig . 5 ) , shkolnik et al . ( 2009 ; fig . although there are several chromospheric activity indicators ( ca ii h&k , balmer lines , ca ii infrared ) , we used the h@xmath0 emission since it is included in the spectra presented here ( with the exception of the sips2039 - 1126 spectrum ) . young m dwarfs generally have h@xmath0 in emission , with the average equivalent width increasing to later spectral types ( e.g. west et al . generally the level of chromospheric activity is measured by comparison of the diagnostic line s flux , but for m dwarfs equivalent width is also used . we derived h@xmath0 pseudo - equivalent widths by direct integration of the line profile using the splot task in iraf , however , we will utilise them as @xmath54s . taking into account that @xmath54(h@xmath0 ) will reflect any activity variability , often present in young objects , and that moving groups with different ages will show different activity saturation levels as well as a higher dispersion in the level of activity when they are older ( see e.g. stauffer et al . 1997 , pizzolato et al . 2003 ; ryan et al . 2005 ; lpez - santiago 2005 ; martnez - arnaz et al . 2011 ) , we compared the @xmath54(h@xmath0 ) versus spectral type of our candidates with objects of known age ( obtained from other methods ) . we compiled a sample of m dwarfs with known age and @xmath54(h@xmath0 ) from the literature . these were taken from hyades and pleaides members in terndrup et al . ( 2000 ) , young and old disk and field stars in mohanty & basri ( 2003 ) and from a compilation of targets with ages determined by different methods in shkolnik et al . when using targets with mg or cluster memberships , we assigned them the age of the cluster . from mohanty & basri ( 2003 ) , we assigned an averaged value of 500 myr to yd , 600 myr to yo ( old - young ) , 1 gyr for od and more than 1200 myr for halo targets . the @xmath54(h@xmath0 ) of our objects is given in column 3 of table [ tab : resultfinal ] . figure [ fig : ewhacomp ] left panel presents @xmath54(h@xmath0 ) versus spectral type for our candidates , with blue crosses and red filled circles for sample a and b respectively , in comparison with data from the literature . in the right panel of figure [ fig : ewhacomp ] , only three ranges of ages are plotted to better discriminate the possible age interval of our samples . sample a targets are plotted as filled squares and sample b as filled triangles , where targets classified with ages @xmath53300 myr are plotted in blue and targets with ages @xmath1300 myr are plotted in red . although theoretically equal , equivalent widths measured in lower resolution spectra can be slightly larger than measured in higher resolution spectra ( due to line blending ) , and pseudo - equivalent width measurements show variation with resolution , therefore there is a possibility that variability effects are seen in our observations . taking into consideration that only one data point is available for each of our candidates ( and therefore variability can not be determined ) , that older mg members can present higher scatter , and that for sample b , values of @xmath54(h@xmath0 ) have been measured at low - resolution , we have determined a rough age interval for 18 of the candidates ( we have included targets in the limiting spectral type m7 and m7.5 ) . we have assumed that the candidates are single objects . any stellar or substellar companion would probably increase the activity levels making the targets look younger when we measure @xmath54(h@xmath0 ) ( e.g. basri et al . 1995 ; ruten 1987 ; schrijver & zwaan 1991 ; glvez - ortiz 2005 ) . all targets in sample a and b with spectral types under m7.5 , have @xmath54(h@xmath0 ) values inside the limits of an age that agree with at least one of their mg candidature ( see columns 3 and 4 of table [ tab : resultfinal ] and figure [ fig : ewhacomp ] ) . it should be noted that the sips2039 - 1126 spectrum does not include the h@xmath0 region , thus the activity criterion has not been used in this case . the li i doublet at 6708 is an important diagnostic of age in young late - type stars . lithium is destroyed in fully convective low - mass objects , with mass from 0.3 - 0.06m@xmath56 ( e.g. rebolo et al . 1996 ) as all acquire core temperatures as they heat up during early contraction . convection ensures that atmospheric lithium abundance reflects core lithium content , and the observable lithium doublet thus acts as a gauge for the core temperature and contraction age . the so - called `` lithium edge '' that separates objects with lithium from those without , can be seen in co - eval populations ( e.g. in open clusters and moving groups ) , through the spectral type ( or mass ) . the lithium edge advances towards later spectral type as a population ages ( see fig . 9 and caption of paper i ) , and the presence or absence of lithium thus provides a critical age constraint as a function of an object s spectral type . due to the low resolution and the presence of artifacts in the li i spectral region for sample b , we searched for the li i doublet only in sample a candidates . the li i region in the sample a spectra show s / n values of no more than 10 in all the 12 candidates which hampers the search for the absorption line . we found the li i absorption line in sips2045 - 6332 and sips2039 - 1126 only , but we do not discard the possibility of other fainter candidates with no signal in this region , presenting some absorption when better spectra can be achieved . we also measured li i in lp944 - 20 with the parameters found here , obtaining logn(li ) = 3.00 @xmath43 0.5 dex , using synthetic spectra with t@xmath2=2400 k , @xmath31 = 4.0 , dusty atmosphere structure and a rotational velocity of 35 km s@xmath36 ( measured by us ) . this is within the uncertainties of other literature findings ( see section 4.4.1 ) . clarke ( 2010 ) found the lithium signature in sips2045 - 6332 and sips2039 - 1126 suggesting youth and a bd nature inside the mg membership . here , we obtained the theoretical li i pseudo - equivalent widths , relative to the computed pseudo - continuum formed by molecular absorption , via direct integration of the line profile over the spectral interval 6703.0 - 6710.8 . figure [ fig : lithium ] shows the li i area for sips2045 - 6332 and sips2039 - 1126 with their respective fits . li abundance of logn(li ) = 3.5 @xmath43 0.5 dex for sips2045 - 6332 and logn(li ) = 3.0 @xmath43 0.5 dex for sips2039 - 1126 provides age estimations as young as 7 - 100 myrs or the possible bd condition . the lithium test was first proposed by rebolo , martn & maggazz ( 1992 ) and developed by magazz , martn & rebolo ( 1993 ) to distinguish bds from stellar objects . while stars and low mass stars deplete lithium with time ( see explanation in section 5.3 ) , substellar objects with m@xmath10.06 m@xmath57 mass can not achieve the temperature needed to destroy lithium and so it should be preserved independently of the object s age ( chabrier & baraffe 2000 ; basri et al . since young low mass objects have not yet depleted all lithium , the discrimination between stars and bd through the lithium test should take age into account . * * sips2045 - 6332 : * sips2045 - 6332 , was classified through several spectral indices as an m8.5 spectral type ( paper ii ) . it shows a clear li i absorption line ( figure [ fig : lithium ] ) . we used the parameters found in the best fit of the synthetic models and calculated its lithium abundance , logn(li ) = 3.5 @xmath43 0.5 dex . the dusty best fit @xmath31 values of 4.0@xmath430.5 and this lithium provide an age that confirms sips2045 - 6332 as probable castor member . + we noticed that @xmath54(h@xmath0 ) value , 2.1 , is quite small for a young object . but with a m8.5 spectral type , this value agrees with the findings of other authors , e.g. mohanty & basri ( 2003 ) , reiners & basri ( 2010 ) , barnes et al . ( 2013 ) , that found that the h@xmath0 emission is roughly constant from mid- to late m , but there is a sharp drop between m8-l0 ( see e.g. fig . 5 of mohanty & basri 2003 ) , from which they show little or significantly reduced emission in spite of significant rotation . + at the castor mg s age , and with a spectral type of m8.5 ( m@xmath10.06m@xmath57 ) , lithium confirms it also as a bd . * * sips2039 - 1126 : * similarly , sips2039 - 1126 , was classified through several spectral indices as an m8.0 spectral type ( paper ii ) . it shows a li i absorption line ( figure [ fig : lithium ] ) although the s / n of the area is only @xmath58 . using the parameters from the best fit of the synthetic models , we obtained logn(li ) = 3.0 @xmath43 0.5 dex ( section 5.3 ) . the dusty best fit @xmath31 values of 4.5@xmath430.5 suggest it is probably still in contraction . in section 5.1 we saw that @xmath54(na i ) of sips2039 - 1126 was high considering its pleiades membership . we have used gravity criteria for targets up to 200 myr but as seen in fig . [ fig : modelo ] , following schlieder et al . ( 2012 ) , some pleiades members may lie outside these limits . we thus can not discard this target as a pl member taking into account the rest of the evidence . + at the pleiades age or in the age interval suggested by the gravity values , the lithium depletion boundary occurs at m6.5 ( m=0.075m@xmath57 ; barrado y navascus et al . thus , with an m8.0 spectral type , lithium confirms sips2039 - 1126 as a bd . we have presented a study of the spectral signatures of 25 low - mass objects that were candidate members of five young moving groups . we studied different typical age - constraining spectroscopic criteria utilising high and low resolution spectra , and combined the results to extract a final membership assessment for each target . we took into account spectral classification and thus approximated atmospheric temperature to apply the most appropriate model or models in the search of the best physical parameter constraints . when we used the synthetic stellar atmospheric models , the good agreement between observed and theoretical energy distributions suggests that semi - empirical models describe well the impact of dust in the atmospheres of m dwarfs . @xmath37-function analysis shows that the inclusion of dust effects makes it possible to achieve better fits for objects with @xmath59 k. in both samples , although youth can generally be established , @xmath31 and @xmath54(na i ) were not useful for discriminating between different mgs . some targets of sample a that were candidates to the hy moving group having lower @xmath31 values than candidates to younger mgs , and similarly in @xmath54(na i ) values . in sample b the values are more consistent but most of the targets have several candidatures , making the situation more complicated . however , looking at the final membership results , only one target of sample a is classified as a member of a mg with age @xmath1200 myr , which makes this criterion non applicable for most of them . for sample b , in spite of the multiple membership candidatures in the final results , we can discriminate between the lower values of @xmath31 ( or lower values of @xmath54s(na i ) ) for targets that are members of younger mgs , medium values for targets that are members of intermediate age mgs and the higher values for targets that are members of oldest mgs , which finally is consistent with the @xmath31-age relation . the comparison of @xmath54(na i)s is probably useful to discriminate very young targets ( less than 100 myr ) in a homogeneous sample ( as suggested by schlieder et al . 2012 ) while @xmath31 values are probably useful to discriminate very young ( less than 100 myr ) and very old ( more that 600 myr ) objects , serving as complementary study to other criteria . the activity - age relation is a reasonably useful age criterion up to @xmath30m7 spectral type but activity variability , binarity , and the comparison of different instruments / resolutions can introduce considerable uncertainties in age estimation . the presence of lithium discriminates between young and old low mass stars but an ambiguity between young low mass stars and bds limits its use in this cool temperature region . gravity sensitive features and rotational velocity can be useful youth indicators in support of activity and lithium diagnostics , but clearly have large dispersion and they need to be applied in conjunction with similar data . for a reliable constraint of cool object ages , a combination of different criteria are needed . we find 10 of the 13 objects from sample a to be probable members of one of the mgs , and also that 2mass0020 - 2346 is a young disk member . sips0007 - 2458 , a candidate member of the ic2391 mg , shows positive candidature under most criteria but does not show the expected lithium ( associated with the age of the ic2391 mg ) . due to the age constraint from @xmath60 emission and surface gravity , we are inclined to think that sips0007 - 2458 is probably a young object from the yd class , but is not in the ic2391 mg , although it could also be a contaminant field object . 2mass0334 - 2130 is a similar case , with an m4.5 - 6 spectral type , it should contain lithium if it were an ic2391 mg member . except for the @xmath60 criterion , that could be reflecting a maximum in activity level or the influence of an unseen companion , everything indicates that 2mass0334 - 2130 is older than the ic2391 mg . thus we also conclude that 2mass0334 - 2130 is not an ic2391 member , although it could be a young object from the yd , or a contaminant target from the field . in sample b , 10 objects could be mg members , although more information is needed to discriminate between the possible mg membership of objects with several candidatures . we could not obtain any results for 2mass1909 - 1937 , so this remains an astrometric candidate to the castor mg . 2mass1734 - 1151 gave contradictory results so it also remains an astrometric pleiades mg candidate until further analysis can be performed . we also confirm two moving group candidates , sips2045 - 6332 and sips2039 - 1126 , as bds . with all the acquired information , we find that 85% , 83% and 84% of samples a , b , and both combined , show spectroscopic signatures of youth in agreement with the age of the moving group to which they present kinematic membership . this result suggests that there is a fairly low rate of contamination in such kinematic candidate samples . also , there are candidates that can not be confirmed or dismissed with the information obtained , which might ( in the future ) increase the final confirmation rate . we note that additional diagnostics to assess membership may be measured in the future , such as chemical tagging . recently , in a kinematical - chemical investigation of the ab dor moving group `` stream '' , barenfeld et al . ( 2013 ) shows that kinematics , color - magnitude positions , and stellar youth indicators alone can be insufficient for testing whether a kinematic group of stars actually shares a common origin . future chemical study of our mg candidates will complement this study . in addition , this study has also helped to test and improve the atmosphere models for cool dwarfs ( see kuznetsov et al . 2012 and kuznetsov et al . youth can be an advantage for various reasons . because young objects are brighter they can be detected and studied out to greater distance than older , fainter objects . also , the proximity of the mgs can allows the exploration of the faint circumstellar environment at relatively small distances from the star . indeed , we will target our strongest mg candidates to search for lower mass companions using high resolution imaging techniques ( e.g. adaptive optics ) . the discovery of planetary systems around young and low - mass objects provides crucial information for the understanding of planetary and stellar formation . from our complete study of the 80 objects considered ( within papers i , ii and this paper ) , we have found a total of 45 new possible mg members ( with 8 of them showing more than one mg candidature ) and around 21 possible young disk objects with no clear membership of the 5 mgs considered . of these we find 26 , 13 , 6 , 4 and 5 possible members of the hyades , castor , ursa major , pleiades and ic2391 mgs respectively . tables [ tab : comptod ] and [ tab : compothers ] of the appendix compile all the m - l dwarfs investigated in our study . lllccc name & @xmath0 ( 2000 ) & @xmath61 ( 2000 ) & j - mag & moving group@xmath62 & @xmath63@xmath64 + & ( h m s ) & ( @xmath65 ) & & candidature & ( km s@xmath36 ) + sips0007 - 2458 & 0 7 7.800 & -24 58 3.80 & 13.11@xmath430.02 & ic & 18 + 2mass0020 - 2346 & 0 20 23.155 & 23 46 5.38 & 12.35@xmath430.02 & yd & 12 + denis0021 - 4244 & 0 21 5.896 & -42 44 43.33 & 13.52@xmath430.03 & ic , ca & 11 + sips0027 - 5401 & 0 27 23.240 & 54 1 46.20 & 12.36@xmath430.02 & hy & 27 + sips0153 - 5122 & 1 53 11.430 & 51 22 24.99 & 13.45@xmath430.03 & ic , hy & 14 + sips0214 - 3237 & 2 14 45.440 & -32 37 58.20 & 14.01@xmath430.02 & hy & 18 + sips0235 - 0711 & 2 35 49.470 & 7 11 21.90 & 12.45@xmath430.03 & hy & 22 + 2mass0334 - 2130 & 3 34 10.657 & 21 30 34.35 & 11.91@xmath430.02 & ic & @xmath110@xmath66 + sips2039 - 1126 & 20 39 13.081 & -11 26 52.30 & 13.79@xmath430.03 & pl & @xmath4715 + sips2045 - 6332 & 20 45 2.278 & -63 32 5.30 & 12.62@xmath430.03 & pl , ca & @xmath4715 + lehpm4908 & 22 36 42.656 & -69 34 59.30 & 12.68@xmath430.02 & hy & - + 2mass2254 - 3228 & 22 54 58.110 & 32 28 52.20 & 13.58@xmath430.03 & pl , ca & 19 + lehpm6542 & 23 57 54.822 & 19 55 1.89 & 13.31@xmath430.02 & hy & 16 + @xmath62 mg to which targets are kinematic candidate ( paper ii ) . hyades mg ; si= ursa major group ; ca= castor mg ; pl= pleiades ; ic = ic 2391 mg ; yd= other young disk object . + @xmath64 from paper ii . + @xmath66 measused with low s / n . + lllcccc name & @xmath0 ( 2000 ) & @xmath61 ( 2000 ) & j - mag & moving group + & ( h m s ) & ( @xmath65 ) & & candidature + 2mass0814 - 4020 & 08 14 35.46 & -40 20 49.26 & 14.356@xmath430.023 & hy + 2mass1146 - 4754 & 11 46 51.04 & -47 54 38.17 & 14.897@xmath430.042 & si + 2mass1236 - 6536 & 12 36 32.38 & -65 36 35.6 & 15.277@xmath430.057 & pl , ca , ic , si + 2mass1326 - 5022 & 13 26 53.48 & -50 22 27.04 & 14.715@xmath430.037 & ic , ca + 2mass1433 - 5148 & 14 33 41.95 & -51 48 03.70 & 14.206@xmath430.034 & ca , pl , si , ic + 2mass1557 - 4350 & 15 57 27.39 & -43 50 21.47 & 14.224@xmath430.028 & pl , ca , ic + 2mass1618 - 3214 & 16 18 08.92 & -32 14 36.17 & 14.920@xmath430.040 & ic , ca , si + 2mass1734 - 1151 & 17 34 30.53 & -11 51 38.83 & 13.110@xmath430.028 & pl + 2mass1736 - 0407 & 17 36 56.09 & - 4 07 25.84 & 15.516@xmath430.070 & si , ca + 2mass1745 - 1640 & 17 45 34.66 & -16 40 53.81 & 13.646@xmath430.026 & si , ca , hy + 2mass1756 - 4518 & 17 56 29.63 & -45 18 22.47 & 12.386@xmath430.019 & ca + 2mass1909 - 1937 & 19 09 08.21 & -19 37 47.96 & 14.520@xmath430.026 & ca + ccccccc number & date & telescope & instrument & spect . range & dispersion + & & & & ( ) & ( ) + 1 & 28/03 - 21/06/08 & eso - vlt - u2 & uves@xmath62 & 6700 - 10425@xmath64 & 0.027 - 0.041 + 2 & 01 - 09 - 2009 - 09/01/2010 & eso - vlt - u2 & uves@xmath62 & 5700 - 7530 & 7650 - 9470 & 0.027 - 0.041 + 3 & 03 - 05/05/2010 & 6.5 m baade - maguellan & imacs short - camera@xmath67 & 6550 - 10000 & 1.98 + 4 & 15 - 16/02/2011 & 6.5 m baade - maguellan & imacs short - camera@xmath67 & 4300 - 10800 & 1.97 + @xmath62 uves : ultraviolet and visual echelle spectrograph . + @xmath64 effective range . + @xmath67 imacs : the inamori magellan areal camera and spectrograph . + lcccccccccccccc & & & & + object & @xmath68 & @xmath69 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & best fit@xmath62 + & & ( k ) & & ( k ) & ( cm s@xmath70 ) & dex & & & ( k ) & cm s@xmath70 & dex & & + gl876 & m4.0 & 3100 & & 3400 & 5.5 & 0.0 & 2.9 & & - & - & - & - & nextgen + sips0007 - 2458 & m7.5 & 2550 & & 2700 & 4.5 & -1.0 & 7.0 & & 2700 & 4.5 & 0.0 & 6.2 & s .- e . + 2mass0020 - 2346 & m6.0 & 2800 & & 2800 & 5.5 & -0.5 & 4.8 & & 2800 & 5.0 & 0.0 & 4.56 & s .- e . + sips0027 - 5401 & m6.5 & 2650 & & 2700 & 4.5 & -1.0 & 3.35 & & 2700 & 4.5 & 0.0 & 2.70 & s .- e . + sips0153 - 5122 & m6.0 & 2800 & & 2800 & 5.5 & -0.5 & 2.42 & & 2800 & 5.0 & 0.0 & 2.31 & s .- e . + sips0214 - 3237 & m6.5 & 2650 & & 2800 & 5.0 & -1.0 & 1.44 & & 2800 & 5.0 & 0.0 & 1.2 & s .- e . + sips0235 - 0711 & m6.0 & 2800 & & 2700 & 5.0 & -1.0 & 0.97 & & 2700 & 4.5 & 0.0 & 0.91 & s .- e . + 2mas0334 - 2130 & m4.5-m6.0 & 3000 - 2800 & & 2800 & 5.5 & -0.5 & 4.65 & & 2800 & 5.0 & 0.0 & 5.65 & @xmath66 + lehpm4908 & m6.0 & 2800 & & 2800 & 4.5 & -1.0 & 7.02 & & 2900 & 5.0 & 0.0 & 5.70 & s .- e . + 2mass2254 - 3228 & m5.5 & 2850 & & 2700 & 5.0 & -1.0 & 1.37 & & 2700 & 4.5 & 0.0 & 1.25 & s .- e . + lehpm6542 & m6.0 & 2800 & & 2800 & 5.0 & -0.5 & 13.5 & & 2700 & 4.5 & 0.0 & 4.03 & s .- e . + @xmath62 nextgen or semi - empirical ( s .- e . ) + @xmath66 both models are equally likely . + lccccccccccccccc & & & & + object & @xmath71 & @xmath68 & @xmath72 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & best fit + & & ( k ) & ( k ) & & ( k ) & ( cm s@xmath70 ) & dex & & & ( k ) & cm s@xmath70 & dex & & + 2mass1146 - 4754 & m8.5 & m7 - 8 & @xmath232500 & & 2800 & 5.5 & -0.5 & 0.39 & & 2700 & 4.5 & 0.0 & 0.18 & s .- e . + 2mass1236 - 6536 & m7.5 & m4.0 & 3100 & & 3200 & 5.5 & 0.0 & 0.64 & & 2900 & 4.0 & 0.0 & 1.14 & nextgen + 2mass1326 - 5022 & m9.0 & m7.5 & 2550 & & 2800 & 5.0 & 0.0 & 0.57 & & 2700 & 4.0 & 0.0 & 0.22 & s .- e . + 2mass1433 - 5148 & m9.0 & m6 - 7 & @xmath232650 & & 3000 & 5.5 & 0.0 & 0.58 & & 2900 & 5.0 & 0.0 & 0.27 & s .- e . + 2mass1618 - 3214 & m7.0 & m6.5 & 2650 & & 2800 & 5.5 & -0.5 & 0.69 & & 2700 & 4.5 & 0.0 & 0.41 & s .- e . + 2mass1736 - 0407 & m8.5 & m4 - 5 & @xmath233000 & & 3200 & 5.0 & 0.0 & 0.49 & & 2900 & 4.0 & 0.0 & 0.55 & nextgen + lccccccccccccccc & & & & & + object & @xmath68 & @xmath69 & & @xmath32 & @xmath31 & @xmath46 & & @xmath32 & @xmath31 & @xmath46 & & @xmath32 & @xmath31 & @xmath46 & best fit + & & ( k ) & & ( k ) & ( cm s@xmath70 ) & & & ( k ) & cm s@xmath70 & & & ( k ) & cm s@xmath70 & & + sips2039 - 1126 & m8.0 & 2500 & & 2800 & 5.5 & 22.45 & & 2600 & 5.0 & 39.96 & & 2600 & 4.5 & 40.12 & + semi - empirical & & & & 2700 & 4.5 & 22.90 & & 2600 & 4.5 & 20.00 & & 2600 & 4.5 & 20.01 & s .- e . dusty + lccccccccccccccc & & & & & + object & @xmath71 & @xmath68 & & @xmath32 & @xmath31 & @xmath46 & & @xmath32 & @xmath31 & @xmath46 & & @xmath32 & @xmath31 & @xmath46 & best fit + & & & & ( k ) & ( cm s@xmath70 ) & & & ( k ) & cm s@xmath70 & & & ( k ) & cm s@xmath70 & & + 2mass0814 - 4020 & l4.0 & m7 - 8 & & 2700 & 5.0 & 0.74 & & 2600 & 5.5 & 1.15 & & 2600 & 5.0 & 1.22 & + semi - empirical & & & & 2700 & 4.5 & 0.35 & & 2600 & 4.5 & 0.28 & & 2600 & 4.5 & 0.28 & s .- e . dusty @xmath73 + 2mass1557 - 4350 & l0.0 & m7.5 - 8.5 & & 2800 & 4.5 & 0.74 & & 2600 & 5.5 & 0.93 & & 2400 & 5.5 & 1.11 + semi - empirical & & & & 2800 & 4.0 & 0.31 & & 2500 & 4.0 & 0.29 & & 2600 & 4.0 & 0.29 & s .- e . dusty @xmath62 + 2mass1756 - 4518 & m7.5 & m8 - 9 & & 2600 & 5.5 & 0.21 & & 2600 & 5.5 & 0.18 & & 2600 & 5.0 & 0.20 & + semi - empirical & & & & 2600 & 4.0 & 0.09 & & 2600 & 5.5 & 0.08 & & 2600 & 5.0 & 0.10 & s .- e . dusty + @xmath62 dusty and cond have the same @xmath46 , but dusty model is more appropriate in the temperature range . + @xmath64 we could appreciate parts of the spectrum were nextgen was the best fit . + lcccccccccccccc & & & & + object & @xmath68 & @xmath69 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & best fit + & & ( k ) & & ( k ) & ( cm s@xmath70 ) & dex & & & ( k ) & cm s@xmath70 & dex & & + denis0021 - 4244 & m9.5 & 2300 - 2400 & & 2000 & 5.5 & 0.0 & 7.85 & & 2400 & 5.5 & 0.0 & 13.40 & + semi - empirical & & & & 2400 & 4.0 & 0.0 & 5.89 & & 2500 & 4.0 & 0.0 & 6.62 & s .- e . dusty + sips2045 - 6332 & m8.5 & 2300 - 2400 & & 2000 & 5.5 & 0.0 & 11.39 & & 2400 & 4.0 & 0.0 & 17.19 & + semi - empirical & & & & 2400 & 4.0 & 0.0 & 8.14 & & 2500 & 4.0 & 0.0 & 10.30 & s .- e . dusty + lp944 - 20 & m9.0 & 2300 - 2400 & & 2000 & 5.5 & 0.0 & 0.42 & & 2400 & 5.5 & 0.0 & 0.57 & + semi - empirical & & & & 2400 & 4.0 & 0.0 & 0.29 & & 2400 & 4.0 & 0.0 & 0.35 & s .- e . dusty + lccccccccccccccc & & & & + object & @xmath71 & @xmath68 & @xmath69 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & & @xmath32 & @xmath31 & [ m / h ] & @xmath46 & best fit + & & & ( k ) & & ( k ) & ( cm s@xmath70 ) & dex & & & ( k ) & cm s@xmath70 & dex & & + 2mass1734 - 1151 & l0.0 & m9.0 & 2300 - 2500 & & 2600 & 5.5 & 0.0 & 1.34 & & 2600 & 5.0 & 0.0 & 1.43 & + semi - empirical & & & & & 2400 & 4.0 & 0.0 & 0.79 & & 2400 & 4.0 & 0.0 & 0.86 & s .- e . dusty + 2mass1745 - 1640 & l8.0 & m9-l2 & @xmath232300 - 2000 & & 2000 & 5.5 & 0.0 & 0.06 & & 2200 & 5.5 & 0.0 & 0.07 & + semi - empirical & & & & & 2000 & 5.5 & 0.0 & 0.05 & & 2200 & 5.5 & 0.0 & 0.05 & s .- e . dusty + 2mass1909 - 1937 & m9.0 & l0.0 & @xmath232200 & & 2100 & 3.5 & 0.0 & 0.55 & & - & - & - & - & + semi - empirical & & & & & 2000/2100 & 6.0/3.5 & 0.0 & 0.45 & & - & - & - & - & s .- e . + lcccccccccc object & @xmath54(na i)@xmath62 & @xmath54(h@xmath0)@xmath64 & age@xmath64 & mg memb . & m. from & m. from & m. from & m. from & m. from & final mg + & ( ) & ( ) & ( myr ) & ( kinem / astrom)@xmath67 & @xmath54(na i)@xmath62 & @xmath54(h@xmath0)@xmath64 & @xmath31@xmath74 & li i@xmath75 & @xmath49@xmath76 & + sips0007 - 2458 & 8.2 & 11.0 & 1 - 300 & ic & n & y & y & n & y & _ yd _ + 2mass0020 - 2346 & 8.1 & 6.1 & @xmath23300 & yd & y & y & y & ? & y & _ yd _ + denis0021 - 4244 & 8.4 & @xmath66 & n / a & ic , ca & ca & n / a & y & - & n & _ ca _ + sips0027 - 5401 & 7.7 & 8.1 & @xmath23300 & hy & n / a & y & y & y & y & _ hy _ + sips0153 - 5122 & 8.0 & 7.5 & @xmath23300 & ic , hy & hy & hy & hy & hy & y & _ hy _ + sips0214 - 3237 & 7.9 & 8.4 & @xmath23300 & hy & n / a & y & y & y & y & _ hy _ + sips0235 - 0711 & 7.9 & 7.7 & @xmath53300 & hy & n / a & y & y & y & y & _ hy _ + 2mass0334 - 2130 & 7.3 & 11.5 & 1 - 300 & ic & n & y & n & n & ? & _ yd _ + sips2045 - 6332 & 6.3 & 2.1 & n / a & pl , ca & ca & n / a & y & y & ? & _ ca _ + sips2039 - 1126 & 7.3 & - & - & pl & n & - & y & y & y & _ pl _ + lehpm4908 & 7.3 & 6.5 & @xmath23300 & hy & n / a & y & y & y & - & _ hy _ + 2mass2254 - 3228 & 8.5 & 5.8 & @xmath23300 & pl , ca & ca & ca & y & - & y & _ ca _ + lehpm6542 & 7.6 & 4.9 & @xmath53300 & hy & n / a & y & y & y & y & _ hy _ + & & & & & & & & & + 2mass0814 - 4020 & 9.0 & 7.2 & @xmath53300 & hy & n / a & y & y & n / a & n / a & _ hy _ + 2mass1146 - 4754 & 7.9 & 5.2 & @xmath53300 & si & n / a & y & y & n / a & n / a & _ si _ + 2mass1236 - 6536 & 5.8 & 1.0 & 300 - 1200 & pl , ic , ca , si & ca , si & ca , si & n & n / a & n / a & _ ca , si _ + 2mass1326 - 5022 & 4.2 & 23.6 & 1 - 100 & ic , ca & ic & ic & y & n / a & n / a & _ ic _ + 2mass1433 - 5148 & 7.9 & 19.8 & 1 - 100 & pl , ic , ca , si & ca , si & pl , ic & si & n / a & n / a & _ ca ? , si _ + 2mass1557 - 4350 & 4.0 & 16.4 & 1 - 100 & pl , ic , ca & pl , ic & n / a & y & n / a & n / a & _ pl , ic _ + 2mass1618 - 3214 & 8.0 & 7.8 & 1 - 300 & ic , ca , si & ca , si & y & y & n / a & n / a & _ ic ? , ca , si _ + 2mass1734 - 1151 & 7.7 & 5.5 & n / a & pl & n & n / a & y & n / a & n / a & _ pl ? _ + 2mass1736 - 0407 & 6.1@xmath77 & 4.2 & 50 - 300 & ca , si & n / a & y & y & n / a & n / a & _ ca , si _ + 2mass1745 - 1640 & 5.6 & 1.4 & @xmath53300 & ca , si , hy & y & n / a & hy & n / a & n / a & _ ca ? , si ? , hy _ + 2mass1756 - 4518 & 7.5 & 3.2 & n / a & ca & n / a & n / a & y & n / a & n / a & _ ca _ + 2mass1909 - 1937 & 6.3 & @xmath66 & n / a & ca & n / a & n / a & - & n / a & n / a & _ ca ? _ + @xmath62 equivalenth width of the na i doublet as expained in section 5.1 . + @xmath64 see section 5.2 . + @xmath67 mg membership from kinematic ( sample a ) or from photometric and astrometric criteria ( sample b ) . see section 2 . + @xmath74 from value obtained in the synthetic fits , given in columns 5 and 9 of tables [ tab : resulta ] , [ tab : resultb ] and [ tab : resultab ] . + @xmath75 only applicable for high resolution data . see section 5.3 . + @xmath76 from paper ii . + @xmath66 absorption line filled in with emission . + @xmath78 absorption line . + @xmath77 measured with a cosmic ray in the line . glvez - ortiz acknowledges financial support by the spanish micinn under the consolider - ingenio 2010 program grant csd2006 - 00070 : first science with the gtc ( http://www.iac.es/consolider-ingenio-gtc ) . she also acknowledges the support of a jae - doc csic fellowship co - funded with the european social fund under the program _ junta para la ampliacin de estudios_. m.k . kuznetsov and ya.v . pavlenko work was supported by eu pf7 marie curie initial training networks ( itn ) ropacs project ( ga n 213646 ) . financial support was also provided by the spanish ministerio de ciencia e innovacin and ministerio de economa y competitividad under aya2011- 30147-c03 - 03 grant . s.l.f . acknowledges funding support from the eso - government of chile mixed committee 2009 , and from the gemini conicyt grant # 32090014/2009 . j.s.j . acknowledges funding by fondecyt through grant 3110004 and partial support from cata ( pb06 , conicyt ) , the gemini - conicyt fund and from the comit mixto eso - gobierno de chile . financial support was also provided by the mexican research council ( conacyt ) grant . based on observations made with eso telescopes , with uves high - resolution optical spectrograph at vlt kueyen 8.2-m telescope at paranal observatory ( chile ) in programmes 081.c-0222(a ) and 084.c-0403(a ) . and observations obtained at the magellan telescope at las campanas ( chile ) observatory . 99 anders e. , grevesse n. , 1989 , geochim . acta 53 , 3273 antoja t. , figueras f. , fernndez d. , torra j. , 2008 , a&a , 490 , 135 allard f. , hauschildt p. h. , alexander d. r. , tamanai a. , schweitzer a. , 2001 , apj , 556 , 357 allard n. f. , allard f. , hauschildt p. h. , kielkopf j. f. , machin l. , 2003 , a&a , 411 , 473 allard f. , homeier d. , freytag b. , 2011 , aspc , 448 , 91 barenfeld s. a. , bubar e. j. , mamajek e. e. , young p. a. , 2013 , apj , 766 , 6b barnes j. r. , et al . , 2013 , mnras , accepted barrado y navascus d. , stauffer j. r. , jayawardhana r. , 2004 , aj , 614 , 386 barrado y navascus d. , 2006 , a&a , 459 , 511 basri g. , laurant r. , walter f. m. , 1985 , apj , 298 , 761 basri g. , 2000 , ara&a , 38 , 485 basri g. , mohanty s. , allard f. , hauschildt p. h. , delfosse x. , martn e. l. , forveille t. , goldman b. , 2000 , apj , 538 , 363 berger e. , et al . , 2008 , apj , 676 , 1307b bovy j. , hogg d. w. , 2010 , apj , 717 , 617 burrows a. , hubbard w. b. , lunine j. i. , liebert j. , 2001 , rev . , 73 , 719 burrows a. , volobuyev m. , 2003 , apj , 583 , 985b chabrier g. , baraffe i. , 2000 , ara&a , 38 , 337 clarke j. r. a. , 2010 , phd thesis , university of hertfordshire clarke j. r. a. , et al . , 2010 , mnras , 402 , 575 ( paper i ) correia a. c. m. , couetdic j. , laskar j. , et al . 2010 , a&a , 511 , a21 cruz k. l. , reid i. n. , 2002 , aj , 123 , 2828 dahn c. c. , harris h. c. , vrba f. j. , guetter h. h. , canzian b. , henden a. a. , levine s. e. , luginbuhl c. b. et al . 2002 , aj , 124 , 1170 deacon n. r. , hambly n. c. , cooke j. a. , 2005 , a&a , 435 , 363 deacon n. r. , hambly n. c. , 2007 , a&a , 468 , 163 eggen o. j. , 1984a , aj 89 , 1358 eggen o. j. , 1984b , apjs , 55 , 597 eggen o. j. , 1989 , pasp 101 , 366 famaey b. , pont f. , luri x. , udry s. , mayor m. , jorissen a. , 2007 , a&a , 461 , 957 famaey b. , siebert , a. , jorissen a. , 2008 , a&a , 483 , 453 folkes s. l. , 2009 , phd thesis work , university of hertfordshire folkes s. l. , et al . , 2012 , mnras , 427 , 3280 francis c. , anderson e. , 2009 , rspsa , 465 , 3425 glvez - ortiz m. c. , 2005 , phd thesis , universidad complutense de madrid glvez - ortiz m. c. , et al . , 2010 , mnras , 409 , 552 ( paper ii ) gray d.f . the observation and analysis of stellar photospheres , 2005 , new york : cambrige university press gorlova n. i. , meyer m. r. , rieke g. h. , liebert j. , 2003 , apj , 593 , 1074 hauschildt p. , allard f. , 1992 , apj , 512 , 377 jenkins j. s. , ramsey l. w. , jones h. r. a. , pavlenko y. , gallardo j. , barnes j. r. , pinfield d. j. , 2009 , apj , 704 , 975 jones h. r. a. , longmore a. j. , allard f. , hauschildt p. h. , 1996 , mnras , 280 , 77 jones h. r. a. , tsuji t. , 1997 , apj , 480 , 1400 johnson d. r. h. , soderblom d. r. , 1987 , aj , 93 , 864 kirkpatrick j. d. , kelly d. m. , rieke g. h. , liebert j.,allard f. , wehrse r. , 1993 , apj , 402 , 643 kirkpatrick j. d. , henry t. j. , simons d. a. , 1995 , aj , 109 , 797 klement r. , fuchs b. , rix h. w. , 2008 , apj , 685 , 261 kupka f. , piskunov n. , ryabchikova t. a. et al . , 1999 , a&as , 138 , 119 kuznetsov m. k. , pavlenko ya . v. , glvez - ortiz m. c. , 2012 , kinematics and physics of celestial bodies ( kpcb ) journal , 28 , 280 kuznetsov m. k. , pavlenko ya . v. , glvez - ortiz m. c. , 2013 , epj web of conferences , volume 47 , 04002 lpez - santiago j. , 2005 , phd thesis , universidad complutense de madrid lpez - santiago j. , micela g. , montes d. , 2009 , a&a , 499 , 129 lyubchik yu . , pavlenko ya . , 2000 , kfnt suppl . 3 , 360 luhman k. l. , liebert j. , rieke g. h. , 1997 , apj , 489 , l165 luyten w. j. , 1978 , iau symposium no . 80 , 63 magazz a. , martn e. l. , rebolo r. , 1993 , apj , 404l , 17 marcy g. w. , butler r. p. , vogt s. s. , fischer d. a. , lissauer j. j. , 1998 , apj , 505 , l147 martn e. l. , rebolo r. , zapatero osorio m. r. , 1996 , apj , 469 , 706 martn e. l. , delfosse x. , basri g. , goldman b. , forveille t. , zapatero osorio m. r. , 1999 , aj , 118 , 2466 martn e. l. , delfosse x. , guieu s. , 2004 , aj , 127 , 449 martn e. l. , et al . , 2010 , a&a 517 , a53 martnez - arnaz r. , lpez - santiago j. , crespo - chacn i. , montes d. , 2011 , mnras , 414 , 2629 mcgovern m. r. , kirkpatrick j. d. , mclean i. s. , burgasser a. j. , prato l. , lowrance p. j. , 2004 , aj , 600 , 1020 mochnacki s. w. , et al 2002 , aj , 124 , 2868 mohanty s. , basri g. , 2003 , aj , 583 , 451 murgas f. , jenkins j. s. , rojo p. , jones h. r. a. , pinfield d. j. , a&a , 552 , 27 pavlenko ya . v. , 1997 , ap&ss , 253 , 43 pavlenko ya . v. , jones h. r. a. , martn e. l. , guenther e. , kenworthy m. a. , zapatero osorio m. r. , 2007 , mnras , 380 , 3 , 1285 pavlenko ya . v. , zhukovska s. v. , volobuev m. , 2007 , arep , 51 , 282 pinfield d. j. , jones h. r. a. , lucas p. w. , kendall t. r. , folkes s. l. , day - jones a. c. , chappelle r. j. , steele i. a. , 2006 , mnras , 368 , 1281 pizzolato n. , maggio a. , micela g. , sciortino s. & ventura p. , 2003 , a&a , 397 , 147 plez b. , 1998 , a&a , 337 , 495 pokorny r. s. , jones h. r. a. , hambly n. c. , 2003 , a&a , 397 , 575 pokorny r. s. , jones h. r. a. , hambly n. c. , pinfield d. j. , 2004 , a&a , 421 , 763 rajpurohit a. s. , reyle c. , schultheis m. , leinert c. , allard f. , 2011 , in sf2a-2011 : proceedings of the annual meeting of the french society of astronomy and astrophysics , ed . g. alecian , k. belkacem , r. samadi , d. valls - gabaud , 339 - 343 rajpurohit a. s. , reyl c. , schultheis m. , allard f. , scholz r. , homeier d. , 2012 , in sf2a-2012 : proceedings of the annual meeting of the french society of astronomy and astrophysics . s. boissier , p. de laverny , n. nardetto , r. samadi , d. valls - gabaud and h. wozniak , 383 - 388 rebolo r. , martn e. l. , magazz a. , 1992 , apj , 389l , 83 rebolo r. , martn e. l. , basri g. , marcy g. w. , zapatero - osorio m. r. , 1996 , apj , 469l , 53 reiners a. , basri g. , 2008 , aj , 684 , 1390 reiners a. , basri g. , 2010 , apj , 710 , 924 reyl c. , rajpurohit a.s . , schultheis m. , allard f. , 2011 , aspc , 448 , 929 ribas i. , 2003 , a&a , 400 , 297 rutten r. g. m. , 1987 , a&as , 177,131 ryan r. d. , neukirch t. , jardine m. , 2005 , a&a 433 , 323 schlieder j. e. , lpine s. , rice e. , simon m. , fielding d. , tomasino r , 2012 , aj , 143 , 114s schrijver c. j. , zwaan c. , 1991 , a&as 251 , 183 shanklad p. d. , rivera e. j. , laughlin g. , blank d. l. , price a. , gary b. , bissinger r. , ringwald f. , white g. , henry g. w. , mcgee p. , wolf a. s. , carter b. , lee s. , biggs j. , monard b. , ashley m. c. b. , 2006 , apj , 653 , 700 shkolnik e. , liu m. c. , reid i. n. , 2009 , apj , 699 , 649s silvestri n. m. et al . 2006 , aj , 131 , 1674 skumanich a. 1972 , apj , 491 , 856 stauffer j. r. , liebert j. , giampapa m. , 1995 , aj , 109 , 298 stauffer j. r. , balachandran s. c. , krishnamurthi a. , pinsonneault m. , terndrup d. m. , stern r. a. , 1997 , apj , 475 , 604 terndrup d. m. , stauffer j. r. , pinsonneault m. h. , sills a. , yuan y. , jones b. f. , fischer d. , krishnamurthi a. , 2000 , aj , 119 , 1303 tinney c.g . , 1998 , mnras , 296 , l42 west a. a. , et al . , 2004 , aj , 128 , 426 west a. a. , hawley s. l. , bochanski j. j. , covey k. r. , reid i. n. , dhital s. , hilton e. j. , masuda m. , 2008 , the astronomical journal , 135 , 785 west a. a. , et al . , 2011 , aj , 141 , 97 tsuji t. , ohnaka k. , aoki w. , 1996 , a&as , 305 , l1 zapatero osorio m. r. , bjar v. j. s. , pavlenko ya . , rebolo r. , allende prieto c. , martn e. l. , garca lpez r. j. , 2002 , a&a , 384 , 937 zhao j. , zhao g. , chen y. , 2009 , apj , 692l , 113 [ lastpage ] lllll name & @xmath0 ( 2000 ) & @xmath61 ( 2000 ) & @xmath79 & note + & ( h m s ) & ( @xmath65 ) & & + hyades & & & & + sips0004 - 5721 & 0 4 18.970 & -57 21 23.30 & m7.0 & also ic candidate ; kinematic + sips0027 - 5401 & 0 27 23.240 & -54 1 46.20 & m6.0 & kinematics + age features + 2mass0123 - 3610 & 1 23 0.506 & -36 10 30.67 & m4.5 & kinematics + sips0153 - 5122 & 1 53 11.430 & -51 22 24.99 & m6.0 & kinematics + age features + sips0214 - 3237 & 2 14 45.440 & -32 37 58.20 & m6.5 & kinematics + age features + sips0235 - 0711 & 2 35 49.470 & -7 11 21.90 & m6.0 & kinematics + age features + sips0440 - 0530 & 4 40 23.328 & -5 30 7.85 & m7.5 & kinematics + 2mass0600 - 3314 & 6 0 33.750 & -33 14 26.84 & m7.0@xmath80 & probable member from kinematics + 2mass0814 - 4020 & 8 14 35.46 & -40 20 49.26 & m7 - 8 & photometry + astrometry + age features + 2mass1745 - 1640 & 17 45 34.66 & -16 40 53.81 & m9-l2 & also ca and si candidate ; photometry + astrometry + age features + sips2014 - 2016 & 20 14 3.523 & -20 16 21.30 & m7.5 & kinematics + 2mass2031 - 5041 & 20 31 27.495 & -50 41 13.49 & m5.0 & kinematics + sips2049 - 1716 & 20 49 52.610 & -17 16 7.80 & m6.5 & kinematics + sips2100 - 6255 & 21 0 30.227 & -62 55 7.31 & m5.0@xmath80 & kinematics + denis2200 - 3038 & 22 0 2.022 & -30 38 32.71 & m9.0 & probable member from kinematics + sips2200 - 2756 & 22 0 16.838 & -27 56 29.70 & m6.0 & probable member from kinematics + 2mass2207 - 6917 & 22 7 10.313 & -69 17 14.25 & m6.5 & kinematics + 2mass2231 - 4443 & 22 31 8.657 & -44 43 18.43 & m4.5 & kinematics + lehpm4908 & 22 36 42.656 & -69 34 59.30 & m5.5 & kinematics + age features + 2mass2311 - 5256 & 23 11 30.330 & -52 56 30.17 & m5.5 & kinematics + sips2318 - 4919 & 23 18 45.952 & -49 19 17.79 & m6.5 & kinematics + sips2322 - 6357 & 23 22 5.332 & -63 57 57.60 & m6.5 & kinematics + sips2347 - 1821 & 23 47 16.662 & -18 21 50.60 & m6.5 & kinematics + sips2350 - 6915 & 23 50 3.948 & -69 15 24.39 & m6.5 & kinematics + lehpm6375 & 23 52 49.138 & -22 49 29.54 & m6.5 & probable member from kinematics + lehpm6542 & 23 57 54.822 & -19 55 1.89 & m6.0 & kinematics + age features + ursa major & & & & + 2mass1146 - 4754 & 11 46 51.04 & -47 54 38.17 & m7 - 8 & photometry + astrometry + age features + 2mass1236 - 6536 & 12 36 32.38 & -65 36 35.6 & m4.0 & also ca candidate ; photometry + astrometry + age features + 2mass1433 - 5148 & 14 33 41.95 & -51 48 03.70 & m6 - 7 & also ca candidate ; photometry + astrometry + age features + 2mass1618 - 3214 & 16 18 08.92 & -32 14 36.17 & m6.5 & also ca candidate ; photometry + astrometry + age features + 2mass1736 - 0407 & 17 36 56.09 & - 4 07 25.84 & m4 - 5 & also ca candidate ; photometry + astrometry + age features + 2mass1745 - 1640 & 17 45 34.66 & -16 40 53.81 & m9-l2 & also hy and ca candidate ; photometry + astrometry + age features + castor & & & & + denis0021 - 4244 & 0 21 5.896 & -42 44 43.33 & m9.5 & kinematics + age features + 2mass1236 - 6536 & 12 36 32.38 & -65 36 35.6 & m4.0 & also si candidate ; photometry + astrometry + age features + 2mass1433 - 5148 & 14 33 41.95 & -51 48 03.70 & m6 - 7 & also si candidate ; photometry + astrometry + age features + 2mass1618 - 3214 & 16 18 08.92 & -32 14 36.17 & m6.5 & also si candidate ; photometry + astrometry + age features + 2mass1736 - 0407 & 17 36 56.09 & - 4 07 25.84 & m4 - 5 & also si candidate ; photometry + astrometry + age features + 2mass1745 - 1640 & 17 45 34.66 & -16 40 53.81 & m9-l2 & also hy and si candidate ; photometry + astrometry + age features + 2mass1756 - 4518 & 17 56 29.63 & -45 18 22.47 & m8 - 9 & photomerty + astrometry + age features + 2mass1909 - 1937 & 19 09 08.21 & -19 37 47.96 & l0.0 & photometry + astrometry + sips2000 - 7523 & 20 0 48.171 & -75 23 6.58 & m8.0 & probable member from kinematics + sips2045 - 6332 & 20 45 2.278 & -63 32 5.30 & m9.0 & kinematics + age features + sips2114 - 4339 & 21 14 40.928 & -43 39 51.20 & m6.5 & kinematics + 2mass2242 - 2659 & 22 42 41.294 & -26 59 27.23 & m5.5 & also hy candidate ; kinematics + 2mass2254 - 3228 & 22 54 58.110 & -32 28 52.20 & m5.5 & kinematics + age features + pleiades & & & & + 2mass1557 - 4350 & 15 57 27.39 & -43 50 21.47 & m7.5 - 8.5 & also ic candidate ; photometry + astrometry + age features + 2mass1734 - 1151 & 17 34 30.53 & -11 51 38.83 & m9.0 & photometry + astrometry + sips2039 - 1126 & 20 39 13.081 & -11 26 52.30 & m7.0 & kinematics + age features + hb2124 - 4228 & 21 27 26.133 & -42 15 18.39 & m7.5 & kinematics + ic 2391 & & & & + sips0004 - 5721 & 0 4 18.970 & -57 21 23.30 & m7.0 & also hy candidate ; kinematics + 2mass1326 - 5022 & 13 26 53.48 & -50 22 27.04 & m7.5 & photometry + astrometry + age features + 2mass1557 - 4350 & 15 57 27.39 & -43 50 21.47 & m7.5 - 8.5 & also pl candidate ; photometry + astrometry + age features + 2mass1618 - 3214 & 16 18 08.92 & -32 14 36.17 & m6.5 & also ca and si candidate ; photometry + astrometry + age features + sips2341 - 3550 & 23 41 47.497 & -35 50 14.40 & m7.0 & kinematics + lllll name & @xmath0 ( 2000 ) & @xmath61 ( 2000 ) & @xmath79 & note + & ( h m s ) & ( @xmath65 ) & & + other young disk & & & & + sips0007 - 2458 & 0 7 7.800 & -24 58 3.80 & m7.5 & kinematics + age features + 2mass0020 - 2346 & 0 20 23.155 & -23 46 5.38 & m6.0 & kinematics + age features + sips0039 - 2256 & 0 39 23.250 & -22 56 44.90 & m7.5 & probable yd from kinematics + denis0041 - 5621 & 0 41 35.390 & -56 21 12.77 & m7.5 & kinematics + sips0054 - 4142 & 0 54 35.300 & -41 42 6.20 & m5.0 & probable yd from kinematics + sips0109 - 0343 & 1 9 51.040 & -3 43 26.30 & m9.0 & not conclusive from kinematics + lehpm1289 & 1 9 59.579 & -24 16 47.82 & m6.0 & not conclusive from kinematics + sips0115 - 2715 & 1 15 26.610 & -27 15 54.10 & m5.0 & not conclusive from kinematics + sips0126 - 1946 & 1 26 49.980 & -19 46 5.90 & m6.0 & not conclusive from kinematics + lehpm1563 & 1 27 31.956 & -31 40 3.18 & m8.5 & kinematics + sips0212 - 6049 & 2 12 33.580 & -60 49 18.40 & m6.5 & not conclusive from kinematics + 2mass0334 - 2130 & 3 34 10.657 & -21 30 34.35 & m4.5 & kinematics + age features + 2mass0429 - 3123 & 4 29 18.426 & -31 23 56.81 & m7.5 & not conclusive from kinematics + 2mass0502 - 3227 & 5 2 38.677 & -32 27 50.07 & m5.5 & not conclusive from kinematics + 2mass0528 - 5919 & 5 28 5.623 & -59 19 47.17 & m5.5 & not conclusive from kinematics + sips1039 - 4110 & 10 39 18.340 & -41 10 32.00 & m6.5@xmath80 & probable yd from kinematics + sips1124 - 2019 & 11 24 22.229 & -20 19 1.50 & m7.0@xmath80 & probable yd from kinematics + sips2049 - 1944 & 20 49 19.673 & -19 44 31.30 & m7.0 & kinematics + sips2128 - 3254 & 21 28 17.402 & -32 54 3.90 & m6.5 & kinematics + sips2321 - 6106 & 23 21 43.418 & -61 6 35.37 & m5.0 & kinematics + sips2343 - 2947 & 23 43 34.731 & -29 47 9.50 & m8.0 & kinematics + old disk & & & & + 2mass0204 - 3945 & 2 4 18.036 & -39 45 6.48 & m7.0 & not conclusive from kinematics + 2mass0445 - 5321 & 4 45 43.368 & -53 21 34.56 & m7.5@xmath80 & kinematics + denis1250 - 2121 & 12 50 52.654 & -21 21 13.67 & m7.5 & kinematics + sips1329 - 4147 & 13 29 0.872 & -41 47 11.90 & m9.5@xmath80 & kinematics + sips1341 - 3052 & 13 41 11.561 & -30 52 49.60 & l0@xmath80 & kinematics + 2mass1507 - 2000 & 15 7 27.799 & -20 0 43.18 & m7.5 & kinematics + sips1632 - 0631 & 16 32 58.799 & -6 31 45.30 & m8.5 & kinematics + sips1758 - 6811 & 17 58 59.663 & -68 11 10.50 & m5.0 & probable od from kinematics + sips1949 - 7136 & 19 49 45.527 & -71 36 50.89 & m7.0 & kinematics + 2mass2001 - 5949 & 20 1 24.639 & -59 49 0.09 & m6.0 & probable od from kinematics + 2mass2106 - 4044 & 21 6 20.896 & -40 44 51.91 & m6.0 & probable od from kinematics + sips2119 - 0740 & 21 19 17.571 & -7 40 52.50 & m7.0 & probable od from kinematics + lehpm4480 & 22 15 10.151 & -67 38 49.07 & m5.5 & probable od from kinematics + 2mass2222 - 4919 & 22 22 3.684 & -49 19 23.45 & m6.5 & kinematics +

by now , it has been a few years since it was realized that noncommutative gauge theories emerge from string theory through various decoupling limits @xcite . nonetheless , the understanding of the dynamics of noncommutative gauge theories remains in a relatively primitive state compared to ordinary gauge theories . the main difficulty is that noncommutative quantum field theories are highly nonlocal . in fact , one can think of the nonlocality as arising from elementary dipole degrees of freedom whose transverse length is proportional to their center of mass momentum @xcite . the novel behavior of these dipole quanta gives rise to uv - ir mixing : uv dipoles grow long in spatial extent and mediate instantaneous long distance interactions that dominate in the ir . not surprisingly , uv - ir mixing leads to difficulties in the naive application of conventional field theory techniques to noncommutative theories . in particular , the non - decoupling of uv and ir states leads to ambiguities in the distinction between short and long distance physics . consequently , there has been some confusion regarding such things as the renormalization of uv divergences @xcite , the treatment of ir divergences @xcite , and wilsonian integration @xcite . what is more , the noncommutative gauge invariance is not preserved separately in each diagram of the standard perturbative expansion ; rather , gauge invariance is achieved by an infinite resummation of diagrams @xcite . finally , there is a sort of naturalness problem with the conventional approach in that the intrinsic dipole structure of the elementary field quanta is not completely clear , although some suggestive results have been obtained @xcite . on the other hand , in the matrix formulation of noncommutative gauge theory , the noncommutative gauge invariance is manifest @xcite , as is the dipole character of the elementary quanta @xcite . in fact , as shown in @xcite , the matrix approach allows for a clear separation between the quantum effects of uv and ir dipoles . in particular , it was shown that one could make sense of wilsonian integration despite uv - ir mixing . for example , the interactions that result from integrating out the uv states were explicitly calculated at both the one and two loop order . the resulting interactions were found to dominate the long distance behavior , which shed some light on uv - ir mixing in noncommutative gauge theory , as well as the nonanalytic dependence of the quantum theory on the noncommutativity parameter , @xmath0 . for a different point of view on how the matrix formulation naturally leads to a bi - local representation , . however , while some progress has been made , there are still many unanswered questions . for one thing , the analysis of @xcite was limited to perturbation theory , which generically breaks down due to strong quantum corrections . although it was conjectured in @xcite that the perturbative analysis of the wilsonian quantum effective action is valid for supersymmetric theories whose uv behavior is softer , this remains to be proven . furthermore , it is not even clear if noncommutative quantum theories are renormalizable , although there have been arguments for the affirmative @xcite . in this work , we continue to develop noncommutative gauge theory in the matrix formulation . in section ii , after reviewing the basic ingredients of the dipole interpretation and the matrix formulation of noncommutative gauge theory , we determine the general form of the gauge invariant perturbative corrections to the quantum effective action that result from integrating out uv states . based on this structure , we then show that the leading long distance interactions cancel at each order in perturbation theory for supersymmetric theories . in section iii , we first discuss renormalizability in the context of nonsupersymmetric noncommutative theories and argue that , beyond one loop order , uv - ir mixing generally introduces divergences that spoil the consistency of the theory . we are led to conjecture that only supersymmetric noncommutative theories that do not get renormalized beyond one loop can be consistent quantum mechanically . we end with some discussion of our results and some concluding remarks . the appendices contain a review of the important results of @xcite and some other technical details . ultimately , we would like to further develop our intuition for how uv - ir mixing affects the quantum mechanical consistency of noncommutative gauge theories . to this end , it will prove convenient to separate the degrees of freedom into an uv region , defined by momenta much greater than any other scale in the problem , and an ir region defined by all of the rest of momenta . the strategy will be to first integrate out the uv modes in the wilsonian sense , and then to study the effective action in the ir regime . in this section , we will discuss the general structure of perturbative corrections to the effective action resulting from integrating out uv states . as discussed in @xcite , the structure of perturbative corrections is strongly constrained by noncommutative gauge invariance , which is a much larger symmetry group than ordinary gauge symmetries . in fact , gauge invariant quantities receive contributions from an infinite number of diagrams in terms of the conventional perturbative expansion . for this reason , it is most convenient to work in the matrix formulation of noncommutative gauge theory , which is manifestly gauge invariant . the technical machinery useful for performing perturbative calculations in the matrix approach was developed in @xcite , although the main technical results from the matrix formulation , which will be used extensively in what is to follow , are briefly reviewed in appendix [ ap : review ] . we will begin by recalling the interpretation of the matrix degrees of freedom in terms of dipoles , since this viewpoint is robust and will clearly generalize to all orders of perturbation theory . the dipole interpretation follows naturally from the representation of the matrix degrees of freedom in terms of the noncommutative gauge group . as discussed in @xcite , the degrees of freedom of noncommutative gauge theory are represented as infinite dimensional hermitian matrices @xmath1 where @xmath2 are time - independent hermitian matrices satisfying the algebra of the noncommuting @xmath3-plane @xmath4=i\theta^{ij}{\rlap{1 } \hskip 1.6pt \hbox{1}}.\ ] ] the essential point is that @xmath5 lives in the adjoint representation of the group noncommutative @xmath6 which , as indicated by the tensor product structure of eq . ( [ eq : xmatrix ] ) , can be thought of as the product of ordinary groups @xmath7 . the representations of the ordinary @xmath6 group will play the role of chan - paton factors joined to the ends of the dipoles , and their treatment will come naturally in our framework . the crucial observation is that the adjoint representation of @xmath8 , furnished by functions of @xmath2 , should be thought of in t hooft double line notation . in this language , the feynman diagrams that appear in the matrix formulation have a simple and intuitive interpretation in terms of dipole degrees of freedom : _ the double lines literally represent the spatial trajectory that is traced out by the endpoints of the dipole quanta as they propagate around the loops _ @xcite . in order to demonstrate and reinforce this dipole picture , we shall review the results of explicit calculations and their interpretations @xcite . for example , consider the one loop diagram . a simple matrix calculation gives an interaction term in the effective lagrangian proportional to @xmath9 where @xmath10 and @xmath11 are the numbers of bosonic and fermionic polarization states , respectively , and the rest of the notation is reviewed in appendix [ ap : review ] . note that although we have discussed only gauge field degrees of freedom explicitly , fermions and other matter fields will generally contribute quantities of the same form as the pure gauge field calculation , but they will differ in the constant of proportionality as in eq . ( [ eq : trace ] ) above . the @xmath12 factor that appears in the integrand of eq . ( [ eq : trace ] ) is familiar from field theory if we identify the momentum of the virtual quantum with @xmath13 . in fact , this is precisely the relation between the center of mass momentum and the end points of the dipoles that is expected @xcite ! the most obvious consequence of this relation is that uv dipoles are long in spatial extent . therefore , if we are to associate the double lines of the matrix diagrams with the ends of the dipoles , we should think of the one loop diagram above stretched out " into a long thin cylinder as depicted in fig . [ fig : cyli ] . [ fig : cylinder ] and @xmath14.,title="fig:",height=37 ] the quantity @xmath15 which appears in eq . ( [ eq : trace ] ) is a gauge invariant functional of the low momentum background field that is well localized in space around @xmath16 , given the fact that we are only integrating out high momentum states , as discussed in @xcite . evidently , we are to associate a trace , which produces the gauge invariant functional @xmath17 , to each boundary of the cylinder . in fact , this leads to the connection between eq . ( [ eq : trace ] ) and the field theory diagrams that appear in the conventional perturbative expansion . if we expand @xmath18 , then as discussed in @xcite , we identify the constant term as the contribution from field theory diagrams with no background field insertions on the corresponding boundary , while the term with nontrivial position dependence descends from diagrams that include background gauge field insertions on the corresponding boundary . this interpretation is supported by the fact that @xmath19 for field configurations gauge equivalent to @xmath20 . using this reasoning , we can conclude that the vacuum diagram , which involves no insertions on either boundary , is proportional to @xmath21 of course , this is just the expected result from field theory . the contribution from diagrams with planar background field insertions vanishes because @xmath22 ; however , the nonplanar diagrams yield a nontrivial interaction term proportional to @xmath23 apparently , eq . ( [ eq : delta ] ) describes the instantaneous interaction between distant background fluctuations at @xmath24 and @xmath14 , as illustrated in fig . [ fig : cyli ] . in fact , this interaction term demonstrates the novel feature of noncommutative quantum field theories : uv dipoles grow long in spatial extent and mediate instantaneous long distance interactions , which is how we can think about uv - ir mixing . however , it should be noted that , in the case of eq . ( [ eq : delta ] ) , the interactions grow strong at long distances because @xmath25 as discussed in @xcite , the strong interaction expressed in eq . ( [ eq : strong ] ) is due to the leading ir poles in external momenta that result from the uv region of loop integrals involving nonplanar background insertions . the appearance of nonanalytic behavior in external momenta has also been discussed from the field theory perspective in @xcite . in the particular case of eq . ( [ eq : strong ] ) , we will require that @xmath26 so that this term vanishes and perturbation theory remains valid . however , the appearance of strong ir corrections from the quantum effects of virtual uv states and their treatment in general will be a central theme of this work . furthermore , as discussed in @xcite , it is important to realize that the one loop matrix calculation ( [ eq : trace ] ) gives only the leading term in a derivative expansion involving the background field . the next to leading order one loop diagram involves two extra insertions of the background field strength . the contribution from this diagram was shown to be of the form @xmath27\int\frac{d\omega}{2\pi } \widetilde{g}({\omega},{\theta}^{-1}x_{12})\widetilde{g}({\omega},{\theta}^{-1}x_{21})\nonumber,\end{aligned}\ ] ] where @xmath28 and @xmath29 are constants depending on the dimensionality and the matter content of the theory , as previously discussed . notice that a nontrivial contribution from planar field theory diagrams is contained in the first term above . we will discuss eq . ( [ eq : ntlo ] ) in much more detail in the next section , but for now , we shall just remark on its basic features . evidently , the field strength operators that were inserted into the matrix diagram above appear as operators attached to the end of wilson lines . this structure seems to reinforce the correspondence between the boundaries of the matrix diagram and traces which lead to gauge invariant wilson lines . in general , higher order terms in the derivative expansion involve higher dimensional operator insertions into the loop diagram , and hence , the wilson lines . however , the double trace structure and the corresponding interpretation in terms of long distance two - body interactions remains the same . moreover , it is interesting that the appearance of the propagator @xmath30 in eq . ( [ eq : ntlo ] ) is again consistent with our field theory intuition . higher loop diagrams , on the other hand , will involve more traces , and will therefore lead to multi - body interactions . for example the leading two - loop contribution to the effective lagrangian comes from @xmath31 . note that we have not included the nonplanar matrix diagrams because their contribution to the wilsonian integration is exponentially suppressed due to moyal phase factors @xcite . as indicated by the double line diagrams above , the leading two - loop diagrams give triple trace contributions . the first diagram gives an interaction term in the effective lagrangian proportional to @xcite @xmath32 + and , as calculated in appendix [ ap : twoloop ] , the second gives a term proportional to @xmath33 indeed , as expected based on our intuition , eqs . ( [ eq : threebody ] ) and ( [ eq : cubicthreebody ] ) describe long distance interactions that arise from high momentum dipoles growing large in spatial extent and `` stretching out '' the matrix diagrams , as depicted in fig . [ fig : twolooppicture ] . furthermore , these expressions bear a close resemblance to what is expected from ordinary field theory , and in fact , the general structure of perturbative corrections is starting to emerge . to each boundary of the double line diagram we should associate a point in space and a trace which yields a gauge invariant wilson line . the position dependence of the interaction strength between the wilson lines follows from an integral over frequencies with the integrand being given by a particular function of both the dipole frequencies and momenta as defined by the separation between the space points associated to the boundaries of the double line diagram . in fact , the particular function of dipole frequencies and momenta corresponds precisely to the structure of momentum space field theory propagators that appear in the analogous process in ordinary field theory . the subleading terms in the derivative expansion will have a similar trace structure but wilson lines modified to include higher derivative operator insertions as in eq . ( [ eq : ntlo ] ) . naturally , the subleading terms will include a different function of dipole frequencies and momenta that reflects the extra propagators that are required by the operator insertions . note that for dipole degrees of freedom , powers of momentum are equivalent to powers of separation , as discussed in @xcite and in appendix [ ap : review ] . therefore , the inclusion of more powers of momentum in the denominator leads to faster falloff with distance , which is expected from subleading terms in a derivative expansion . however , quantum corrections coming from higher loop orders are generally strong , as we will discuss in the next section . of more importance to our immediate goal , is the determination of the form of the leading long distance interactions at each order in perturbation theory . it is clear from the general structure discussed above that at @xmath34-loop order the leading long distance interactions will involve @xmath35 wilson lines with no operator insertions , such as eq . ( [ eq : trace ] ) in the case @xmath36 and both eq . ( [ eq : threebody ] ) and eq . ( [ eq : cubicthreebody ] ) in the case @xmath37 . the crucial observation is that since the vacuum diagrams are contained in interactions of this form , as shown explicitly by eq . ( [ eq : vacuum ] ) in the one - loop case , these terms must vanish entirely if the vacuum diagrams vanish . thus , _ in supersymmetric noncommutative theories , the leading long distance interactions must cancel at each order in perturbation theory_. this statement is a generalization to all loop orders of the cancellation that we have discussed in the case of the leading one - loop interactions given by eq . ( [ eq : trace ] ) . it reflects the fact that supersymmetric theories are softer in the uv , and hence , the ir behavior that is generated by uv - ir mixing is not as strong . in fact , the cancellation of the leading ir interactions occurs even in minimally supersymmetric theories . the relation between the amount of supersymmetry and the cancellation of other terms has been discussed in @xcite from the field theory perspective , and we will touch on this important point as well in the next section . in summary , we have determined the gauge invariant form of perturbative corrections to the wilsonian quantum effective action for noncommutative gauge theories . it followed immediately from the structure of perturbation theory that the leading long distance interactions cancel order by order in the supersymmetric theories . since nonsupersymmetric theories do not enjoy these cancellations , they are strongly coupled at long distances . in fact , what we are beginning to discover is that noncommutative quantum theories are highly sensitive to the uv , which we will further explore in the next section . in the previous section , we determined the gauge invariant form of the perturbative quantum corrections to the effective action that result from integrating out uv states . in this section we will take a closer look at these perturbative corrections and their divergence properties . in addition to the expected types of uv divergences , we will find new divergences that have dual uv and ir interpretations and for which there are no analogues in ordinary quantum field theories . in fact , we will argue that these new uv - ir divergences , if they are allowed , will spoil the renormalizability of the theory . in the end , we will find that only supersymmetric theories with the softest uv behavior can be consistent quantum mechanically . because it will set the ground work for the discussion of the more interesting supersymmetric case , we will start this section by discussing divergences in the context of nonsupersymmetric noncommutative gauge theories . we will begin with the familiar procedure of gauge coupling renormalization . in fact , in this case , it will turn out that renormalizability is threatened already at the two loop level . to see how this happens , we first recall from @xcite the form of the one - loop quantum correction to the gauge coupling . the only candidate for this type of term is eq . ( [ eq : ntlo ] ) because it comes from a one - loop graph with two insertions of the field strength operator . to isolate the uv divergence in the planar diagrams , we must use the splitting scheme @xmath18 and focus on the first term , as previously discussed . the integral then factorizes into @xmath38}^{2}\right ) n\int\frac{d{\omega}d^{2p}p}{2\pi(2\pi)^{2p}}{\widetilde{g}(\omega , p)}^{2}.\end{aligned}\ ] ] this quantity is easily recognized as contributing to the renormalization of the operator @xmath39^{2}$ ] , as expected . therefore , the one - loop quantum correction to the gauge coupling is proportional to the dimensionless quantity @xmath40 where we have restored the gauge coupling @xmath41 explicitly . if we are to consistently absorb the one - loop correction ( [ eq : gaugerenorm ] ) into a new quantum corrected gauge coupling , we must find the same quantity at higher loops correcting each factor of @xmath41 that appears . otherwise , renormalizability will be threatened , because we will be forced to introduce new couplings into the effective lagrangian in order to cancel all of the resulting divergences . however , in our conventions , the dependence on the gauge coupling is particularly simple powers of @xmath41 count the number of loops . thus , powers of the gauge coupling only appear multiplying the loop corrections ; the wilson lines , which encode the background gauge field insertions into the loops , are independent of the coupling . therefore , to test renormalizability of the gauge coupling at the two loop order , we should look for a term of the form @xmath42 because it corresponds to the one - loop correction to @xmath41 given by eq . ( [ eq : gaugerenorm ] ) multiplying the leading one - loop interaction ( [ eq : trace ] ) . in order to find this sort of term , we consider the leading two - loop interactions . as discussed in appendix [ ap : twoloop ] , these contributions can be combined into the form of eq . ( [ eq : threebody ] ) . it is now straight forward to isolate the divergent quantum corrections to the two - body interactions . the constant term in either @xmath43 or @xmath44 gives @xmath45 apparently , eq . ( [ eq : propcorrection ] ) involves the leading one - loop uv divergent contribution to the mass term of the field theory propagator @xmath30 appearing in eq . ( [ eq : trace ] ) . in fact , this type of correction is familiar from ordinary quantum field theories . however , it will turn out that in the supersymmetric theories of ultimate interest to us , corrections of this form will not appear . in any case , eq . ( [ eq : propcorrection ] ) is not the term we are looking for , so we move on . the only other candidate for the appearance of eq . ( [ eq : required ] ) in the leading order two - loop diagrams is the interaction that comes from the constant term of @xmath46 @xmath47 needless to say , eq . ( [ eq : twoloopcorrection ] ) is not the term we need either ; thus , the renormalizability of the gauge coupling is threatened already at the two - loop order . however , a closer look at eq . ( [ eq : twoloopcorrection ] ) reveals an entirely new problem facing the renormalizability of noncommutative theories that we will now discuss . evidently , eq . ( [ eq : twoloopcorrection ] ) represents a two - loop quantum correction to the leading one - loop two - body interaction ( [ eq : trace ] ) . the growth in @xmath48 of the interaction strength of eq . ( [ eq : twoloopcorrection ] ) is given by @xmath49 clearly , eq . ( [ eq : intcorrection ] ) is uv divergent for @xmath50 , so we must introduce an uv cutoff @xmath51 . the integral can now be calculated in a straight forward fashion and the leading dependence on separation goes as @xmath52 . however , since there is no two - body interaction that grows as @xmath53 at one loop , this logarithmic divergence can not be absorbed into the gauge coupling . therefore , a new coupling for a two - body interaction of the form @xmath54 must be introduced into the effective lagrangian so that the divergence in eq . ( [ eq : twoloopcorrection ] ) can get cancelled . although one can arrange for the cancellation of divergent corrections to any interaction strength by continuing to introduce new interactions with divergent couplings , as above , it is not hard to see that this procedure is a serious threat to the renormalizability of the theory . to see this , we need only note that , at higher orders , there will be quantum corrections to the interaction strengths that come in the form of powers of the dimensionless quantity @xmath55 therefore , to cancel all of the divergences that appear in the various powers of ( [ eq : running ] ) , we would have to introduce an infinite number of new interactions the reason is that there are an infinite number of distinct functions of @xmath56 that can appear , each requiring a different interaction term with its own coupling in order to cancel the required divergences in the effective lagrangian . thus , it seems that divergent corrections to the interaction strength between wilson lines , a new feature of noncommutative theories having no analogue in ordinary quantum field theories , will generally spoil the renormalizability of the theory . a crucial assumption in the argument above is that if divergent quantum corrections to interaction strengths occur at the two - loop order , they will appear at all higher orders as well . in fact , this is easy to see graphically by recalling the relation between matrix diagrams and field theory diagrams . using the correspondence that we discussed earlier in section ii , we find that eq . ( [ eq : twoloopcorrection ] ) descends from the leading order terms in the external momenta expansion of field theory diagrams of the form and . evidently , the dangerous divergences are not due to planar subdiagrams , which have the standard uv interpretation , but are due to purely nonplanar field theory diagrams . we will develop more understanding for what this means in the moment , but for now let us note that higher order graphs of this type exist at each loop order in perturbation theory . for example , fig . [ fig : threeloop ] depicts some three - loop field theory diagrams that give rise to divergent quantum corrections to three - body and two - body interactions . thus , dangerous corrections to interaction strengths that first appear at some given order in perturbation theory will receive quantum corrections from all higher loop orders , generating corrections in the form of powers of ( [ eq : running ] ) . however , by the arguments above , the divergences appearing in powers of eq . ( [ eq : running ] ) will inevitably spoil renormalizability . therefore , any renormalizable noncommutative theory can not allow the leading order terms in the derivative expansion , which as we have seen , contain these dangerous corrections . we are forced to conclude that _ only supersymmetric noncommutative theories can be renormalizable _ since , in this case , the leading order interactions always vanish . before discussing the supersymmetric theories , let us take some time to understand the nature of the dangerous divergences by recasting them into the form of momentum integrals . for example , consider the transformation of eq . ( [ eq : twoloopcorrection ] ) to momentum space @xmath57 we can first perform the integrals over @xmath58 and @xmath59 by introducing a schwinger parameter and an uv cutoff @xmath51 , as in @xcite . then the integral over @xmath60 and @xmath61 is regular , and we are left with a quantity proportional to @xmath62 now the @xmath63 limit results in an ir divergence instead of an uv divergence , and @xmath64 plays the role of an ir cutoff ! the transform back into position space just gives the same correction to the interaction strength that we have already determined directly from eq . ( [ eq : intcorrection ] ) . apparently , these types of divergences have a dual uv - ir interpretation . the origin of this duality can be traced to the moyal phase factor @xmath65 in eq . ( [ eq : moyal ] ) , which represents the nonplanarity of the background insertions and gives rise to uv - ir mixing . in particular , the uv loop integrals over virtual states , labelled by momenta @xmath59 and @xmath61 , produce nonanalytic dependence on the background momentum @xmath66 that becomes important in the ir . thus , on one hand , the divergence comes from integrating over uv states , but on the other hand , the divergence cast be recast into the form of an ir singular fourier transform . actually , as discussed in @xcite , this type of uv - ir mixing has a simple interpretation in terms of dipole degrees of freedom : the separation between the endpoints of the dipoles is the dual variable to the external momenta @xmath66 ; therefore , the ir singular fourier transform gives rise to corrections to interaction strengths which grow strong with large separation ; but , since powers of separation are equivalent to powers of dipole momentum , these strong corrections can also be thought of as arising from uv divergences in the theory . thus , the dipole intuition that emerges from the matrix formulation seems to shed new light on the proposal of @xcite concerning the dual interpretation of divergences due to nonplanar diagrams . in fact , at this point , let us take an aside to remark on the virtues of the matrix formulation of noncommutative gauge theory . first of all , the matrix degrees of freedom naturally encode the elementary dipole structure of the field quanta , and hence , the essence of uv - ir mixing . this is a very convenient aspect of the matrix approach , which leads to great conceptual clarity in the physics of noncommutative quantum theories , as we have seen . perhaps , the most important advantage of the matrix description , however , is the manifest nature of the noncommutative gauge invariance . this crucial property allows us to distinguish between various long distance interactions , which in turn , forms the basis for our interpretation of the uv - ir divergences that appear in the theory . back to the issue of divergences in supersymmetric noncommutative theories , we now consider interactions that are of next to leading order in the derivative expansion . as we have discussed , these are the lowest order terms that do not generally vanish in the case of supersymmetric theories . however , we will find that uv - ir divergences do generally occur in the next to leading order interactions , and therefore , supersymmetric theories generally suffer from uv - ir divergences , as well . for example , as derived in appendix [ ap : ntlotl ] , the next to leading order two - loop interactions include dangerous terms such as eq . ( [ eq : dangerous ] ) , which contain a quantum correction to the interaction strength of eq . ( [ eq : ntlo ] ) that goes as @xmath67 following an analysis parallel to the one above for eq . ( [ eq : intcorrection ] ) , it is straight forward to show that uv - ir divergences appear in eq . ( [ eq : div ] ) for @xmath68 ( note that by @xmath59 equal to a half - integer we really mean that , in addition to @xmath69 noncommutative two - planes , there is an additional commutative spatial direction . the proper way to include this case in our analysis is to let @xmath59 remain the integer valued number of noncommutative two - planes and replace @xmath70 with @xmath71 to represent the two commuting spacetime directions . all of the conclusions that we have drawn up to this point are true also in the case when the space dimensionality is odd ) . furthermore , the graphical argument that we employed earlier in the case of the leading order interactions , can be easily generalized to include the next to leading order terms in the derivative expansion . we need only consider adding higher derivative operator insertions in the matrix framework or , in the field theory picture as in fig . [ fig : threeloop ] , taking the next to leading order terms in the expansion in external momenta . in any case , we are forced to conclude that if dangerous uv - ir terms appear at two loops , then there will be more dangerous quantum corrections that emerge from all higher loop orders . therefore , if eq . ( [ eq : dangerous ] ) appears in @xmath72 dimensions or higher , uv - ir divergences will proliferate and renormalizability will fail . in order to determine what corrections are allowed by supersymmetric theories , let us take some time to discuss the proportionality constants that we have so far ignored . this will be fruitful because we will find that the nonrenormalization theorems enjoyed by ordinary supersymmetric theories have a natural extension to noncommutative theories . we can then use these nonrenormalization theorems to show that @xmath72 dimensional noncommutative theories with a sufficiently high degree of supersymmetry do not allow uv - ir divergences , and therefore , renormalizability does not face any obvious obstruction in these special theories . we begin by considering the content of the matrix diagrams from a field theory point of view . first of all , the double line matrix diagrams themselves only contain information pertaining to the contraction of gauge indices , which encodes the planarity and nonplanarity of both the external operator insertions and the internal propagators . thus , the double line diagrams contain purely topological information , not specific to any particular gauge theory ordinary or noncommutative . in terms of the framework that we have discussed up to this point , this topological information only fixes the ratio of coefficients between terms with a similar propagator structure but different wilson line structure . for example , @xmath73 is fixed by topological considerations , as are both @xmath74 and @xmath75 . furthermore , the coefficients coming from nonplanar matrix diagrams , which we have not discussed here , are fixed relative to the coefficients of the corresponding planar diagrams ( note that factors of @xmath76 are contained in the traces of the wilson lines , not the coefficients that we are referring to ) . however , topological quantities , such as the ratio between coefficients , are independent of any characteristic specific to the theory in question such as the field content , the allowed couplings between fields , and whether the theory is noncommutative or not . although the ratios of certain coefficients are topological quantities , independent of the theory in question , the coefficients coming from planar and nonplanar matrix diagrams involving a similar propagator structure all share a common normalization factor that depends on the number and type of modes that propagate around the loops . thus , the overall normalization factor does depend on characteristics specific to a particular theory , such as the field content , the allowed couplings between fields , and the coupling constants . however , the noncommutativity of the theory enters only through the moyal phase factors inside the momentum integrals of nonplanar field theory diagrams ; noncommutativity does not affect the coefficient outside the integral . it follows that a given coefficient in a particular noncommutative gauge theory is identical to the same coefficient in the corresponding ordinary gauge theory . after all , with the exception of moyal phase factors in the the noncommutative theories , the contraction of fields is the same in both cases . therefore , _ the nonrenormalization theorems of ordinary supersymmetric theories generalize immediately to noncommutative supersymmetric theories_. as an application of this useful result , let us return to the problem of dangerous uv - ir divergent terms in the next to leading order two - loop contribution to the quantum effective action , which are given by eq . ( [ eq : dangerous ] ) . according to our reasoning above , @xmath77 , @xmath78 , and @xmath79 are all proportional to the two - loop contribution to the beta function . therefore , we can guarantee their vanishing by imposing @xmath80 supersymmetry in @xmath72 dimensions . in this case , the dangerous two - loop interactions cancel , as do all higher loop corrections . moreover , the two - loop terms given by eq . ( [ eq : leading ] ) cancel because no mass terms are allowed to appear perturbatively , given any amount of supersymmetry . thus , renormalizability for @xmath72 dimensional noncommutative gauge theories requires at least @xmath80 supersymmetry , in which case the next to leading order interactions are restricted to the one - loop result given by eq . ( [ eq : ntlo ] ) , along with the two - loop terms containing the one - loop correction to the gauge coupling ( [ eq : beta ] ) . in the case of @xmath81 supersymmetry , none of the next to leading order interactions are allowed what so ever . moreover , it is also worth mentioning that @xmath82 supersymmetry also implies that most , if not all , of the next to leading order nonplanar matrix diagrams cancel as well . the reason is that , as we have discussed , the coefficients of these contributions differ from their planar counterparts only by a combinatoric factor . thus , the only next to leading order nonplanar remnant is which is proportional to the one - loop beta function . of course , all of the leading order nonplanar diagrams cancel , given any amount of supersymmetry . the advantage of eliminating the leading and next to leading order nonplanar matrix diagrams , is that they are divergent when the wilsonian cutoff is removed . in fact , these divergences resulting from the ir region of integration in nonplanar loop integrals , have posed a major challenge to our understanding of the quantum effects of ir states in noncommutative theories . therefore , it is fortunate that renormalizability of uv quantum fluctuations turns out to also constrain most of these ir divergences to vanish as well . although a thorough treatment of the quantum effects of ir states is beyond the scope of this work , it seems that our results here are already a help in this direction . while we have shown that @xmath82 supersymmetry is sufficient to prevent uv - ir divergences from appearing in diagrams containing two insertions of the field strength , we should still worry about dangerous terms appearing from diagrams involving other operator insertions . for example , all orders of wave function renormalization for matter hypermultiplets is generally allowed by @xmath80 theories , although @xmath81 theories do not allow any renormalization at all . therefore , @xmath80 gauge theories generally involve new uv divergences at each loop order , which will inevitably lead to the proliferation of dangerous uv - ir divergences via uv - ir mixing . after all , the divergent integral @xmath83 that appears embedded in both eqs . ( [ eq : intcorrection ] ) and ( [ eq : div ] ) will certainly appear at two loops and beyond in diagrams containing the insertion of any marginal operator , unless new divergences beyond one loop are not allowed . thus , divergent quantum corrections beyond one loop are inherently dangerous in noncommutative theories . we are led to conjecture that _ the only renormalizable , and hence , quantum mechanically consistent noncommutative gauge theories are supersymmetric and do not require renormalization beyond the one loop order_. it is striking how strong this statement is . apparently , noncommutative quantum theories are extremely sensitive to the uv , and only the softest uv behavior can be tolerated . in particular , the allowed @xmath84 dimensional theories include any supersymmetric gauge theory , and the allowed @xmath72 dimensional theories include both @xmath80 and @xmath81 noncommutative super yang - mills ( ncsym ) theories plus some @xmath80 gauge theories that include matter hypermultiplets in special representations . it is interesting that @xmath85 supersymmetry , or even @xmath80 in most cases , does not appear to be enough to render @xmath72 dimensional noncommutative gauge theories renormalizable . moreover , it is tempting to extend our conjecture to noncommutative theories other than gauge theories . after all , the intuition that we have gained appears to be quite generic and most likely applies to any noncommutative theory . for example , it seems unlikely that noncommutative scalar theory is consistent quantum mechanically , when nonsupersymmetric gauge theories are not . it even seems reasonable that renormalizability could fail in @xmath85 and @xmath80 supersymmetric noncommutative theories that do not include gauge degrees of freedom , in light of the fact that any extra hypermultiplets of matter that are added to @xmath80 ncsym theory must form special representations which do not allow wave function renormalization , so as not to spoil the theory . nonetheless , arguments have been put forward for the renormalizability of many noncommutative theories that are excluded by our conjecture @xcite . although , these works have involved different approaches which have encoded uv - ir mixing in one way or another , none have employed a manifestly dipole construction such as the matrix formulation . since the dipole behavior of the elementary quanta is the fundamental origin of uv - ir mixing in noncommutative theories , it seems that the physical interpretation that naturally emerges from the matrix approach is the most reliable . we believe that this is a tremendous advantage when discussing renormalizability , because traditionally , the proper treatment of divergences has resulted from a sound physical interpretation for their meaning . of course , the physical content of noncommutative theories is independent of the language used to discuss them ; we are simply suggesting that the physics is more clear in the matrix representation . in any case , the difference between our work and @xcite is the interpretation and treatment of the dual uv - ir divergences that occur in noncommutative quantum theories . before closing this section , let us make a few comments concerning the scope of our results from the perspective of string theory . our analysis has been limited to noncommutative gauge theories , and given these dipole degrees of freedom , we have shown that both a sufficiently high degree of supersymmetry and low spacetime dimensionality is necessary to ensure the quantum mechanical consistency of the theory . however , this result does not imply that other degrees of freedom can not be added to the theory to give a consistent uv completion . for example , @xcite shows that there is an uv completion of some higher dimensional ncsym theories in the form of noncommutative open string theories , in which the closed string sector has decoupled but there are still stringy modes from the open string sector that remain . yet another distinct possibility is that , in the case of theories with a lesser degree of supersymmetry , there could be some closed string modes that survive the decoupling limit and render the theory consistent @xcite . the point is that any noncommutative theory that emerges from a decoupling limit of string theory will be consistent ; our results only imply that the decoupled theory can conceivably be a noncommutative gauge theory in only very special cases . string theory aside , however , the renormalizability of noncommutative quantum field theories is interesting in its own right . in this work , we have continued to develop noncommutative gauge theory in the matrix formulation . after reviewing the dipole interpretation of the matrix approach , we determined the general gauge invariant form of perturbative corrections to the quantum effective action that result from integrating out uv states . we then studied the divergence structure of these quantum corrections , which revealed new divergences , having dual uv and ir interpretations , that are unique to noncommutative theories . these uv - ir divergences , which appear at two loops and beyond , were found to represent dangerous quantum corrections that generally spoil the renormalizability of the theory . thus , we were led to conjecture that only supersymmetric noncommutative theories that do not receive quantum corrections beyond one - loop are renormalizable , and hence , quantum mechanically consistent . furthermore , beyond specific results concerning renormalizability , it is interesting that the consequences of uv - ir mixing in noncommutative quantum theories so profoundly affect the long distance behavior : in @xmath72 dimensions or higher , uv quantum corrections generally introduce new long distance interactions that grow stronger and stronger , even if the gauge coupling is small . in the end , the lesson that we learn is that noncommutative quantum theories exhibit an extreme sensitivity to the uv , such that only the softest uv behavior can be tolerated . however , there still remain a number of open questions concerning noncommutative quantum field theories . perhaps , the most glaring omission in our understanding is the nature of quantum effects due to ir states in the theory . this important problem represents an interesting avenue to pursue in future studies . in particular , it would be very satisfying if both the uv and ir dynamics could be encoded together in a unified framework . surely , this would shed some more light on the fascinating structure of noncommutative theories . we thank li jiang , sonia paban , and willy fischler for many stimulating discussions . this work was supported by nsf grant phy-0071512 . for an introduction to the formulation of field theories on noncommutative spaces , see @xcite , and for a complete review of the matrix formulation of noncommutative gauge theory , see @xcite . here we will review only the main results that we will need for the calculations that appear in this work . furthermore , we will focus only on the treatment of the gauge degrees of freedom , the generalization to matter fields , including fermions , being obvious . the basic idea of the matrix formulation is that the degrees of freedom are contained in infinite dimensional time dependent hermitian matrices @xmath86 and @xmath5 . in order to separate the quantum effects from uv and ir states , it will prove convenient to work in the background field gauge . in this approach , we expand the fields @xmath87 and @xmath88 where the background fields , @xmath89 and @xmath90 , are interpreted as the low momentum degrees of freedom , while the fluctuating fields , @xmath91 and @xmath92 , contain the high momentum degrees of freedom . to make the connection with noncommutative gauge theory , we think of @xmath93 and @xmath94 where @xmath2 are infinite dimensional time - independent hermitian matrices satisfying the algebra of the noncommuting @xmath3 plane @xmath4=i\theta^{ij}{\rlap{1 } \hskip 1.6pt \hbox{1}}.\ ] ] after a suitable choice of gauge fixing , which includes setting @xmath95 , the lagrangian takes the form @xmath96 where @xmath97[b^{i},b^{j}]\right ) ; \nonumber\\ l_2 & = & { \rm tr}\left(\frac{1}{2}\dot{y}^{j2}+\frac{1}{2}[b^{i},y^{j}]^{2}-\frac{1}{2}{\dot{a}}^{2}-\frac{1}{2}[b^{i},a]^{2}+\dot{\bar{c}}\dot{c } \right . \nonumber\\ & & \left.{}+[b^{i},\bar{c}][b^{i},c]+[b^{i},b^{j}][y^{i},y^{j}]-2i\dot{b}^{i}[a , y^{i } ] \right ) ; \nonumber\\ l_3 & = & { \rm tr}\left([b^{i},a][a , y^{i}]+[b^{i},y^{j}][y^{i},y^{j}]+[b^{i},\bar{c}][y^{i},c ] \right . \nonumber\\ & & \left.{}-i{\dot{y}}^{i}[a , y^{i}]-i\dot{\bar{c}}[a , c ] \right ) ; \nonumber\\ l_4 & = & { \rm tr}\left(\frac{1}{4}[y^{i},y^{j}][y^{i},y^{j}]-\frac{1}{2}[a , y^{i}]^{2}\right).\end{aligned}\ ] ] + notice that we have neglected the linear term in the fluctuating fields , which generically , is not consistent because we expect that uv quantum corrections will contribute to the dynamics of the background field through tadpoles and other effects resulting from the linear interactions . in the language of perturbation theory , this amounts to corrections that are both higher order in the gauge coupling and higher order in derivatives of the background field . therefore , in order to consistently ignore the linear term , we must require both that the coupling is sufficiently weak so that the loop corrections are suppressed and that the states we integrate out are of sufficiently high momenta relative to the scale set by the background so that the higher derivatives are suppressed , as well . we will find that both of these conditions are met in our approach , so for simplicity sake , we have neglected the linear interactions . moving along , from @xmath98 , we see that up to commutators and time derivatives of the background fields , all of the fluctuating fields , @xmath99 , have similar quadratic terms of the form @xmath100^{2}\right)\\ & & = \frac{1}{2}\phi^{t}\left(-{\rlap{1 } \hskip 1.6pt \hbox{1}}\otimes { \rlap{1 } \hskip 1.6pt \hbox{1}}\frac{d^{2}}{dt^{2}}+(b^{i}\otimes { \rlap{1 } \hskip 1.6pt \hbox{1}}-{\rlap{1 } \hskip 1.6pt \hbox{1}}\otimes b^{i})^{2}\right).\nonumber\end{aligned}\ ] ] for both consistency with the neglect of the linear terms in the fluctuating fields , as discussed above , and convenience in the definition of the propagator for the fluctuating fields , we will treat the commutator and time derivative terms perturbatively while absorbing the remaining background field dependence into the definition of the propagator . as discussed in @xcite , this choice corresponds to a derivative expansion of the background field , which makes sense because we will only integrate out high momentum states . to lowest order in the derivative expansion , the propagator for the high momentum modes is @xmath101 where @xmath102 is the momentum space field theory propagator for a massless bosonic state and @xmath103 is the inverse spatial metric , which we will discuss at the end of this section . were we dealing with fermions instead of bosons , we would simply replace @xmath104 above with the appropriate propagator for fermi fields . as discussed in @xcite , eq . ( [ eq : prop ] ) manifestly describes dipole degrees of freedom whose transverse length is proportional to their center of mass momentum . in particular , the identification is that @xmath16 and @xmath105 are the endpoints of the dipole in space , and the center of mass momentum is given by @xmath106 . it is also important to understand that there is a lower cutoff on the integrals over @xmath107 and @xmath16 , which reflects the fact that the wilsonian integration is only performed over states of high energy and momentum compared to the background . in practicality , when we perform calculations , we frequently encounter traces of the form @xcite @xmath108 + where @xmath109 is some operator in the adjoint representation of the gauge group . we immediately recognize this object as an open wilson line with @xmath110 attached at the end . in fact , this structure was essentially guaranteed by the noncommutative gauge invariance @xcite . furthermore , we will also frequently encounter the fourier transform of the open wilson lines above , which we shall denote @xmath111 + finally , we should mention that we are using a convention of units in which @xmath112 . this convention could potentially lead to some unfamiliar notions of dimension . what is more , the situation is further complicated by the fact that the inverse spatial metric is @xmath113 . nonetheless , the quantities that will appear in our calculations will all have simple physical interpretations . the rule to follow is that the magnitude squared of distance @xmath114 has dimensions of _ energy _ squared . the reason is that @xmath115 where we have used the relation between the center of mass momentum and the position of the end points of the dipole quanta @xmath116 . likewise , by the same argument , the magnitude squared of momenta has dimensions of length squared . the leading two - loop contribution to the effective action comes from treating the interaction terms in @xmath117 to first order in perturbation theory and the interaction terms in @xmath118 to second order . in @xcite , the contribution from both the @xmath117 terms as well as the time derivative terms in @xmath118 was computed . in this section , we will comput the contribution to the effective action from the remaining terms in @xmath118 , which can be represented schematically as @xmath119[\phi,\phi](t_1)\right)\nonumber\\ & & \qquad\qquad\times\int dt_2 \rm{tr}\left([b,\phi][\phi,\phi](t_2)\right)\rangle,\end{aligned}\ ] ] where @xmath120 is the low momentum background field and @xmath121 represents the high momentum fluctuating fields to be integrated out . the angled brackets denote the vacuum expectation value , which in our background field language , is the product of the expectation value of the low momentum background fields " and the expectation value of the high momentum fluctuating fields " . since we are working in the wilsonian scheme , we evaluate only the latter quantity and think of the residual expression involving an expectation value of background fields as originating from an effective lagrangian of background fields only . furthermore , in the evaluation of eq . ( [ eq : expect ] ) , we neglect the tadpole contractions , although generally , there will be tadpole contributions from the linear interaction term in the lagrangian as well as both the cubic and quartic interaction terms . however , the linear terms are suppressed by higher derivatives of the background as we have already discussed in appendix [ ap : review ] , and in the supersymmetric theories that are ultimately of interest to us , the tadpoles from interaction terms cancel as well . after a straight forward but somewhat tedious calculation , up to subleading time derivative and commutator terms we have @xmath122.\end{aligned}\ ] ] since we only integrate out states with high energy and momentum , time derivatives and commutators involving the background field are suppressed . therefore , to lowest order , we obtain @xmath123 after performing the integral over @xmath124 and @xmath125 and integrating by parts in the @xmath66 integrals . note that we have dropped the single trace term , since as discussed in @xcite , it corresponds to a nonplanar matrix diagram , which is suppressed in the domain of wilsonian integration . finally , upon fourier transforming back to position space , we are left with @xmath126 we can now combine the above expression with the result that was calculated in @xcite to obtain the leading two - loop contribution from the second order treatment of all of the cubic interactions in @xmath118 . we get a result proportional to @xmath127 note that we have used the @xmath128 symmetry that is present in the vacuum diagrams in order to determine that the integrand of the combined expression can only depend on the invariant product @xmath129 . with this factor present in the integrand , as in the commutative theories , the two - loop cubic interactions given by eq . ( [ eq : totaltwoloop ] ) can be reduced to the form of eq . ( [ eq : threebody ] ) . as in the next to leading order one - loop calculation , to get the precise result we must retain higher order commutators and time derivatives that were dropped in the derivation of the matrix propagator ( [ eq : prop ] ) as well as the field strength terms in @xmath98 that were also excluded from the the propagator @xcite . however , after developing some intuition in section ii for the structure of terms that can appear , a lengthly calculation is not necessary . we simply need to keep track of all the distinct possibilities in which the field strength insertions can appear in the diagrams . for example , in the case of the quartic diagram , both insertions can go into one loop or each loop can get a single insertion of course , each insertion requires an extra field theory propagator to appear in the corresponding loop , and for each loop , there are two boundaries in which the insertion can go . therefore , the first diagram corresponding to both insertions going into the same loop contributes two terms @xmath130,\nonumber\end{aligned}\ ] ] while the second diagram corresponding to one insertion going into each loop contributes three terms @xmath131.\nonumber\end{aligned}\ ] ] there are many more possibilities in the case of the cubic graph . for simplicity , we will enumerate them in four propagator form , which is obtained after the cancellation of one propagator by the momenta in the numerator of the integrand ( see eq . ( [ eq : totaltwoloop ] ) ) . it is not difficult to see that there are three combinations of field theory propagators that emerge : the first is similar to eq . ( [ eq : leading ] ) , the second is similar to eq . ( [ eq : beta ] ) , and the third is given by @xmath132.\nonumber\end{aligned}\ ] ] let us now discuss the meaning of these expressions in the context of supersymmetric theories . first of all , consider the terms of the form eq . ( [ eq : leading ] ) . the planar subdiagram corresponding to the constant term of @xmath44 , contributes one - loop divergent mass corrections to the propagators appearing in eq . ( [ eq : ntlo ] ) . as discussed in section iii , in supersymmetric theories , contributions of this type will always cancel . another obvious group is the combination given by eq . ( [ eq : beta ] ) . by isolating the constant term of @xmath44 , we recover the uv divergent one - loop correction to the gauge coupling that multiplies eq . ( [ eq : ntlo ] ) . of course , these types of corrections are required by renormalizability and pose no threat to the consistency of the theory . moreover , the constant term of @xmath46 gives rise to a benign quantum correction to the @xmath133 interaction in eq . ( [ eq : ntlo ] ) . finally , we arrive at the dangerous group of terms given by eq . ( [ eq : dangerous ] ) . although , the constant term of @xmath44 gives a harmless quantum correction to the interaction strength of eq . ( [ eq : ntlo ] ) , the constant term of @xmath46 leads to the same type of dangerous uv - ir divergent correction that was discussed at length in section iii . therefore , renormalizability demands that these contributions vanish . n. seiberg and e. witten , j. high energy phys . * 09 * , 032 ( 1999 ) . m. m. sheikh - jabbari , phys . b * 455 * , 129 ( 1999 ) ; d. bigatti and l. susskind , phys . d * 62 * , 066004 ( 2000 ) ; z. yin , phys . b * 466 * , 234 ( 1999 ) ; h. liu and j. michelson , phys . d * 62 * , 066003 ( 2000 ) . s. minwalla , m. van raamsdonk and n. seiberg , j. high energy phys . * 02 * , 020 ( 2000 ) . martin and d. sanchez - ruiz , phys . lett . * 83 * , 476 ( 1999 ) ; t. krajewski and r. wulkenhaar , int . j. mod a * 15 * , 1011 ( 2000 ) . martin and f.r . ruiz , nucl . b597 * , 197 ( 2001 ) . i. chepelev and r. roiban , j. high energy phys . * 05 * , 037 ( 2000 ) ; h. grosse , t. krajewski , and r. wulkenhaar ; h. girotti , m. gomes , v. rivelles , and a. da silva , nucl . phys . * b587 * , 299 ( 2000 ) ; i. chepelev and r. roiban , j. high energy phys . * 03 * , 001 ( 2001 ) ; a. bichl , _ et al _ , j. high energy phys . * 06 * , 013 ( 2001 ) ; s. sarkar , j. high energy phys . * 06 * , 003 ( 2002 ) . m. hayakawa , phys . lett . b * 478 * , 394 ( 2000 ) . ruiz , phys . b * 502 * , 274 ( 2001 ) . v. v. khoze and g. travaglini , j. high energy phys . * 01 * , 026 ( 2001 ) . l. griguolo and m. pietroni , j. high energy phys . * 05 * , 032 ( 2001 ) . g.h . chen and y.s . yu , nucl . phys . * b622 * , 189 ( 2002 ) . y. kiem , s. lee , s. j. rey and h. t. sato , hep - th/0110215 ; y. kiem , s. j. rey , h. t. sato and j. t. yee , phys . d * 65 * , 026002 ( 2002 ) ; hep - th/0107106 ; y. kiem , s. s. kim , s. j. rey and h. t. sato , hep - th/0110066 . a. armoni and e. lopez , nucl * b632 * , 240 ( 2002 ) . h. liu and j. michelson , nucl . phys . * b614 * , 279 ( 2001 ) ; h. liu , _ ibid_. * b614 * , 305 ( 2001 ) . m. van raamsdonk , j. high energy phys . * 11 * , 006 ( 2001 ) . l. jiang and e. nicholson , phys d * 65 * , 105020 ( 2002 ) . s. iso , h. kawai , and y. kitazawa , nucl . b576 * , 375 ( 2000 ) . n. ishibashi , s. iso , h. kawai and y. kitazawa , nucl . * b573 * , 573 ( 2000 ) ; s. j. rey and r. von unge , phys . b * 499 * , 215 ( 2001 ) ; s. das and s. j. rey , nucl . b590 * , 453 ( 2000 ) ; d. gross , a. hashimoto and n. itzhaki , adv . phys . * 4 * , 893 ( 2000 ) . n. seiberg , j. high energy phys . * 09 * , 003 ( 2000 ) . a. matusis , l. susskind and n. toumbas , j. high energy phys . * 12 * , 002 ( 2000 ) . d. zanon , phys . b * 502 * , 265 ( 2001 ) . r. gopakumar , s. minwalla , n. seiberg , and a. strominger , j. high energy phys . * 08 * , 008 ( 2000 ) . a. rajaraman and m. rozali , j. high energy phys . * 04 * , 033 ( 2000 ) . r. gopakumar , s. minwalla and a. strominger , j. high energy phys . * 05 * , 020 ( 2000 ) .

consider different data providers holding vertically partitioned data . each data provider holds different information about the same set of individuals and there is a common identifier ( social security number , or any other sort of i d ) that allows one to cross - reference individuals across data providers . for example , the data providers contain a data of a set of individuals : gene `` banks '' hold genetic information , police departments hold criminal records , financial institutions keep record of credit history , hospitals record health history of patience for future diagnosis , etc . those data providers might be geographically spread , belong to different organizations and have their specific privacy and security requirements . for instance , a data provider of genetic information might not allow a public access to its data while allowing access for medical doctors to their patients data , or police departments that allow data access for international law enforcements , but deny access to anyone outside the departments . in such cases , it is usually impossible to gather all the data in one place , either due to the size of the data or due to privacy restrictions on data sharing . a client issuing a query to a different data providers in public cloud settings , might be required to pay for utilized cpu time and network traffic . in many cases , such client will be willing to trade performance ( in other words , cost ) for accuracy of the answer . especially , if the trade off is controlled by the client , which can define the accepted error in the received answer . in this paper we propose a method of trading off performance for accuracy using sampling . we suggest a specific way to perform sampling in vertically split datasets and calculate sample size given acceptable error . performance improvements of the method are shown on a calculation of intersection set cardinality , together with a proposed heuristic algorithm . we also discuss a case of non - cooperative data providers and adaptation of the proposed algorithm for differential privacy calculations . * our goal * : given a set of @xmath0 data providers @xmath1 which data records share a common identifier @xmath2 and a set of predicates @xmath3 , find the size of intersection @xmath4 where @xmath5 is a set of all records in @xmath6 that satisfy predicate @xmath7 . we assume that the number of records that are not present in all datasets is very small compared to the total number of records and thus , for the sake of simplicity , the data providers in @xmath8 are assumed to contain the same records with size denoted by @xmath9 , i.e. vertical split of the data . the size of the intersection might not be calculated precisely , but up to a given accuracy @xmath10 provided by the client querying the data providers . for example , the client that performs a query is interested in an integer percentage of people from the entire population having both criminal record and specific genetic mutation , i.e. the results can be rounded to a closest percentage . intuitively , this relaxed requirement should result in more computationally and network efficient algorithms . the optimization of computation time and network traffic are important , as mentioned above , in the commonly used public - cloud deployment ; the user is charged for computation time ( also called cpu time ) and network traffic . thus , minimization of those parameters is a very attractive algorithm property and present a trade - off with accuracy of the intersection set size . * related work and our contribution : * conjunctive queries over distributed databases were extensively researched ( see as an example @xcite ) . previous research has references both the aggregation techniques and performance issues of such queries . however , those works are based on an assumption that databases freely share the data , which does not hold in a settings of non - cooperative data providers . there were a number of works describing privacy preserving calculations of intersection size , see @xcite . most of those works described a protocol that performs an exact calculation of the intersection set using secure multiparty computations ( mpc ) , first investigated by @xcite and later generalized to multiparty computation . performing mpc protocols allows efficient calculation of the intersection size , however , when used by itself , it is also posssible to leak information about specific items in the datasets . for instance , the user can issue a query knowing with a single possible answer and check whether the intersection set is empty or not . such technique allow the user to identify the presence or absence of a specific item in the datasets . a technique that can be used to hide a presence of individual items in the dataset is differential privacy ( @xcite ) . while some of the above works ( @xcite ) performed inexact calculations of the intersection set , the source of approximation was to preserve differential privacy of the results . for instance , @xcite applied oblivious transfer ( ot ) protocol to approximate the size of the intersection of two databases , where ot has lead to the inexact result . our approach is to utilize the relaxed requirement of providing inexact result to improve the computational complexity of queries . sampling as a way to cope with huge volumes of data or to increase computation performance were considered in a number of different contexts . ( @xcite ) and section 4.2 of @xcite discussed sampling as a means to cope with massive datasets by creation of a histogram - based estimators . once the estimator is created , when possible , the database performed estimation calculations on the histogram . reservoir stream sampling algorithms ( @xcite ) were developed to provide a sample of online data . the techniques fill in the sample ( `` reservoir '' ) by knocking out existing sample items from the sample with reducing probability as the sample begins to fill . ( @xcite ) initiated usage of document sampling for document similarity comparison on internet scale . ideas were later developed into min - wise sampling techniques ( @xcite ) , which appear to be more suited for horizontal split datasets . in addition , a considerable corpus of work exists on concentration inequalities in scope of machine learning . concentration inequalities investigate the relation between the size of the sample and its statistical similarity to the entire dataset . pac - bayes bounds on hypothesis error as a function of sample size were provided in @xcite . in this paper , we consider the case of approximate calculations in distributed databases where the data is split vertically . we suggest an algorithm that takes advantage of a lack of accuracy in a distributed answer to considerably speed up the queries . we show a use - case of an algorithm on intersection set cardinality calculations both with and without privacy considerations . the algorithm adopts sampling techniques from ( @xcite ) and uses techniques similar to pac - bayes bounds from ( @xcite ) to decide on a mimimal , representative sample size . performance improvement might be especially significant in privacy preserving setting , where the calculations take considerable time . specifically our contributions are : * suggested efficient method for approximate , distributed calculations with vertical - split datasets . * proposed a method of choosing sampling size given a required error level and provided simulation results of comparing the error level of different sample sizes . * suggested a heuristic algorithm of speeding up the intersection set size calculation based on bounds convergence and showed its adoptation for privacy preserving intersection set size calculation . the rest of the article is structured in the following way . section [ sec : naive algorithm ] defines a nave algorithm for calculation of exact intersection set size , section [ sec : sampling ] relaxes the requirement to calculate the exact size of intersection set and defines theoretical bounds for _ inexact _ intersection . section [ sec : experiments ] presents results of experiments that validate the proposed sampling method . section [ sec : bounded estimation ] describes a heuristic algorithm for calculation of intersection based on the bounds of intersection size . finally , section [ sec : privacy ] discusses privacy issues in the described algorithms . the paper is concluded in section [ sec : concluding remarks ] . the idea of the nave algorithm is simple : iterate over data providers @xmath11 and exchange a monotonously decreasing set of keys that satisfy all predicates up to the current iteration . the data provider then checks which of the received keys answer the corresponding predicate and returns a new list to the client . the client continues the process until all data providers were queried and the intersection set is found . in this way , the client calculates the intersection set iteratively . denote this set by @xmath12 where @xmath13 is the iteration number . in other words , @xmath14 where @xmath5 is a set of records in @xmath6 that satisfy predicate @xmath7 , and @xmath15 . the nave algorithm can also be performed in a parallel way , where the client receives a set of records that satisfy the corresponding predicate from each data provider . the client then calculates the intersection set . comparing the sequential and parallel nave algorithms , sequential algorithm optimizes the network load and also improves cpu time , but not a wall - clock time of the algorithm ( i.e. , the time between the beginning of the operation and the time the client gets the final answer ) . unless stated otherwise , in the rest of the section we consider only the sequential version of the algorithm . despite its virtue of being simple , the nave algorithm has a few drawbacks . * the amount of information transferred between the client and the data providers is relatively large , as the entire set of keys in the intersection set is sent . * the first data providers evaluate the given predicate over their entire dataset , which is expensive . * the wall - clock time of the sequential algorithm is linear in a number of data providers and the size of the data . * the algorithm completely exposes information both between data providers and between data providers to the client . as mentioned above , those drawbacks are even more extreme in a current common practice , where a public cloud infrastructure is used for calculations , as in a public cloud infrastructure the client is charged for computation time ( cpu time ) and for network traffic . thus , there is a monetary incentive to minimize those two values . both the sequential and the parallel nave algorithms described above calculate the intersection set exactly , thus having a relatively high network traffic and cpu time demands . the next section discusses a way to reduce computation and network complexity by calculation of an estimation of the intersection size . relaxing the requirement for exact calculation of the intersection , this section discusses a way to calculate a smaller intersection set whose size can be extrapolated to the intersection size of the entire population . instead of calculating the intersection of the entire datasets , perform a sample and calculate the intersection of the sample . then , scale up the size of the sample intersection to the size of the intersection for the entire population . the results of scale up will be an estimate for the intersection size . the requirements from the sample are clear : the sample should be as small as possible while truthfully representing the entire population ( given the definition and the error of representation ) . those two requirements present a trade - off between sample size and the accuracy of algorithm results . a number of different sampling techniques were developed over the years : on - line ( reservoir ) stream sampling algorithms ( @xcite ) , histogram - based estimators ( @xcite ) and min - wise sampling techniques ( @xcite ) which are more suited for horizontal split datasets . in our settings , it seems natural to choose hash function based technique , which in a sense a randomized version of histogram . to provide sufficient improvements in performance , the size of the sample ( denoted by @xmath16 ) should be much smaller than the average size of the datasets ( @xmath17 ) : @xmath18 . therefore , if each data provider performs its own sampling , the overlap between the samples will be very small . in other words , the probability of the data record ( person in our example ) appearing in the sample is low . for every data provider @xmath13 @xmath19 where @xmath20 is a sample from the dataset of the data provider @xmath13 . leading to the probability of the item appearing in all samples being ( assuming that all data providers contain the same records ) : @xmath21 where @xmath22 and @xmath23 are a sample size and a dataset size of a data provider @xmath24 . assuming , for simplicity , that the sample size and dataset size are equal for all data providers , the probability becomes : @xmath25 as this probability is low , the size of the intersection will be essentially zero for most of the samples . the solution is to make all data providers create the same sample . in order to do that , we will use a sampling technique based on hashing ( @xcite ) . the technique works in the following way : 1 . the client defines an accuracy threshold for the calculation . 2 . using techniques described below , the size of the sample is determined : @xmath16 . the number of hash buckets is @xmath26 . 4 . pick a hash function @xmath27 that will distribute datasets of data providers @xmath6 into @xmath28 buckets . @xmath27 hashes only the i d of the records and not the entire data , as the i d is the shared information across different data providers . 5 . the client sends each data provider 3 parameters : @xmath27 , the number of the bucket that was chosen randomly and predicate @xmath7 . this will ensure that all data providers use the same sample . each data provider evaluates its predicate on the records in a given bucket ( @xmath5 ) and sends the results to the client . the client performs the intersection calculation on the received results . as client acceptable error defines sample size , which is the bucket size , it is possible to pre - calculate a number of buckets for different error value . this will eliminate the need for sequential scan of the entire database on each query . however , it will also mean that the client will get approximately the required error . notice that just like the nave algorithm sampling can be done both in parallel and sequential ways . following the similar idea of the nave algorithm , in the parallel version of sampling algorithm each data provider evaluates the predicate over the entire given bucket , where in the sequential version , the client sends the current intersection set ( which is a subset of the given bucket ) to the next data provider , which evaluates the given predicate only over the received subset . however , in contrast to the nave algorithm , where the sequential variant significantly reduces the network load and cpu time , in sampling algorithm the reduction is much smaller , as the core underlying idea of the sampling algorithm is minimization of the amount of records participating in the intersection calculation . importantly , the accuracy of the intersection size estimation is the same in both sequential and parallel versions of the sampling algorithm . sampling improves both the computation time and network traffic over the nave algorithm by the factor of @xmath29 . thus , one important question still remains unanswered : what should be the size of the sample given an accuracy threshold ? the following sections define bounds on the sample size such that with high probability it will represent behavior similar to the entire dataset . two different cases are considered : sample size is small compared to the dataset size ( section [ subsec : small sample ] ) and sample size is comparable to dataset size ( section [ subsec : big sample ] ) . notice that in both cases the sampling is performed in the same way as explained above . however the bound used to calculate the sample size is different for those two cases . selection of the sample size is driven by the trade - off between the accuracy of the sample and the performance of calculations . the bigger the sample , the more accurate it is , but also the larger the computation time required to calculate the intersection . using a bound on the error as a function of the sample size , it is possible , given a limit on an acceptable error , to choose the size of the sample set . when a sample size is very small compared to the dataset size , the sampling can be approximated by independent sampling of the same size with replacement . the approximation can be done , as the probability of any record being a part of the sample goes from @xmath30 for the first pick to @xmath31 for the last pick . if @xmath32 then @xmath33 , and therefore , the probability is approximately ( @xmath30 ) , which is the same probability for a record to be picked when sampling with replacement . the reason to perform such approximation is due to a fact that it is simpler to provide a bound for independent samples with replacement rather than for sampling without replacement . the size of the sample should be big enough that the sample will be a good representative of the entire dataset for the intersection calculations . how can the size be estimated ? let us consider a sampling from the dataset @xmath6 according to the uniform distribution . notice that even though the sampling is done according to uniform distribution , as described in the sampling algorithm ( section [ sec : sampling ] ) , the datasets themselves might be drawn from other distributions . the provided bounds still hold , as the sample `` similarity '' to the entire dataset depends only on its size . if a different , non - uniform sampling technique is used , the bound might be changed to use more general bernstein inequalities ( @xcite ) . define @xmath34 to be a random variable that the sampled item @xmath35 satisfies condition @xmath36 . then , the average value of the sample , denoted by @xmath37 , should be close to the mean value , @xmath38 , of the entire dataset . notice that in the binary case , the average is simply the number of data records that satisfy a given predicate divided by the data size . in order to bound the difference between the average value of the sample and the mean value of the dataset , we can use concentration inequalities . namely using results by hoeffding ( @xcite ) if @xmath39 are independent and @xmath40 , then : @xmath41 for simplicity , for now we will consider only the right - side of the equation , i.e. @xmath42 where @xmath16 is the size of the sample and @xmath10 measures the `` resemblance '' of the sample to the mean of the dataset . notice that the equation does not depend on the size of the dataset , as it is considered much larger than the sample . the bound in equation [ eq : short hoeffding ] is used by the client when it issues queries to the data providers by defining both the acceptable error ( @xmath10 ) and the target probability of exceeding the acceptable error . in cases where sampling size is comparable to the dataset size , it is possible to develop bounds directly for sampling without replacement . in the rest of the section , we show two bounds on sampling without replacement , one is based on a reduction of `` randomness '' of the data and the second one is based on a direct counting technique . those bounds can be expected to be tighter than those based on reduction to independence or bounds for sampling with replacement . the reason for this as follows . assume that @xmath0 points were sampled out of @xmath43 points without replacement . the next point is to be sampled from a set of @xmath44 rather than @xmath43 points , which would be the case in sampling with replacement . the successive reduction in the size of the sampled set reduces the `` randomness '' of the newly sampled point as compared to the independent case , and also introduces dependency between samples . whereas , the bound provided in equation [ eq : short hoeffding ] does not depend on the dataset size and does not take the reduction in population size into consideration . this intuition is at the heart of serfling s improved bound ( @xcite ) which is stated next . the result holds for general bounded loss functions and is established by a careful utilization of martingale techniques combined with chernoff s bounding method . [ theorem : serfling ] ( serfling ) let @xmath45 be a finite set of non - negative bounded real numbers , @xmath46 . let @xmath47 be a random variables obtaining their values by sampling @xmath48 uniformly at random * without * replacement . set @xmath49 . then , @xmath50 similar bounds hold for @xmath51 . compared to the bound in equation [ eq : short hoeffding ] , the above bound is always tighter when @xmath52 , i.e. when @xmath53 . in our case @xmath54 s are binary variables and the bound could be improved further by using a proof based on a counting argument ( @xcite ) . figure [ fig : bounds comparison ] presents a single example of comparison between the above bounds ( [ eq : short hoeffding ] and [ eq : serfling ] ) . as expected , the serfling bound is tighter than the hoeffding bound and thus , using this bound for calculation of sample size will result in a smaller sample set . , @xmath55 . the error value is preset and then the sample size that fits this value is picked . in tighter bounds , the sample size will be smaller.,scaledwidth=50.0% ] now we can briefly describe the process of using those bounds . we assume that the client attempts to minimize the size of the sampling sets , as this also minimizes the network load and cpu time , both of which are chargeable in public cloud environments . the client defines a value of an acceptable error for the combined distributed query ( @xmath56 ) and a `` confidence '' of the error . the bounds defined in ( [ eq : short hoeffding ] ) or ( [ eq : serfling ] ) provide a `` confidence '' that the error will be smaller than @xmath10 given a sample size . now the client uses the bound from ( [ eq : short hoeffding ] ) or ( [ eq : serfling ] ) to determine the minimum sampling size that allow the required error with `` high enough '' confidence . using a predefined confidence and a chosen bound , the client can find the minimum value of sample size that will provide the required error with the required confidence . for example , consider figure [ fig : bounds comparison ] , where for dataset size of @xmath57 the client choose @xmath58 and confidence of @xmath59 . thus , using the simplified hoeffding bound ( [ eq : short hoeffding ] ) the minimal sample size is @xmath60 and using the serfling bound ( [ eq : serfling ] ) the minimal sample size is @xmath61 . of a sample size . ] after the intersection of sample sets is calculated , there is a need to estimate the size of the intersection of the entire datasets . as mentioned above , the mean of the binary random variable is also the number of records that belong to a set defined by the predicate divided by the size of the sample . thus , the ratio of the records in a sample that satisfy a given predicate is close to the ratio of the records that satisfy the predicate in the entire dataset of a given data provider . this leads to the conclusion that the ratio of the sample intersection set size to the sample size should be the same as the ratio of the intersection set size to the size of entire dataset . let @xmath62 be a sample of @xmath6 of size @xmath16 , then @xmath63 notice that even though the absolute error in the intersection estimation of the entire dataset is proportionally larger than the error in estimation of a sample , the relative error will remain the same . the reason for the error to remain the same in a process of intersection calculation is due to the sampling method . even though the sample from each data provider has the same ( potential ) relative error , the error is unique for the sample and thus , it does not accumulates when the intersection is calculated . as @xmath10 in the above bounds ( equation [ eq : short hoeffding ] and [ eq : serfling ] ) is a bound on difference in estimation , it is a relative error . therefore , if the client defines an acceptable error ( @xmath56 ) as an absolute error , it can be easily translated : @xmath64 . we have performed a number of experiments to show the utility and error resulted from sampling . the experiments were performed on simulated data and validated on the adult dataset ( @xcite ) . the described sampling technique is targeted at large datasets with many records , such that calculation of predicates over the entire dataset is wasteful . for such datasets , it is much more practical to test the algorithm on simulated data , which can be generated on any required scale . for experiments on simulated data , a number of datasets were generated with random values of the predicates and then the intersection size was calculated according to each algorithm . the methodology works as follows . * generate a dataset of a given size . all data providers assumed to share the same set of records with possibly a different data per record . * for each data provider generate a set of predicate values randomly , such that the frequency of `` true '' values is as defined . * calculate the intersection iteratively , i.e. by addition of a single datasets at a time . this is a both a simpler way of implementation ( as opposite to parallel calculation ) and also allows observation of convergence rate of the intersection size . all experiments were averaged over 10 runs , standard deviation was calculated and drawn on the resulting graphs . figure [ fig : algorithms comparison ] shows the intersection size as calculated by the sequential nave algorithm and estimation by sampling of various sizes . the graph provides a high - level , visual practical validity of the approach . even a relatively small sample sizes ( @xmath65 of the dataset size ) estimate intersection size close to the real value . in order to focus on the error caused by sampling and show the difference in sample size more clearly , we have performed a number of experiments showing relative error between estimation and the real intersection size . the nave algorithm calculated was taken as a baseline for error calculations . the sampling algorithm was executed with a number of sample sizes , each one showing the relative error from the nave algorithm results . figures [ fig : algorithms error comparison ] , [ fig : algorithms error comparison 1 m ] , [ fig : algorithms error comparison 100k r05 ] , and [ fig : algorithms error comparison 1 m r05 ] show the relative error of the sampling algorithm with various sample sizes . the y axis shows the error of the sampling , i.e. |sampled estimation - intersection size| / ( intersection size ) , while the x axis is the number of datasets in the intersection . the graphs compare datasets sizes of 100,000 ( figures [ fig : algorithms error comparison ] and [ fig : algorithms error comparison 100k r05 ] ) and 1,000,000 ( figures [ fig : algorithms error comparison 1 m ] and [ fig : algorithms error comparison 1 m r05 ] ) records with predicate satisfaction frequency of @xmath66 and @xmath67 , i.e. in each data provider , @xmath68 or @xmath69 of the records satisfy the corresponding predicate . as mentioned above , each graph is an average of 10 different runs and standard deviation is depicted by error bars on the graph . the graphs show that , as expected , larger samples result both in smaller error and smaller standard deviation . however , it also can be seen , that the error quickly converges as a function of sample size . in some cases , even for sample of @xmath65 from entire dataset , it is possible to achieve reasonable error . overall , sampling @xmath70 or @xmath71 of the dataset resulted in errors of approximately @xmath59 from the intersection size , which is an order of a single percent . thus , queries of a type : `` what is the approximate percentage of people ... ? '' can be answered using only tenth of a data . , number of datasets is @xmath72 , ratio of predicate satisfaction in each dataset is @xmath66 . the y axis shows the size the intersection estimation . the x axis show the number of the data sets that participated in the intersection . error bars show standard deviation over 10 different executions.,scaledwidth=50.0% ] , number of datasets is @xmath72 , ratio of predicate satisfaction in each dataset is @xmath66 . error bars show standard deviation over 10 different executions.,scaledwidth=50.0% ] , number of datasets is @xmath72 , ratio of predicate satisfaction in each dataset is @xmath66 . error bars show standard deviation over 10 different executions.,scaledwidth=50.0% ] , number of datasets is @xmath72 , ratio of predicate satisfaction in each dataset is @xmath67 . error bars show standard deviation over 10 different executions.,scaledwidth=50.0% ] , number of datasets is @xmath72 , ratio of predicate satisfaction in each dataset is @xmath67 . error bars show standard deviation over 10 different executions.,scaledwidth=50.0% ] in addition to experiments on simulated datasets , we have tested the methodology on adult dataset from uci machine learning repository ( @xcite ) . the dataset contains data from census bureau , where each record has data about different person . this fits a description of the setup we have described . the dataset contains 32,562 instances . as a test case we have used an intersection of the following predicates : age @xmath73 30 , marital status : never - married , sex : female or male in two different tests and income @xmath74 50k . the exact intersection set size is 252 for males and 139 for females . since the dataset was not used for classification task , it allowed us to use income as one of the predicates . in addition , the intersection was of 4 predicates , as addition of more predicates resulted in a small or empty intersection set due to a limited dataset size . samples of different sizes were drawn with sample size ratio going from 0.1 to 0.5 . the intersection set size was calculated from the drawn sample . figure [ fig : adult sampling ] depicts the results of the testing . while the accuracy of the estimation does not necessary improves by taking larger samples , the standard deviation becomes smaller . the accuracy improvement is most probably caused by the small size of the intersection , while improvement in standard deviation fits the results shown in simulated datasets . overall , the accuracy of the intersection set size calculation verify the validity of the approach . previous sections described exact and approximate ways of calculating the size of the intersection . this section presents a heuristic algorithm that attempts to optimize the calculation of the intersection by not performing the calculations for all datasets . the simplest case of heuristic is when the intersection is calculated iteratively and at some point the intersection set is empty . clearly , the calculation can be stopped at this stage . following is a heuristic algorithm for the intersection calculation that attempts to stop at the earliest possible point . the size of the intersection depends on the sizes of the sets from each data provider that answer the corresponding predicate and on the correlation between those sizes . below we suggest a heuristic algorithm for estimation of the intersection set size . the algorithm starts with an accuracy parameter , the ratio of records that satisfy the relevant predicate in each data set , and the lower and the upper bounds on the intersection size . the algorithm iteratively tightens the bounds until the difference between the bounds is smaller or equal to the accuracy parameter . then the algorithm stops and returns the middle value of the range between the lower and the upper bound . let @xmath7 be the predicate that is associated with a data provider @xmath6 . then , @xmath75 is the set of members in @xmath6 that satisfy @xmath7 . also , @xmath76 will denote the fraction of the members that satisfy @xmath7 in @xmath6 , i.e. @xmath77 . now , let us define the bounds on the size of the intersection set @xmath78 for the sake of the algorithm iterations . for simplicity , we assume that data provider datasets @xmath79 contain _ exactly _ the same records . the bounds will hold when the difference in members between datasets is small . as the intersection size is at most the size of the smallest predicate set , the size of the intersection is bounded from above by the following bound : @xmath80 the lower bound on the intersection of two sets @xmath81 and @xmath82 is ( @xcite ) @xmath83 in case of 3 sets , the lower bound becomes : @xmath84 which in general is @xmath85 where @xmath86 is the average size of the datasets ( as they contain the same set of records ) . as described above , the iterative step of the algorithm is to tighten the bounds on the intersection size . the algorithm will then stop once the difference between bounds is smaller than the required accuracy . as the upper bound ( equation [ eq : upper bound ] ) is the minimal predicate set , an iteration step will attempt to make the minimum value smaller by calculating the intersection between the two smallest predicate sets . the intersection of two sets will be smaller or of the same size as minimal predicate set and will replace those two sets in the bound . calculating the intersection of two minimal predicate sets is also a good technique for increasing the lower bound ( equation [ eq : lower bound ] ) as well .. in this case the lower bound of their intersection is 0 . ] notice that each iteration step of the algorithm also decreases the value of @xmath0 and thus removes the subtractive member of the lower bound . the algorithm steps are as follows . 1 . define a required accuracy threshold : @xmath87 as @xmath88 is a relative error , the iterations continue until an absolute error is smaller than @xmath89 . when the iterations stop , the middle of the bounds range is returned , therefore , the iterations stop when the distance between the bounds is less or equal to @xmath90 2 . calculate @xmath5 for every data provider @xmath13 . 3 . calculate the lower ( @xmath91 ) and the upper bounds ( @xmath92 ) , given @xmath7 . [ step : iteration ] while @xmath93 and the number of sets is larger than 1 : 1 . let @xmath94 and @xmath95 . in other words , pick two minimal predicate sets : @xmath96 and @xmath97 . [ step : intersect ] calculate the intersection between those two sets . 3 . replace @xmath98 and @xmath99 with the new predicate set : @xmath100 . 4 . update new values of @xmath92 and @xmath91 . if only one set remains @xmath101 ( @xmath102 ) , then the size of its predicate set , @xmath103 , is the size of the intersection set . 6 . if there is more than one set and @xmath104 , then the size of the intersection set to the middle value of @xmath105 $ ] range : @xmath106 . the algorithm clearly converges , as the iterations stop when the intersection size is calculated exactly . at this point the lower bound is equal to the upper bound and thus , the accuracy requirement will be satisfied . notice that if step [ step : intersect ] results in an empty set , the algorithms also stops , as the lower and the upper bounds will be equal and zero . * example : * [ example : algorithm example ] as an example of the algorithm execution , assume @xmath107 , equal size of all data sets @xmath43 and the required accuracy of @xmath108 . let the respective ratios be @xmath109 and @xmath110 . in this case , the upper bound on the intersection set size will be @xmath111 , whereas the lower bound will be @xmath112 , as the sum of all ratios is @xmath113 less than @xmath114 . the first iteration of the algorithm will be to find the intersection of @xmath115 and @xmath116 . let us assume that their intersection results in @xmath117 . now , the upper bound of the intersection becomes @xmath118 , whereas the lower bound increases to @xmath119 . the iterations continue until the bounds converge to within @xmath120 . in general , the heuristic algorithm will work fine in cases where the ratio of records that satisfy the predicate is high across most data providers , but the intersection size might decrease relatively fast . see figure [ fig : heuristic algorithm ] for example of convergence of the algorithm bounds on simulated data . , number of datasets is @xmath72 , ratio of predicate satisfaction in each dataset is @xmath121 ( the high ratio is required for the lower bound to be larger than zero for @xmath72 ) . notice that in simulated case , the intersection size is co - located with the upper bound of the intersection.,scaledwidth=50.0% ] clearly , both the computation time and network traffic depend heavily on the calculation of the intersection between the two sets in step [ step : intersect ] of the heuristic algorithm . to improve the performance of this step it is possible to use the sampling technique from section [ sec : sampling ] . instead of performing intersection between two sets @xmath13 and @xmath96 , the algorithm will calculate the intersection of two samples of those sets . however , sampling introduces an error into calculation of the intersection , which has to be related to the accuracy threshold @xmath88 of the heuristic algorithm . assuming that the error in sampling distributes proportionally among datasets and noting that each iteration `` eliminates '' one dataset , we can define the following changes to the algorithm . every sample introduces error of @xmath122 . to accommodate this error , the algorithm will decrease the required accuracy threshold by this amount on each iteration . this will ensure that required by user accuracy threshold will be honored . two steps are changed in the algorithm : 1 . step [ step : iteration ] then becomes : + _ while @xmath123 where @xmath13 is the iteration number , and the number of sets is larger than 1 . step [ step : intersect ] becomes : + _ calculate the intersection between those two sets : @xmath96 and @xmath24 by sampling with error threshold @xmath124 . _ the rest of the algorithm remains the same , including required accuracy threshold value . in the described setting there are 2 different types of actors : data providers and client . there are different privacy concerns between those actors . one privacy concern is preserving the privacy of data providers . the client querying the data providers should not learn the identity of the records that are part of the intersection set or the number of records that satisfy a specific predicate in any single data provider . notice that the client should be limited to a reasonable number of queries ( sub - linear in a data size ) , as it is not possible to answer a large number of arbitrary queries while preserving privacy ( @xcite ) . another privacy concern is in keeping a privacy of records in a specific data provider from other data providers . for instance , a data provider might be interested in hiding the presence of specific record from other data providers . first , it is imperative to note that the sampling ( see section [ sec : sampling ] ) provides a very nave form of privacy . as every data set record has a @xmath125 chance of not being included in sample , the presence of the specific record in the data set is not immediately observable . yet , the fact that the client gets record identifiers of each data provider is undesirable . data providers can encrypt the sampling indices to hide the identity of the records that are sent to the client . even though an additional encryption will obscure the identity of those items from the client , it still will expose the presence of individual records to other data providers . in addition , using an additional knowledge , the client can easily learn the presence of a specific record in the data providers . one way to preserve privacy is to use secure multiparty computations to calculate the intersection set ( @xcite ) a simple method that uses commutative cryptography ( another approach is to use secret sharing ) is described in @xcite . each data provider encrypts record identifiers using its own commutative key and passes the key to other data providers . once all data providers have encrypted all keys , it is possible to find the intersection using encrypted identifiers utilizing commutativity of the encryption and calculate its size . there are a few downsides to using secure multi party computations . first , performance is heavily impacted by performing secure computations using these methods . it is possible to aleviate the performance issues by using the sampling to reduce a number of records in the calculated intersection , as described in section [ sec : sampling ] . thus , the idea is to perform the sampling algorithm and then calculate the intersection in a secure way using the mpc computations between the data providers . this method however , will still require calculation of the sampled intersection exactly , without the ability to halt when required accuracy was reached , as it is possible in heuristic algorithm . the second drawback is specific for the described use - case where the data is kept in different organizations . secure multi party computations require direct communication between data providers , which in some cases is very challenging in a real - world deployments due to both security and technological reasons . while it is still possible for the client to act as an intermediary between different data providers , such setup also doubles the network traffic . a different approach is only to allow communication between client and data provider with adopted algorithms . however , the most significant drawback of mpc is that the exact intersection size is calculated . this might allow the client to use additional information to infer a presence of a specific record in the dataset . the current de - facto privacy standard is differential privacy ( @xcite ) . informally , the idea of differential privacy is to protect the privacy of individual records in the dataset without any assumption on the additional knowledge . differential privacy is preserved if it is practically impossible to identify whether the record is present or absent in the dataset from the result of database queries . the most common methods of ensuring differential privacy rely either on addition of a carefully chosen amount of noise to the result or by using an exponential mechanism that chooses the output according to some specific probability distribution . the protocols that are based on secure multiparty computations perform the exact calculation of the set , which is impossible in differential - privacy settings . this led to several works describing differential - privacy - preserving calculations of the intersection size , see @xcite . even though those algorithms preserve differential privacy , the requirement to provide an approximate answer in our case allows our scheme to optimize the algorithm for performance by performing secure computations on as little number of the records as possible due to sampling . the most common method of ensuring differential privacy is an addition of a specifically crafted random noise to the released data ( @xcite ) . moreover , there are two different approaches for the private data release : interactive ( @xcite)and non - interactive ( @xcite ) . interactive data release is when the client sends data query to a data provider . the data provider then releases the data to the client while ensuring differential privacy of records in its database . the data provider might decide not to answer a specific queries or to stop answering queries from a given client , as this might impact the privacy . in non - interactive data release , the noise is applied only once and then the data is released to the client . the client can perform any number and any type of queries on the data . intuitively , non - interactive release requires to add more noise to the released data , as it has to be ready for any query , while interactive release can adopt added noise to the results of each query ( @xcite ) . in the described above sampling algorithm ( section [ sec : sampling ] ) , there are two intuitive locations to add random noise . first , it is possible to add noise in hashing function that assigns records to buckets . such noise will preserve privacy of individual records , as it will be impossible to distinguish whether a specific record is within the bucket or not present at all in the dataset . a second location to add the noise is during a query processing . since the data provider exposes only the number of records that satisfy a given predicate , the noise can be added to the output after the predicate was evaluated or to the predicate itself [ [ ehud : this is not clear , the noise should be fake ids ] ] . the difference between those two locations is exactly the difference between interactive and non - interactive data release . adding noise to the hashing function is a one - time operation and has no relationship to the result of the specific predicate . thus , this noise addition can be seen as non - interactive data release . on the other hand , if the noise is added after the predicate is evaluated over records in a bucket , then the noise value can be adapted to the results of predicate evaluation . thus , the amount of noise can be directly related to the results of a specific query , like in interactive data release settings . due to the above reasoning , the next section describes an addition of a noise to the predicate function results . differentially private calculation of a counting function ( referred in the paper as `` noisy sum '' ) , i.e. counting a number of records that satisfy a given predicate , was considered previously in ( @xcite ) . ( @xcite ) showed that adding a laplacian noise @xmath126 to the sum query output : @xmath127 , ensures @xmath128differential private function . notice that the sensitivity of counting function is @xmath129 and thus , the distribution standard deviation is @xmath130 . the fast algorithm for privacy - preserving intersection calculation is described below . heuristic algorithm from section [ sec : bounded estimation ] is used as a basis for privacy preserving algorithm . 1 . use a hash function @xmath27 to divide data into buckets . in case only a single bucket is used for all queries , only the agreed bucket is kept . 2 . when a predicate is received from the client , evaluate the predicate over the chosen bucket . 3 . add random noise according to laplace distribution to the size of the predicate set . the noisy set - size is shared with the client . the client gathers results from all data providers and calculate upper and lower bounds . [ step : chosen data providers ] the client continues to execute heuristic algorithm and picks two data providers , @xmath96 and @xmath0 , to perform intersection according to the algorithm step [ step : intersect ] . perform privacy - preserving , secure intersection between two chosen providers using the below algorithm . 7 . update the bounds and continue until the bounds are close enough . the algorithm used for privacy - preserving , secure intersection set calculation is based on a work of ( @xcite ) , which allows a differentially private calculation of an intersection over multiple databases . the algorithm provides computational differential privacy , as it relies on homomorphic encryption and makes some assumptions on computational hardness . this section describes how the algorithm from ( @xcite ) can be used in our heuristic algorithm from section [ sec : bounded estimation ] for inexact computations . the algorithm describes a basic operation _ private set - intersection cardinality ( psi - ca ) _ and then adds noise to ensure differential privacy , thus resulting in bn - psi - ca . psi - ca operation is based on usage of homomorphic cryptosystem that allows addition and multiplication by constant ( for instance , paillier s cryptosystem ( @xcite ) ) . when two data providers , @xmath13 and @xmath96 , attempt to calculate intersection , @xmath13 defines a polynomial whose roots are the members of its set , the polynomial coefficients are then encrypted and transfered to @xmath96 data provider , which calculates @xmath131 . in other words , it evaluates the polynomial on its set members , multiplies by a constant number @xmath132 and adds a special string of zeros @xmath133 . when transfered back to the first data provider ( @xmath13 ) , it calculates the number of zero string . in order to add differential privacy , a random number of dummy values are added to the transfered set by both parties ( for details see ( @xcite ) rounding the set size to the bucket size in our case . in the end , data provider @xmath13 learns the noised cardinality and data provider @xmath96 knows the amount of noise it added to the intersection set size . adopting this algorithm for our case , the client performs the following actions . assume that two data providers are chosen for intersection in step [ step : chosen data providers ] : @xmath96 and @xmath0 . 1 . perform bn - psi - ca algorithm to calculate a noisy intersection set . assume that without loss of generality , @xmath96 will hold the size of a noisy set , where @xmath0 will hold the amount of added noise . 2 . @xmath96 sends the size of a noisy set to the client . in the following iterations of the heuristic algorithm , bn - psi - ca might be invoked to find intersection size of @xmath96 , @xmath0 and @xmath132 , and more additional data providers . as due to privacy issues , no single data provider holds the intersection dataset , it is necessary to perform intersection size calculations over again for any additional data provider . notice that in cases where data providers do not communicate directly , the client acts as an intermediary for the protocol . in the end , the client has a value of intersection size which is still preserving differential privacy . the client also keeps record that @xmath0 holds the amount of noise of the last intersection . as described in ( @xcite ) , the noise added in one intersection , might be removed if there is a need to calculate another intersection of @xmath96 and @xmath0 with additional data provider . ( @xcite ) performance results show approximately linear dependency of an algorithm run - time in the number of records in the participating datasets . using results from section [ sec : experiments ] , with a reasonable error , it is possible to reduce the number of records by a factor of 5 - 10 ( according to acceptable error ) , thus tens of minutes to single minutes . such a reduction might make a difference between practical and impractical system . the algorithm accelerate approximate intersection calculation by : * computationally intensive calculations of intersection sizes are performed on samples . those calculations are demanding computationally , thus , any reduce in the size of datasets is important . [ [ ehud : we should try to find a third - year student to implement this ] ] * the intersection is not calculate exactly , but rather the algorithm stops when the difference between bounds is close enough . a steady and rapid increase in the amount of data has resulted in an abundance of various data providers . more data is kept for longer and in more places . in parallel , the awareness of data privacy and security is also on a rise both from government regulation and personal perspective . this trend requires new algorithms and protocols for dealing with distributed data . those new methods should be both efficient and privacy preserving . the main practical downside of the current privacy preserving computations is the run - time complexity . both mpc , oblivious - transfer and private - information - retrieval methods that are used to perform distributed calculations in privacy - preserving manner significantly increase the running time of the query . while some advances in improving performance were made , the running time still remains a limiting factor . our proposed method uses the ability to provide an approximate answer to the supplied query and thus to reduce considerably the dataset that is used for computations . this decouples the use of sampling technique for reducing the size of the dataset used for calculations from the protocols designed to preserve privacy of individual records or data providers . as such , most of the above protocols can be used in conjunction with sampling techniques to perform approximate calculations with improved performance . as showed in this work , when acceptable , calculating an approximate answer can significantly improve computation time . in those cases we have suggested a way to use this possibility for approximate answering by using sampling of the datasets . the sampling leads to a dramatic reduction in the size of data required for calculation : @xmath134 of the data as observed in simulations . such reduction of the data size is much more significant when calculations are done in a secure and privacy - preserving way . r. agrawal , a. evfimievski , and r. srikant . information sharing across private databases . in _ proceedings of the 2003 acm sigmod international conference on management of data _ , sigmod 03 , pages 8697 , new york , ny , usa , 2003 . acm . c. dwork , f. mcsherry , k. nissim , and a. smith . calibrating noise to sensitivity in private data analysis . in _ theory of cryptography , third theory of cryptography conference , tcc 2006 , new york , ny , usa , march 4 - 7 , 2006 , proceedings _ , pages 265284 , 2006 . p. b. gibbons . distinct sampling for highly - accurate answers to distinct values queries and event reports . in _ in proceedings of the 27th international conference on very large data bases _ , pages 541550 . n. mohammed , r. chen , b. c. fung , and p. s. yu . differentially private data release for data mining . in _ proceedings of the 17th acm sigkdd international conference on knowledge discovery and data mining _ , kdd 11 , pages 493501 , new york , ny , usa , 2011 . a. narayan and a. haeberlen . djoin : differentially private join queries over distributed databases . in _ in proceedings of the 10th usenix symposium on operating systems design and implementation _ , 2012 . k. nissim , s. raskhodnikova , and a. smith . smooth sensitivity and sampling in private data analysis . in _ proceedings of the thirty - ninth annual acm symposium on theory of computing _ , stoc 07 , pages 7584 , new york , ny , usa , 2007 . r. pagh , m. stckel , and d. p. woodruff . is min - wise hashing optimal for summarizing set intersection ? in _ proceedings of the 33rd acm sigmod - sigact - sigart symposium on principles of database systems , pods14 , snowbird , ut , usa , june 22 - 27 , 2014 _ , pages 109120 , 2014 . d. woodruff and q. zhang . when distributed computation is communication expensive . in y. afek , editor , _ distributed computing _ , volume 8205 of _ lecture notes in computer science _ , pages 1630 . springer berlin heidelberg , 2013 .

models with more than four dimensions are interesting , because all the physical ingredients of the universe can be unified in a higher dimensional theory . string theory is the most promising scenario where quantum gravity is included by the requirement of additional dimensions and supersymmetry . the idea of large extra dimension@xcite is important , because it may solve the hierarchy problem . in this case , the observed planck mass is obtained by the relation @xmath3 , where @xmath4 and @xmath5 denote the fundamental scale of gravity and the volume of the @xmath6-dimensional compact space , respectively . in the scenarios of large extra dimension , the fields in the standard model are expected to be localized on wall - like structures ( `` our branes '' ) , while the graviton propagates in the bulk . in the context of string theory , a natural embedding of this picture is realized by brane construction . the brane models are interesting from both phenomenological and cosmological viewpoints . analyses of cosmological defect formation are important in brane models . for the cosmological scenarios , one can consider at least three different types of defects : * defects are branes . + in this case , cosmological defects are the branes that have less than three spatial dimensions in the uncompactified space - time . in the previous discussions@xcite , in which the dynamical effect of `` our brane '' has been neglected , it was concluded that the cosmological production of monopoles and domain walls are negligible in the models of brane inflation . later in ref.@xcite and @xcite , however , it was shown that cosmological formation of such monopoles and domain walls are quite natural in generic cosmological scenarios . * defects are deformed branes . + in this case , the cosmological defects are formed by the continuous deformation of branes . first in ref.@xcite , and later in ref.@xcite , it was shown that the fields that parameterize the position of branes can fluctuate in spatial directions to form cosmological brane defects . cosmic strings are constructed in ref.@xcite , where the singularity in the core is resolved by smearing the wall - like structure . monopoles , strings and domain walls are constructed in @xcite . * localized fields are shifted in the cosmological defects . + in ref.@xcite , localization of matter fields on a fat domain wall is discussed to explain small interactions in the effective four - dimensional action . this idea is important in constructing realistic models in which proton must be stabilized . along the line of this argument , new defect configurations are constructed in ref.@xcite , which induce shifts of the localization . due to the shifts of the localization , small interactions can be enhanced in the core of these defects . therefore , these defects can assist the generation of baryon number asymmetry of the universe@xcite . first , we shall review the essence of the `` previous arguments''@xcite , which were used to exclude the cosmological production of monopoles and domain walls . in the original scenario of brane inflation@xcite , inflationary expansion is driven by the potential between d - brane and anti d - brane evolving in the bulk . then the scenario of the inflating branes at a fixed angle is studied in ref.@xcite , where the slow - roll condition is improved by introducing a small angle . the end of brane inflation is induced by the brane collision where brane annihilation ( or recombination ) proceeds through tachyon condensation@xcite . during brane inflation , tachyon is trapped in the false vacuum , which can result in the formation of lower - dimensional branes after brane inflation . the production of cosmological brane defects was discussed in ref.@xcite , where it was concluded that cosmic strings are copiously produced in these scenarios , while monopoles and domain walls are negligible . to be precise , their arguments are based on the speculation that any variation of fields in the compactified direction must be suppressed during inflation since the compactification radius must be small compared to the horizon size during inflation . then they have concluded that the daughter branes , which are formed by tachyon condensation on the world volume of the mother branes , must wrap the same compactified space as the mother branes . in this case the effect of compactification seems significant . as a result , it was concluded that the codimension of the daughter branes must lie within the uncompactified space . since the number of the codimension must be even in this case , the cosmological defects must be cosmic strings . moreover , later in ref.@xcite , it was discussed that the analysis does not fully account for the effects of compactification , because the directions transverse to the mother branes had not been considered . the effect of the rr fields extended to the compactified dimensions was discussed in ref.@xcite . then it was concluded that the creation of the gradients of the rr fields in the bulk of the compactified space is costly in energy , so that there should be a serious suppression in the creation of the daughter d brane as far as they do not fill all the compactified dimensions . as a result , in the `` previous arguments '' , cosmological brane defects were discussed to suffer from many unnatural constraints . however , here we should recall that no such suppression has been discussed so far in the cosmological production of the conventional defects in four - dimensional effective gauge theory , while the realistic brane models must reproduce the standard model in its effective action . therefore , it is quite natural to think about the following questions . is it really impossible to produce branes that do not wrap the same compactified space as the mother branes ? is it really impossible to produce monopoles and domain walls in brane inflation ? here we should note that the `` previous arguments '' are not fully reliable in more realistic cases . the most obvious example was first discussed in ref.@xcite , where the inconsistency of the string tension is solved in angled inflation . in the original argument in ref.@xcite , the tension of the string did not coincide with the one calculated from the effective action . this inconsistency is clearly due to the assumption that the daughter brane must wrap the same compactified space as the mother brane . in fact the previous arguments might be true in some simplest cases , however we must be more careful about the process of the brane recombination . in ref.@xcite , it was shown that the production of the cosmic strings in angled inflation is realized by the creation of the daughter branes that are * extended * between the splitting mother branes . it should be noted that the creation of such extended branes is not a counter example to the generic mechanism of tachyon condensation . to be precise , one can understand from a careful treatment of the effective action@xcite that the eigenfunction of the tachyonic mode should be localized at the intersection . since the mechanism of this localization is * different * from the kibble mechanism , this argument does not contradict to the `` previous arguments '' . therefore , the `` seed '' must be localized at the intersection . as the recombination proceeds , the @xmath0 brane at the intersection is pulled out of the mother @xmath1 branes , and finally becomes extended between mother branes . in this case the serious problem of the rr field is also avoided , since the length of the extended daughter brane vanishes when it is created at the intersection . it costs energy to pull @xmath0 branes out of the @xmath1 branes , however the cost is paid by the repulsive force between the splitting @xmath1 branes . of course the tension of such extended daughter branes matches precisely to the d - term strings in the effective action . moreover , this conclusion is consistent with the analysis of the string production in the effective action , where the production of the d - term strings is not suppressed . of course , it seems obvious from the string perspective that the brane creation is due to the local dynamics of the open strings at the intersection of the splitting @xmath1 branes , which suggests the obvious localization of the eigenfunction . the daughter branes satisfies the required properties . thus at least in the models of angled inflation , it is obvious that the cosmic strings in the effective action do not wrap the same compactified space as the mother branes . from the above discussions , it is natural to think that the previous arguments are not reliable and should be compensated in more generic cosmological scenarios . obviously , there is no reason that we must believe that in their final state the daughter branes must wrap the same compactified space as the mother branes . since the previous argument can not fully account for the generic process of the cosmological defect formation , one can hardly accept their conclusions . is it really impossible to produce sufficient amount of cosmic monopoles and domain walls by the brane creation ? this question is answered in ref.@xcite , paying careful attention to the brane dynamics after inflation , especially to the dynamics of `` our brane '' . for the extended branes to be produced between mother branes , the distance between mother branes is required to be zero at a moment . this idea is generic and applicable to other conventional cosmological processes . the spontaneous symmetry breaking in the effective action is sometimes described by the recombination of the branes@xcite or by the branes falling apart , which can be induced by the thermal effects@xcite . as a result , contrary to the previous speculation , it was shown@xcite that there is no reason that one should believe that monopoles and domain walls@xcite are suppressed in the brane universe , once the dynamics of `` our brane '' is included . besides the cosmological defects that are formed by the brane creation , one can consider the defects that are formed by the continuous deformation of the branes . the two kinds of the brane defects can be produced by the same process@xcite . therefore , the actual cosmological defects should be * the mixture of these defects . * it should be noted that the analyses of such mixed defects are quite important in understanding the evolution of the brane universe . from the above viewpoints , we shall reconsider the evolution of the cosmic strings in angled inflation . in this paper , we show that the strings in angled inflation can be distinguished from the conventional cosmic strings . in angled inflation , cosmic strings are the daughter @xmath0 branes that are extended between mother @xmath1 branes . the @xmath0 branes have the flat direction ( moduli ) in the compactified space , along which the endpoints of the @xmath0 branes can move freely on the @xmath1 branes . ] then the position of the @xmath0 branes can vary along the cosmic strings , which results in the peculiar type of @xmath2-dimensional kink configurations . one can also find the peculiar winding states that are defined on the world volume of the strings . a schematic picture of the kinks is shown in fig.[fig : necklace ] . the point - like objects that appear on strings are monopoles . since the monopoles are skewered with the strings , the cosmic strings in angled inflation become necklaces . these kinks are produced by the spatial deformation of the @xmath0 branes . therefore , the brane necklaces are the * hybrid * of the brane creation and the brane deformation . the chopped loops of the necklaces can shrink to produce the stable winding states , which looks like coils . a schematic picture of the brane coils is shown in fig.[fig : shrink ] . it should be noted here that the brane coils wind around the compactified space that is * different * from the mother branes . of course the mechanism that induces such windings is different from the kibble mechanism . therefore , our mechanism avoids the serious constraints that have been discussed previously in ref.@xcite . in the early universe there could be a pair of brane anti - brane separated in the compactified space . the potential energy of these branes can drive brane inflation , then they annihilate rapidly at the end . if this process contains only a pair of @xmath7 branes , it might be true that only cosmic strings are produced after inflation . in this case , the cosmic strings are the daughter branes that wrap the same compactified space as the mother branes . at first the strings could move freely in the compactified space , but later their position should be fixed as was discussed in ref.@xcite . then , if one assumes that other branes ( including `` our branes '' ) are decoupled , these strings can not break nor dissolve into other branes . in this case , since other branes are completely decoupled from the cosmic strings , the interactions of the cosmic strings must be quite simplified . the cosmological consequence of this scenario is already investigated in ref.@xcite . on the other hand , in generic cases it seems rather difficult to obtain enough number of `` decoupled '' cosmic strings while reheating the universe to an acceptable temperature . if the moving brane of the brane inflation collides to the stack of the standard model ( sm ) branes , reheating should be successful . however , in this case one can not ignore the interaction between the daughter branes and the other branes at the stack . since the cosmic strings are the lower - dimensional branes that could break or dissolve into other branes , in generic cases one can not simply ignore the interaction between other branes . therefore , it is important to investigate cosmic strings that intersect with other branes . from the above viewpoints , we shall try to understand the evolution of the cosmic strings in angled inflation to show how one can distinguish the cosmic strings in angled inflation from the conventional ones . as we have discussed in the previous section , angled inflation is the most obvious generalization of the simplest brane inflation , where the slow - roll condition is improved and the * cosmic strings intersect with `` our branes''*. if the potential that lifts the flat direction of the cosmic strings is shallow , the brownian motion in the compactified space must be important . the brownian motion in the direction that is perpendicular to the daughter brane can be induced by the brane collision@xcite . even if the potential is not shallow due to the stringy effects , it seems natural to expect that the internal position of the cosmic strings can vary because of the energetic process of the brane collision . in any case , it seems too optimistic to expect that the potential is so steep that one can avoid the generation of the kinks on the strings . moreover , if the angle is small(@xmath8 ) and the final state of the recombined mother brane wraps about @xmath9 times around the compactified space , ] the potential should have many local minima . these minima appear as the `` domains '' on the strings . to be precise , strings can have several distinctive regions(domains ) where the branes are located at different sites in the compactified space . ] then kinks ( i.e. ( 1 + 1)-dimensional domain walls ) will appear on the strings , which interpolates between neighboring domains on the strings . these kinks may have interesting consequences . moreover , it is interesting to think about the configuration where the internal position of the cosmic string is changed so that it finally wraps around a non - trivial circle of the compactified space . then a peculiar type of topological number can be defined * on * the strings , which can stabilize the chopped loops .. ] the winding states of the strings look like coils . even though the brane coils are stabilized by their windings , they can be annihilated with another coils that have the opposite windings the cosmological evolution of the conventional necklaces is already investigated in ref.@xcite . therefore , it is quite interesting to investigate the brane necklaces so that we can find the crucial difference from the conventional necklaces . because of their peculiarity , the brane necklaces can be used to distinguish the brane universe from the conventional one . of course the above speculation seems rather naive , therefore we shall discuss this issue in more detail . to make our discussions simple and convincing , here we consider angled brane inflation with a small angle(@xmath10 ) . in this case , the eigenfunction of the tachyonic mode should be localized at the intersection . it should be noted that this mechanism of the localization is * different * from the conventional kibble mechanism , which suggests that our result does not contradict to the previous arguments . moreover , comparing the effective tension of the daughter branes with the one obtained in the effective action , one can easily confirm that the cosmic strings in angled inflation must be extended between mother branes .. ] if the daughter branes are extended between the mother branes as is shown in fig.[fig : flat ] , it is easy to see that the daughter branes can move freely along the mother branes . the flat direction that corresponds to the free motion in the compactified space is depicted in fig.[fig : flat ] . of course , we know that the flat direction is supposed to be lifted by the potential that is due to the conventional supersymmetry breaking , which induces the mass of @xmath11 . on the other hand , since we are considering small angle @xmath10 , the flat direction @xmath12 is periodic and winds about @xmath13 times around the compactified space . therefore , the number of ( local or global ) minima should be as large as @xmath14 . if the potential that lifts the flat direction is shallow , the brane strings can initially move freely along the flat direction . even if the potential is not so shallow due to the stringy effects , one can naturally expect that the collision is so energetic that the daughter branes can stay at any minima . branes located at different sites in the compactified space can not intersect with each other@xcite , because their separation in the compactified space is larger than the string scale . let us consider the @xmath2 dimensional world on the strings . the @xmath2 dimensional world on the strings will have many domains of the number of @xmath15 , which correspond to the degenerated or quasi - degenerated vacua on the flat direction . then the domain walls will appear on the strings , which interpolate between the two neighboring domains .. ] the width of the domain walls is determined by the shape of the effective potential @xmath16 , which becomes @xmath17 if the potential is induced by the conventional mechanism of supersymmetry breaking , or can become as thin as @xmath18 if the potential is lifted by the stringy dynamics . * in the former case , the height of the effective potential is @xmath19 , where @xmath20 denotes the typical distance between the neighboring vacua . from the dimensional argument one can easily understand that the mass of the monopoles is @xmath21 . dimensional world of the strings must have non - zero cosmological constant due to the tension of the string in the background . therefore , the precise formula of the effective potential should contain the effective cosmological constant , which may or may not dominate the vacuum energy . since the cosmological constant in this case is nothing but the tension of the strings , one can easily understand that kinks are point - like objects that are `` skewered '' with the strings . on the other hand , if one considers point - like objects to which two strings are `` attached '' , the definition of the mass should be obscured because of the contribution from the string tension . in this paper , we shall consider the former ( skewered ) picture for fat monopoles . see also fig.[fig : skew ] ] * from the brane perspective the kinks are corresponding to the @xmath0 branes that are extended along the internal direction .. ] the kinks are nothing but the monopoles , which correspond to the @xmath0 branes that are extended in the @xmath22-dimensional compactified space . then one can easily calculate the effective mass of this brane configuration , which becomes @xmath23 . as we have discussed above , there are two different contributions to the mass of the monopoles . when one considers the sqcd - mqcd correspondence , one should think that the potential in the effective action must properly represent the brane dynamics . of course , in such cases the two must agree@xcite . on the other hand , if the effective potential that lifts the flat direction is due to the undefined mechanism of supersymmetry breaking , the results can not agree . therefore , in our case the mass of the monopoles becomes @xmath24 in the limit @xmath25 , while the mass is dominated by the brane tension if @xmath26 . in this paper we shall consider these two different limits without specifying the origin of the effective potential . let us consider the evolution of the network of the brane necklaces . the strings at different domains can not reconnect because they are placed at a distance in the compactified space . therefore , in the length scale that is much larger than the distance between the monopoles , the reconnection probability of the necklaces is suppressed by the number of the minima : @xmath27 the evolution of such strings is already discussed in ref.@xcite . according to ref.@xcite , one can easily understand that the cosmic strings in @xmath10 angled inflation are quite important in the observation of the gravitational wave signals . however , here we must be cautious because the evolution of the network of the necklaces can be different from the conventional @xmath28 strings . therefore , we should first examine the effect of the monopoles that might or might not affect the evolution of the network of the @xmath29 strings . of course , it is apparent that the details must be investigated using numerical simulations , however here we can modestly assume that one can use the conventional analyses as far as the parameter space of the model is within the range of ref.@xcite . to make our arguments clear , we use the same notations as in ref.@xcite . in ref.@xcite , it was assumed that the monopoles are formed * before * the strings are produced at the second phase transition . in our case , however , the monopoles are formed * after * the strings are formed . since the succeeding evolution of the necklaces is irrelevant to how they are formed , the analysis in ref.@xcite is applicable in our case . the sequence of the phase transition is only relevant to their initial condition . for example , if monopoles are produced long after strings are produced , their initial average separation @xmath30 will be large compared to the original analysis of the conventional necklaces . * if the potential is lifted by the conventional mechanism of supersymmetry breaking , the temperature @xmath31 when monopoles are formed can be estimated as @xmath32 . in this case , the width of the monopoles @xmath33 is as large as @xmath17 . therefore , their initial average separation @xmath30 is inevitability as large as the horizon size . * as we have stated above , one can consider steep potential that could be induced by the stringy effects . in this case , monopoles should be produced at the same time when strings are produced , which suggests that the initial average separation between monopoles can range from the minimal distance @xmath34@xcite to the horizon size . here we introduce an important quantity for the evolution of the necklaces , which is the dimensionless ratio @xmath35 . here @xmath36 and @xmath30 denote the tension of the strings and the separation between the monopoles , respectively . in general , the initial value of @xmath37 can be large or small , depending on the nature of the two phase transitions . as we have discussed above , we are considering two different limits . * when monopoles are thin and are dominated by the brane dynamics , the mass of the monopoles is @xmath38 . in this case we can assume @xmath39 , which results in the initial value of the ratio @xmath40 . * on the other hand , if the potential is almost flat and is lifted by the conventional mechanism of supersymmetry breaking , the mass of the monopoles becomes @xmath41 . in this case , since the width of the monopoles is as large as @xmath42 , we must assume @xmath43 . therefore , the initial value of the ratio is @xmath44 . in both cases we can put a modest assumption that @xmath45 . following the arguments in ref.@xcite , we can obtain the following equation for @xmath46 : @xmath47 where the first term on the right hand side describes the string stretching due to expansion of the universe , while the second term describes the competing effect of string shrinking due to gravitational radiation . in this regime , one can find that if @xmath37 is initially small it will grow at least until @xmath48 . as @xmath37 grows , monopole anti - monopole ( or coil anti - coil ) annihilation should become important and the growth of @xmath37 will terminate . * if the potential is steep and the mass of the monopoles is given by @xmath49 , the width of the strings @xmath50 should be comparable to the width of the monopoles @xmath33 . therefore , in this case one can assume that @xmath51 , which leads to the result @xmath52 . this result suggests that as far as @xmath53 the parameter space of our model is within the range of the original argument in ref.@xcite . if @xmath54 , the loop self - intersections should be frequent and their fragmentation into smaller loops is very efficient . therefore , a loop of size @xmath55 typically disintegrates on a time scale @xmath56 which modifies the conventional analysis of the @xmath29 strings . * if the potential is almost flat and the mass of the monopoles is given by @xmath41 , the maximum value of the ratio becomes @xmath57 , which suggests that the monopoles are negligible . in this case , the standard evolution of the string network is applicable . let us recall the main quantities that are used in the standard analysis of the evolution of a string network . the long string network is characterized by the parameters @xmath58 , @xmath59 , and @xmath60 . here @xmath58 is the coherence length , which is defined as the distance beyond which the directions along the string are uncorrelated . @xmath59 is the average distance between the strings , and @xmath60 is the characteristic wavelength of the smallest wiggles that appear on long strings . the length @xmath61 of the chopped - off loops are characterized by the parameter @xmath62 , @xmath63 where the standard value of @xmath62 is determined by the gravitational radiation losses from loops , which is given by @xmath64 where @xmath65 is a numeriacl coefficiend of o(10 ) . if the gravitational radiation from counter - streaming wiggles on long strings is much less efficient in damping the wiggles than ordinarily thought , @xmath62 should be much smaller than @xmath66@xcite . in our case , if the ratio @xmath37 is small , the equation of motion is not different from the original argument , which means that we can examine the evolution of the network just using the results obtained in ref.@xcite . in the macroscopic scale where the reconnection probability is small , one can estimate the above parameters as @xmath67 and @xmath68 . a significant fraction of the total string length within hubble volume should go into loops each hubble time , whose number density is @xmath69 which is precisely the same as the original result . however , what we should show in this paper is the difference from the standard results , which can be used to distinguish the brane necklaces from the conventional defects . in the original argument in ref.@xcite , the reconnection probability of the strings is @xmath29 at any time during the evolution of the network . however , in our case the small reconnection probability is due to the fact that the strings in different vacua can not reconnect . therefore , in the macroscopic scale which is much larger than the distance @xmath30 , the brane necklaces look precisely the same as the original @xmath29 strings , while in the microscopic scale they look quite different . for example , if the wavelength of the wiggles @xmath70 is much smaller than @xmath30 , the production of the small loops is efficient and at least in the local region the brane necklaces look like conventional @xmath71 strings . on the other hand , if the wavelength of the wiggles @xmath70 is much larger than @xmath30 , one can easily understand that the evolution of the brane necklaces is quite similar to the @xmath29 strings . therefore , in the case of @xmath72 , we can conclude that the evolution of the brane necklaces is precisely the same as the conventional @xmath73 strings as far as @xmath74 is satisfied . the above relation ( [ wigled ] ) puts a cut - off time for the original analysis of the @xmath28 strings , which is not important for the gw burst analysis at the present cosmic time . the numerical simulations of this kind are difficult , because of the two hierarchically different scales . however , what is important for the observation of the gw emitted from strings is the behaviour at the macroscopic scale . therefore , it must be important to note that at later period the behaviour of the brane necklaces is quite the same as the conventional @xmath29 strings , which is already discussed in ref.@xcite . thus we can conclude that the brane necklaces can have significant impact on the observation of the gravitational wave . please be sure that in our model the small reconnection probability is due to the small angle . our next example is the peculiar case of @xmath75 . in this case , the characteristic length scale of the network will be modified@xcite as : @xmath76 the string length per unit volume is @xmath77 . the number of loops formed per hubble volume per hubble time is therefore@xcite @xmath78 the loop self - intersections should be frequent and their fragmentation into smaller loops is very efficient , which suggests that the lifetime of a loop of size @xmath55 is @xmath79@xcite . in this case , it is easy to see that the loops produced by the necklaces of @xmath75 are short - lived and the calculation of the loop density is similar to the cases of small @xmath62 . the corresponding loop density is therefore @xmath80 here we have assumed that the wiggles are small compared to the typical scale of the network , so that @xmath81 . the results obtained here suggests that the analysis of the gw signals emitted by the network of the cosmic necklaces are similar to the original argument@xcite . the difference that we have found is the suppression factor @xmath82 in their loop density . therefore , we may conclude that there is no qualitative difference between the brane necklaces and the conventional strings(or necklaces ) as far as the observation of the gravitational wave signals is concerned . however , as we have discussed above , chopped loops of the brane necklaces can be stabilized due to their non - zero windings around compactified space . the stable relics of such loops are superheavy , which must satisfy the cosmological bound of the conventional cold dark matter . therefore , in order to find qualitative differences , it is quite important to calculate the relic density of the brane coils . to calculate the relic density of the stable relics , it is convenient to introduce the relative abundance @xmath83 , where @xmath84 is the entropy density . in our case , small loops are produced through multiple self - interactions of larger loops , whose sizes are determined by the wavelength of the wiggles @xmath85 . the conventional analysis of @xmath86 string network is reliable as far as @xmath87 . * let us first consider the case where the effective potential @xmath16 is due to the stringy dynamics . in this case , monopoles are produced by the phase transition just after angled inflation . the number of the loops formed per hubble volume per hubble time is already given in ref.@xcite . for @xmath88 , it becomes @xmath89 which cooresponds to the number of the coils emitted from the network of the brane necklaces . therefore , we can write @xmath90 in the most efficient case when all the loops are stabilized by the coils , the relic number density of the coils is obtained by integrating eq.([dotn ] ) . the ratio @xmath91 is : @xmath92 for @xmath93 , and @xmath94 for @xmath75 . here @xmath95 is the time when the monopoles are produced as the kinks on the strings , and @xmath96 is defined by @xmath97 , where @xmath98 is the cut - off time when the wavelength of the wiggles becomes much larger than the typical distance between monopoles . we have assumed that the coils produced before @xmath98 are negligible . * we must be careful if the effective potential @xmath16 is lifted by the conventional mechanism of supersymmetry breaking . in this case , since the sizes of the monopoles are comparable to the hubble radius when they are formed , initially the wiggles can not form coils . the coils are formed at much later period when @xmath99 , which means @xmath100 . as we have discussed above , a rough estimation of the characteristic properties of the networks of the brane necklaces is straightforward . their evolution can be approximated by the evolution of the conventional @xmath29 strings or the necklaces . on the other hand , one may suspect that the `` vacua '' on the strings might not be degenerated , and the brane necklaces might not behave as we have discussed above . we think this claim could be true . if the vacua on the strings are only quasi - degenerated , the tension of the strings in the different domains will be different by a small factor @xmath101 . due to @xmath102 , the monopoles may move along the strings , and may be annihilated or gathered to form coils on the strings . after the `` true vacuum '' dominates the necklaces , the reconnection probability of the necklaces will be enhanced up to @xmath103 , because the other `` false vacua '' have shrunk to a point . then the situation becomes quite the same as the evolution of the conventional @xmath71 necklaces@xcite . the only difference is that the monopoles are replaced by the coils , which can stabilize the loops of the strings . in this paper we have investigated the cosmological evolution of the cosmic strings in angled inflation . we have shown that the cosmic strings in angled inflation turn out to be the brane necklaces . the evolution of the networks of the brane necklaces is similar to the conventional @xmath29 strings . however , the loops of the brane necklaces will shrink to produce superheavy winding states , which look like coils . in angled inflation , it is therefore possible to distinguish the brane necklaces from the conventional cosmic strings and necklaces . moreover , the reconnection probability of the brane necklaces can be enhanced up to @xmath104 if the degeneracy of the vacua is resolved by a small perturbation . we have made a rough estimation of the characteristic properties of the network of the brane necklaces . it is obvious that the numerical simulations are required to obtain more sensible results . here we should discuss what we can say about the conventional brane strings in the simplest scenario of @xmath105 brane inflation . as is already discussed in ref.@xcite , it seems natural to think that the position of the daughter branes in the compactified space should be stabilized . on the other hand , however , there is no sensible reason that we have to believe that the potential has only a unique global minimum in the compactified space . therefore , it is always important to consider daughter branes that have multiple minima in the compactified space . as a result , it is always natural to consider cosmic strings that have many domains on their world volume , which induces kinks interpolating between them . even if the flat directions are stabilized at a unique minimum , winding states ( coils ) can appear if the brownian motion induces windings around a nontrivial circle in the compactified space . naively , the nontrivial circle means a conventional noncontractible circle that winds around compactified space . in generic cases , however , the nontrivial circle could be a valley of the potential that winds around a hill . it is obvious that * any * branes that are produced by brane collision can have flat directions in the internal space . then due to the brownian motion , the cosmic strings will become the brane necklaces or the brane coils . the brane necklaces can therefore be produced in other generic brane models . if the necklaces are produced , they can be used to probe the internal structure of the compactified space . moreover , future numerical simulations will provide us the important information about the undetermined parameters , which may put serious bounds on brane inflation . we wish to thank k.shima for encouragement , and our colleagues in tokyo university for their kind hospitality . 1 i. antoniadis , n. a - hamed , s. dimopoulos , and g. r. dvali , _ new dimensions at a millimeter to a fermi and superstrings at a tev , phys.lett.b436(1998)257_ [ hep - ph/9804398 ] ; i. antoniadis , _ a possible new dimension at a few tev , phys.lett.b246(1990)377_ ; n. a - hamed , s. dimopoulos and g. r. dvali , _ the hierarchy problem and new dimensions at a millimeter , phys.lett.b429(1998)263_ [ hep - ph/9803315 ] . p. kanti and k. a. olive , _ on the realization of assisted inflation , phys . rev . d60(1999)043502 _ [ hep - ph/9903524 ] ; p. kanti and k. a. olive , _ assisted chaotic inflation in higher dimensional theories , phys . b464(1999)192 _ [ hep - ph/9906331 ] ; n. arkani - hamed , s. dimopoulos , n. kaloper , and j. march - russell , _ rapid asymmetric inflation and early cosmology in theories with submillimeter dimensions , nucl.phys.b567(2000)189_ [ hep - ph/9903224 ] ; r. n. mohapatra , a. perez - lorenzana , and c. a. de s. pires , _ inflation in models with large extra dimensions driven by a bulk scalar field , phys.rev.d62(2000)105030_ [ hep - ph/0003089 ] ; a. m. green and a. mazumdar , _ dynamics of a large extra dimension inspired hybrid inflation model , phys.rev.d65(2002)105022_ [ hep - ph/0201209 ] ; d. h. lyth , _ inflation with tev scale gravity needs supersymmetry , phys.lett.b448(1999)191_ [ hep - ph/9810320 ] ; t. matsuda , _ kaluza - klein modes in hybrid inflation , phys.rev.d66(2002)107301_ [ hep - ph/0209214 ] ; t. matsuda , _ successful d term inflation with moduli , phys.lett.b423(1998)35_ [ hep - ph/9705448 ] . t. matsuda , _ topological hybrid inflation in brane world , jcap 0306(2003)007 _ [ hep - ph/0302204 ] ; t. matsuda , _ q ball inflation , phys.rev.d68(2003)127302_ [ hep - ph/0309339 ] . t. matsuda , _ thermal hybrid inflation in brane world , phys.rev.d68(2003)047702_ [ hep - ph/0302253 ] t. matsuda , _ f term , d term and hybrid brane inflation , jcap 0311(2003)003 _ [ hep - ph/0302078 ] t. matsuda , _ nontachyonic brane inflation , phys.rev.d67(2003)083519_ [ hep - ph/0302035 ] g. r. dvali , g. gabadadze , _ nonconservation of global charges in the brane universe and baryogenesis , e phys.lett.b460(1999)47_ [ hep - ph/9904221 ] ; a. masiero , m. peloso , l. sorbo , and r. tabbash , _ baryogenesis versus proton stability in theories with extra dimensions , phys.rev.d62(2000)063515_ [ hep - ph/0003312 ] ; a. pilaftsis , _ leptogenesis in theories with large extra dimensions , phys.rev.d60(1999)105023_ [ hep - ph/9906265 ] ; r. allahverdi , k. enqvist , a. mazumdar and a. perez - lorenzana , _ baryogenesis in theories with large extra spatial dimensions , nucl.phys . b618(2001)377 _ [ hep - ph/0108225 ] ; s. davidson , m. losada , and a. riotto , _ a new perspective on baryogenesis , phys.rev.lett.84(2000)4284_ [ hep - ph/0001301 ] . t. matsuda , _ baryon number violation , baryogenesis and defects with extra dimensions , phys.rev.d66(2002)023508_ [ hep - ph/0204307 ] ; t. matsuda , _ activated sphalerons and large extra dimensions , phys.rev.d66(2002)047301_ [ hep - ph/0205331 ] ; t. matsuda , _ enhanced baryon number violation due to cosmological defects with localized fermions along extra dimensions , phys.rev.d65(2002)107302_ [ hep - ph/0202258 ] ; t. matsuda , _ defect mediated electroweak baryogenesis and hierarchy , j.phys.g27(2001)l103_ [ hep - ph/0102040 ] . t. matsuda , _ hybridized affleck - dine baryogenesis , phys.rev.d67(2003)127302_ [ hep - ph/0303132 ] ; t. matsuda , _ affleck - dine baryogenesis after thermal brane inflation , phys.rev.d65(2002)103501_ [ hep - ph/0202209 ] ; t. matsuda , _ affleck - dine baryogenesis in the local domain , phys.rev.d65(2002)103502_ [ hep - ph/0202211 ] ; t. matsuda , _ electroweak baryogenesis mediated by locally supersymmetry breaking defects , phys.rev.d64(2001)083512_ [ hep - ph/0107314 ] ; t. matsuda , _ curvaton parindent can accommodate multiple low inflation scales , class.quant.grav.21(2004)l11_ [ hep - ph/0312058 ] . t. matsuda , _ brane q ball , branonium and brane q ball inflation _ [ hep - ph/0402223 ] . t. matsuda , _ formation of monopoles and domain walls after brane inflation , jhep 0410(2004)042 _ [ hep - ph/0406064 ] t.matsuda , _ formation of cosmological brane defects _ [ hep - ph/0402232 ] . n. barnaby , a. berndsen , j. m. cline and h. stoica , _ overproduction of cosmic superstrings _ [ hep - th/0412095 ] . n. jones , h. stoica , and s. h. h. tye , _ brane interaction as the origin of inflation , jhep 0207(2002)051 _ [ hep - th/0203163 ] ; s. sarangi , s. h. h. tye , _ cosmic string production towards the end of brane inflation , phys.lett.b536(2002)185_ [ hep - th/0204074 ] ; l. pogosian , s. h. h. tye , i. wasserman and m. wyman , _ observational constraints on cosmic string production during brane inflation , phys.rev.d68(2003)023506_ [ hep - th/0304188 ] ; m. gomez - reino and i. zavala , _ recombination of intersecting d - branes and cosmological inflation , jhep0209(2002)020 _ [ hep - th/0207278 ] . g. dvali , r. kallosh and a. van proeyen , _ d term strings , jhep 0401(2004)035 _ [ hep - th/0312005 ] ; e j. copeland , r. c. myers and j. polchinski , _ cosmic f and d strings _ [ hep - th/0312067 ] . n. dorey , t. j. hollowood and d. tong , g. r. dvali , i. i. kogan and m. a. shifman , _ topological effects in our brane world from extra dimensions , phys.rev.d62(2000)106001_ [ hep - th/0006213 ] . t. matsuda , _ incidental brane defects , jhep 0309(2003)064 _ [ hep - th/0309266 ] . m. eto , y. isozumi , m. nitta , k. ohashi and n. sakai , _ instantons in the higgs phase _ [ hep - th/0412048 ] ; y. isozumi , m. nitta , k. ohashi and n. sakai , _ all exact solutions of a 1/4 bogomolnyi - prasad - sommerfield equation _ [ hep - th/0405129 ] . n.arkani-hamed , m.schmaltz , _ hierarchies without symmetries from extra dimensions , phys.rev.d61(2000)033005_ [ hep - ph/9903417 ] . g. r. dvali and s. h. henry tye _ brane inflation , phys.lett.b450(1999)72_ [ hep - ph/9812483 ] . c. herdeiro , s. hirano and r. kallosh , _ string theory and hybrid inflation / acceleration , jhep0112(2001)027 _ [ hep - th/0110271 ] ; k. dasgupta , c. herdeiro , s. hirano and r. kallosh , _ d3 / d7 inflationary model and m theory , phys.rev.d65(2002)126002_ [ hep - th/0203019 ] , j. garcia - bellido , r. rabadan and f. zamora , _ inflationary scenarios from branes at angles , jhep 0201(2002)036 _ [ hep - th/0112147 ] . a. sen , _ rolling tachyon , jhep 0204(2002)048 _ [ hep - th/0203211 ] . g. dvali and a. vilenkin , _ formation and evolution of cosmic d strings _ [ hep - th/0312007 ] . t. matsuda , _ string production after angled brane inflation , phys.rev.d70(2004)023502_ [ hep - ph/0403092 ] . k. hashimoto and s. nagaoka , _ recombination of intersecting d - branes by local tachyon condensation , jhep 0306(2003)034 _ [ hep - th/0303204 ] . g. aldazabal , s. franco , l. e. ibanez , r. rabadan , a.m. uranga , _ intersecting brane worlds , jhep 0102(2001)047 _ [ hep - ph/0011132 ] ; r. blumenhagen , d. lust , s. stieberger , _ gauge unification in supersymmetric intersecting brane worlds , jhep 0307(2003)036 _ [ hep - th/0305146 ] ; i. r. klebanov , e. witten , _ proton decay in intersecting d - brane models , nucl.phys.b664(2003)3-20_ [ hep - th/0304079 ] ; m. cvetic , p. langacker , j. wang , _ dynamical supersymmetry breaking in standard - like models with intersecting d6-branes , nucl.phys.b642(2002)139_ [ hep - th/0303208 ] ; m. cvetic , i. papadimitriou , _ more supersymmetric standard - like models from intersecting d6-branes on type iia orientifolds , phys.rev.d67(2003)126006_ [ hep - th/0303197 ] ; s.a . abel , a.w . owen , _ interactions in intersecting brane models , nucl.phys.b663(2003)197_ [ hep - th/0310257 ] ; d. bailin , g.v . kraniotis , a. love , _ standard - like models from intersecting d5-branes , phys.lett.b547(2002)43_ [ hep - th/0210219 ] ; c. f. doran , m. faux , _ intersecting branes in m - theory and chiral matter in four dimensions , jhep 0208(2002)024 _ [ hep - th/0208030 ] ; c. kokorelis , _ exact standard model compactifications from intersecting branes , jhep 0208(2002)036 _ [ hep - th/0206108 ] ; c. kokorelis , _ new standard model vacua from intersecting branes , jhep 0209(2002)029 _ [ hep - th/0205147 ] ; d. cremades , l.e . ibanez , f. marchesano , _ standard model at intersecting d5-branes : lowering the string scale , nucl.phys.b643(2002)93_ [ hep - th/0205074 ] . g. r. dvali , _ infrared hierarchy , thermal brane inflation and superstrings as superheavy dark matter , phys.lett.b459(1999)489_ [ hep - ph/9905204 ] . t. matsuda , _ weak scale inflation and unstable domain walls , phys.lett.b486(2000)300_ [ hep - ph/0002194 ] ; t. matsuda , _ on cosmological domain wall problem in supersymmetry models , phys.lett.b436(1998)264_ [ hep - ph/9804409 ] ; t. matsuda , _ domain wall solution for vector - like model , phys.lett.b423(1998)40_ [ hep - ph/9710230 ] . m. g. jackson , n. t. jones and j. polchinski , _ collisions of cosmic f- and d - strings _ [ hep - th/0405229 ] , j. polchinski , _ cosmic superstrings revisited _ [ hep - th/0410082 ] , t. w. b. kibble , _ cosmic strings reborn ? _ [ astro - ph/0410073 ] . t. damour and a. vilenkin , _ gravitational radiation from cosmic ( super)strings : bursts , stochastic background , and observational windows _ [ hep - th/0410222 ] , a. avgoustidis and e. p. s. shellard , _ cosmic string evolution in higher dimensions _ [ hep - ph/0410349 ] . v. berezinsky and a. vilenkin , _ cosmic necklaces and ultrahigh energy cosmic rays , rev.lett.79(1997)5202_ [ astro - ph/9704257 ] . amihay hanany and david tong , _ vortex strings and four - dimensional gauge dynamics , jhep 0404(2004)066 _ [ hep - th/0403158 ] x. siemens and k. d. olum , _ gravitational radiation and the small - scale structure of cosmic strings , nucl.phys.b611(2001)125_ [ gr - qc/0104085 ] ; x. siemens , k. d. olum and a. vilenkin , _ on the size of the smallest scales in cosmic string networks , phys.rev . d66 ( 2002 ) 043501 _ [ gr - qc/0203006 ] .

soon after the experimental discovery @xcite of carbon nanotubes , it was suggested @xcite that there might also exist carbon _ nanotori _ , carbon nanotubes in which the ends of the tube are identified and sewn together . within a few years , experimental evidence for such structures emerged @xcite , and since then they have attracted a great deal of experimental @xcite and theoretical @xcite attention . there are a variety of reasons for this intense interest . for example , certain species of carbon nanotori exhibit unusual magnetic properties @xcite , including persistent magnetic moments at nearly zero flux @xcite and colossal paramagnetic moments @xcite . these objects can also display a diverse variety of electric properties : some nanotori are inherently metallic , while others are semiconducting and still others are insulators . as we shall discuss , the most general nanotorus can be parametrized by four integers @xmath0 . together and in various combinations , these describe the radius of the underlying nanotube , the chirality of the underlying nanotube , the length of the underlying nanotube , and the relative twist @xcite that might occur upon sewing opposite ends of the tube together to form the torus . the important point , however , is that nanotori with different @xmath0 are fundamentally physically distinct : they have different sizes , different shapes , and different twisted `` honeycomb '' patterns of carbon atoms laid out on their surfaces . of course , there are certain symmetries of the underlying graphene sheet which lead to trivial equivalences amongst these nanotori . for example , @xmath1 rotations of the underlying graphene sheet will produce identical nanotori . such nanotori are therefore not physically distinct . in this paper , however , we shall demonstrate that there are additional symmetries that relate _ physically distinct _ nanotori to each other , forcing such nanotori to exhibit identical energy spectra and electrical properties . these so - called `` modular '' symmetries therefore transcend the traditional hexagonal lattice symmetries of the graphene sheet , and arise solely in the process of compactifying the graphene sheet in order to form the nanotorus . in some sense , the appearance of these modular symmetries is entirely expected , for they result directly from the geometry of the compactification . however , to the best of our knowledge , the significance and consequences of these symmetries have not been fully appreciated thus far in the nanotorus literature . as we shall demonstrate , these symmetries have a number of profound effects . first , we shall see that use of these symmetries allows us to partition the set of carbon nanotori into distinct equivalence classes as far as their spectral properties are concerned , and leads to a dramatic reduction in the numbers of _ spectrally distinct _ carbon nanotori as compared with the numbers of _ physically distinct _ nanotori . moreover , as we shall show , many of the standard rules of thumb advanced in the literature in order to describe the conductivity properties of these nanotori actually fail to respect these symmetries . such rules of thumb are therefore incomplete as descriptions of the physics of these nanotori , and must be replaced by statements which respect the full symmetry structure of the compactified graphene sheet . finally , we demonstrate that these symmetries even extend to situations in which the carbon nanotori enclose different types of magnetic flux . they thus should have applicability for many of the fascinating magnetic properties of carbon nanotori , including the possibility of persistent currents . in order to explain the origins of these spectral symmetries , we begin with a brief review which will also serve to highlight our notation and conventions . 3.0 truein for the uncompactified graphene sheet , all electron wavevectors @xmath9 are allowed . however , this situation changes when we `` roll up '' one dimension of the graphene sheet to produce a single - walled carbon nanotube , or equivalently when we identify any two carbon atoms on the graphene sheet whose positions differ by an arbitrary lattice vector @xmath14 with @xmath15 . imposing the bloch condition on the electron wavefunctions @xmath16 in addition to the new periodicity condition @xmath17 then restricts @xmath18 to the set of wavevectors satisfying the condition @xmath19 where @xmath20 and @xmath21 . these allowed values of @xmath22 therefore form parallel lines in the @xmath23 plane , and the locations at which these lines intersect the bragg planes in fig . [ fig : energycontours ] determine whether the corresponding @xmath24 nanotube is metallic , semiconducting , or insulating . nanotubes for which @xmath25 ( , @xmath26 ) or @xmath27 ( , @xmath28 ) are dubbed `` zigzag '' or `` armchair '' respectively ; nanotubes with other values of @xmath24 are generally referred to as `` chiral '' . in general , the vector @xmath29 perpendicular to @xmath30 is the tube axis . 2.15 truein we now consider imposing _ two _ non - parallel identifications on the graphene sheet , as illustrated in fig . [ fig : torusparam ] . in general , we consider two arbitrary lattice identification vectors @xmath31 and @xmath32 ; without loss of generality we shall assume that @xmath33 . we shall also assume without loss of generality that the quantity @xmath34 is positive , or equivalently that the relative angle @xmath35 from @xmath30 to @xmath36 lies in the range @xmath37 , as illustrated in fig . [ fig : torusparam ] . such identifications together result in a so - called @xmath0 nanotorus which may be viewed as a @xmath24 nanotube in which opposite ends are joined to each other , potentially with a relative twist angle @xmath35 . this description is especially appropriate if @xmath38 . note that @xmath39 describes the number of hexagonal cells which tile the surface of the resulting donut , and is equal to precisely half the number of carbon atoms in the nanotorus . imposing the double periodicity condition @xmath40 on the electron wavefunction @xmath16 in conjunction with the bloch condition then leads to the constraint in eq . ( [ eq : tubek ] ) as well as the additional constraint @xmath41 where @xmath42 . solving eqs . ( [ eq : tubek ] ) and ( [ eq : tubek2 ] ) simultaneously then yields @xmath43 \nonumber\\ k_y & = & \frac{2\pi}{l_1l_2\sin\theta } \left[-\ell_1 l_2\cos\delta + \ell_2 l_1\cos\beta\right ] \label{eq : torik}\end{aligned}\ ] ] or equivalently @xmath44\nonumber\\ k_y & = & { 2\pi \over \sqrt{3 } n_{\rm hex } } \left [ -\ell_1 ( 2p + q ) + \ell_2 ( 2m+n ) \right]~. \label{eq : torik2}\end{aligned}\ ] ] thus , we see that the allowed electron wavevectors now form a two - dimensional grid of points in the @xmath23 plane . note that there are always exactly @xmath39 allowed wavevectors lying inside the fundamental region bounded by the bragg planes in fig . [ fig : energycontours ] , where points lying on a single bragg plane itself are counted as half and where points lying at the intersections of two bragg planes are counted as a third . whether such nanotori are metallic , semiconducting , or insulating depends on whether these points hit or come particularly close to the intersections of the bragg planes in fig . [ fig : energycontours ] . it is found that nanotori in which both @xmath45 and @xmath46 are multiples of three are metallic , with solutions to eqs . ( [ eq : torik ] ) and ( [ eq : torik2 ] ) that lie precisely on the fermi surface . for practical applications , it proves useful to focus on only those nanotori with @xmath47 , as this condition allows one to bend the nanotube into a nanotorus without excessive strain or deformation of the underlying graphene sheet ( the effects of which would otherwise distort the dispersion relation in fig . [ fig : energycontours ] , and hence the band structure of the torus ) . within this limit , it is then conventional to regard nanotori with @xmath48 ( mod 3 ) but @xmath49 ( mod 3 ) as semiconducting , since the allowed wavevectors come `` close '' to the fermi surface in this limit . all other nanotori are then considered insulating . all of the above results are completely standard , and are well known in the carbon - nanotorus literature . in particular , tori with different values of @xmath0 are physically distinct : they have entirely different arrangements of hexagons tiling their surfaces , with different values of the physical radii @xmath50 , chiral angle @xmath51 , and twist angle @xmath35 . we shall take this to be our definition of `` physically distinct '' . there are , of course , certain trivial identifications which relate different nanotori to each other : for example , the @xmath0 torus and the @xmath52 torus are actually identical , since they correspond to @xmath53 pairs which are related to each other by a uniform @xmath1 rotation . such trivial identifications reflect the underlying hexagonal lattice symmetries of the graphene sheet from which these nanotori are constructed , and result in identical carbon nanotori with identical patterns of carbon atoms on their surfaces . such nanotori are therefore not physically distinct . by contrast , an important question is whether there exist physically _ distinct _ nanotori ( , tori which are _ not _ related by symmetries of the hexagonal lattice ) which nevertheless yield identical grids of allowed wavevectors @xmath23 . if so , we will have found cases of physically distinct tori with different values of @xmath54 which are nevertheless spectrally identical . in other words , such tori will have identical energy spectra and electrical conducting properties . at first sight , it might appear that no such spectral equivalences exist for carbon nanotori . after all , the results in eq . ( [ eq : torik ] ) for carbon nanotori may initially appear to be nothing more than a two - dimensional generalization of the results in eq . ( [ eq : tubek ] ) for carbon nanotubes , and no such spectral equivalences exist for carbon nanotubes . however , it is important to realize that this is not true for carbon nanotori . in particular , it turns out that the set of allowed values of @xmath23 in eq . ( [ eq : torik2 ] ) is actually invariant under two additional symmetry transformations which we shall denote @xmath55 and @xmath56 : @xmath57 under these transformations , it is straightforward to demonstrate that @xmath34 is invariant , while the physical parameters @xmath58 , @xmath59 , @xmath35 , and @xmath51 transform in the following manner : @xmath60 it is important to stress that the individual equations in eqs . ( [ eq : torik ] ) and ( [ eq : torik2 ] ) are not invariant under @xmath55 or @xmath56 ; rather , what is invariant is the _ set _ of values of @xmath23 to which these equations lead . moreover , since these sets of solutions for @xmath23 are invariant under @xmath55 and @xmath56 individually , they are also invariant under any sequence of @xmath55 and @xmath56 transformations . for example , under the @xmath61 transformation we find @xmath62 , and this too is a symmetry of the solutions to eq . ( [ eq : torik2 ] ) . it then follows from these observations that any two tori whose defining vectors @xmath30 and @xmath36 differ through @xmath55 and @xmath56 transformations share the same electronic spectra but are intrinsically different from each other , that they are isospectral but physically distinct . that they are isospectral follows from the fact that the same set of @xmath23 solutions are selected in each case . by contrast , that they are physically distinct follows from the fact that the fundamental identifications between carbon atoms on the graphene sheet are altered by @xmath55 and @xmath56 transformations in a manner that transcends trivial hexagonal lattice symmetries . this is perhaps easiest to see in the case of the @xmath56 transformation , which corresponds to the action @xmath63 . this changes not only @xmath59 but @xmath35 , and thus produces a new torus which has a greater `` twist '' when the ends of the nanotube are joined . it may be less obvious that the @xmath55 transformation also connects physically distinct tori . indeed , this transformation corresponds to the action @xmath64 , and at first glance it might appear that this is merely a trivial relabeling of the two independent periodicities , along with a reflection ( sign flip ) . however , we must remember that the @xmath55 transformation also changes the associated @xmath35 and @xmath51 angles in non - trivial ways . alternatively , we can also appreciate the non - trivial nature of @xmath55 by considering the combination t s t^-1 s^-1 = t s t [ tprimedef ] which corresponds to the action @xmath65 . [ the second equality in eq . ( [ tprimedef ] ) follows as a result of the identity @xmath66 , or equivalently @xmath67 . ] note that @xmath68 is in some sense `` dual '' to @xmath56 : each performs a full twist around a different cycle of the torus . in other words , as illustrated in fig . [ tori ] , while @xmath56 corresponds to cutting through one side of the torus and re - attaching the two edges along with a twist , @xmath68 corresponds to cutting the torus along the other cycle and then twisting along the cut in an analogous manner before attaching the newly adjacent bonds . thus , @xmath68 results in a physically distinct torus , just as @xmath56 does , and furthermore these resulting tori are not related to each other through hexagonal lattice symmetries . it then follows from eq . ( [ tprimedef ] ) that @xmath55 also can not correspond to a lattice symmetry , @xmath55 must connect physically distinct tori as well . indeed , @xmath55 and @xmath68 are interchangeable in the sense that the two generators of the modular group can be considered to be either @xmath69 or @xmath70 . 3.5 truein note that if we define the complex quantity @xmath71 , then the @xmath55 and @xmath56 transformations correspond to @xmath72 and @xmath73 respectively . together , these transformations generate the so - called `` modular group '' , which is one of the primordial symmetries associated with toroidal compactifications . in general , any transformation which takes the form @xmath74 where @xmath75 and @xmath76 is a modular transformation . what we have shown , then , is that any two carbon nanotori whose defining parameters are related by a modular transformation are spectrally identical even though they are physically distinct . in other words , any transformation which can be generated through repeated actions of the @xmath55 and @xmath56 generators leads to a spectral equivalence between physically distinct tori . it is important to emphasize that the appearance of these modular symmetries is not a total surprise . modular symmetries often arise in the presence of toroidal compactifications , and are a direct consequence of the underlying geometry of the carbon nanotorus . indeed , the zone - folding of a macroscopic lattice necessarily introduces periodic equivalences of the sort discussed here , and in the case of two dimensions with non - parallel identifications , such periodicities generically lead to modular symmetries . in other words , the underlying toroidal graph is not altered by modular symmetries , and thus the adjacencies of the graph ( and the isospectrality which results ) are preserved intact . however , when applied to the case of carbon nanotori , these symmetries are _ physical _ , in the sense that the nanotori they relate are isospectral but physically distinct . as we shall see , this is particularly significant , allowing these symmetries to have profound effects on the physics of these nanotorus systems and in particular on the range of behaviors they may exhibit . in a sense , these modular symmetries can be viewed as a generalization of a similar set of symmetries that emerge for mbius graphs . to see this , let us imagine taking a long strip of hexagons and gluing opposite ends of this strip together with an arbitrary number of half - twists . at first glance , one might suspect that the resulting spectrum should depend on the actual number of half - twists . however , it turns out that the resulting spectrum depends only on whether this number is even or odd @xcite . this mbius case is , of course , a relatively trivial example of this phenomenon , since it deals with only a single possible kind of twist corresponding to the single periodicity that defines the mbius strip . indeed , for the case of carbon nanotori , our point is that there exists a whole _ group _ of transformations the modular group which is built upon the actions of _ two _ non - commuting underlying generators ( @xmath55 and @xmath56 ) which can be combined and performed in a variety of inequivalent ways . this leads to `` sewing '' and `` gluing '' configurations on the torus which are much more complex than on the mbius strip , and which require greater mathematical machinery to study and this in turn leads to distinctions between the notions of `` physically distinct '' and `` spectrally distinct '' which are even more pronounced than in the mbius case . however , even the relatively simple mbius case illustrates one of our basic themes : there can be lots of physically distinct twisted hexagon strips , yet very few of them are spectrally distinct . it is the preservation of the underlying graph in each case which makes this possible . as a result of these modular symmetries , we can henceforth establish an unambiguous convention for uniquely describing the spectral properties of a given carbon nanotorus : we calculate its complex parameter @xmath77 as defined above , and then use a sequence of @xmath55 and @xmath56 modular transformations as needed in order to bring @xmath77 into a special region of the complex @xmath77-plane known as the `` fundamental domain '' @xmath78 defined by : -0 , ||1where @xmath79 and @xmath80 . note that the condition @xmath81 already ensures that @xmath82 . we can then use the underlying symmetries of the hexagonal graphene lattice in order to bring the resulting angle @xmath51 into the range @xmath83 . note that this convention is tantamount to describing the spectrum of a given carbon nanotorus in terms of that spectrally equivalent nanotorus which is as close to being rectangular as possible . following this procedure therefore provides an unambiguous test of whether any two physically distinct carbon nanotori in fact have the same spectral properties . the existence of these modular symmetries and the spectral equivalences they induce has profound implications for the physics of carbon nanotori . indeed , as we shall see , the use of these modular symmetries will sharpen our ability to classify nanotori on the basis of those physical properties ( such as their metallicities , _ etc . _ ) which ultimately stem from their band structures . furthermore , use of these modular symmetries can even provide _ statistical _ insights into the properties of randomly produced carbon nanotori insights which may ultimately have practical consequences for the manufacture of such objects in environments in which their resulting properties often can not be controlled or engineered in advance . as discussed above , the first and most direct implication of these modular symmetries is that there exist physically distinct nanotori often with markedly different radii and chiral angles which nevertheless possess identical energy spectra and conductivity properties . as an example of this phenomenon , let us consider the @xmath84 , @xmath85 , and @xmath86 carbon nanotori . clearly , we see that these nanotori are physically distinct and are characterized by different sets of cycle lengths @xmath50 , different chiral angles @xmath51 , and different twist angles @xmath35 . as a result , these three tori have entirely different patterns of carbon atoms tiling their surfaces . yet , the latter two nanotori are in fact related to each another by @xmath55 and @xmath56 transformations , and consequently they each yield the same spectrum of allowed @xmath22 vectors shown in fig . [ fig : energyspecthreetori ] . similarly the first nanotorus is related to the second two through not only @xmath55 and @xmath56 transformations , but also trivial 60@xmath87 rotations of the underlying graphene sheet . thus , these tori all share identical spectra , and according to the convention specified above , we can describe this spectrum uniquely as having @xmath88 , @xmath89 , and @xmath90 . 3.0 truein second , as a corollary to this observation , we also learn that the usual notions of `` zigzag '' and `` armchair '' notions which are critical for describing the chirality and metallicity of carbon nanotubes no longer hold as special indicators of the spectral properties of corresponding carbon nanotori . in other words , nanotori built from zigzag or armchair nanotubes might have spectral properties which are identical to those of nanotori built from non - zigzag or non - armchair nanotubes ; moreover , these spectral properties might or might not correspond to the special zigzag or armchair angles @xmath91 . as an example , the @xmath92 nanotorus is built from a zigzag nanotube while the @xmath93 nanotorus is built from an armchair nanotube and the @xmath94 and @xmath95 nanotori are built from chiral nanotubes with different chiralities . yet all four nanotori have identical spectral properties which are the same as those of a chiral nanotorus with @xmath96 , @xmath97 , and @xmath98 ( or @xmath99 ) . indeed , using the mathematical results of ref . @xcite , we can show that any carbon nanotorus is spectrally equivalent to one built from a zigzag carbon nanotube . conversely , a carbon nanotorus can have a spectrum which exhibits a zigzag or armchair property that is lacking in the original nanotube from which it is constructed . as an example , the @xmath100 nanotorus is built from a zigzag nanotube and the @xmath101 nanotorus is built from a chiral nanotube . yet both nanotori have the spectral properties of an armchair nanotorus , with @xmath96 , @xmath102 , and @xmath103 . likewise , the @xmath104 nanotorus is built from an armchair nanotube and the @xmath105 nanotorus is built from a chiral nanotube . yet both nanotori have the spectral properties of a zigzag nanotorus , with @xmath96 , @xmath106 , and @xmath107 . third , it turns out that not every possible spectral signature @xmath108 can be realized , even in principle . instead , these three quantities experience internal constraints and correlations which ultimately reflect the fixed hexagonal lattice structure of the underlying graphene sheet and which require that any allowed spectral signature @xmath108 originate from four real integers @xmath0 . these correlations amongst @xmath108 can take a variety of different forms . for example , for odd values of @xmath39 , it turns out that there exist no self - consistent spectral signatures with purely imaginary values of @xmath77 . similarly , for @xmath109 , there are solutions with @xmath110 , but no other inverse integers ; likewise , all of these except for @xmath111 have @xmath26 . in fact , these internal constraints amongst the four real parameters embodied in @xmath112 are sufficiently strong that they effectively eliminate one real degree of freedom from within this parametrization . in other words , although four real degrees of freedom are required in order to parametrize a given physically distinct nanotorus , only three real degrees of freedom are required in order to uniquely describe its spectral properties @xcite . it is easy to understand why such constraints arise . ordinarily , as a question of topology , tori can exist with all shapes and volumes , for there are literally an infinite number of ways in which we can construct a torus by rolling up an unmarked sheet of paper . however , in the case of carbon nanotori , our original `` sheet of paper '' is not unmarked : it is actually a graphene sheet of carbon atoms which has its own hexagonal lattice structure . the existence of such a lattice structure has a number of critical consequences : it forces our toroidal defining vectors @xmath30 to @xmath36 to be _ lattice vectors _ ; it restricts the resulting possible combinations of @xmath39 and @xmath77 to those values consistent with the periodicity of the lattice ; and it breaks the rotational symmetry of our original uncompactified two - dimensional sheet and necessitates the introduction of a new measurable parameter , the angle @xmath51 defined in fig . [ fig : torusparam ] , which describes the orientation of the torus relative to the underlying lattice . conversely , the presence of the hexagonal lattice gives a clear meaning to the twist angle @xmath35 . this quantity would have had no meaning when rolling up an unmarked sheet of paper . fourth , it is clear that the existence of spectral equivalences between different nanotori implies that the number of _ spectrally _ distinct nanotori in any set will necessarily be smaller than the number of _ physically _ distinct nanotori in that set . however , it turns out that the magnitude of this truncation can easily become quite staggering . as an example , let us consider the set of physically distinct @xmath0 nanotori with fixed @xmath113 which can be formed from integers in the range @xmath114 for some cutoff @xmath115 . for concreteness , we shall take @xmath116 . in order to count only physically distinct nanotori , we shall require that @xmath24 be chosen such that @xmath117 . we shall also require , as a rough measure of their physical consistency in three - dimensional space , that each such nanotorus have an outer circumference @xmath59 which is at least triple its inner cross - sectional circumference @xmath58 ; note that other similar mathematical conditions may alternatively be imposed , but the qualitative results to follow are essentially unchanged . given these constraints , we then find through direct enumeration that there are exactly @xmath118 physically distinct carbon nanotori which have @xmath88 . yet , as a result of these spectral equivalences , it turns out that these @xmath118 physically distinct carbon nanotori give rise to only @xmath119 distinct energy spectra ! these distinct energy spectra are listed in table [ nhex18table ] , along with a representative carbon nanotorus in each class . .for @xmath88 and @xmath116 , there are @xmath118 physically distinct carbon nanotori @xmath0 with @xmath120 . however , these exhibit only 14 spectrally distinct energy spectra and band structures , and only four of these correspond to metals . these 14 spectrally distinct values of @xmath121 and @xmath51 are listed above , along with a sample @xmath0 nanotorus in each class . [ cols="^,^,^,^,^",options="header " , ] this is clearly a major truncation . even with the relatively small value @xmath88 and relatively small cutoff @xmath116 , each spectral signature listed in table [ nhex18table ] is shared by literally hundreds or thousands of physically distinct carbon nanotori . fifth , it is also worth noting that this truncation does not treat metallic and non - metallic nanotori equally . in general , a given @xmath0 carbon nanotorus will be metallic if at least one of its allowed wavevectors @xmath9 lies on the fermi surface , or in this case on one of the six points at which two bragg planes intersect . as is well known , this occurs only when the differences @xmath45 and @xmath46 are each a multiple of three . this implies that in any large set of physically distinct carbon nanotori , approximately one - ninth of the nanotori should be metallic . however , we see from table [ nhex18table ] that four out of the fourteen possible distinct spectral signatures are metallic . this is almost triple what would have been expected , implying that the fraction of spectrally distinct nanotori which are metallic is nearly triple the fraction of physically distinct nanotori which are metallic . in other words , the truncation from the four - parameter space of physically distinct nanotori to the three - parameter space of spectrally distinct nanotori is remarkably sensitive to the metallicity of the nanotori in question . one might argue that both of these effects the huge truncation in the number of distinct spectral signatures and the relative abundance of those which are metallic merely reflect the fact that we restricted our integers @xmath0 to lie within a fixed range bounded by @xmath122 , or that we took a relatively small value of @xmath39 . however , it is easy to demonstrate that the second of these effects is relatively insensitive to the choice of @xmath115 , and that the first of these effects only becomes even more dramatic as @xmath115 is increased . for example , if we restrict our attention to carbon nanotori with @xmath96 , we find that the number of physically distinct nanotori and the number of spectrally distinct nanotori both rise as a function of @xmath115 . however , we see from fig . [ panels ] ( left panel ) that the number of physically distinct nanotori with fixed @xmath39 grows as @xmath123 , as expected , while the number of spectrally distinct nanotori quickly hits a plateau ( right panel ) which remains flat for an increasingly long interval in @xmath115 before a new , hitherto - unrealizable spectral signature becomes possible and a new plateau develops . [ this growth in the number of physically distinct nanotori as a function of @xmath115 can easily be deduced from the observation that the number of quadruplets of integers @xmath0 grows as @xmath124 , while the corresponding number of attainable values of @xmath125 grows as @xmath123 . ] as a result , the number of physically distinct nanotori quickly outpaces the number of spectrally distinct nanotori . moreover , we see from fig . [ panels ] that the number of distinct metallic spectral signatures remains roughly one third ( and not one ninth ) the total number of distinct spectral signatures . we may also consider how these results vary with the choice of @xmath39 . this is shown in fig . [ nhexplot ] where , as functions of @xmath39 and for @xmath116 , we have plotted the number of spectrally distinct carbon nanotori as well as the number of those spectrally distinct carbon nanotori which are metallic . it is clear from these results that the numbers of spectrally distinct nanotori remain relatively small , even though they rise with @xmath39 , as expected . 3.4 truein 0.2truein 3.4 truein -0.1 truein 3.25 truein it is also easy to understand the jagged , oscillating nature of the results in fig . [ nhexplot ] . nanotori must have @xmath126 in order to be metallic , while they must have @xmath127 if they are rectangular , with @xmath128 . thus , one difference between tori with @xmath129 and those with @xmath130 is the existence of additional non - rectangular carbon nanotori with @xmath131 . however , it must also be borne in mind that a random selection of four integers @xmath0 is @xmath132 times more likely to result in an even value for @xmath133 than an odd one . as we see from fig . [ nhexplot ] , the combined effect from these two features is fairly significant . finally , another important implication of these spectral equivalences between physically distinct carbon nanotori concerns the traditional rules of thumb which allow us to determine whether a given carbon nanotorus is metallic , semiconducting , or insulating . throughout the existing literature on this topic , one finds what we shall call the `` rule of three '' : a given @xmath0 carbon nanotorus with @xmath134 will be metallic if both @xmath45 and @xmath46 are multiples of three , semiconducting if @xmath45 is multiple of three while @xmath46 is not , and insulating in all other cases . it is , of course , easy to verify that this characterization of a metallic nanotorus is modular invariant . specifically , if a given nanotorus has @xmath0 parameters which are metallic according to the rule of three , then all other nanotori to which it is spectrally equivalent will also have parameters which are metallic according to the rule of three . in other words , if @xmath45 and @xmath46 are both multiples of three , modular transformations of these parameters will not disturb this property . by contrast , the rule - of - three definition of semiconductors is _ not _ modular invariant even when we preserve the condition that @xmath135 . as a graphic illustration of this point , consider the @xmath136 torus . note that indeed @xmath135 for this torus . according to the standard rule of three , such a torus can be identified as a semi - conductor because @xmath45 is a multiple of three . by contrast , let us now consider the @xmath137 nanotorus . this torus also clearly has @xmath135 . however , because @xmath45 is not a multiple of three , we would expect this nanotorus to be an insulator . indeed , the rule of three tells us to expect this even though our first impression might be that the second nanotorus has larger radii in both directions , and therefore might have energy levels which are more closely spaced . however , even though these tori have very different physical parameters , it turns out that they are related through modular transformations and therefore have identical energy spectra . they therefore also have identical metallicity properties . this provides a graphic illustration that as a mathematical statement , the standard `` rule of three '' fails to characterize the conductivity properties of such nanotori because it is inconsistent with the modular transformations which reflect the additional symmetries of the compactified graphene sheet . in other words , the standard definition for a semiconducting carbon nanotorus fails to be modular invariant , and thus can not be complete as a description of the underlying conductivity properties of the nanotorus . this is yet another consequence of the fact that modular invariance in this context is an actual physical symmetry relating the spectra of physically distinct nanostructures . we close this section with an important comment . throughout this section , our goal has been to illustrate various mathematical ramifications of the modular symmetries which govern the spectra of different carbon nanotori . the specific examples we have provided throughout this section were therefore chosen for their mathematical simplicity as opposed to their phenomenological practicality . for example , we restricted our attention in this section to nanotori with relatively small values of @xmath39 , while realistic carbon nantoroi can be expected to have @xmath138 . likewise , realistic carbon nanotori will generally be quite long and thin , with quantities such as @xmath139 exceeding @xmath58 by an order of magnitude or more . however , all of the conclusions we have drawn in this section continue to hold even when more realistic tori are considered . as an example , let us restrict our attention to carbon nanotori with @xmath140 for which @xmath141 . the latter condition guarantees that our nanotorus remains relatively long and thin even if there are multiple windings of the hexagonal carbon lattice around the tube axis of the torus . taking @xmath142 , we find there are @xmath143 physically distinct nanotori which satisfy these conditions , but only @xmath144 of these are spectrally distinct . furthermore , of these @xmath144 spectral equivalence classes , @xmath119 correspond to metals . we see , then , that even for realistic nanotori , our modular symmetries continue to lead to large classes of spectrally equivalent nanotori and a relative overabundance of classes which are metallic . indeed , in the limit @xmath145 , it is straightforward to show that the number of physically distinct carbon nanotori in each spectral equivalence class also grows to infinity . with an eye towards potential implications of modular symmetries for the magnetic phenomena associated with twisted carbon nanotori @xcite , we now consider the introduction of magnetic fluxes . for full generality , we consider the possibility of _ two _ distinct fluxes : one which travels all the way around ( and through ) the length of the nanotube which forms the torus ; and another , namely the usual aharonov - bohm flux , which pierces the plane of the nanotorus , coming up through the donut hole . we shall denote these fluxes @xmath146 and @xmath147 respectively , as their associated vector potentials @xmath148 and @xmath149 lie parallel to the vectors @xmath30 and @xmath29 in fig . [ fig : torusparam ] , respectively . as is typical for such systems , we then find that we can incorporate the effects of these fluxes by keeping our previous band - structure energy function @xmath150 in eq . ( [ eq : bandstightbinding ] ) unchanged , and simply modifying our constraint equations for @xmath23 in eqs . ( [ eq : tubek ] ) and ( [ eq : torik ] ) so that the integers @xmath151 are shifted according to _ i _ i + _ i_0 where @xmath152 is the flux quantum and where _ 2 _ t + _ 1 _ 1 . note that it is the possible existence of a non - trivial twist angle @xmath35 which is responsible for the distinction between @xmath153 and @xmath147 . we then find that the resulting system continues to exhibit a spectral equivalence under the modular transformations in eqs . ( [ eq : stonindices ] ) and ( [ eq : stonvars ] ) as long as we allow @xmath154 to remain invariant under the @xmath56 transformation and to mix with each other under the @xmath55 transformation : s : . thus , a carbon nanotorus parametrized by @xmath155 will have the same spectral properties as one parametrized by @xmath156 , where these two sets of parameters are related through the modular transformations discussed above . in this paper , we have highlighted and investigated the implications of a geometric symmetry modular invariance which emerges upon the compactification of a graphene sheet to form a carbon nanotorus . although not traditionally considered in the carbon - nanotorus literature , modular invariance plays a critical role in describing the spectral properties of these nanotori and leads to spectral equivalences between physically distinct nanotori . as we have shown , this has profound implications for the classification of carbon nanotori , indicating that large numbers of seemingly unrelated nanotori are in fact completely identical in terms of their spectral properties . along the way , we also showed that the traditional `` rule of three '' classification rubric is incomplete , as it is based on quantities which are do not respect these spectral equivalences . we also found that the fraction of spectrally distinct carbon nanotori which are metals is approximately three times greater than would naively be expected on the basis of standard results in the literature . finally , we also showed that these spectral equivalences can easily be extended to cases in which non - trivial magnetic fluxes are present . the existence of these spectral symmetries also provides a deeper theoretical underpinning to certain results which already exist in the literature . for example , it is well known that many carbon nanotori exhibit persistent currents in the presence of a non - zero magnetic flux @xmath157 . as functions of @xmath158 , these currents typically follow complicated `` sawtooth '' patterns which have a natural periodicity under shifts @xmath159 . it has also separately been observed ( see , , ref . @xcite ) that any such sawtooth pattern is preserved but shifted horizontally upon the introduction of a nanotorus twist in which @xmath160 , where @xmath161 is chosen such that @xmath162 is also a lattice vector . indeed , when @xmath163 , the magnitude of this horizontal shift exactly matches the periodicity of the sawtooth pattern and the net result is unchanged . remarkably , this coincidence is now easy to understand from the point of view of modular transformations : when @xmath163 , the mapping from @xmath164 is nothing but the @xmath56 modular transformation , and as we have shown , modular transformations preserve the electrical properties of the torus , including its persistent currents . we thus see that the periodicity of the sawtooth pattern for persistent currents under shifts @xmath159 a periodicity which can be understood on elementary grounds having nothing to do with modular transformations can now also be interpreted as a spectral equivalence under the @xmath56 modular transformation . moreover , we now see that this is merely the tip of the iceberg , and that these sorts of spectral equivalences actually have a richer structure and context that not only corresponds to the entire modular group but also transcends the specific example of persistent currents . as we have discussed above , the traditional `` rule of three '' is not formulated in a modular - invariant way and is therefore incomplete as a description of the electronic properties of carbon nanotori . however , the full implications of these modular symmetries are significantly broader than just the rule of three : no theoretical result concerning the electronic properties of carbon nanotori can be correct unless it respects these modular symmetries . in other words , no calculation of any electronic property of a given carbon nanotorus in terms of its fundamental defining parameters @xmath0 can be correct unless it yields a result which is invariant under the modular transformations in eq . ( [ eq : stonindices ] ) . in this respect , modular invariance functions for carbon nanotori in much the same way as gauge invariance functions for electromagnetic systems : no theoretical result can be correct unless it can be phrased in terms of quantities which are invariant under the symmetry in question . of course , at a mathematical level , both modular invariance and gauge invariance rest on relatively simple algebraic identities . however , they both provide powerful organizing principles , and have significant physical manifestations and implications for the symmetry structures of the systems in which they appear . needless to say , several additional comments are in order . first , it should be noted that in this paper we have focused on what might called the `` ideal '' nanotorus . in particular , we have not accounted for the fact that the actual physical construction of such a torus in three - dimensional space requires that we introduce both an intrinsic and extrinsic curvature onto our otherwise flat graphene sheet . in this sense , the construction of a carbon nanotorus is different from that of a carbon nanotube ( in which only the extrinsic curvature is non - vanishing ) . the introduction of intrinsic curvature requires that we subject our underlying graphene sheet to considerable strain , deforming not only the positions of carbon atoms but also their relative spacings . these effects have been addressed by a number of authors , using a variety of different techniques @xcite . that said , these spectral equivalences should continue to hold , even in the presence of such deformations . there are several reasons for this . first , in the limit @xmath165 , all effects due to these deformations will be suppressed . however , this is precisely the limit in which nanotori can be constructed from purely hexagonal graphene sheets without the introduction of curvature - inducing pentagonal or heptagonal carbon rings . second , it can be shown that even when such strain is present , maximum toroidal stability occurs when this strain is uniformly distributed along the nanotorus @xcite , and this is precisely the situation in which the techniques of ref . @xcite can be used in order to mathematically rewrite the effects of such strain as arising due to a fictitious magnetic flux . as we have seen , the spectral equivalences we have found continue to exist even when such fluxes are introduced . but most importantly , while _ any _ deformations of the underlying graphene sheet can be expected to have effects on the corresponding band - structure energy function @xmath8 shown in fig . [ fig : energycontours ] , such deformations will not disturb the symmetries inherent in the constraint equations for @xmath166 derived in eqs . ( [ eq : tubek ] ) and ( [ eq : tubek2 ] ) . indeed , these equations reflect nothing more than the effects of toroidal compactification , and their structure leads directly to the modular symmetries inherent in eqs . ( [ eq : torik ] ) and ( [ eq : torik2 ] ) . thus , since these spectral equivalences ultimately stem from these symmetry properties , we are assured that any two tori related by modular transformations will sample the same set of wavevectors @xmath9 . such tori will therefore continue to be spectrally identical regardless of the function @xmath8 , provided the deformations to @xmath8 for the two tori are themselves identical . moreover , as we have argued above , even in cases where these deformations are not identical , they can at most differ by terms which are suppressed by factors of the presumably large nanotori radii . such differences can therefore be safely neglected . a similar conclusion also holds for thermal effects . it might seem , at first glance , that thermal effects could also destroy the spectral equivalences , since they too can have a dramatic effect on the band structure of the underlying graphene sheet @xcite . however , as noted above , these spectral equivalences are a consequence of the geometric symmetries that arise upon _ compactifying _ this sheet ; they are largely independent of the symmetries of the sheet itself . indeed , these modular symmetries exist for _ any _ choices of identification vectors @xmath30 and @xmath36 , even if those vectors are altered by other effects . therefore , as long as the temperature is sufficiently low that the electron coherence length exceeds both the inner and outer toroidal circumferences , the spectral equivalences we have been discussing should remain intact up to terms suppressed by the large nanotori radii . these comments notwithstanding , it still remains true that not all such nanotori are equally likely to appear in nature . in particular , those nanotori whose constructions implicitly involve large numbers of twists are likely to be rather difficult to construct or stablize , as the carbon atoms that constitute the underlying graphene sheet are likely to experience significant strain , leading to major deformations of the underlying carbon bond lengths and angles away from their ideal values . this is especially relevant , given that our statistical discussions in sect . iv implicitly assume that each of the relevant @xmath0 nanotori can appear with equal probability . at a mathematical level , it is difficult to draw a firm boundary between those carbon nanotori which are realizable in nature as _ bona - fide _ molecules and those which are not . however , in this paper we have restricted ourselves to statistical examinations of only those nanotori which already obey certain critical constraints . for example , we have limited ourselves to nanotori in which @xmath58 and @xmath59 are both significantly bigger than the carbon - carbon bond length @xmath7 . thus each twist experienced for the whole nanotorus has only a minor effect on the bond lengths and angles corresponding to each individual carbon atom . in addition , we have further limited ourselves to nanotori which also satisfy @xmath167 . this ensures that each such nanotorus can be realized in three - dimensional space , without the unphysical self - overlapping that would arise if this condition were not met . together , these conditions help to ensure that the strains induced by these modular transformations are not too severe . needless to say , there are many implications of these spectral equivalences which we have not yet explored . it would be interesting , for example , to consider the implications of these symmetries for the existence or absence of persistent currents as well as the existence or absence of colossal magnetic moments . this work is currently in progress . it would also be interesting to understand these modular transformations in terms of fictitious fluxes , using analogues of the techniques presented in ref . @xcite . first , as we have seen , modular invariance is the symmetry which underlies most of the results we have presented in this paper . at a mathematical level , modular invariance is normally just a relabelling symmetry in the sense that two sets of torus parameters related by a modular transformation normally correspond to the same physical torus . this is certainly the case in string theory , and in most other situations in theoretical high - energy physics in which modular transformations have played a significant role ( see , , ref . @xcite ) . however , the case of carbon nanotori is quite different . here , modular transformations relate parameters corresponding to carbon nanotori which are physically distinct . as we have discussed , this is because we are not merely rolling up an unmarked sheet of paper when we subject it to two non - parallel identifications ; we are rolling up a graphene sheet which already has a hexagonal carbon lattice imprinted on it . viewed from this perspective , it is therefore somewhat remarkable that modular transformations continue to play a role , indicating when two distinct nanotori will have the same spectral properties . indeed , in the case of carbon nanotori , we see that modular invariance is thus promoted from a mere relabeling symmetry to something far deeper : _ modular invariance becomes an outright physical symmetry between physically distinct entities_. we know of no other physical situation in which modular invariance plays such a role . second , it is also exciting at a practical level that physically distinct carbon nanotori can have identical energy spectra and electrical properties . since these nanotori are physically distinct , some are likely to be far more complicated to construct in the laboratory than others . nevertheless , these spectral equivalences suggest that it may not be necessary to fabricate a very complex nanotorus ( or generate sizable magnetic fluxes ) in order to obtain a desired spectral property ; there are likely to be far simpler nanotori and/or fluxes which can perform the same function . this could have significant implications for the production and use of such nano - materials . this work was supported in part by the department of energy under grant de - fg02 - 04er-41298 . we are happy to thank p. mceuen , d. ralph , c. stafford , and h. tye for discussions , and we wish to acknowledge the hospitality of the kavli institute for theoretical physics ( kitp ) in santa barbara , california , where this paper was completed . the opinions and conclusions expressed here are those of the authors , and do not represent either the department of energy or the national science foundation . m. terrones , w. k. hsu , j. p. hare , h. w. kroto , h. terrones , and d. r. m. walton , phil . a * 354 * , 2025 ( 1996 ) ; h. r. she , r. martel , and p. avouris , phys . * 84 * , 4441 ( 2000 ) ; m. sano , a. kamino , j. okamura , and s. shinkai , science * 293 * , 1299 ( 2001 ) ; h. watanabe , c. manabe , t. shigematsu , and m. shimizu , appl . phys . lett . * 78 * , 2928 ( 2001 ) . m. f. lin , r. b. chen , and f. l. shyu , solid state commun . 107 , 227 ( 1998 ) ; a. a. odintsov , w. smit , and h. yoshioka , europhys . lett . * 45 * ( 5 ) , 598 ( 1999 ) ; k. sasaki , phys . b * 65 * , 155429 ( 2002 ) ; s. latil , s. roche , and a. rubio , phys . rev . b * 67 * , 165420 ( 2003 ) ; z. zhang , z. yang , x. wang , j. yuan , h. zhang , m. qiu , and j. peng , j. phys . condens . matter * 17 * , 4111 ( 2005 ) ; c. feng and k. m. liew , carbon * 47 * , 1664 ( 2009 ) .

various approaches have been used to understand ageing experiments @xcite @xcite @xcite : droplet picture @xcite , mean field model @xcite @xcite , and trap models @xcite @xcite @xcite @xcite . ageing is by definition the property that the two points correlation function @xmath1 depends explicitly on the waiting time @xmath2 . @xmath3 is a macroscopic observable , for instance the magnetization in the zero field cooled experiment @xcite or the thermoremanent magnetization experiment @xcite . on the theoretical level , dynamics can be implemented in several ways . one possibility is langevin dynamics which was used for instance in @xcite where mean field ageing was first obtained , or in @xcite where ageing in the presence of flat directions was studied . another possibility is glauber dynamics @xcite . within the framework of glauber dynamics and given the probability distribution @xmath4 , for the system to be in any microcanonical state @xmath5 at time @xmath6 , the probability distribution at time @xmath7 is given by @xmath8 where @xmath9 is the glauber matrix . by `` microcanonical states '' , we mean a state such as the system is in a given microcanonical configuration with probability one . this article is devoted to the study of the glauber dynamics of various models , phenomenological or microscopic , in connection with ageing . more precisely , for a finite size system of size @xmath10 , we calculate from the glauber matrix a time @xmath0 such as the finite size system equilibrates on a time of the order of @xmath0 , starting from out of equilibrium initial conditions at time @xmath6 . in terms of ageing , the correlation function ( [ eqcorr ] ) depends explicitely on the waiting time @xmath2 provided the system is out of equilibrium at time @xmath2 , namely @xmath2 is smaller than @xmath0 . in other words , ageing for a finite size system is interupted after @xmath0 [ see @xcite where the denomination `` interrupted ageing '' was proposed ] . in order to understand ageing in the thermodynamic limit ( @xmath11 ) , the variations of @xmath0 with the system size have to be carefully studied . if @xmath0 goes to a constant in the thermodynamic limit , ageing is interrupted , whereas if @xmath0 diverges ageing is not interrupted in the thermodynamic limit : in the thermodynamic limit , the system reaches equilibrium on infinite time scales . we also discuss the possibility of metastability without ageing on experimental time scales , as it is the case for instance in diamond . in general , @xmath0 depends on the eigenvalues of the glauber matrix , but also on the overlaps between microcanonical states , and the transformation matrix elements between the eigenstates of the glauber matrix and the microcanonical states . however , in some simple cases , with only self overlaping states , the time @xmath0 can be expressed only in terms of the density of eigenvalues of the glauber matrix . the form of the glauber matrix will be given in each case in the core of the paper . whatever the system , the eigenvalues of @xmath9 are negative . the spectrum is bounded above by zero ( zero is always an eigenvalue , corresponding to the equilibrium boltzmann distribution ) . the eigenvalues of the glauber matrix have the dimension of the inverse of a time , so that they can be viewed as the inverse relaxation times of the system . since we are interested in long time behavior , a special attention will be paid to the smallest eigenvalues . this article is organized as follows . in section [ gm ] , we present a detailed study of the average density of small eigenvalues of the random energy model ( rem ) like trap model proposed and studied in @xcite and inspired by the rem model @xcite . in the case of the rem like trap model , two regimes are predicted and checked numerically for the density of small eigenvalues in the glass phase : an arrhenius like regime and a non arrhenius regime for the smallest eigenvalues . by arrhenius like regime , we mean a regime in which the density of eigenvalues is the same as the density of eigenvalues calculated from the arrhenius law , assuming a correspondence between the eigenvalues of the glauber matrix and the inverse relaxation times of the system . the eigenvalue density in the arrhenius regime are given in section [ arrh - def ] . we also establish a connection between the behavior of the density of small eigenvalues of rem like trap models and the spectral dimension of the associated diffusion process . next , in section [ general ] , we give a general relation for the existence of ageing from a glauber dynamics point of view . this section is the main part of the present article . when applied to the rem like trap model , our criterium is equivalent to the fact noticed in @xcite that the average trapping time diverges below @xmath12 . the rest of the paper is devoted to examine the case of other models : brownian motion which was shown in @xcite to be an ageing phenomenon , a model introduced by barrat and mzard @xcite ( bm model ) , the one dimensional ising chain ( in this case , our results are consistent with the existence of interrupted ageing ) , and the sherrington kirkpatrick model ( sk model ) . in all these cases , we find consistent results between the existence of ageing calculated from the behavior of the two times autocorrelation functions and our criterium based on the glauber matrix . we end up with some final remarks . trap models can be defined without a priori reference to a microscopic model . however , some trap models , such as the rem like trap model @xcite are related to the parisi solution of spin glasses : even though these models are toy models , some of them retain some feature of the dynamics of microscopic models . the rem like trap model consists of a trap model in infinite dimension , that is , the system can `` hop '' from one trap to any other trap , according to the rules of the dynamics . other toplogies will be analyzed below , in reduced dimension . the rem like trap model corresponds to a single replica symmetry breaking step . generalizations of the rem like trap model including multi replica symmetry breaking steps were studied in @xcite and their relevance to explain experimental data was shown . the models of this section consist of @xmath10 traps of depth @xmath13 , with the @xmath14 chosen among the distribution @xmath15 . the reason why this particular energy trap distribution is chosen is that there exists a finite transition temperature between a high and a low temperature glassy phase @xcite . the dynamics rules are given by @xmath16 for @xmath17 if the traps @xmath18 and @xmath19 are connected and zero otherwise . the diagonal terms are defined by @xmath20 which enforces probability conservation . equation ( [ propcons ] ) is a general constraint for any glauber dynamics . another constraint is the detailed balance condition @xmath21 it is clear that the aforementioned rules for the dynamics of the rem like trap model satisfy detailed balance . however , as we shall see below , other types of dynamics also satisfy detailed balance and probability conservation . as a consequence of detailed balance , the eigenvectors of the glauber matrix are not physical states ( in the sense that they are not probability distributions ) except for the boltzmann distribution ( the stationary eigenvector corresponding to the @xmath22 eigenvalue ) . notice that if @xmath9 is a glauber matrix @xmath23 , with @xmath24 a positive number , is also a glauber matrix . multiplying @xmath9 by @xmath24 amounts to changing the unit time . the proportionality constant is imposed by physical considerations . for the models considered in this section , this constant is chosen such as the diagonal elements of the glauber matrix do not scale with the number @xmath10 of traps in the large @xmath10 limit . for instance , in the infinite dimensional case , we choose @xmath25 . this dynamics will be studied with different phase space topologies : infinite dimension , where all the traps are connected ( section [ infinite ] ) , one dimensional topology , where a given trap is connected only to its two nearest neighbors ( section [ disor ] ) , and also one dimensional topology with `` ordered '' traps , in the sense that the trap energies have been ordered @xmath26 , and the trap at site @xmath18 has an energy @xmath14 . a useful property that we will use throughout this paper is that glauber matrices can be mapped onto symmetric matrices with the same spectrum . let us call @xmath27 the diagonal matrix @xmath28 . then , because of detailed balance , the matrix @xmath29 is symmetric and has the same spectrum as @xmath9 . using a quantum mechanical analogy , we denote by @xmath30 the microcanonical state corresponding the system in trap @xmath18 with a probability unity . the analogy to quantum mechanics is useful , even though there is no hermitian structure . in particular , any linear combination of microcanonical states is not a probability distribution . we propose ourselves to carry out a detailed study of the density of eigenvalues of the rem like trap model and related models in reduced dimension . since the arrhenius law is introduced `` by hand '' in the dynamics , it is interesting to question in which regime the density of eigenvalues is given by the arrhenius law , assuming a correspondence @xmath31 between the eigenvalues of the glauber matrix and relaxation times . we will naturally call _ arrhenius regime _ such a regime . as we will see below , the density of eigenvalues is not always in the arrhenius regime . in this arrhenius regime , and by definition , the off - diagonal terms are neglectable as far as the density of eigenvalues is concerned . the diagonal coefficients of the glauber matrix are proportional to @xmath32 , so that the distribution of eigenvalues is @xmath33 this distribution is the same as the one expected from an arrhenius law with life times @xmath34 . if one assumes an arrhenius picture , the distribution of relaxation times is @xmath35 which is equivalent to ( [ aha ] ) . as noticed in @xcite , the average trapping time diverges if @xmath36 , leading to ageing . distribution of trapping times with no first moment were already studied in @xcite , and shown to lead to anomalous diffusion , namely to a non trivial spectral dimension , even above the upper critical dimension . we will compare our results for the rem like trap model with the results of @xcite derived of the context of anomalous diffusion on site disordered hypercubic lattices . the density of eigenvalues is thus described by the arrhenius law provided the off - diagonal terms of the glauber matrix are a small perturbation , so that the density of eigenvalues of the glauber matrix is given by the diagonal terms only . we now examine several cases with different phase space topologies and determine in the different cases on which condition the density of eigenvalues is determined by the arrhenius picture . we also compare our analytical approach to numerical diagonalizations of the symmetrized matrix @xmath37 . in this section , the system can hop from one trap to any other trap ( infinite dimension ) . the diagonal coefficients are @xmath38 and the off diagonal coefficients of the symmetrized glauber matrix are @xmath39 we treat the off diagonal coefficients @xmath40 ( @xmath17 ) as a perturbation . this perturbative treatment is consistent since we are looking for a regime in which the off diagonal coefficients do not modify the density of eigenvalues with respect to the purely diagonal case . the second order correction to the eigenvalue @xmath41 is @xmath42 the first order corrections being identically zero since the perturbation consists only of off diagonal elements . we assume that the density of trap energies in the vicinity of @xmath14 is large enough for the discrete summation to be replaced by an integral . if this condition is not satisfied , the second order corrections @xmath43 may have large fluctuations because the divergence for @xmath44 is no more cancelled . the condition that the unperturbed eigenvalue density is large in the vicinity of @xmath14 simply reads @xmath45 . we assume in the rest of this section that this condition is fulfilled , and we can safely replace the discrete sum by an integral : @xmath46 the singularities for @xmath47 and @xmath48 cancel each other , and we can approximate @xmath43 as @xmath49 with @xmath50 . we have used the following approximation : if @xmath51 is small , then @xmath52 and we use @xmath53 as a cut off . clearly ( [ app ] ) is only an approximation , but we are only interested in the dominant behavior of @xmath54 as a function of @xmath14 . now , we want to compare the approximation ( [ app ] ) of @xmath43 to the typical level spacing between @xmath41 and @xmath55 , where we assume that the traps are ordered such as @xmath56 . to to so , we first calculate the distribution of @xmath57 and we obtain @xmath58 , with @xmath59 notice here that @xmath60 diverges as @xmath61 goes to zero if @xmath36 . this accumulation of small relaxation times is related to the existence a glass transition for this model for @xmath62 . let us now calculate the level spacing statistics @xmath63 , which is the density probability that the spacing between the absolute value of the eigenvalue @xmath9 and the absolute value of the next eigenvalue is @xmath64 . because of the statistical independence of the trap depths , @xmath63 is readily obtained as @xmath65 in this expression , @xmath66 is the number of levels in the interval @xmath67 $ ] ( with @xmath68 ) . the other terms correspond to the condition that there is no eigenvalue in the interval @xmath69 $ ] . assuming that @xmath61 is small compared to unity , and that @xmath10 is large , we get @xmath70 the off - diagonal terms are irrelevant to the density of eigenvalues provided @xmath71 is much smaller than unity . this condition reads @xmath72 , with @xmath73 defined in ( [ deltai ] ) . since we are looking for a necessary condition valid if @xmath36 , we neglect the temperature dependent prefactor and conclude that the density of states is in the arrhenius regime only for traps such that @xmath74 . notice that the traps such as @xmath75 are deeper than the typical deepest trap ( of the order of @xmath76 ) . the arrhenius picture is thus valid for a _ typical _ trap realization , with a lowest trap not deeper than @xmath76 . however , non typical disorder realizations may be generated . for these realizations , the lowest eigenvalues are not described by an arrhenius picture . the existence or absence of the arrhenius regime even for a very large number of traps is thus sample dependent . this non self averaging behavior is reminiscent on the sample dependence of the quantity @xmath77 with @xmath78 the probability to find the system in trap @xmath18 @xcite . the density of small eigenvalues of the rem like trap model is pictured on figure [ fig1 ] . for practical purposes , we do not calculate the density @xmath79 since this quantity is too small for small eigenvalues ( see ( [ aha ] ) ) . we rather calculate @xmath80 , where @xmath81 is the boltzmann weight associated to the eigenvalue @xmath82 , which is calculated as the average of the boltzmann weight operator @xmath83 over the eigenstate associated to the eigenvalue @xmath82 . in the arrhenius picture , we have @xmath84 since @xmath36 , @xmath85 increases exponentially as a function of @xmath86 when @xmath87 . we see on figure [ fig1 ] the existence of the arrhenius regime , in good agreement with the condition @xmath74 . we can now fit the density of eigenvalues in the arrhenius regime to the form @xmath88 and see if we recover @xmath89 , as predicted from the arrhenius law . the fit to the arrhenius law is shown on figure [ fig2 ] . the variations of the exponent @xmath90 as a function of the inverse temperature are plotted on figure [ fig3 ] , and we conclude to a good agreement with the arrhenius law . random walks on a finite @xmath91dimensional lattice with a broad distribution of waiting times have been analyzed in the past ( see @xcite and references therein ) . as we shall see , we can relate our results concerning the existence of a arrhenius like regime for the rem like trap model to the results of @xcite , where the long time behavior of random walk on such lattices was investigated by means of the real space renormalization method introduced in @xcite . the rem like trap model falls in the universality class described by machta since the ingredients are the same in the two cases , rather than the universality class of random walks in a random field force @xcite . in order to make contact with the work of machta , we first calculate the probability @xmath92 of being at the origin : @xmath93 where @xmath94 is the probability to find the system in the trap @xmath18 at time @xmath7 , starting from the trap @xmath18 at time @xmath95 , and @xmath96 is the set of eigenvalues of the glauber matrix . we have shown that , in the infinite dimensional case , the typical small eigenvalue density ( relevant to the long time behavior of @xmath97 ) has the arrhenius form @xmath98 . a simple scaling analyzis then shows that @xmath99 in the long time limit . this result was derived for a graph such as any site is connected with any other site , that is we started from the begining from an infinite dimensional situation . in order to extrapolate this result to a finite dimension @xmath91 larger than the upper critical dimension 2 ( as shown in @xcite ) we first notice that , within the framework of the rem like trap model , the microscopic time @xmath100 scales like @xmath101 , with @xmath102 the lattice spacing , whereas , in the context of random walks , the correct scaling is @xmath103 [ an illustration of the validity of such a scaling will be given in section [ brownian - motion ] ] . this implies that @xmath104 , which gives the correct prescription to go from the rem like trap model in infinite dimension to the finite @xmath91 dimensional random walk model : one should replace the time @xmath7 in ( [ p0rem ] ) by @xmath105 . this leads to @xmath106 for @xmath91 larger than the upper critical dimension @xmath107 . this is exactly what was found in @xcite using a real space renormalization group approach , with a non trivial spectral dimension @xmath108 . we aim to calculate the structure of the eigenstates of the rem like trap model . to do so , we decompose a given microcanonical state @xmath30 into the eigenvectors @xmath109 , and calculate the absolute value of the coefficients of this decomposition . the result is plotted on figure [ fig4 ] . we see that a microcanonical state @xmath30 with an eigenvalue @xmath110 is mainly a linear combination of eigenstates with eigenvalues close to @xmath110 . this observation can be understood in the framework of perturbation theory as follows . the first order correction to the eigenvectors are @xmath111 so that , at first order @xmath112 at the first order of perturbation theory , the pure states @xmath30 is thus mainly a linear combination of eigenvectors with an eigenvalue close to the eigenvalue of @xmath113 . starting from a pure state @xmath30 at time @xmath6 , we can calculate the evolution of the system : @xmath114 \nonumber , \end{aligned}\ ] ] where we have used ( [ per1 ] ) and ( [ per2 ] ) . as expected from the arrhenius law , the relaxation time is of the order of @xmath115 , which diverges in the limit @xmath116 . we will compare below the structure of the eigenvectors of the rem like trap model to the structure of eigenvectors of another model ( bm model ) . assuming periodic boundary conditions , the diagonal terms of the glauber matrix are given by @xmath117 the perturbation induced by the off - diagonal terms does not affect significatively the density of eigenvalues provided the perturbation @xmath118 is small compared to the level spacing between @xmath41 and @xmath55 . since @xmath119 is uncorrelated from @xmath14 , the matrix element @xmath120 must be averaged over @xmath119 , so that @xmath40 is of the order of @xmath121 , that is @xmath122 the distribution of eigenvalues of the unperturbed system is @xmath123 from what we deduce the level spacing statistics in the long time ( small @xmath61 ) approximation @xmath124 the off - diagonal terms do not affect the density of eigenvalues provided @xmath125 which leads to the condition @xmath126 . the average number of traps generated among a sequence of @xmath10 independent traps chosen among the distribution @xmath127 and fulfilling the aforementioned condition is @xmath128 which is vanishing in the large @xmath10 limit . we thus conclude that the arrhenius regime with this topology does not exist , that is off - diagonal terms will always modify significatively the density of eigenvalues with respect to the arrhenius case . as we will see below , the arrhenius regime is replaced by a renormalized arrhenius regime , where the density of eigenvalues is given by ( [ lawarr ] ) , but with @xmath90 lower than @xmath129 . again , we generate numerically the density of small eigenvalues @xmath85 . the fit is shown on figure [ fig2 ] , and the variations of the exponent in the arrhenius regime are plotted on figure [ fig3 ] . we conclude to the existence of a renormalized arrhenius regime , in the sense that the density of eigenvalues is given by ( [ lawarr ] ) but with an exponent @xmath90 smaller than @xmath129 , indicating a slowing down of the dynamics due to the one dimensional topology compared to the infinite dimensional case : in one dimension , to go from one trap to a deeper trap , the system has to jump to intermediate traps . section [ relation1 ] was devoted to the connection between the spectrum of the glauber matrix in infinite dimension and the long time behavior of random walks on finite @xmath91 dimensional trapped hypercubic lattices above the upper critical dimension . we now analyze the analogy in one dimension . we have shown numerically that the density of eigenvalues behaves like @xmath130 for small @xmath82 , where @xmath131 is different from @xmath9 . in terms of random walks , this result becomes , with the notations of section [ relation1 ] @xmath132 namely the spectral dimension is @xmath133 . we now use the results of @xcite where @xmath134 was calculated : @xmath135 from what we conclude that the coefficient @xmath90 plotted on figure [ fig3 ] is @xmath136 this analytic form is consistent with numerical calculations of the density of eigenvalues . we now assume that @xmath10 energy traps were generated among the distribution @xmath127 , and that these energies have been ordered in such a way that @xmath56 . a site @xmath66 with a trap of energy @xmath137 is connected to the sites @xmath138 ( with an energy @xmath139 ) and @xmath140 ( with an energy @xmath141 , except for the sites @xmath142 and @xmath10 which have no neighbor . intuitively , such a trap ordering should make the dynamics faster with respect to the case analyzed in section [ disor ] since the system can decrease its energy continuously during the relaxation . we have @xmath143 since the density of level spacings @xmath63 ( [ p(s ) ] ) is large for small values of @xmath61 , we expect that @xmath55 is close to @xmath41 at least if @xmath144 is small , so that we can use the following approximation : @xmath145 . the off - diagonal terms will not modify the density of eigenvalues with respect to the arrhenius case provided @xmath146 that is @xmath75 . the number of traps fulfilling this condition in a distribution of @xmath10 traps is on average equal to unity , so that only a small number of eigenvalues contribute to the arrhenius regime . nevertheless , it is still possible to make a statistics over a large ensemble of @xmath10 traps distribution and to test numerically if the arrhenius regime can be observed . we generated numerically the density of small eigenvalues @xmath85 . as shown on figure [ fig2 ] , we still can fit @xmath147 to the form @xmath148 , and the exponent @xmath90 is close to @xmath129 , as in the arrhenius regime . ( see figure [ fig3 ] ) . we conclude that ordering the traps in one dimension restores the arrhenius regime for the lowest eigenvalues . again , the existence of the arrhenius regime is sample dependent . we are now going to give a general relation between ageing and the glauber matrix . this relation is a generalization of a criterium given in @xcite . in the context of the rem like trap model , the average trapping time @xmath149 diverges if @xmath150 . on the other hand , ageing exists below @xmath12 only , in the limit where the microscopic time scale @xmath100 goes to zero . we are now going to argue that a similar criterium can be found for any statistical mechanical model , provided the master equation is linear . this is not always the case . for instance , the master equation of the backgammon model in a mean field dynamics at zero temperature is not linear @xcite . let us now start our general argument . the idea is to start with a system in a given microcanonical state @xmath30 and write the glauber dynamics equation for the evolution of this pure state . first , we call @xmath151 the transformation matrix @xmath152 where @xmath30 is a microcanonical state and @xmath153 are the eigenstates of the glauber matrix . by microcanonical state , we mean that if the system is in the state @xmath30 , the system is with probability one in the microcanonical state @xmath5 . in order to avoid confusion , all the labels related to the eigenvectors will be denoted by greek letters whereas the labels for pure states will be denoted by latin letters . the evolution of the pure state ( [ pure2 ] ) is readily obtained as @xmath154 in other words , the probability to be in microcanonical state @xmath155 at time @xmath7 starting from the microcanonical state @xmath5 is @xmath156 we now consider the time dependent correlation function @xmath157 where @xmath158 is the number of microcanonical states . we use @xmath158 instead of @xmath10 to avoid confusion between the number of microstates @xmath158 and the number of sites @xmath10 in a spin model . in a trap model , @xmath159 , whereas in an ising spin system , @xmath160 . in ( [ correlation ] ) , we have carried out an average over all the initial microcanonical states , that is the system is first quenched from an infinite temperature , and the correlation ( [ correlation ] ) is averaged over the realizations of the infinite temperature states . in ( [ correlation ] ) , @xmath161 is the overlap between the pure states @xmath5 and @xmath155 , which has to be defined within each model . as we will see below in section [ brownian - motion ] , even within a given dynamics , the choice of the overlap depends on what type of observables are considered in the two times correlation functions . we now want to calculate from ( [ correlation ] ) a typical time by integrating ( [ correlation ] ) . however , the boltzmann eigenstate leads to a trivial divergence of this integral . we thus need a prescription to eliminate the equilibrium distribution in ( [ correlation ] ) . we first discuss this prescription in the absence of metastability and postpone for later discussion the case of metastability . we define the typical time @xmath162 by @xmath163 where the summation over the eigenstates excludes the boltzmann distribution . the overline denotes an average over the disorder realizations , if necessary . the significance of @xmath0 is clear : for a finite size system with @xmath10 sites , the out of equilibrium dynamics occurs before @xmath0 and , for times larger than @xmath0 , the finite size system has equilibriated . in other words , @xmath0 is the cross over time before which the finite size system exhibits ageing . notice that in this argument , the system is a finite size system . one should thus carefully study finite size effects . two cases may occur : first , @xmath0 tends to a constant in the thermodynamic limit . in this case , ageing is interrupted @xcite in the thermodynamic limit . as we shall see , this is the case for the ising chain at a finite temperature . on the other hand , if @xmath0 diverges in the thermodynamic limit , ageing is not interrupted . in this case , the exitence of a divergent relaxation time with respect to times @xmath95 and @xmath7 implies the existence of a divergent relaxation time with respect to any finite waiting time @xmath2 and @xmath164 , and thus the existence of ageing in the thermodynamic limit . the relation between the existence / absence , the nature of ageing in the thermodynamic limit and the behavior of @xmath0 was derived on heuristic basis . we could not find a rigourous derivation using a general argument . however , in what follows , we investigate the behavior of several models ( rem like trap model , random walks , bm model , one dimensional ising chain and sk model ) and we find in all these cases that our criterium based on the behavior of @xmath0 is correct . as far as non interrupted ageing in the thermodynamic limit is concerned , we will see below that this is the case for the bm and sk models . this is also the case for the rem like trap model . we have shown in section [ infinite ] that for all the typical realizations of the trap configurations ( typical is defined in section [ infinite ] ) , the density of eigenvalues is given by the arrhenius law , which means that , in the case of the rem like trap model , our criterium for the existence of ageing is equivalent to the criterium of bouchaud @xcite that we have recalled at the beginning of this section , provided the overlap between traps is taken to be @xmath165 , where @xmath166 is the edwards anderson order parameter ( within the one replica symmetry breaking step picture ) . notice that the existence of a non arrhenius regime generated by non typical trap configurations does not modify our conclusion regarding the divergence of @xmath167 since the non arrhenius regime only occurs for traps such as @xmath168 if @xmath11 . the correlation @xmath169 is @xmath170 which is non integrable if @xmath171 . the typical time @xmath172 only depends on the density of eigenvalues @xmath173 for a system of @xmath10 traps , and is independent on the transition matrix , which motivates the study of the density of eigenvalues carried out in section [ gm ] . in general , in any model with only self overlapping pure states , the existence of ageing is coded only in the behavior of the density of small eigenvalues : there is ageing in the thermodynamic limit if @xmath174 tends to a constant if @xmath82 goes to zero ( an exemple will be provided in section [ bm - model ] ) or if @xmath174 diverges if @xmath82 goes to zero ( this is the case for the rem like trap model already studied in section [ gm ] ) . let us now discuss the behavior of the correlation function ( [ correlation ] ) in the presence of metastability . we have in mind the case of diamond for instance , which is metastable but in which case there is no ageing . we do not have a glauber matrix description of diamond , but take this case as a generic situation for metastability and draw qualitative conclusions . in this case we expect that for a finite size system , ( [ correlation ] ) will first decay to a finite value over a first time scale @xmath175 . after @xmath175 , the system reaches its metastable configuration ( for instance diamond ) and thermalizes in a portion of its phase space . we thus expect that , for times larger than @xmath175 , the correlation ( [ correlation ] ) reaches a plateau . on larger time scales , the correlation ( [ correlation ] ) should decay to its equilibrium value with a cross over time @xmath176 , corresponding for instance to the transition from diamond to graphite . the time scale @xmath176 should increase with the number of sites , so that , with a macroscopic number of sites , no transition from diamond to graphite is observable on experimental time scales . on the other hand , @xmath175 may remain small even in the thermodynamic limit , signaling the absence of ageing in the graphite phase . in such a situation where the cross over time scale @xmath176 is not observable , the only relevant time scale is @xmath175 . in order to calculate @xmath175 one should not use the prescription ( [ prescription ] ) but rather substract to the correlation ( [ correlation ] ) its plateau value for @xmath177 . brownian motion was shown in @xcite to be an ageing phenomenon . this model is simple and thus provides a simple test of the criterium of section [ general ] . brownian motion in an euclidian space of dimension @xmath91 has the following glauber dynamics : @xmath178 and @xmath179 if @xmath18 and @xmath19 are nearest neighbors on a hypercubic lattice . @xmath100 is the microscopic hopping time . in this case , the glauber matrix is symmetric and , because of translational invariance , its eigenstates are plane waves @xmath180 where @xmath181 ( we have assumed cyclic boundary conditions with a lattice spacing @xmath102 ) . the matrix elements of the glauber matrix on the plane wave basis are @xmath182 where @xmath90 runs over the @xmath91 directions and @xmath183 is the component of the wave vector @xmath184 on the @xmath90 direction . the long time behavior of the random walk is readily obtained by expanding the cosine up to the second order , and the probability to find the brownian particle at time @xmath2 on site @xmath185 is , in the long time limit : @xmath186 where the particle is at the origin at time @xmath6 ( out of equilibrium state ) . as noticed first in @xcite , the two times correlation functions of the position are proportional to the waiting time : @xmath187 in the large @xmath2 limit . in order to make contact with @xcite , one should adopt the scaling @xmath188 in the thermodynamic / long time limits , which allows to recover the langevin dynamics results and thus the existence of ageing in any dimension for the position position correlation function . in order to make contact between the correlation ( [ c(1 ) ] ) and the general form of the correlation ( [ correlation ] ) , we should choose the overlap @xmath189 . with this choice of the overlap , ( [ correlation ] ) and ( [ c(1 ) ] ) are both related to the same physical quantity : the mean square displacement . using this specific form for the overlap , one can calculate @xmath190 , integrate with respect to @xmath7 and show that @xmath191 , in agreement with the existence of ageing in this system for the position correlations . one could also be interested in the following correlation : @xmath192 in the large @xmath7 , @xmath2 limit , this correlation is found to be @xmath193 in the thernodynamic limit , @xmath194 , @xmath188 , so that @xmath195 is vanishing in the thermodynamic limit , signaling the absence of ageing for the correlation ( [ c(2 ) ] ) . the overlap associated to the correlation ( [ c(2 ) ] ) is the local one : @xmath196 . we can thus calculate @xmath197 using the integration ( [ taurem ] ) over the density of eigenvalues . in order to calculate the behavior of @xmath0 , we consider a finite system of volume @xmath198 with a lattice spacing @xmath102 . the number @xmath10 of sites is @xmath199 . by thermodynamic limit , we mean that @xmath200 is fixed , @xmath194 and @xmath188 . using ( [ taurem ] ) , we have then @xmath201 where the summation is restricted over the wave vectors @xmath184 such as @xmath202 . clearly , the dominant contribution to ( [ tau(2 ) ] ) comes from the smallest eigenvalues only . it is thus legitimate to expand the eigenvalues up to the quadratic order in @xmath184 : @xmath203 one then obtains @xmath204 if @xmath205 , @xmath206 since @xmath188 , @xmath207 in the thermodynamic limit in two dimensions . now if @xmath208 , @xmath209 since @xmath188 , this expression goes to zero in the thermodynamic limit whatever the dimension . we have thus shown that , in the case of random walks , and in the case of the correlations ( [ c(1 ) ] ) and ( [ c(2 ) ] ) , the existence / absence of ageing in the thermodynamic limit is consistent with the general argument of section [ general ] . we assume that the traps have been ordered such as @xmath56 . unlike the rem like trap model , we do not make any asumption about the trap distribution . the finite temperature dynamics is @xmath210 if @xmath17 and satisfies detailed balance . the zero temperature form of the glauber matrix of this model is very simple since @xmath211 , @xmath212 if @xmath213 and @xmath214 if @xmath215 at zero temperature . the spectrum is @xmath216 . the eigenvector @xmath217 for the eigenvalue @xmath22 is @xmath218 . it is easy to show that the eigenvector @xmath219 associated to the eigenvalue @xmath220 is @xmath221 we can now express the pure states in terms of eigenvectors @xmath222 this means that , at zero temperature , a pure state @xmath223 [ @xmath66 is the @xmath66-ieth traps , with @xmath26 ] overlaps only with eigenstates associated to eigenvalues @xmath225 , and this overlap does not depend on the eigenstate . compared to the result of section [ eigi ] , the structure of the decomposition of a pure state into eigenstates of the glauber dynamics is very different from the rem like trap model . if one starts from a pure state @xmath226 at time @xmath6 , the state of the system at time @xmath7 is given by @xmath227 we recover the fact that this model can reach its ground state in a time of order @xmath10 at zero temperature , whereas in the limit of zero temperatures , and for the rem like trap model , the time spent in a trap @xmath18 is proportional to @xmath228 , which goes to infinity if @xmath229 and @xmath10 fixed . the two times autocorrelation functions were calculated in @xcite by means of a laplace transform formalism . the result is @xmath230 if @xmath231 . we rederive this result in the appendix using a treatment different from @xcite , and based only on the diagonalization of the glauber matrix . the bm dynamics thus exhibits ageing . provided only self overlaps are non zero ( @xmath232 ) , the correlation function introduced in section [ general ] is @xmath233 , provided @xmath234 . the typical time @xmath235 below which ageing occurs is thus logarithmically divergent . on the other hand , this can be also shown directly from the spectrum : @xmath236 which diverges logarithmically . as far as the finite temperature behavior of the two times autocorrelation function is concerned , this model exhibits interrupted ageing @xcite since at any finite temperature the correlation length is finite , of the order of @xmath237 , ageing does not exist after a time independent of the system size . it is thus interesting to apply to this case the general argument of section [ general ] . we first begin with the definition of the glauber dynamics . we use the dynamics introduced in @xcite . the master equation of this single spin flip dynamics is @xmath238 as a consequence of detailed balance , @xmath239 notice that , unlike the case of the rem like trap or bm models , there is no need to multiply the matrix elements by @xmath240 . it is shown in @xcite that in the infinite temperature limit , the eigenvalues of the glauber matrix are negative integers between @xmath241 and @xmath95 , with degeneracies given by the binomial coefficients . the largest relaxation time in the infinite temperature limit is equal to unity whatever the system size , and there is thus no need to rescale the coefficients of the glauber matrix . in the case of a spin model , the overlap @xmath242 is given by @xmath243 since ageing is interrupted , we should observe a saturation of the time @xmath244 as @xmath10 increases . moreover , @xmath244 should increase as the temperature decreases . in practise , we work with an open chain and we take advantage of the existence of the global @xmath245 symmetry ( invariance of the glauber dynamics under the @xmath246 transformation in a zero magnetic field ) , and the reflection symmetry of the chain . the glauber matrix is diagonalized in each of these four symmetry sectors . the results are plotted on figure [ fig5 ] . the difference @xmath247 decreases as @xmath10 increases , in agreement with the existence of interrupted ageing . this model was proposed in 1975 as an `` exactly solvable '' model of spin glasses @xcite . for reviews on this model and its developments , we refer the reader to the reviews @xcite . the hamiltonian of the sk model is @xmath248 where the sum is carried out over all the pairs of sites ( infinite dimensional model ) and the quenched random interactions are distributed according to a gaussian distribution @xmath249 this model is a spin glass if @xmath250 and a paramagnet otherwise . the two times correlation function were shown in @xcite to exhibit ageing in the glassy phase . we are going to apply our general argument of section [ general ] to the case of the rem like trap model . to do so , we calculate numerically @xmath251 for a large number of disorder realizations . the result is plotted on figure [ fig6 ] , as well as the differences @xmath252 for @xmath253 sites . we observe that @xmath254 increases at low temperatures and decreases at high temperature . the crossing point on the insert of figure [ fig6 ] should be identified to the glass temperature . this temperature is not equal to unity since the system sizes are very small . notice that even though the sizes are quite small , a large number of disorder configurations is required to have a significant statistics ( of the order of 500000 independent disorder configurations ) . the computer time required was equivalent to 1000 hours of cpu time on a single processor . our results are thus consistent with the existence of ageing in the low temperature phase , and no ageing in the high temperature phase . we have thus carried out a detailed analysis of ageing from a glauber dynamics point of view , with an emphasis on finite size effects . we have shown how to extract information from the glauber matrix about ageing , and how to analyze finite size effects . in some simple cases with only self overlaps , the behavior of the time after which there is no ageing only involves the knowledge of the average density of eigenvalues @xmath173 when @xmath87 . this is the case for the rem like trap model , in which case we have carried out a careful analysis of the density of small eigenvalues . our analysis was confirmed by numerical diagonalizations of glauber matrices of small clusters . in this case , we were able to related the typical behavior of the density of eigenvalues to the spectral dimension of the corresponding diffusion process . : r.m . acknowledges interesting discussions with a. barrat , j.p . bouchaud , s. franz and n. wingreen . the authors acknowledge b. douot for his interest to this work and his encouragements . r.m . also acknowledges the hospitality of nec research institute at princeton where part of this work was done . we are going to calculate the two times autocorrelation function @xmath255 using a method based only on the exact diagonalization of the glauber matrix at a zero temperature . starting from an infinite temperature state at time @xmath6 , @xmath256 with @xmath30 a pure state , the natural definition of the two times autocorrelation function within this model is @xmath257 where @xmath258 is the probability to find the system in the trap number @xmath18 at time @xmath2 and @xmath259 is the escape rate from trap @xmath18 , that is the diagonal element of the glauber matrix @xmath260 . this definition of the waiting time dependent autocorrelation function is similar to the one used in @xcite . using the expression ( [ eig ] ) of the eigenvectors and the decomposition ( [ dec1 ] ) and ( [ dec2 ] ) of the pure states into eigenvectors , we can calculate in a straightforward fashion the evolution of the system , starting from the state ( [ initial ] ) : @xmath261 |e_k \rangle .\ ] ] the probabilities @xmath258 are easily recovered : @xmath262 . after straightforward calculations , the dominant behavior of @xmath255 is obtained , in the limit @xmath263 : @xmath264 which is equivalent to the result of @xcite . notice that if @xmath2 is of the order of the number of traps , the system has reached equilibrium : the dynamics is stationary ( the system is stuck in the lowest trap ) and there is no ageing . m. alba , m. ocio , j. hamman , europhys . lett . , 45 ( 1986 ) ; j. phys . ( 1985 ) l-1101 ; m. alba , j. hamman , m. ocio , p. refregier , j. appl . phys . , 3683 ( 1987 ) ; e. vincent , j. hamman , m. ocio , _ recent progress in random magnets _ , d.h . ( world scientific , singapore , 1992 ) . m. mzard , g. parisi and m. virasoro , _ spin glass theory and beyond _ , world scientific ed . ( 1993 ) ; k.h . fisher and j.a . hertz , _ spin glasses _ , cambridge studies in magnetism ( 1991 ) ; v.s . dotsenko , m.v . feigelman and l.b . ioffe , _ spin glasses and related problem _ , physics reviews , hardwood academic publishers ( 1990 ) . density of eigenvalues in the arrhenius regime for the rem like trap model and fit to the exponential form ( @xmath265,@xmath266 traps ) . the insert shows the same quantity , but with also the smallest eigenvalues . for these small eigenvalues , the arrhenius picture is not valid . density of small eigenvalues of the models of section [ gm ] , for @xmath267 . ( a ) : infinite dimensional case ( rem like trap model ) ; ( b ) : one dimensional case , with `` disordered '' traps ; ( c ) : one dimensional case with `` ordered '' traps . the lines in cases ( b ) and ( c ) are parallel , and in good agreement with the arrhenius law , whereas we observe a significant deviation to the arrhenius law in ( b ) case . exponent @xmath90 deduced from the fit ( [ lawarr ] ) of @xmath85 . the circles correspond to the infinite dimensional case ( rem like trap model ) , the squares to the one dimensional `` disordered '' case , and the diamonds to the `` ordered '' one dimensional case . the solid line corresponds the the arrhenius regime ( in other words , to site disordered random walks above the upper critical dimension 2 ) , and the dashed line corresponds to the one dimensional renormalized arrhenius regime ( in other words , to site disordered random walks in one dimension ) . structure of the eigenstates of the rem like trap model for 20 traps . the label @xmath18 is an eigenvalue label , with eigenvalues in growing order and @xmath19 labels the traps ( the depth increases if @xmath19 increases ) . ( a ) represents @xmath268 for the @xmath269 pure states . ( b ) represents the depth of the traps . ( c ) represents the 20 eigenvalues between @xmath270 and @xmath95 for 4 different temperatures ( circles : @xmath271 , squares : @xmath267 , diamonds : @xmath272 , triangles : @xmath273 ) .

quantum spin chains are physically realized in quasi - one - dimensional magnetic compounds . these are crystalline materials with magnetic ions arranged in exchange - coupled chains that are isolated from each other by non - magnetic ions . the elementary magnetic moments are localized on the sites @xmath3 of a one - dimensional lattice , which makes them distinguishable . the associated spin operators thus commute if they belong to different sites of that lattice , @xmath4=\imath\hbar \epsilon_{\alpha\beta\gamma } s_{l}^{\gamma } \delta_{ll'}$ ] . the hilbert space of a spin-1/2 chain with @xmath5 sites is conveniently spanned by product basis vectors @xmath6 , @xmath7 . prominent among the many models employed in the context of quantum spin chain compounds is the spin-1/2 @xmath0 model , @xmath8 it describes a uniform nearest - neighbor exchange coupling with uniaxial anisotroy . periodic boundary conditions are assumed . we distinguish ferromagnetic coupling @xmath9 from antiferromagnetic coupling @xmath10 , and the planar regime @xmath11 from the axial regime @xmath12 . important special cases are the @xmath1 model @xmath13 , the @xmath2 model @xmath14 , and the ising model @xmath15 . one persistent challenge through decades of experiments on quantum spin chain compounds has been the interpretation of the observed intensity spectrum in terms of constituent quasiparticles @xcite . the chief motivation of the work reported here is to shed new light on this very issue . the approach taken is eclectic in nature , combining older and more recent results for the unified purpose of understanding the quasiparticle composition of the @xmath0 spectrum more thoroughly . the strong interest in the @xmath0 model is sustained not only by its direct relevance in experimental physics , but also by its amenability to exact analysis via bethe ansatz . the solution of the @xmath2 model was , in fact , the very problem for which bethe originally invented the method in the early days of quantum mechanics @xcite . the bethe ansatz allows for a characterization of all many - body eigenstates as composed of quasiparticles that scatter off each other or form bound states and thus turn into different quasiparticles . over the years the same basic idea has been successfully applied to many different kinds of systems and situations including interacting boson and fermion gases with contact interactions @xcite , the hubbard model for electrons on a lattice with on - site repulsion @xcite , and for bosonic and fermionic quantum field theories including the quantum sine - gordon and thirring models @xcite . the @xmath0 spectrum can be generated from different pseudovacua by the systematic creation of quasiparticles with different exclusion statistics . in most cases these pseudovacua are states of lowest energy ( physical vacua ) . an overview of the different kinds of quasiparticles that have emerged from analytic work on the @xmath0 model is shown in fig . [ fig : qpzoo ] . the boxes in the top row represent quasiparticles whose names are derived from jargon used to describe coordinate bethe ansatz solutions . these quasiparticles are characterized by strings of complex momenta with common real part and different imaginary parts @xcite . one - strings are unbound magnons , two - strings are bound magnon pairs etc . small ( large ) imaginary parts indicate loose ( tight ) binding @xcite . at @xmath16 the strings are constituent particles of ( degenerate ) multiplets of eigenstates with total spin @xmath17 . these multiplets split up energetically at @xmath18 . model . for @xmath19 , the pseudovacua of spinons and solitons are at the bottom of the spectrum . the pseudovacuum of the string particles is at the top in the axial regime ( @xmath20 ) and moves downward in the planar regime ( @xmath21 ) , reaching the center in the @xmath1 limit and coinciding there with the pseudovacuum of jordan - wigner fermions.,width=453 ] at @xmath22 the energy levels join up in new degenerate configurations , reflecting different symmetries @xcite . a more natural classification scheme for the @xmath1 spectrum is then based on string fragments . hence the name _ broken strings_. these fragments are closely related to the free lattice fermions that emerge from the jordan - wigner representation of spin-1/2 operators @xcite . at @xmath23 the imaginary parts of the string solutions grow in magnitude and diverge as @xmath24 . hence the name _ stretched strings_. the tightly bound strings are related , in the ising limit , to localized domains of reversed spins on successive lattice sites . the exclusion statistics of these domains is similar to yet subtly different from that of the strings . the boxes in the bottom row of fig . [ fig : qpzoo ] represent quasiparticles that are complementary to the string particles . the string pseudovacuum contains the maximum number of spinons or solitons . the spinon and soliton vacua contain strings at maximum capacity . whereas the string particles have integer - valued exclusion statistics , the spinons and solitons are realizations of fractional statistics @xcite . they are both semions but with different pseudovacua . the exclusion principle for semions is , roughly speaking , halfway between those applicable for fermions and bosons . if it takes @xmath25 particles to lower the number of orbitals in a band available for occupancy by one , then @xmath26 describes fermions , @xmath27 semions , and @xmath28 bosons ( as a limit ) . the string particles and the semionic particles are natural building blocks for the systematic construction of a complete @xmath0 eigenbasis from different pseudovacua . the configurations of string particles and semionic particles are both unique and preserved in every @xmath0 eigenstate . the ground state ( physical vacuum ) of the @xmath0 antiferromagnet @xmath10 coincides with the pseudovacua of the semions . the spinon vacuum is unique and coincides with the non - degenerate ground state of the @xmath0 model at @xmath29 for even @xmath5 . the soliton vacuum is twofold and coincides with the ground state of the @xmath0 model at @xmath23 for @xmath30 . let us briefly illustrate the relation between the particles from the top and bottom rows in fig . [ fig : qpzoo ] with two simple scenarios , one in configuration space and the other in momentum space . in the ising limit , @xmath31 has simple product eigenstates . the ferromagnetic state @xmath32 is the unique pseudovacuum for domains of consecutive flipped spins , @xmath33 , as shown in top part of fig . [ fig : qpis22ar ] . the twofold nel state , @xmath34 , @xmath35 , by contrast , is the pseudovacuum for antiferromagnetic domain walls of the kind @xmath36 , @xmath37 as shown in the bottom part of fig . [ fig : qpis22ar ] . these domain walls ( named solitons ) have effective spin @xmath38 . the integer - valued exclusion statistics of domains is associated with the fact that neighboring particles must be separated by any positive integer number of lattice sites . the fractional exclusion statistics of domain walls , on the other hand , is associated with the fact that one of two lattice sites may be shared by neighboring particles . at @xmath24 in real space ( lattice of @xmath5 sites ) . they are either domains of flipped spins embedded in the unique vacuum with all spins up ( top ) or domain walls embedded in the twofold vacuum with all spins alternatingly up / down ( bottom ) . , width=453 ] in the @xmath1 limit , @xmath31 is equivalent to a system of @xmath39 free lattice fermions in a band @xcite . the ground state corresponds to the intact fermi sea at @xmath40 as shown in fig . [ fig : xxfermvspin ] . the entire spectrum can be generated systematically via particle excitations @xmath41 , hole excitations @xmath42 , and particle - hole excitations @xmath43 . the fermi - sea ground state can be viewed as the pseudovacuum for semionic spinons . we introduce a threshold momentum @xmath44 that varies with @xmath39 . when a fermion is removed from the intact sea , @xmath44 slightly increases to generate not just one but two vacancies in the interval @xmath45 of the band . these two holes are then identified with a pair of spin - up spinons . when a fermion is added outside the intact fermi sea , @xmath44 slightly decreases to produce two particles in the region @xmath46 . they are identified with a pair of spin - down spinons . a fermionic particle - hole excitation leaves @xmath44 unchanged and is interpreted as a pair of spinons with opposite spin orientation . the entire spectrum can thus be described in the form of spinon configurations . at @xmath22 in reciprocal space ( fermionic band with fermi momentum @xmath47 ) . they are either spinless jordan - wigner fermion or semionic spinons with spin @xmath38.,width=453 ] the main goal of this work is to identify the relationship between the string particles and the semionic particles for @xmath31 at @xmath48 in particular and to illuminate how this relationship can be tracked between the three points in parameter space . in sec . [ sec : ba ] we set the stage for this investigation by reviewing the relevant bethe - ansatz representations that best serve our purpose . we then discuss the quasiparticle composition of the @xmath0 spectrum in the @xmath2 limit ( sec . [ sec : xxx ] ) , the ising limit ( sec . [ sec : ising ] ) , and the @xmath1 limit ( sec . [ sec : xx ] ) . the coordinate bethe ansatz for @xmath31 uses the magnon vacuum @xmath49 with energy @xmath50 as its reference state . the bethe form @xcite of any eigenvector in the invariant subspace with magnetization @xmath51 , @xmath52 has coefficients of the form @xmath53 determined by @xmath54 magnon momenta @xmath55 and one phase angle , @xmath56 , for each magnon pair . the sum @xmath57 is over the permutations of the indices @xmath58 . the @xmath55 and @xmath59 satisfy the bethe ansatz equations ( bae ) , @xmath60 the energy and the wave number of an eigenstate with magnon momenta @xmath61 are @xmath62 reasons of practicality dictate the use of different transformations of the bae in the regimes of axial anisotropy @xmath12 , planar anisotropy @xmath63 , and isotropic exchange @xmath14 . it is convenient to introduce rapidities @xmath64 , @xmath65 which renders the limit @xmath66 smooth from both sides . the bae thus transform into @xmath67 where @xmath68 the bethe quantum numbers ( bqn ) @xmath69 of integer or half - integer value reflect the multivaluedness of the logarithm used in the transformation . they are subject to restrictions that will be discussed case by case . the energy expression in becomes @xmath70 note that we are , effectively , dealing with a single anisotropy parameter , @xmath71 , that is real in one regime , imaginary in the other , and zero at the isotropy point . this parametrization is particularly useful for tracking the spectrum between the axial and planar regimes across the point of higher rotational symmetry . slightly different parametrizations are more adequate for the exploration of the limits @xmath24 in the axial regime and @xmath72 in the planar regime . at @xmath23 we use the transformation @xcite , @xmath73 instead of ( [ eq:6 ] ) , to bring the bae ( [ eq:16plj3 ] ) into the form @xmath74 - \imath } { \coth(\eta ) \tan \left[(z_i - z_j)/2\right ] + \imath } , \qquad i = 1,\dots , r.\ ] ] the associated trigonometric bae are @xmath75 where @xmath76 the second term in ( [ eq:13 ] ) ensures that the set @xmath77 remains the same as an eigenstate is tracked across the axial regime @xcite . here @xmath78 is the floor function ( integer part of @xmath79 ) . for the analysis of solutions that include complex magnon momenta we invoke the string hypothesis for the rapidities @xcite : @xmath80 the index @xmath3 distinguishes rapidities belonging to the same string ( of size @xmath81 ) . the index @xmath82 distinguishes different strings of the same size . the string ansatz ( [ eq:104plj3 ] ) produces only asymptotic solutions of the bae ( [ eq:19plj3 ] ) for @xmath30 . the finite-@xmath5 corrections are , in general , exponentially small , and not all finite-@xmath5 solutions fit the string template ( [ eq:104plj3 ] ) @xcite . however , neither corrections nor exceptions affect macroscopic systems . in the ising limit , where the spread of imaginary parts in ( [ eq:104plj3 ] ) diverges , all corrections and exceptions disappear even for finite @xmath5 . a given string solution of ( [ eq:19plj3 ] ) with magnetization @xmath51 is described by @xmath54 rapidities that breaks down into configurations of strings such that the constraint , @xmath83 is satisfied , where the set @xmath84 identifies those sizes of strings that occur in a given eigenstate . with the functions @xmath85 we rewrite eqs . ( [ eq:19plj3 ] ) in the form @xmath86^{n } = \prod_{j \neq i}^r \varphi_{2}(z_{i}-z_{j } ) , \quad i = 1,\dots , r.\ ] ] when we substitute the string ansatz ( [ eq:104plj3 ] ) into eqs . ( [ eq:106plj3 ] ) we obtain @xmath87^{n } = \prod_{(m',\beta ) \neq ( m,\alpha)}\prod_{k=1}^{m'}\ ; \varphi_{2}(z_{\alpha}^{m , l}-z_{\beta}^{m',k } ) \prod_{k\neq l}^{m } \varphi_{2}(z_{\alpha}^{m , l}-z_{\alpha}^{m , k}),\ ] ] for @xmath88 and @xmath89 , where @xmath90/2\big ) } { \sin\big([z_{\alpha}^{m}+\imath\eta(m+1+\nu-2l)]/2\big)}.\ ] ] to determine the real parts , @xmath91 , we form the product of all eqs . ( [ eq:107plj3 ] ) for fixed @xmath92 : @xmath93^{n } = \prod_{(m',\alpha ' ) \neq ( m,\alpha ) } \left[\prod_{k=1}^{m'}\prod_{l=1}^{m } \varphi_{2}(z_{\alpha}^{m , l}-z_{\alpha'}^{m',k})\right ] \left[\prod_{l=1}^{m}\prod_{k\neq l}^{m } \varphi_{2}(z_{\alpha}^{m , l}-z_{\alpha}^{m , k})\right].\ ] ] each expression in square brackets can be simplified massively , producing the bae for the @xmath91 , @xmath94^{n } = \prod_{(m',\alpha ' ) \neq ( m,\alpha ) } \varphi_{m'-m}(z_{\alpha\alpha'}^{mm ' } ) \varphi_{m'+m}(z_{\alpha\alpha'}^{mm ' } ) \prod_{l=1}^{m-1 } \left[\varphi_{m'+m-2l}(z_{\alpha\alpha}^{mm'})\right]^{2 } , \end{aligned}\ ] ] with @xmath95 . the associated trigonometric bae , @xmath96 @xmath97 depend on a set @xmath98 of bqn that reflects the specific string combination of any given eigenstate . the energy and wave number of that state are @xmath99 \mathrm{mod}(2\pi).\end{aligned}\ ] ] the range of the @xmath100 will be discussed first for the case @xmath16 in sec . [ sec : isotropic ] and then for the axial regime including the ising limit in sec . [ sec : ising ] . at @xmath16 we retain the rapidities from ( [ eq:6 ] ) , the trigonometric bae in the form ( [ eq:15 ] ) , and the energy expression ( [ eq:16 ] ) . the string hypothesis now reads @xmath101 the bae for the real parts @xmath91 are eqs . with ( [ eq:108plj3 ] ) replaced by @xmath102 the associated trigonometric bae for the @xmath91 then take on the form with @xmath103 from ( [ eq:8 ] ) and @xmath104 from ( [ eq:7 ] ) . the energy and wave number of an eigenstate specified by the set @xmath98 are @xmath105 \mathrm{mod}(2\pi).\ ] ] the string hypothesis sets the range of the @xmath100 ( with @xmath106 implied ) as follows @xcite : @xmath107 where @xmath108 is the number of @xmath81-strings ( distinguished by running index @xmath109 ) in the eigenstate . note that the range of allowed values becomes narrower for all sizes if a string of any size is added . the @xmath100 for a given combination @xmath110 and a given value of @xmath81 are either all integers or all half - integers such that the border values of the range ( [ eq:10 ] ) are realized . the bethe state characterized by a set @xmath98 is the highest - weight component ( i.e. the state with @xmath111 ) of an @xmath112-multiplet with total spin @xmath113 the other components of any given @xmath112-multiplet are generated by the addition of magnons with zero momentum . these have no effect on the energy or the wave number . consider the case of a highest - weight state with total spin ( [ eq:11 ] ) that only contains 1-strings . suppose this state is specified by the following set of bqn subject to the constraint ( [ eq:10 ] ) : @xmath114 the rapidities @xmath115 derived from the bae ( [ eq:1 ] ) , @xmath116 are real and , with rare exceptions , finite . the member state with @xmath117 of the same multiplet has one extra rapidity , @xmath118 , representing the additional magnon with @xmath119 . the bae ( [ eq:1 ] ) for this state are then satisfied with the same @xmath115 , @xmath120 and with @xmath121 if we set the bqn as follows : @xmath122 in the more general case , where the highest - weight state under consideration contains strings with @xmath123 , the shifts in the already existing bqn , @xmath100 , and the value of the new bqn , @xmath124 , will be different . in sec . [ sec : xxx ] we will treat the strings as interacting particles and examine their exclusion statistics . the string ansatz at @xmath125 @xcite will not be used here . simplifications and residual complications that occur in the limit @xmath72 can be seen in the raw form ( [ eq:16plj3 ] ) of the bae . there are two categories of solutions , both of which are ubiquitous . regular and singular solutions are distinguished by the absence or presence of pairs of critical magnon momenta with @xmath126 , which make both the numerator and the denominator in ( [ eq:16plj3 ] ) vanish as @xmath72 . all regular solutions produce real magnon momenta from @xmath127 , whereas singular solutions include critical magnon pairs that are either both real or form a complex - conjugate pair @xcite . the latter can be interpreted as fragments of strings that exist throughout the planar regime . since all critical pairs are associated with a twofold degeneracy of eigenstates with equal wave number , the singular features as imposed by the bae in the limit @xmath72 can be removed by unitary transformations . this erases , at @xmath22 , all traces of the string nature , at @xmath128 , of complex solutions . the magnon momenta thus regularized are @xmath129 the associated bqn are integers for odd @xmath54 and half - integers for even @xmath54 with range @xmath130.\ ] ] their relation to the bqn ( [ eq:10 ] ) and ( [ eq:30 ] ) of the @xmath112-multiplet states will be discussed in sec . [ sec : xx ] . the regularized bae solutions ( [ eq:33 ] ) describe hard - core bosons . the phase shift is @xmath131 for all two - particle interactions . these hard - core bosons are equivalent to the jordan - wigner fermions @xcite that have been instrumental in most studies of the @xmath1 model . we are now ready to explore the relationship between the complementary quasiparticle compositions of the @xmath0 spectrum . we begin in sec . [ sec : xxx ] with string particles and the complementary spinon particles for the @xmath2 case . in sec . [ sec : ising ] we then discuss the effects of axial anisotropy on the strings and their relationship to ferromagnetic domains in the ising limit . complementary to these domains are the soliton particles in the shape of antiferromagnetic domain walls . in the planar regime the strings evolve differently . what remains of them in the @xmath1 limit are fragments that act like hard - core bosons or , equivalently , free jordan - wigner fermions . complementary to the latter are again the spinons as will be discussed in sec . [ sec : xx ] . the very structure of the coordinate bethe ansatz suggests that the strings ( [ eq:5 ] ) can be interpreted as quasiparticles . there exists a universal energy - momentum relation as implied by ( [ eq:9 ] ) . the particle interaction is encoded in the set of momenta ( or rapidities ) dictated by the bae and in the phase shifts associated with elastic two - particle collisions . the exclusion statistics of strings is determined by the rule ( [ eq:10 ] ) governing the range of bqn and by the relation ( [ eq:11 ] ) governing the capacity for strings in a highest - weight state of given @xmath112 . the total number of @xmath112-multiplets with string content @xmath132 becomes the solution of a standard combinatorial problem : @xmath133 where @xmath134 are statistical capacity constants and statistical interaction coefficients , respectively , that are specific to the string particles @xcite . taking into account the @xmath135-fold degeneracy of each multiplet , this classification accounts for the complete spectrum , @xmath136 with the dependence of @xmath112 on @xmath110 given in ( [ eq:11 ] ) . to illustrate the string composition of the @xmath2 spectrum and to explain its relationship to the complementary spinon composition we consider a chain of @xmath137 sites . in table [ tab : comstrn6 ] we list all combinations @xmath110 permitted by ( [ eq:11 ] ) . also listed is the range of all @xmath100 present in each combination as inferred from ( [ eq:10 ] ) . the distinct configurations of @xmath100 thus allowed produce the highest - weight states of all @xmath112-multiplets as shown in fig . [ fig : stringn6 ] ( left ) . the quantum number @xmath112 depends on the combinations @xmath110 via ( [ eq:11 ] ) and the wave number @xmath138 depends on the configurations @xmath98 via ( [ eq:9 ] ) . each row of @xmath100 in fig . [ fig : stringn6 ] ( left ) represents the distinct string motif of an @xmath112-multiplet . the solution of the bae ( [ eq:1 ] ) thus specified is for the highest - weight component of the multiplet . the non - highest - weight components are characterized by additional bqn as described at the end of sec . [ sec : isotropic ] . the role of these additional bqn will be discussed in sec . [ sec : ising ] for @xmath23 ( see fig . [ fig : stringn6plus ] ) and in sec . [ sec : xx ] for @xmath125 ( see fig . [ fig : stringn6minus ] ) . .string composition @xmath110 of all @xmath112-multiplets for @xmath137 . each row describes a distinct combination for a total of 7 . each combination produces @xmath139 multiplets for a total of 20 . each multiplet represents @xmath140 states for a total of 64 . associated with each string in a given combination is a bqn @xmath100 of the range shown . [ cols="^,<,<,^,^,<",options="header " , ] -multiplets of the @xmath2 model for @xmath137 in the string representation ( left ) and spinon representation ( right ) . the quantum numbers @xmath112 and @xmath141 of each multiplet are stated on the far left . the positions of the numbers @xmath142 and the vertical lines ( left ) mark the values of the @xmath100 and their range . the @xmath143 , emphasized by circles , are markers for the spinon configurations ( right ) , where they are reproduced as small circles . each square on the right represents a spinon . the set of squares is complementary to that of circles with the same range . the number @xmath144 inside each square marks the spinon orbital . the energy of each @xmath112 multiplet is stated on the far right.,width=377 ] elements of the string motif serve as the template of the spinon motif of the same @xmath112-multiplet . spinons are specified by their spin and momentum quantum numbers . the spinon interaction depends on the particle momenta and ( in general ) also on the particle spins . the unique @xmath2 ground state for even @xmath5 is the spinon vacuum . the ground state for odd @xmath5 is fourfold degenerate and contains exactly one spinon . the total number @xmath145 of spinon is restricted to be even ( odd ) for even ( odd ) @xmath5 and can assume the values @xmath146 . counting the spin - up spinons @xmath147 and spin - down spinons @xmath148 separately produces the relations @xmath149 it is useful to introduce spinon orbitals associated with distinct spinon momentum quantum numbers @xmath150 . the wave number of any @xmath2 multiplet can be expressed in terms of the spinon orbital momenta as follows : @xmath151 the range of @xmath150 depends on both @xmath5 and @xmath145 : @xmath152 the number of available orbits to @xmath145 spinons , @xmath153 , thus decreases by one for every two spinons added , @xmath154 , which is one way of recognizing the semionic nature of spinon particles . the exact exclusion statistics of spinons is encoded in the number of ways @xmath155 spin - up spinons and @xmath156 spin - down spinons can be distributed among @xmath157 accessible orbitals . the spinon multiplicity expression @xcite , @xmath158 @xmath159 is cast in the same general form as its string counterpart ( [ eq:19 ] ) . summation of @xmath160 over all allowed values of @xmath161 accounts for all @xmath162 states . the complementary relationship between the string and spinon particles is reflected in their motifs . in fig . [ fig : stringn6 ] we show the string motif and spinon motif of all 20 @xmath112-multiplets for @xmath137 side by side . the number of spinons contained in each eigenstate of a given @xmath112-multiplet is equal to the number of vacancies left by the 1-string bqn , @xmath143 , across the range ( [ eq:10 ] ) . in fig . [ fig : stringn6 ] ( right ) we have marked the positions of the @xmath143 by circles and the vacancies by squares . the number inside each square denotes the orbital @xmath150 to which every spinon belongs . the available orbitals depend on @xmath145 via ( [ eq:23 ] ) . the rules for assigning spinons of particular momentum quantum numbers to orbitals must ensure ( i ) that the @xmath150 reproduce , via ( [ eq:24 ] ) , the wave number @xmath138 already known via ( [ eq:9 ] ) and ( ii ) that the permissible spin orientations are consistent with the quantum number @xmath112 . the allowed combinations of spinon orbitals are encoded in those @xmath112-multiplets that do not contain any strings with @xmath123 . here the positions of the 1-strings , i.e. the circles in the spinon motifs , play the role of delimiters between successive spinon orbitals . a spinon orbital with @xmath163 spinons has an orbital spin @xmath164 , @xmath165 . if all @xmath145 spinons are in the same orbital such as in the first two motifs ( numbered from top to bottom ) this spinon configuration represents a multiplet with @xmath166 . any distribution of spinons into multiple orbitals represents more than one @xmath112-multiplet . the multiplets represented are determined via the decomposition of tensor products of orbital spins . for example , if we have one spinon with @xmath167 and three spinons with @xmath168 , that decomposition reads @xmath169 . the associated multiplets with @xmath170 and @xmath171 are found in motifs three and thirteen , respectively . the configuration with two spinons in each of the same two orbital represents three @xmath112-multiplets according to @xmath172 . they are found in motifs four , fourteen , and twenty , respectively . the systematic application of these rules assigns a unique spinon momentum and spin content to every @xmath112-multiplet . the energetic split of different @xmath112-multiplets associated with the same spinon momenta is caused by a coupling between orbital spins . in the haldane - shastry model , which has higher symmetry , that coupling is absent and these particular @xmath112-multiplets remain degenerate @xcite . in the @xmath1 model , which has lower symmetry , even the @xmath112-multiplets split up energetically as will be discussed in sec . [ sec : xx ] @xcite . the translation and reflection symmetries of @xmath31 dictate that every state with wave number @xmath138 can be transformed into a state with wave number @xmath173 and the same energy , implying that any state with @xmath174 is its own image . these transformation properties are reflected in the string motif @xmath175 and in the spinon motif @xmath176 as is evident in fig . [ fig : stringn6 ] . in the following we describe three ways of generating the spectrum of the ising chain , each producing a distinct set of quasiparticles with different exclusion statistics . we begin with the analysis of the string solutions of the bae from sec . [ sec : axial ] . to arrive at a non - divergent spectrum in the ising limit of @xmath31 we rescale the exchange coupling as follows : @xmath177 the rescaled energy expression ( [ eq:17plj3 ] ) of the bethe ansatz in raw form thus becomes @xmath178 evidently , real magnon momenta contribute only summarily to the energy , namely via the first term in ( [ eq:31plj3 ] ) . non - vanishing terms @xmath179 can only come from complex @xmath55 with infinite imaginary parts . we first track the highest - weight states of the @xmath112-multiplets from the @xmath2 limit to the ising limit . for these states , all solutions that are real at @xmath16 stay real and all solutions that are complex at @xmath16 have imaginary parts that diverge as @xmath24 . the non - highest - weight states evolve far less uniformly . the additional rapidities , which are all equal to @xmath180 at @xmath16 , become , in general , finite and thus contribute to the energetic split of the @xmath112-multiplets . in some states , the extra @xmath64 stay real , in other states they combine to form complex pairs . it appears that this transformation from real to complex rapidities takes place throughout the axial regime . here we focus on the end product at @xmath181 , where the ferromagnetic domains in ising product eigenstates become the natural quasiparticles . expanding the trigonometric bae ( [ eq:1 ] ) of the axial regime about the ising limit produces , in leading order , a set of linear bae for the real parts of the rapidities , @xmath182 where the new set @xmath183 of bqn , related to the original set @xmath98 by a shift that depends on @xmath5 and @xmath108 , has the same range ( [ eq:10 ] ) @xmath184 which guarantees the correct number of states from the highest - weight type . as the first step in the solution of the linear bae ( [ eq:87 ] ) for these states we introduce the quantities @xmath185 equations ( [ eq:87 ] ) summed over @xmath109 can thus be brought into the form @xmath186 with @xmath187 guaranteed to be positive . equation ( [ eq:90 ] ) is solved by matrix inversion . substitution of the solution @xmath188 into ( [ eq:87 ] ) yields the rapidities @xmath189 more explicit solutions in compact form for the situations where only 1-strings or only 2-strings are present read @xmath190 @xmath191 if a single @xmath81-string with @xmath192 is present we have @xmath193 with @xmath194 . eigenstates with @xmath195 for @xmath137 in the string representation ( two columns ) . the quantum numbers @xmath196 , @xmath112 and @xmath141 of each state are stated on the left . the vertical lines indicate the range ( [ eq:10 ] ) of the @xmath143 . the positions of the numbers @xmath142 between the vertical lines represent the values of the @xmath100 for the highest - weight states ( with @xmath111 ) . the additional bqn of the non - highest - weight states ( with @xmath197 ) are all located on the outside . in the ising limit each state evolves into a translationally invariant linear combination of ising product states with domains of lengths as indicated on the far right in each column.,width=491 ] in fig . [ fig : stringn6plus ] we have reproduced the bqn of all 20 highest - weight states for @xmath137 at @xmath16 from fig . [ fig : stringn6 ] and have added the bqn for all 22 non - highest - weight states with @xmath195 . the additional @xmath143 of non - highest - weight states , obtained for @xmath16 as described at the end of sec . [ sec : isotropic ] , are located beyond the range of the @xmath143 for the highest - weight states ( vertical lines ) . the choice of the extra @xmath143 and the associated shift of the @xmath100 already present is not unique in most cases . our choice was guided by the aim to avoid multiple @xmath143 of the same value and to restore the symmetry relations described at the end of sec . [ sec : xxx ] . for states with @xmath174 , which are their own images under the symmetry transformation , neither goal was fully achievable for @xmath137 . the @xmath143 configurations are asymmetric in several instances and two identical @xmath143 are unavoidable in one instance . these aesthetic flaws pose no problem for what follows . to the right of each string motif in fig . [ fig : stringn6plus ] we have added another motif in the shape of a ket . it identifies each string solution as tracked from @xmath16 to @xmath181 with an ising eigenstate as composed of linear combinations of ferromagnetic domains . we have already introduced these domains in the context of fig . [ fig : qpis22ar ] . here we use a notation conducive to linking ising product eigenstates with obvious domain composition to bethe eigenstates with specific string structure in the limit @xmath181 . let the set of @xmath198 product vectors with @xmath199 flipped spins that are generated from one such state , @xmath200 , via translations be represented by @xmath201 . let @xmath202 represent the set of @xmath203 translationally invariant linear combinations of all product states that contain @xmath204 domains of length @xmath205 subject to the constraint @xmath206 . any set of the latter kind is constructed from one or several sets of the former kind with matching domain content . for @xmath137 there are seven sets of translationally invariant ising eigenstates at @xmath195 thus constructed : [ eq:91sub ] @xmath207 in fig . [ fig : stringn6plus ] we have identified each of these 42 states with a particular bae solution in the limit @xmath24 . the identification is straightforward for the highest - weight states . each bqn @xmath100 represents exactly one domain of size @xmath208 . however , the evolution of the non - highest - weight states between @xmath16 and @xmath24 is far less predictable . the extra bqn are all of the type @xmath143 at @xmath16 and the associated rapidities are real and infinite . in some cases all extra rapidities stay real . the domain structure of these states is determined by the @xmath100 just as in highest - weight states . however , for the majority of non - highest - weight states some of the rapidities that start out real at @xmath16 turn into complex - conjugate pairs at @xmath23 with imaginary parts that diverge as @xmath24 . this metamorphosis necessitates a reconfiguration of the associated bqn ( not shown in fig . [ fig : stringn6plus ] ) . of the 22 non - highest - weight states for @xmath137 only six do not acquire additional complex rapidities between @xmath16 and @xmath24 , namely those states for which the string content encoded in the @xmath100 matches the domain content encoded in the ket . the identification of the bae solutions in the limit @xmath24 with ising eigenstates of specific domain content may be less certain in longer chains where states with different @xmath209 but equal @xmath196 are degenerate . that is the case for the two sets of states @xmath210 and @xmath211 in a chain with @xmath212 sites , for example . however , it seems reasonable to assume that this degeneracy is absent at @xmath213 in most if not all cases and thus guarantees that the bae solutions have , in general , a unique domain content in the limit @xmath24 . even though the systematics of the transformation of some bae solutions between @xmath16 and @xmath24 appears elusive at present , the systematics of the endproduct , namely the domain structure of the ising spectrum for arbitrary @xmath5 , is well known . an ising chain of length @xmath5 can accommodate domains with @xmath214 consecutive flipped spins . domains of size @xmath215 are treated as distinct species of independent particles . the capacity for domains is restricted by the relation @xmath216 the total number of ising eigenstates with domain content @xmath217 is governed by a multiplicity expression somewhat similar to ( [ eq:19 ] ) yet different @xcite : @xmath218 @xmath219 the capacity of the system for domains is controlled by @xmath220 and the statistical interaction between domains by @xmath221 . all domains have the same energy , @xmath222 . the statistical mechanics of domains for the ising chain has been carried out exactly @xcite , reproducing familiar results . in the context of fig . [ fig : qpis22ar ] we have already qualitatively described the antiferromagnetic domain walls ( solitons ) that are complementary to the ferromagnetic domains . among the four distinct bonds in the general product state @xmath223 , the bonds @xmath36 , @xmath37 represent solitons with spin @xmath224 , @xmath225 , respectively , and @xmath226 , @xmath227 are vacuum bonds . close - packed solitons with like spin orientation reside on successive bonds ( e.g. @xmath228 ) , whereas close - packed solitons with opposite spin orientation are separated by one vacuum bond ( e.g. @xmath229 ) . each of the seven sets of ising eigenstates ( [ eq:91sub ] ) contains a specific number of spin - up and spin - down solitons , @xmath230 all solitons have the same energy , @xmath231 . the multiplicity expression for solitons @xcite , @xmath232 @xmath233 is similar to expression ( [ eq:25 ] ) for spinons . solitons and spinons are both semions but have different pseudovacua . whereas the spinon vacuum was found to be unique , the soliton vacuum is twofold , consisting of the two product nel states , @xmath234 and @xmath235 or linear combinations thereof . the soliton vacuum , like the spinon vacuum , is realized only in chains with even @xmath5 . here we investigate what happens to the highest - weight and non - highest - weight states of the @xmath112-multiplets in the presence of planar exchange anisotropy , particularly in the limit @xmath22 . we have seen that in the axial regime there exists a tendency for real bae solutions to become complex . as @xmath24 all imaginary parts diverge , binding the magnons tightly into domains . in the planar regime , there exists a trend in the opposite direction . at @xmath22 all bae solutions can be regularized as stated in sec . [ sec : planar ] , making all magnon momenta real . all strings with @xmath123 break up into 1-strings . the configurations allowed by the set ( [ eq:34 ] ) of bqn , which describe regularized solutions , produce the complete spectrum for arbitrary @xmath5 . eigenstates with @xmath195 for @xmath137 in the string representation ( two columns ) . the quantum numbers @xmath196 , and @xmath141 of each state are stated on the left . also shown on the left is the quantum number @xmath112 of the multiplet at @xmath16 from which the state evolved . the states are listed in the same sequence as those in fig . [ fig : stringn6plus ] . the positions of the 1 represent the values of the @xmath143 from ( [ eq:34 ] ) and constitute the motifs of the ( broken ) string configuration of the @xmath1 eigenstates . the motif of the complementary spinon configuration is marked by the squares ( spin - up spinons ) and diamonds ( spin - down spinons ) as explained in the text.the momentum quantum number @xmath150 and the spin orientation @xmath236 of each spinon present in that state are given to the right of the dual string / spinon motif . , width=491 ] in fig . [ fig : stringn6minus ] we show the configurations of @xmath143 of all 42 eigenstates with @xmath195 at @xmath22 for @xmath137 in the same sequence as the corresponding @xmath112-multiplet states at @xmath16 have been listed in fig . [ fig : stringn6plus ] . all multiplet states at @xmath16 that contain only 1-strings keep the same configuration at @xmath22 . in the other states , all @xmath143 stay in the same position except the pair of @xmath143 that have identical values in fig . [ fig : stringn6plus ] . they are replaced in fig . [ fig : stringn6minus ] by a pair of distinct @xmath143 . furthermore , each @xmath237 in fig . [ fig : stringn6plus ] is replaced by two @xmath143 in fig . [ fig : stringn6minus ] and the one occurrence of a @xmath238 is replaced by three @xmath143 . whereas the general rules for these substitutions still elude us , the constraints imposed by symmetry and conservation laws eliminate any ambiguity for the case @xmath137 shown here . the resulting configuration of @xmath143 produces , via ( [ eq:33 ] ) , the exact configurations of jordan - wigner fermion momenta in a periodic chain @xcite . the statistical mechanics of the regularized 1-strings is that of a system of free jordan - wigner fermions @xcite . each @xmath1 eigenstate thus identified by its composition of 1-strings has a unique composition of spinons . the rules for inferring the spinon motif from the string motif are consistent with the rules described in sec . [ sec : isotropic ] for the case @xmath16 but there are some noteworthy differences . the lower rotational symmetry at @xmath22 , which splits up the @xmath112-multiplet degeneracy , makes it possible to assign to each @xmath1 state not only a spinon configuration with unique momentum quantum numbers but to the spinons in each orbital also a unique spin orientation . the spinon motif is encoded in the string motif of each state shown in fig . [ fig : stringn6minus ] as described in the following . ( i ) consider the vertical lines dividing the space of the @xmath143 into two domains the inside and the outside , the latter wrapping around at @xmath239 . ( ii ) every @xmath143-vacancy ( small circle ) inside represents a spin - up spinon ( marked by a square ) and every @xmath143 ( 1 ) outside represents a spin - down spinon ( marked by a diamond ) . ( iii ) any number of adjacent spinons in the motif are in the same orbital , i.e. have the same momentum quantum number @xmath150 from the set ( [ eq:23 ] ) . two squares or diamonds that are separated by @xmath240 consecutive 1 s have spinon quantum numbers separated by @xmath241 . ( iv ) the spinon momentum quantum numbers are sorted in increasing order from the line on the left toward the right through the inside domain @xmath242 and toward the left with wrap - around through the outside domain @xmath243 . the spinon motif of fig . [ fig : stringn6minus ] establishes the much needed link between spinon motif of the fermion representation at @xmath22 introduced in refs . @xcite and the spinon motif of the string representation at @xmath16 introduced in fig . [ fig : stringn6plus ] . unlike the solitons at @xmath24 , the spinons at @xmath22 are not free . nevertheless , an exact statistical mechanical analysis of spinons at @xmath22 is possible @xcite . away from the ising limit , the solitons become interacting particles as well . they can be tracked all the way from @xmath24 to @xmath22 . a comparison between the spinon composition and soliton composition of the @xmath1 eigenstates can be found in ref . the long - established integrability of @xmath31 imposes stringent constraints on the nature of the interaction between quasiparticles of any kind , reducing it , effectively , to two - body scattering events . as a consequence , these quasiparticles have infinite lifetimes and can thus be regarded as structural elements of the @xmath0 eigenstates . this makes it possible to systematically generate the complete spectrum from the pseudovacuum of one or the other set of quasiparticles . in this work we have discussed two complementary sets of quasiparticles , one set being the basic elements of string solutions of the coordinate bethe ansatz and the other the semionic spinons or solitons . our focus has been on developing interlocking motifs for each set of quasiparticles for the purpose of tracking them across the axial and planar regimes of @xmath31 . starting from the symmetry point @xmath16 in parameter space we have identified opposite trends , in the two regimes , in the evolution of string particles , which can be thought of as bound clusters of magnons . in the axial regime , the strings show a tendency of increasing tightness in the binding within each cluster ( rapidities with growing imaginary parts ) and a tendency of merging clusters ( real rapidities becoming complex ) . the endproduct at @xmath181 are ferromagnetic domains of flipped spins . in the planar regime , the strings appear to loosen up and break apart . at @xmath22 it is possible to transform away all complex rapidities and with them any trace of magnon clustering . the free magnons behave like hard - core bosons or free fermions . for @xmath19 the pseudovacuum of string particles is located at the top of the spectrum throughout the axial regime and then migrates down toward the center of the spectrum in the planar regime . the ground state ( physical vacuum ) of @xmath31 is unique in the planar regime and twofold in the axial regime . it can be identified as the pseudovacuum of spinons or solitons , respectively . we have given detailed descriptions of spinons at @xmath22 and @xmath16 and of solitons at @xmath181 , including their relationship to the string particles . what remains to be investigated is the metamorphosis of the string particles between the special points @xmath48 in parameter space . this is work in progress . of particular interest is an analysis of the transformations for which we have given here only qualitative descriptions and the impact of these transformations on the motifs such as displayed in figs . [ fig : stringn6]-[fig : stringn6minus ] including the effects on the complementary spinon and soliton particles .

thick discs are defined to be disc - like components with a scale height larger than that of the `` canonical '' or `` classical '' discs . they are most easily detected in close to edge - on galaxies in which they appear as a roughly exponential excess of light a few thin disc scale heights above the midplane . they were first detected and described by tsikoudi ( 1979 ) and burstein ( 1979 ) . the milky way was soon found to also host a thick disc ( gilmore & reid 1983 ) made of old ( bensby et al . 2005 ) , metal - poor , and @xmath2-enhanced stars ( fuhrmann 1998 ; prochaska et al . observations now suggest that thick discs in galaxies in the local universe are ubiquitous ( yoachim & dalcanton 2006 ; comern et al . 2011a ) . we recently showed evidence of thick discs being significantly more massive than previously considered with stellar masses ranging from 0.3 to 3.0 times that of the thin disc depending on the galaxy mass and the assumed mass - to - light ratios ( comern et al . 2011b ; hereafter co11b ) . this study confirmed that galaxies dominated by the thick disc are usually those with lower masses ( yoachim & dalcanton 2006 ) . hosting such a large fraction of the stellar mass makes thick discs likely reservoirs of some of the `` local missing baryons '' . a few galaxies , such as ngc 4013 , host an additional extended component which could be interpreted as a second thick disc or a flattened halo ( comern et al . 2011c ) . the properties of those more massive than expected thick discs give some constraints on their origin . higher mass galaxies could have their thick discs with a low relative mass explained by internal secular heating ( see , e.g. , bovy et al . 2012 ) , by the heating caused by the interaction with a satellite galaxy ( see , e.g. , qu et al . 2011 ) , by the accretion of stars from satellite galaxies ( see , e.g. , abadi et al . 2003 ) , and/or because of the radial migration of stars ( see , e.g. , schnrich & binney 2009 ; loebman et al . 2011 ) , but the very large thick disc mass fraction found in low - mass galaxies indicates an in situ origin during or shortly after the buildup of the galaxy for a large fraction of their thick disc stars . two in situ and high - redshift thick disc formation models have been put forward . elmegreen & elmegreen ( 2006 ) and bournaud et al . ( 2009 ) proposed a scenario where the thick disc forms from the primordial galaxy discs , which contained large clumps of star formation whose scale height was comparable to that of present - day thick discs . those large clumps would have dissolved into what we know as thick discs . brook et al . ( 2007 ) proposed thick discs to be made of stars in gas - rich proto - galactic fragments which merged and of stars created during that merger . in both scenarios , the thin disc would have formed afterwards from gas remaining from the initial galaxy formation processes and from new gas accreted through cold flows . these two models predict different properties for the thick discs . if they have been created from the dissolution of giant star - forming clusters , one would expect them to be genuine discs , with a high rotation speed and a luminosity profile similar to that of thin discs . if thick discs result from the merger of proto - galactic fragments they would , at least in some cases , host a significant fraction of lagging and/or counterrotating stars . our comprehensive detection and modeling of thick+thin discs in edge - on galaxies started in co11b allow us to study breaks , both truncations and antitruncations , in the radial profile of the thick discs , hitherto not possible . a radical truncation in the radial light profile was first discussed in edge - on galaxies by van der kruit ( 1979 ) and van der kruit and searle ( 1981a , b ; 1982 ) . the break phenomenon is widely observed in local galaxies ( van der kruit and searle 1981b ; barteldrees and dettmar 1994 ; pohlen et al . 2000 ; de grijs et al . 2001 ; florido et al . 2001 ; kregel et al . 2002 ; kregel and van der kruit 2004 ; pohlen and trujillo 2006 ; florido et al . 2006a , b ) and at high redshift ( prez 2004 ; trujillo and pohlen 2005 ) . see for reviews of these optical studies van der kruit ( 2001 ) and pohlen et al . ( 2004b ) . galaxies have been divided into three types depending on their disc properties ; type i , with an unbroken exponential ( freeman 1970 ) , type ii ( truncation ) , hosting a downbending break , and type iii ( antitruncation ) hosting a transition to a shallower exponential profile ( pohlen & trujillo 2006 ; erwin et al . 2008 ) , based on averaged radial light profiles . this opens the possibility that some of the observed characteristics may be due to the superposition of the thin and thick discs in these galaxies . due to the low - surface - brightness nature of the thick disc , this was left mostly unexplored . pohlen et al . ( 2004a , 2007 ) do suggest a weakening of the truncation with height moving from the thin to the thick disc - dominated regime . this has been confirmed by hubble space telescope observations over two truncations for all stellar populations ( de jong et al . 2007 , radburn - smith et al . 2012 ) . the origin of truncations and antitruncations is under debate . the former are hypothesized to be formed due to star - formation thresholds ( kennicutt 1989 ) , the point of maximum angular momentum of the original primordial cloud from which the galaxy has formed ( van der kruit 1987 ) , and/or by bar angular momentum redistribution ( debattista et al . laurikainen & salo ( 2001 ) proved that many m 51-like galaxies have an antitruncation caused by the stripping of stars and gas from the disc during the interaction . they numerically showed that these antitruncations could last for several gyr . also , a few antitruncations ( @xmath3 ) seem to be caused by the contribution of an extended bulge to the luminosity profile ( maltby et al . 2012 ) . the main purpose of this paper is to obtain the luminosity profiles of thin and thick discs of nearly edge - on galaxies parallel to their midplane to further constrain their origin . we also study truncations and antitruncations in thin and thick discs as separate features by fitting their luminosity profiles with a generalization of the function proposed by erwin et al . ( 2008 ) for the description of discs with breaks . we used images from the spitzer survey of stellar structure in galaxies ( s@xmath4 g , sheth et al . 2010 ) which has imaged more than 2000 galaxies representative of the nearby universe ( radial velocity @xmath5 ) in @xmath6 and @xmath7 using the infrared array camera ( irac ; fazio et al . the s@xmath4 g reaches a typical surface brightness of @xmath8 with a pixel size of @xmath9 and traces mostly the light of old stars with little dust absorption , which makes it ideal for the study of edge - on galaxies . this paper is structured as follows . we present the selected sample in section [ secsample ] and we describe the fitting procedure in section [ secfitting ] . we continue in section [ seccaveat ] by addressing what would happen if some assumptions made when producing the fits were not applicable to our sample of galaxies . we show and discuss our results in section [ secresults ] and section [ secdiscussion ] and we present our conclusions in section [ secconclusions ] . we looked at each of the 2132 galaxies available in the s@xmath4 g archive by february 15 , 2012 and we manually selected disc galaxies appearing to be edge - on with morphological types @xmath10 ( @xmath11 from hyperleda ; paturel et al . later type galaxies were rejected because of their generally ill - defined structure . we have assumed that @xmath11 can be `` reasonably reliably '' determined for edge - on galaxy as claimed by buta ( 2012 ) . we also rejected galaxies with distorted morphologies , with indications of being not quite edge - on ( with visible spiral arms or resonance rings ) , or discs that are too dim to certainly assess their orientation . when possible , sloan digital sky survey data release 8 ( aihara et al . 2011 ) imaging was used in order to ensure close to edge - on galaxy orientation by looking at the position of the midplane dust . the resulting sample has 169 galaxies which includes 29 out of 30 galaxies studied in detail in co11b . the remaining galaxy , ic 1970 , was excluded due to some hint of spiral structure . all those galaxies underwent the image processing and fitting procedure described in sections [ secmakever ] and [ secverfit ] . in section [ sectrim1 ] we describe how , after fitting thin+thick+gas disc profiles to galaxy luminosity profiles perpendicular to their midplanes , we applied additional selection criteria aiming to remove faint galaxies , galaxies with extended envelopes , and/or badly fitted galaxies , in order to ensure the data quality and we built up a final sample with 70 galaxies . in order to obtain luminosity profiles of thin and thick discs parallel to galaxy midplanes and to study them we followed a procedure that is described in detail in the next subsections : * we prepared a grid of synthetic luminosity profiles perpendicular to the galaxy midplanes , as done in co11b . the selected profile was that of a combined thin , thick , and gas disc gravitationally coupled in local isothermal equilibrium for galaxies with @xmath12 . for earlier - type galaxies we only considered a thin and a thick disc . ( section [ sectsynt ] ) . * we extracted the observed luminosity profiles perpendicular to galaxy midplanes using an average of 3.6 and 4.5 @xmath13-band s@xmath4 g images ( vertical luminosity profiles ; section [ secmakever ] ) . * we fitted the observed luminosity profiles by comparing them with the grid of synthetic models as done in co11b ( section [ secverfit ] ) . * using the vertical luminosity profile fits , we determined the height , @xmath14 , above which the thick disc dominates the light emission ( section [ secdefrs ] ) . * by looking at the properties of vertical luminosity profile fits and the resulting @xmath14 we trimmed the 169 galaxy sample down to 70 in order to ensure the quality of the data . the galaxy rejection criteria were aimed to remove galaxies with bad fits , having extended envelopes , or being too faint and thus having noisy luminosity profiles ( section [ sectrim1 ] ) . * knowing @xmath14 has allowed us to determine the region dominated by the thin and thick disc light ; thus , we have been able to prepare luminosity profiles of the thin and thick discs parallel to the galaxy midplanes . we also prepared a profile including light from both discs ( horizontal luminosity profiles ; section [ secmakehor ] ) . * the horizontal luminosity profiles of both the thin and the thick discs , as well as those including light from both , were fitted with a generalization of the function used by erwin et al . ( 2008 ) to describe discs with breaks ( section [ secfithor ] ) . in order to prepare thin and thick disc luminosity profiles parallel to the midplane of a galaxy , it is fundamental to know in which range in heights above the midplane each disc dominates . to do so , we fitted luminosity profiles perpendicular to galaxy midplanes with thin+thick+gas disc functions for galaxies with @xmath12 . for earlier - type galaxies we set the gas disc to have a zero density , which is equivalent to assuming a thin+thick disc function . assuming little or no gas in these galaxies solves one of the fitting biases discussed in section [ secnogas ] , namely that of gas - depleted galaxies with a thick disc mass overestimated because being fitted with a function accounting for gas . as in co11b , we assumed that galaxy discs are relaxed structures whose particles behave like those of a fluid in equilibrium . we also assumed that discs are made of three gravitationally coupled baryonic discs gas disc , thin disc and thick disc which feel the effect of something acting like a dark matter halo . then we can write following the equation from narayan & jog ( 2002 ) : @xmath15 where @xmath16 refers to the thin disc , @xmath11 to the thick disc and @xmath17 to the gas disc ; the subindex @xmath18 can be either @xmath16 , @xmath11 , or @xmath17 , @xmath19 stands for the mass density , and @xmath20 for the vertical velocity dispersion of the component . @xmath21 is the term describing dark matter effects . this is a set of three coupled second order differential equations which we solved by using the newmark-@xmath22 method with @xmath23 and @xmath24 . @xmath22 and @xmath25 are internal parameters of the algorithm which have been set to make it unconditionally stable ( newmark 1959 ) . when integrating equation ( [ equilibrium ] ) we assumed that the isothermal hypothesis is true . this hypothesis implies that for a given galactic radius , @xmath26 , the vertical velocity dispersion is constant with height , @xmath27 ( @xmath28 ) . we also assumed that the effect of dark matter is negligible ( @xmath29 ; see justification in co11b and section [ seccaveat ] ) . line - of - sight effects were avoided by assuming that all of the discs have similar scale lengths and that the scale heights remain roughly constant with varying @xmath26 . this last assumption is based on the fact that s@xmath4 g discs do not flare significantly within the optical radius and has been widely used in the literature ( van der kruit & searle 1981a ; yoachim & dalcanton 2006 ; co11b ) . we also assumed that the face - on gas surface mass density for galaxies with @xmath12 is 0.2 times that of the thin disc at all radii . a @xmath30 gas fraction is slightly higher than that found in the milky way at solar radius ( banerjee & jog 2007 ) and we have shown in co11b that it yields slightly better fits than when not accounting for gas . it is also a fraction of gas representative of what is found in our galaxy sample according to our gas fraction calculations presented in section [ secv120 ] . we set @xmath31 for galaxies with @xmath32 . finally , we assumed @xmath33 in rough accordance with milky way velocity dispersion measurements ( spitzer 1978 ; stark 1984 ; clemens 1985 ; lewis & freeman 1989 ) . equation ( [ equilibrium ] ) provides mass density profiles . for obtaining luminosity profiles some assumptions about @xmath34 and @xmath35 , the mass - to - light ratios of the thin and the thick discs , were needed . different @xmath36 values can be calculated from different star formation histories for the thin and the thick discs . reasonable star formation histories yield @xmath37 ( see in discussion in section 3.3 in co11b ) . in this paper we use the most conservative value that with lower relative thick disc masses @xmath38 . when integrating equation ( [ equilibrium ] ) , six boundary conditions are needed . since the maximum density is found in the midplane , we set @xmath39 . the other boundary conditions we used are the mass densities of each disc in the midplane @xmath40 ( hereafter @xmath41 ) . we solved equation ( [ equilibrium ] ) for a grid of models with different central density ratios ( @xmath42 ) and different vertical velocity dispersion ratios , @xmath43 . we made integrations for 150 values of @xmath42 , equally spaced from 0.015 to 2.25 , and for 150 values of @xmath44 , equally spaced from 1.1 to 16.0 . the solutions in this grid were then compared to observed luminosity profiles as described in the following subsections . we selected the 3.6 and 4.5 @xmath45m - band images for the 169 galaxies in our original sample and we subtracted the background from each image . the background was measured by selecting several tens of @xmath46 pixel boxes far away from the midplane of the galaxy and bright stars and then by obtaining the median value of the pixels in the box . the selected background value was that of the median of the individual medians found for each box . the angular distance between the boxes in which we measured the background and the midplane of the galaxy for galaxies fitting in the irac chip ( @xmath47 ; sheth et al . 2010 ) is limited by irac s field of view ( @xmath48 ) and is typically around 100 . for the few galaxies which required a mosaic , the sky has been measured in boxes as close to the border of the mosaic as possible in the direction perpendicular to the midplane . the median distance of the galaxies for which we studied the breaks is 29.8mpc ( see section [ sectrim1 ] for details on the original sample trimming ) . at that distance , 100 correspond to 14.4kpc , which may cause our estimated sky value to be actually the stellar halo level . bakos & trujillo ( 2012 ) found that haloes start affecting the luminosity profiles of galaxies at @xmath49 . because we have not found color indexes including irac filters for stellar haloes in the literature , we have used globular cluster colors as a proxi for halo colors in order to test whether this error in the sky subtraction could affect our analysis . the reddest globular cluster in the sfample of spitler et al . ( 2008 ) has @xmath50 . this globular cluster has @xmath51 ( spitler et al . 2008 ) and using the color transformations for population ii stars in jordi et al . ( 2006 ) , we can afirm that @xmath52 within @xmath53mag . that implies that the halo would start affecting our luminosity profiles at @xmath54 . since our break detection limit is within the range @xmath55 ( see section [ sectrunc ] ) stellar haloes are not likely to affect our analysis of the background for subtraction . typical one-@xmath20 errors of the background determination of the addition of the 3.6 and 4.5@xmath45 m images are around 0.0015counts , corresponding to a surface brightness level of @xmath56 . this error bar was measured by finding the standard deviation of different background values obtained using bootstrapping over the local background values in the individual @xmath46 pixel boxes . the value of the measured background does not vary significantly when increasing the box size from @xmath46 to @xmath57 pixels . however , the background error rarely is the limiting factor when fitting the luminosity profiles ; the actual main limiting factor is the quality of the mask as explained at the end of section [ secverfit ] . using the average of the sky - subtracted 3.6 and 4.5 @xmath45m - band images we produced four luminosity profiles for each galaxy : at each side of the bulge , central cluster or presumed center along the disc long axis for projected galactocentric distances @xmath58 and @xmath59 ( top panel in figure [ example ] ) . we used @xmath60 values from hyperleda except for ngc 4111 , for which we preferred the third reference catalog of bright galaxies value ( rc3 ; de vaucouleurs et al . 1991 ) due to the hyperleda value being clearly underestimated . the profiles were prepared by adding the counts above and below the midplane in order to get a unique profile for each bin and averaging for each @xmath27 over non - masked pixels . we used manually refined masks from the s@xmath4 g pipeline 2 ( sheth et al . the flux was transformed into magnitudes using a zero point @xmath61 . we increased the signal - to - noise at low surface brightness levels by smoothing the luminosity profiles . to do so , we found the height of the first profile data point for which the statistical error of the photometry was larger than 10% of the signal . we named that height @xmath62 . then , for every height , the photometry was obtained by averaging over the range of heights @xmath63 , where @xmath64 is the pixel size . thus , smoothing was small close to the midplane ( @xmath65 ) , but significant for low signal - to - noise regions , where @xmath66 . c c + & + & both synthetic and observed luminosity profiles were scaled in the same way before the actual fitting . in both cases the midplane surface brightness was set to be one . in co11b we had stretched the @xmath27 axis in such a way that the surface brightness at @xmath67 was equal to one tenth of the surface brightness at @xmath68 , @xmath69 . thus , the synthetic and the observed luminosity were forced to intersect at the point for which the luminosity is one tenth of the midplane one . in the present paper , we made four different vertical scalings , @xmath69 , @xmath70 , @xmath71 , and @xmath72 . for practical purposes , what this stretching does is to set the luminosity at which the synthetic and the observed luminosity profiles intersect . the synthetic fits were convolved with a gaussian kernel with a fwhm of @xmath73 , which is the approximate fwhm of both 3.6 and 4.5@xmath13-band s@xmath4 g point - spread functions made by stacking several hundreds non - saturated stars ( s@xmath4 g `` super - psf '' ; sheth et al . 2010 ) . the fitting of the observed luminosity profiles was done by following the procedure described in section 3.5 and figure 4 in co11b . basically , after being scaled , the observed luminosity profiles for each radial bin in each galaxy were compared to the grid of synthetic models . that was done by minimizing the differences of the brightest section of the profiles starting at a dynamic range of @xmath74 and then going to fainter levels in steps of @xmath75 down to @xmath76 . we selected as the `` correct '' fit the one with the largest @xmath77 having a @xmath78 , where @xmath79 is the mean squared difference in magnitudes between both profiles . this procedure was only done down to @xmath80 in co11b . the lower fitted surface brightness for a given fit is @xmath81 and the midplane luminosity is @xmath82 . thus , for a given fit , the dynamical range @xmath77 over which a fit has been produced is @xmath83 . @xmath81 is reached at a height @xmath84 which is typically on the order of the smoothing scale height , @xmath62 . in some cases , no @xmath78 fit was found for a given bin and thus this bin was flagged to be non - fittable . the code also includes a module for computing the effect of some dust absorption in the midplane as described in co11b . an example of a vertical luminosity profile fit is presented in figure [ example ] and all the vertical luminosity profiles for our final sample galaxies appear in figure [ verticalfits ] of appendix [ appendixver ] . the errors in the photometric profile depict the statistical error of the averaging used at obtaining each photometric point . in spite of the care which has been taken at determining the background level , several profiles show a significant flux at large height ( at a level of @xmath85 ) . there are several possible explanations for this : in some cases , the galaxies are surrounded by a crowd of globular clusters and/or foreground stars whose extended wings are hard to distinguish from thick disc light and may have not been properly masked . in the case of galaxies with a larger angular size , large - scale background variations as those described in comern et al . ( 2011a ) may be present and may cause artifacts in some profiles . also , large saturated stars affect some of the profiles . finally , some galaxies may have faint extended haloes at a brightness level comparable to our detection threshold . the restrictions applied at choosing right fits ( mainly selecting as good fits those with @xmath78 ) , precludes this residual light from affecting our results . a key parameter needed for producing the luminosity profiles of the thin and the thick discs parallel to midplanes is the range of heights for which each disc dominates the light emission . we defined @xmath14 to be the height above which a fraction @xmath86 of the light is emitted by the thick disc according to the fits to the vertical luminosity profiles . this value was calculated for each of the correctly fitted bins . in the example fit presented in figure [ example ] , the @xmath14 for each bin is indicated by a solid vertical line in each of the middle and lower row panels . in order for a fit made for a given bin in a given galaxy to be further considered for use in this paper , several conditions were implemented : * a few galaxies , such as ngc 4013 , have relatively bright components in addition to the thin and thick discs ( co11b ; comern et al . these components are likely to be a second thick disc or a bright squashed halo . in order to avoid including galaxies with wrong fits due to these components , we excluded fits for which @xmath87 . this threshold ensures that , for most cases , possible additional components are dim enough not to significantly affect the fit . * the statistical uncertainties for fits made in very faint galaxies are large . to minimize this problem , we have only flagged as valid those fits for which the midplane surface brightness is @xmath88 . * in a few cases , as described in co11b , the fit is compatible with a single disc . this is most likely to happen for the fits done over a small @xmath77 , for which the fitted dynamical range is not large enough to allow distinguishing between both discs . we rejected fits compatible with a single disc . the criterion to detect those fits is slightly more restrictive than in co11b , namely that the mean squared difference between the fitted solution and a single - disc profile is @xmath89 . one of our assumptions when producing the fits is that the scale lengths of both thin and thick discs are similar and that their scale heights do not vary too much with radius . this implies that @xmath14 should be roughly constant with the projected radius , @xmath90 . as a final quality check for our fits , we selected those galaxies having more than one radial bin with a `` good '' fit according to the criteria expressed in the list and we looked for the largest and smallest @xmath14 among those fits . if the ratio between these two was smaller than 1.5 , the galaxy was included in the final sample . if the ratio was larger than 1.5 , the galaxy was only included if the two @xmath58 had a `` good '' fit and the ratio between their @xmath14 was smaller than 1.5 . the fits for the bins with @xmath59 were then rejected and only those with @xmath58 were considered when measuring global properties of the galaxy ( such as the average @xmath14 and the disc relative masses , @xmath91 ) . this was the case for 10 galaxies and was made so as not to exclude those galaxies with flares or truncations potentially happening at @xmath59 or galaxies with noisy profiles at this range of projected radii . galaxies with varying @xmath14 ( and thus excluded from the final sample ) , have not necessarily intrinsically largely varying scale lengths . for example , in the case of galaxies with a small angular size , they are likely to have a foreground star or a globular cluster covering a large fraction of the fitting bin at some height . in some cases , masking those features may be as harmful for the vertical luminosity profiles as not masking them at all , because it will cause that different heights will have their surface brightness measured in significantly different ranges of projected radii . this would result into unreliable fits which are excluded by the @xmath14 stability criterion . the @xmath14 stability selection criterion is also useful at removing galaxies whose thin disc is so thin that it is barely or not resolved . applying all these criteria trimmed the original sample from 169 to 70 galaxies . the properties of these 70 galaxies are presented in table [ sampledata ] . these restrictive criteria cause eleven out of thirty galaxies included in the co11b final sample not to appear in this paper . fits in galaxies of the final sample which have not been used are labeled as `` not used '' in figure [ verticalfits ] of appendix [ appendixver ] . the distance distribution of the 70 selected galaxies is presented in figure [ distances ] . the median distance of the sample is 29.8mpc . for each of the 70 galaxies in our sample , as done in co11b , the vertical luminosity profile fits were used for estimating the ratio of the stellar mass of the thick and the thin disc for each galaxy , @xmath91 , using the following expression : @xmath92 where @xmath93 refers to the edge - on column mass densities , the subindex @xmath94 refers to the different bins in galactocentric distance for which `` good '' fits have been obtained and @xmath95 is the midplane surface brightness for a given bin in magnitudes . [ cols="^ " , ] as seen in section [ secantitrunc ] and figure [ anti ] , the scale length of the thin disc profile after an antitruncation break seems to correlate with the thick disc scale length at that projected radius . a visual inspection of the fits in figure [ verticalfits ] shows that , in many cases , the shape of the luminosity profiles of type iii , type ii+iii , and type ii+iii+ii thin discs after the antitruncation break radius , @xmath96 , are very similar to that of the thick disc in the same projected radius range . it is therefore possible that it is not the thin disc itself which antitruncates and that what we are actually seeing is that the thick disc dominates the light emission even in the range of heights @xmath97 for @xmath98 . in order to test that possibility , we used the fits to vertical luminosity profiles and we calculated the mean flux per pixel emitted by the thick disc in the range @xmath97 ( height range for thin disc horizontal luminosity profiles ) divided by the mean flux per pixel emitted at @xmath99 ( height range for thick disc horizontal luminosity profiles ) . this value , which we will call @xmath100 , was averaged over all the valid vertical bins . however , because the shape of the vertical thick disc luminosity profile changes in the absence of a thin disc , we calculated a second value , @xmath101 , which measures the same ratio if the thin disc was removed but the thick disc face - on column mass density and vertical velocity dispersion were kept constant . as seen in figure [ thindiscremoved ] , @xmath102 . the @xmath100 and @xmath101 values can be used for estimating the brightness of the thick disc in the range of heights @xmath97 , but we have to consider that they have been calculated at low @xmath90 and can only be considered a useful approximation to what happens in galaxy outskirts . in figure [ verticalfits ] , the lower limit of the gray area is the estimate of the thick disc brightness at that range ( @xmath97 ) in the absence of a thin disc calculated by moving the red symbols up by a factor @xmath101 . the upper limit of the gray area represents the thick disc surface brightness estimate for the range @xmath97 in the case of a thin disc whose face - on column mass density relative to the thick disc is similar to that found at low @xmath90 ( moving the red symbols up by a factor @xmath100 ) . thus , for the outskirts of a galaxy , the lower limit of the gray area symbols represents what we have defined to be the thin disc if it was only the low @xmath27 part of a thick disc and no significant genuine thin disc was found there . of the 26 thin discs with antitruncations 14 ( eso 157 - 049 , eso 440 - 027 , ic 1553 , ngc 0522 , ngc 1163 , ngc 1422 , ngc 3454 , ngc 4330 , ngc 4359 , ngc 4607 , ngc 5529 , ugc 06526 , ugc 12518 , and ugc 12692 ) are well fitted by the lower limit of the gray area , meaning that they are compatible with the range of projected radii @xmath98 being dominated by the thick disc at all heights . in six galaxies ( eso 469 - 015 , ic 1576 , ugc 00903 , ugc 01970 , ugc 07086 , and ugc 10288 ) , the thick disc alone does not seem to be bright enough to account for all the light for heights @xmath97 at @xmath98 . in four galaxies ( ngc 1596 , ngc 3628 , ngc 4081 , and ugc 08737 ) our thick disc brightness estimate at @xmath97 overestimates the actual light emission . for ngc 5084 , the antitruncation is an artifact caused by the presence of an extended bulge . finally , for one galaxy ( ugc 10297 ) , the thick disc profile is too dim at @xmath98 to be able to obtain any conclusion . it thus seems that at least 14 out of 26 antitruncated galaxies , the antitruncation is actually an artifact caused by the presence of a thick disc . this number is probably a low estimate , because of the uncertainties of using @xmath101 for inferring the behavior of the disc in its outskirts . a legitimate question is that if for @xmath98 we are mostly detecting the thick disc at all heights , why its horizontal luminosity profile is almost always flatter for @xmath103 ( red horizontal luminosity profiles ) than in the range @xmath97 ( blue horizontal luminosity profiles ) ? this can be explained naturally if we consider that at @xmath104 there is still a small fraction of stars belonging to the thin disc , meaning that the vertical thick disc luminosity profile becomes more peaked ( see figure [ thindiscremoved ] ) than when the thin disc mass fraction becomes negligible for @xmath105 . this progressive broadening of the thick disc vertical luminosity profile with increasing projected radii causes the horizontal luminosity profile for @xmath97 to be steeper than that measured at @xmath99 . additionally , unmasked faint stars and extended psf wings are more likely to affect the much dimmer @xmath99 range , potentially further contributing to make its horizontal luminosity profile shallower . here we have shown that genuine antitruncations are rare ( twelve at maximum out of 70 or less than @xmath106 ) and that most of the features which would be considered an antitrucation in a face - on view are actually an artifact caused by the superposition of a thin and a thick disc with different scale lengths . however , it seems that several of the found antitruncations are genuine and could be a signature of a past interaction as suggested by laurikainen & salo ( 2001 ) . laurikainen & salo ( 2001 ) showed that many m 51-like interacting galaxies have antitruncations . they used numerical simulations to show that the redistribution of material caused by an interaction may last for several gyr , which could explain the presence of genuine antitruncations in isolated galaxies . also , some of the antitruncations which do not seem related to the thick disc could be caused by the presence of outer rings as suggested by erwin et al . ( 2005 ) . in section [ secmasses ] we found that thick discs with @xmath1 have roughly constant relative masses with respect the thin disc ( @xmath0 ) , and that the relative stellar mass of the thick disc increases for @xmath108 . this threshold was not noticed in co11b due to smaller statistics and a larger scatter in data due to less restrictive quality selection criteria . however , it was detected by yoachim & dalcanton ( 2006 ) . if , as indicated in co11b , thick discs would have formed in situ at high redshift and have their thin discs formed from gas remaining from the formation process and posterior cold accretion , at least two factors could contribute to make @xmath91 larger in less massive galaxies : * supernova feedback is more efficient at removing the gas from shallower potential wells . this is because in a high - mass galaxy , gas expelled by supernovae during the thick disc build - up could eventually come back to the galaxy and be used for building the thin disc . gas expelled from low - mass galaxies would be less likely to come back . * lower - mass galaxies are dynamically younger than higher - mass ones . all galaxies have already formed their thick disc , and have later started to form the thin disc . because of the slowest evolution of lower - mass galaxies , their thin discs are younger and less developed , implying a larger fraction of gas in them . it is important to stress the point that less massive galaxies are likely to expel a larger gas fraction through supernova feedback than massive ones . indeed , as pointed out by elmegreen & elmegreen ( 2006 ) , if a thick disc formed at high redshift had only @xmath109 of the total mass of the disc ( as it is the case for today s thick discs in some massive galaxies ) , its scale height would shrink considerably when the gas which formed the thin disc , the remaining 80% in mass , was accreted . this could cause , if the thick disc scale height was not large enough ( several kpc ) that the thick discs would be indistinguishable from the thin discs in vertical luminosity profiles . the thick disc scale height shrinking would be smaller in a low - mass galaxy accreting a lower fraction of gas ( @xmath110 ) , which would make the original thick disc formed at high redshift much easier to detect . thus , what we observe as a thick discs in today s low - mass galaxies , would be mostly the original in situ thick discs formed at high - redshifts . for massive galaxies , which have expelled less gas through supernova feedback , it is possible that this disc would have shrunk so much that it would be undetectable . in this case , there would be no genuine thick discs , but a continuous distribution of scale heights created by the secular heating of the thin disc by its own overdensities in accordance with the claim from bovy et al . ( 2012 ) for the milky way . significant thin disc heating by satellites is likely to be discarded due to the absence of widespread flares as discussed in co11b . the thick disc being mostly made of stars stripped from satellites is also discarded because that mechanism would lead thick discs significantly less massive than what is observed ( see co11b for a full discussion on these two points ) . radial migration ( schnrich & binney 2009 ) could also explain the presence of thick discs in massive galaxies , but fails at explaining the presence of very massive thick discs in low - mass galaxies which are dynamically younger than high - mass ones . a way of testing whether @xmath91 is higher in low - mass galaxies due to a slower dynamical evolution is to recalculate @xmath111 by considering that , in addition to the thin disc stars , it includes the gas disc . if low - mass galaxies were less efficient at forming stars , the new @xmath91 would show a flatter trend with the rotation velocity , @xmath112 . to calculate the mass of gas contained in a disc , we used the 21-cm corrected for self - absorption flux from hyperleda and we converted it to a gas mass using the expression @xmath113 where @xmath114 is the area of the 21-cm line profile expressed in mjykms@xmath115 and @xmath116 is the distance to the galaxy expressed in mpc ( zwaan et al . 1997 ) . as done in yoachim & dalcanton ( 2006 ) , we made the correction @xmath117 to account for helium and metals and we did not apply a correction for molecular gas . the new @xmath91 values have been calculated considering that the absolute @xmath6 magnitude of the sun in ab system is @xmath118 ( oh et al . 2008 ) , that @xmath119 , and that @xmath38 . this has been done for all except three galaxies in the sample . those three galaxies had no @xmath114 value available in hyperleda . the result of including the gas into the thin disc is shown in figure [ plotgas ] in which the diamond symbols represent @xmath91 once the gas has been included in the thin disc . we note a slight tendency for low - mass galaxies to get their relative thick disc masses lowered by a larger factor by this operation than in the case of high - mass galaxies . when including the gas in the thin disc , the decrease in the thick disc relative mass fraction is @xmath120 when @xmath108 and @xmath121 for galaxies with @xmath1 . this would support the idea of lower - mass discs having a slower evolution than higher - mass galaxies and having transformed less gas into thin disc stars . however , a definitive quantification of this effect can not be obtained without knowing the mass - to - light ratio of the discs . we are working on this quantification and we will be publish it in a follow - up paper . recent works suggest that thick discs are formed in situ at high redshift . so far , the easier way the study thin and thick discs as separate features is by looking at edge - on galaxies . breaks ( truncations and antitruncations ) in disc galaxies have been until now mostly studied for the disc as a whole , without distinguishing the light of the thin and the thick disc . several theories compete for explaining the origin of such breaks . truncations have been explained by dynamical arguments related with the conservation of angular momentum during galaxy formation , by star formation thresholds , and by the redistribution of angular momentum by a bar . antitruncations have in some cases been linked to interactions and mergers . in order to study a significant sample of edge - on galaxies ( 70 ) , we produced luminosity profiles of thin and thick discs parallel to galaxy midplanes from the average of the 3.6 and 4.5@xmath122 images from the spitzer survey of stellar structures in galaxies ( s@xmath4 g ) . the vertical height dominated by each disc was derived from luminosity profiles perpendicular to midplanes done in a similar way to that described in comern et al . ( 2011b ) . in brief , we fitted the observed vertical luminosity profiles by comparing them with a grid of models made by solving the equations of three gravitationally coupled vertically isothermal discs ( thin , thick , and gas discs ) . we defined the range of heights for which the thick disc dominates that height ( named @xmath14 ) above which it emits 90% of the light . we defined a height , @xmath123 at which the @xmath80 level was reached for the fitted vertical bins . the thick disc horizontal luminosity profile was made by averaging the light emission for @xmath103 and the thin disc one was made by averaging light in @xmath97 . once the thin and thick disc luminosity profiles parallel to midplanes were created , we fitted them with a integrated over the line of sight generalization of the function used by erwin et al . ( 2008 ) to describe discs with breaks . the conclusions of this paper are : * the position of our breaks are compatible with those fitted in face - on galaxies . * horizontal luminosity profiles for thin discs are often truncated ( 77% of cases ) and also have a significant number of antitruncations ( 40% of cases ) . according to our fits , thick discs truncate with a frequency of 31% , although some extra truncations can not be discarded at a lower surface brightness than our detection threshold . antitruncations are rare in our thick disc sample ( 6% ) and are in three out of four cases artifacts caused by an extended bulge . * when thick discs truncate , they do so at a radius usually compatible with that of the truncation of the thin disc . however , not all truncated thin discs are associated with a truncated thick disc . this dichotomy suggests two different mechanisms creating truncations : one necessarily of dynamical origin affecting both discs and one affecting only the thin disc . another possibility is that the same mechanism creates a truncation in the thin disc or in both depending on some galactic property which has not been studied in this paper ( such as bar or kinematical properties ) . * antitruncations in thin discs seem to be related , in more than half of the cases , to the superposition of a thin and a thick disc with different scale lengths . thus , for those thin discs , the part outside the antitruncation would actually be the thick disc . we estimate the fraction of genuine thin disc antitruncations to be at maximum 15% . this is less than half of those that would be expected from a deep enough fit to a face - on galaxy sample . * because our sample is much larger and better selected than that used in comern et al . ( 2011b ) , we have been able to further constrain the shape of the relation between the stellar mass of the thick disc relative to that of the thin disc , @xmath91 , and the galaxy circular velocity , @xmath112 originally found by yoachim & dalcanton ( 2006 ) ( figure [ plotgas ] ) . we found that galaxies with @xmath1 have a roughly constant @xmath91 . from @xmath107 downwards , @xmath91 increases dramatically with , in some cases , @xmath124 . for these galaxies , the high @xmath91 could be caused by a combination of a high efficiency of supernova feedback that removes gas from the weak potential well of the young galaxy and limits the eventual mass of the thin disc and a slower dynamical evolution causing the stellar thin discs to be younger and less massive than in higher - mass galaxies . in the latter case we showed that , if we assume the same @xmath36 for all galaxies , when gas is included in the mass of the thin disc , galaxies with @xmath108 have @xmath91 values that are more similar to those in galaxies with @xmath1 . this similarity implies that part of the high value for @xmath91 low mass galaxies is the result of thin disc youth : the thin disc mass has not formed stars yet . the authors wish to thank the entire s@xmath4 g team for their efforts in this project . ea and ab thank the centre national dtudes spatiales for financial support . ks , j - cm - m , tk , and tm acknowledge support from the national radio astronomy observatory , which is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . we thank our anonymous referee for comments that helped to improve the paper . this work is based on observations and archival data made with the spitzer space telescope , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . we are grateful to the dedicated staff at the spitzer science center for their help and support in planning and execution of this exploration science program . we also gratefully acknowledge support from nasa jpl / spitzer grant rsa 1374189 provided for the s@xmath4 g project . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . this research has made use 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 . 99 abadi , m. g. , navarro , j. f. , steinmetz , m. , & eke , v. r. 2003 , apj , 597 , 21 aihara , h. , allende - prieto , c. , an . , d. , et al . 2011 , apjs , 193 , 29 bakos , j. , & trujillo , i. 2012 , submitted to mnras , ( arxiv:1207.7023 ) banerjee , a. , & jog , c. j. 2007 , apj , 622 , 335 barteldrees , a. , & dettmar , r .- j . 1994 , a&as , 103 , 475 bensby , t. , feltzing , s. , lundstrm , i. , & ilyin , i. 2005 , a&a , 433 , 185 bigiel , f. , leroy , a. , brinks , e. , et al . 2008 , aj , 136 , 2846 bournaud , f. , elmegreen , b. g. , & martig , m. 2009 , apj , 707 , l1 bovy , j. , rix , h .- w . , & hogg , d. w. 2012 , apj , 751 , 131 brook , c. , richard , s. , kawata , d. , martel , h. , & gibson , b. k. 2007 , apj , 658 , 60 burstein , d. 1979 , apj , 234 , 829 buta , r. j. 2012 , in planets , stars , and stellar systems , vol . series editor t. d. oswalt , volume editor w. c. keel , springer reference buta , r. j. , sheth , k. , regan , m. , et al . 2010 , apjs , 190 , 147 clemens , d. p. 1985 , apj , 295 , 422 comern , s. , elmegreen , b. g. , knapen , j. h. , et al . 2011c , apj , 738 , l17 comern , s. , elmegreen , b. g. , knapen , j. h. , et al . 2011b , apj , 741 , 28 ( co11b ) comern , s. , knapen , j. h. , sheth , k. , et al . 2011a , apj , 729 , 18 dalcanton , j. j. , yoachim , p. , & bernstein , r. a. 2004 , aj , 608 , 189 debattista , v. p. , mayer , l. , carollo , c. m. , et al . 2006 , apj , 645 , 209 de grijs , r. , kregel , m. , & wesson , k. h. 2001 , mnras , 324 , 1074 de jong , r. s. , seth , a. c. , radburn - smith , d. j. , et al . 2007 , apj , 667 , l49 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , et al . 1991 , third reference catalog of bright galaxies ( new york : springer ) ( rc3 ) elmegreen , b. g. , & elmegreen , d. m. 2006 , apj , 650 , 644 erwin , p. , beckman , j. e. , & pohlen , m. 2005 , apj , 626 , l81 erwin , p. , pohlen , m. , & beckman , j. e. 2008 , aj , 135 , 20 fazio , g. g. , hora , j. l. , allen , l. e. , et al . 2004 , apjs , 154 , 10 florido , e. , battaner , e. , guijarro , a. , garzn , f. , & castillo - morales , a. 2006a , a&a , 455 , 467 florido , e. , battaner , e. , guijarro , a. , garzn , f. , & castillo - morales , a. 2006a , a&a , 455 , 475 florido , e. , battaner , e. , guijarro , a. , garzn , f. , & jimnez - vicente , j. 2001 , a&a , 378 , 82 freeman k. c. 1970 , apj , 160 , 811 fuhrmann , k. 1998 , a&a , 338 , 161 gilmore , g. , & reid , n. 1983 , mnras , 202 , 1025 gutirrez , l. , erwin , p. , aladro , r. , & beckman , j. e. 2011 , aj , 142 , 145 haynes , m. p. 1979 , aj , 84 , 1830 iwrin , j. a. , kennedy , h. , parkin , t. , & madden , s. 2007 , a&a , 474 , 461 jordi , k. , grebel , e. k. , & ammon , k. 2006 , a&a , 460 , 339 kennicutt , jr . , r. c. 1989 , apj , 344 , 685 kregel , m. , & van der kruit , p. c. 2004 , mnras , 355 , 143 kregel , m. , van der kruit , p. c. , & de blok , w. j. g. 2004 , mnras , 352 , 768 kregel , m. , van der kruit , p. c. , & de grijs , r. 2002 , mnras , 334 , 646 laurikainen , e. , & salo , h. 2001 , mnras , 324 , 685 lewis , j. r. , & freeman , k. c. 1989 , aj , 97 , 139 loebman , s. r. , rokar , r. , debattista , v. p. , et al . 2011 , apj , 737 , 8 maltby , d. t. , hoyos , c. , gray , m. e. , aragn - salamanca , a. , & wolf , c. 2012 , mnras , 420 , 2475 narayan , c. a. , & jog , c. j. 2002 , a&a , 394 , 89 newmark , n. m. 1959 , proc . asce , 85 , 67 oh , s .- h . , de blok , w. j. g. , fabian , w. , brinks , e. , & kennicutt , jr . , r. c. 2008 , aj , 136 , 2761 paturel , g. , petit , c. , prugniel , ph . , et al . 2003 , a&a , 412 , 45 prez , i. 2004 , a&a , 427 , l17 pohlen , m. , balcells , m. , ltticke , r. , & dettmar , r .- j . 2004a , a&a , 422 , 465 pohlen , m. , beckman , j. e. , httmeister , s. , et al . 2004b , in penetrating bars through masks of cosmic dust , ed . d. l. block et al . ( cambridge : cambridge univ . press ) , 713 pohlen , m. , dettmar , r .- j . , & ltticke , r. 2000 , a&a , 357 , 1 pohlen , m. , & trujillo , i. 2006 , a&a , 454 , 759 pohlen , m. , zaroubi , s. , peletier , r. f. , & dettmar , r .- j . 2007 , mnras , 378 , 594 prochaska , j. x. , naumov , s. o. , carney , b. w. , mcwilliam , a. , & wolfe , a. m. 2000 , aj , 120 , 2513 qu , y. , di matteo , p. , lehnert , m. d. , & van driel , w. 2011 , a&a , 530 , 10 radburn - smith , d. j. , rokar , r. , debattista , v. p. , et al . 2012 , apj , accepted for publication scnrich , r. , & binney , j. 2009 , mnras , 396 , 203 sheth , k. , regan , m. , hinz , j. l. , et al . 2010 , pasp , 122 , 1397 spitler , l. r. , forbes , d. a. , & beasley , m. a. 2008 , mnras , 389 , 1150 spitzer , l. 1978 , physical processes in the interstellar medium ( new york : john wiley ) stark , a. a. 1984 , apj , 281 , 624 trujillo , i. , & pohlen , m. 2005 , apj , 630 , l17 tsikoudi , v. 1979 , apj , 234 , 842 van albada , t. s. , bahcall , j. n. , begeman , k. , & sancisi , r. 1985 , apj , 295 , 305 van der kruit , p. c. 1979 , a&as , 38 15 van der kruit , p. c. 1987 , a&a , 173 , 59 van der kruit , p. c. 2001 , in asp conf . 230 , galaxy disks and disk galaxies , ed . j. g. funes & e. m. corsini ( san francisco : asp ) , 119 van der kruit , p. c. , & searle , l. 1981a , a&a , 95 , 105 van der kruit , p. c. , & searle , l. 1981b , a&a , 95 , 116 van der kruit , p. c. , & searle , l. 1982 , a&a , 110 , 79 yoachim , p. , & dalcanton , j. j. , 2006 , aj , 131 , 226 zwaan , m. a. , briggs , f. h. , sprayberry , d. , & sorar , e. 1997 , apj , 490 , 173 l c c c c c c + galaxy i d & @xmath11 & distance & @xmath125 & @xmath112 & @xmath60 & pa + & & ( mpc ) & ( mag ) & kms@xmath115 & ( ) & ( ) + + galaxy i d & @xmath11 & distance & @xmath125 & @xmath112 & @xmath60 & pa + & & ( mpc ) & ( mag ) & kms@xmath115 & ( ) & ( ) + eso 157 - 049 & 3.0 & 23.9 & 13.55 & 90.48 & 52 & 28.1 + eso 240 - 011 & 4.8 & 42.4 & 11.52 & 267.46 & 16 & 127.2 + eso 292 - 014 & 6.5 & 24.9 & 12.47 & 107.73 & 89 & 83.9 + eso 346 - 001 & 5.1 & 34.9 & 13.29 & 47.68 & 64 & 64.3 + eso 440 - 027 & 6.6 & 21.6 & 11.50 & 117.53 & 10 & 79.7 + eso 443 - 021 & 5.8 & 48.1 & 13.24 & 161.71 & 37 & 159.3 + eso 466 - 014 & 3.8 & 47.7 & 14.23 & 121.00 & 40 & 50.4 + eso 469 - 015 & 3.3 & 27.0 & 13.69 & 93.01 & 56 & 149.6 + eso 533 - 004 & 5.2 & 36.2 & 13.05 & 147.54 & 55 & 150.5 + eso 544 - 027 & 3.4 & 40.3 & 14.20 & 92.00 & 46 & 153.3 + ic 0217 & 5.9 & 24.2 & 13.86 & 100.15 & 60 & 35.7 + ic 0610 & 3.9 & 20.7 & 13.44 & 133.84 & 57 & 29.0 + ic 1197 & 6.0 & 25.7 & 13.14 & 87.90 & 77 & 56.4 + ic 1553 & 5.2 & 39.9 & 13.38 & 67.93 & 40 & 15.0 + ic 1711 & 3.0 & 52.3 & 13.19 & 173.41 & 75 & 43.9 + ic 1913 & 3.4 & 21.4 & 13.38 & 77.88 & 58 & 147.7 + ic 2058 & 6.5 & 19.4 & 12.13 & 82.78 & 102 & 17.4 + ic 2135 & 6.0 & 29.2 & 11.93 & 107.08 & 97 & 108.6 + ic 5176 & 4.5 & 26.4 & 12.02 & 164.38 & 134 & 27.9 + ngc 0489 & 2.7 & 33.5 & 12.42 & 180.91 & 46 & 120.8 + ngc 0522 & 4.1 & 35.0 & 12.28 & 169.10 & 72 & 32.9 + ngc 0678 & 3.0 & 27.1 & 12.15 & 169.03 & 93 & 77.4 + ngc 1032 & 0.4 & 35.6 & 12.13 & 283.94 & 106 & 67.5 + ngc 1163 & 4.1 & 30.9 & 13.40 & 142.07 & 69 & 143.0 + ngc 1422 & 2.3 & 16.9 & 13.22 & 65.87 & 75 & 65.5 + ngc 1495 & 5.0 & 17.3 & 12.19 & 90.73 & 67 & 104.5 + ngc 1596 & -2.0 & 15.6 & 11.94 & 98.32 & 117 & 18.7 + ngc 2732 & -2.0 & 31.8 & 12.73 & 32.30 & 55 & 65.7 + ngc 3098 & -1.5 & 22.4 & 12.68 & 129.73 & 70 & 88.7 + ngc 3279 & 6.5 & 32.5 & 12.37 & 155.85 & 74 & 152.0 + ngc 3454 & 5.5 & 21.3 & 12.44 & 91.25 & 72 & 116.1 + ngc 3501 & 5.9 & 23.3 & 12.09 & 133.61 & 128 & 28.0 + ngc 3592 & 5.3 & 22.7 & 13.44 & 79.72 & 64 & 117.7 + ngc 3600 & 1.0 & 13.6 & 12.78 & 86.90 & 56 & 4.4 + ngc 3628 & 3.1 & 12.2 & 9.150 & 215.16 & 329 & 102.6 + ngc 4081 & 1.0 & 26.1 & 15.12 & 116.53 & 42 & 130.0 + ngc 4111 & -1.4 & 16.0 & 11.60 & 71.63 & 53 & 152.2 + ngc 4330 & 6.3 & 19.5 & 11.99 & 117.85 & 69 & 59.2 + ngc 4359 & 5.0 & 13.3 & 13.43 & 102.83 & 41 & 107.7 + ngc 4437 & 6.0 & 9.8 & 9.620 & 140.13 & 274 & 82.4 + ngc 4565 & 3.2 & 13.3 & 8.970 & 244.94 & 498&135.1 + ngc 4607 & 3.4 & 20.0 & 12.63 & 98.93 & 87 & 2.7 + ngc 4747 & 7.2 & 12.3 & 12.36 & 69.60 & 63 & 31.9 + ngc 5084 & -2.0 & 26.4 & 11.07 & 309.88 & 300 & 81.0 + ngc 5470 & 3.1 & 18.0 & 13.00 & 109.45 & 77 & 62.2 + ngc 5529 & 5.3 & 50.0 & 11.20 & 284.14 & 173&114.3 + ngc 5981 & 4.3 & 29.2 & 12.48 & 251.12 & 83 & 139.8 + ngc 6010 & 0.4 & 21.6 & 12.33 & 148.80 & 55 & 103.3 + ngc 7347 & 4.6 & 38.7 & 13.25 & 115.11 & 46 & 130.8 + pgc 013646 & 5.1 & 32.6 & 12.32 & 168.16 & 102&34.4 + pgc 028308 & 6.8 & 40.5 & 12.87 & 167.27 & 60&125.8 + pgc 030591 & 6.8 & 29.8 & 13.36 & 111.59 & 45&168.4 + pgc 032548 & -0.2 & 37.3 & 15.06 & 93.33 & 27 & 149.7 + pgc 052809 & 5.9 & 24.9 & 12.46 & 100.97 & 95&169.2 + ugc 00903 & 3.9 & 51.6 & 13.09 & 163.23 & 50 & 52.6 + ugc 01970 & 5.9 & 33.9 & 13.32 & 95.20 & 55 & 22.5 + ugc 05347 & 6.5 & 33.8 & 13.53 & 95.64 & 41 & 17.0 + ugc 05689 & 6.4 & 43.4 & 14.07 & 119.67 & 40 & 153.4 + ugc 05958 & 4.0 & 29.4 & 14.26 & 78.15 & 49 & 178.9 + ugc 06526 & 7.0 & 32.3 & 12.64 & 72.57 & 49 & 87.2 + ugc 07086 & 3.1 & 32.8 & 14.02 & 126.01 & 69 & 70.5 + ugc 08737 & 4.0 & 41.9 & 13.27 & 163.31 & 72 & 150.7 + ugc 09448 & 3.2 & 37.8 & 13.95 & 111.50 & 49 & 85.8 + ugc 09665 & 4.0 & 42.5 & 13.57 & 126.05 & 51 & 141.5 + ugc 10043 & 4.1 & 37.3 & 13.80 & 143.59 & 66 & 150.5 + ugc 10288 & 5.3 & 32.4 & 12.09 & 166.59 & 144 & 90.4 + ugc 10297 & 5.1 & 39.2 & 13.43 & 102.82 & 64 & 2.9 + ugc 12518 & 3.0 & 51.6 & 14.11 & 49.58 & 42 & 25.2 + ugc 12692 & 3.9 & 49.6 & 14.21 & 101.34 & 38 & 53.2 + ugc 12857 & 4.0 & 34.5 & 13.24 & 107.94 & 61 & 33.5 + note . @xmath11 values from hyperleda . distance values from the average of redshift - independent measurements in ned when available and if not from virgo infall corrected radial velocities and a hubble - lematre constant @xmath126 . @xmath125 values are hyperleda s internal absorption - corrected values . @xmath112 values from hyperleda . @xmath60 values from hyperleda except for ngc 4111 , for which we used the rc3 value . pa values from hyperleda , except for eso 440 - 027 , ic 5176 , and ngc 3098 , for which we measured our own pa values using ellipse fitting . lccccc + i d & type & @xmath127 & @xmath128 & @xmath129 & @xmath130 + & & ( mag / as@xmath131 ) & ( ) & ( ) & ( ) + + i d & type & @xmath127 & @xmath128 & @xmath129 & @xmath130 + & & ( mag / as@xmath131 ) & ( ) & ( ) & ( ) + eso 157 - 049&type ii+iii&@xmath132&@xmath133&@xmath134&@xmath135 + & & & @xmath134&@xmath136&@xmath137 + & & & @xmath136&@xmath138&@xmath139 + eso 240 - 011&type ii&@xmath140&@xmath141&@xmath142&@xmath143 + & & & @xmath142&@xmath144&@xmath145 + eso 292 - 014&type ii&@xmath146&@xmath147&@xmath148&@xmath149 + & & & @xmath148&@xmath150&@xmath151 + eso 346 - 001&type ii+ii&@xmath152&@xmath153&@xmath154&@xmath155 + & & & @xmath154&@xmath156&@xmath157 + & & & @xmath156&@xmath158&@xmath159 + eso 440 - 027&type ii+iii&@xmath160&@xmath161&@xmath162&@xmath163 + & & & @xmath162&@xmath164&@xmath165 + & & & @xmath164&@xmath166&@xmath167 + eso 443 - 021&type ii&@xmath168&@xmath141&@xmath169&@xmath170 + & & & @xmath169&@xmath171&@xmath172 + eso 466 - 014&type ii&@xmath173&@xmath174&@xmath175&@xmath176 + & & & @xmath175&@xmath177&@xmath178 + eso 469 - 015&type ii+iii&@xmath179&@xmath174&@xmath180&@xmath181 + & & & @xmath180&@xmath182&@xmath183 + & & & @xmath182&@xmath158&@xmath184 + eso 533 - 004&type ii&@xmath185&@xmath186&@xmath187&@xmath188 + & & & @xmath187&@xmath150&@xmath189 + eso 544 - 027&type ii&@xmath190&@xmath191&@xmath192&@xmath193 + & & & @xmath192&@xmath194&@xmath195 + ic 0217&type ii&@xmath196&@xmath153&@xmath197&@xmath198 + & & & @xmath197&@xmath199&@xmath200 + ic 0610&type ii+ii&@xmath201&@xmath141&@xmath202&@xmath203 + & & & @xmath202&@xmath204&@xmath205 + & & & @xmath204&@xmath206&@xmath207 + ic 1197&type ii&@xmath208&@xmath209&@xmath210&@xmath211 + & & & @xmath210&@xmath212&@xmath213 + ic 1553&type ii+iii&@xmath214&@xmath153&@xmath215&@xmath216 + & & & @xmath215&@xmath217&@xmath218 + & & & @xmath217&@xmath194&@xmath219 + + ic 1711&type ii&@xmath220&@xmath221&@xmath222&@xmath223 + & & & @xmath222&@xmath150&@xmath224 + ic 1913&type ii&@xmath225&@xmath226&@xmath227&@xmath228 + & & & @xmath227&@xmath206&@xmath229 + ic 2058&type ii&@xmath230&@xmath174&@xmath231&@xmath232 + & & & @xmath231&@xmath233&@xmath234 + ic 2135&type ii&@xmath235&@xmath147&@xmath236&@xmath237 + & & & @xmath236&@xmath238&@xmath239 + ic 5176&type iii&@xmath240&@xmath241&@xmath242&@xmath243 + & & & @xmath242&@xmath166&@xmath244 + ngc 0489&type i&@xmath245&@xmath241&@xmath246&@xmath247 + ngc 0522&type ii+iii&@xmath248&@xmath186&@xmath249&@xmath250 + & & & @xmath249&@xmath251&@xmath252 + & & & @xmath251&@xmath212&@xmath253 + ngc 0678&type ii&@xmath254&@xmath194&@xmath255&@xmath256 + & & & @xmath255&@xmath257&@xmath258 + ngc 1032&type ii&@xmath259&@xmath177&@xmath260&@xmath261 + & & & @xmath260&@xmath262&@xmath263 + ngc 1163&type ii+iii&@xmath264&@xmath265&@xmath266&@xmath266 + & & & @xmath266&@xmath267&@xmath268 + & & & @xmath267&@xmath212&@xmath269 + ngc 1422&type iii&@xmath270&@xmath133&@xmath271&@xmath272 + & & & @xmath271&@xmath233&@xmath273 + ngc 1495&type ii&@xmath274&@xmath275&@xmath276&@xmath277 + & & & @xmath276&@xmath278&@xmath279 + ngc 1596&type iii&@xmath280&@xmath275&@xmath281&@xmath282 + & & & @xmath281&@xmath283&@xmath284 + ngc 2732&type ii&@xmath285&@xmath286&@xmath287&@xmath288 + & & & @xmath287&@xmath289&@xmath290 + ngc 3098&type iii&@xmath291&@xmath275&@xmath292&@xmath293 + & & & @xmath292&@xmath199&@xmath294 + ngc 3279&type ii&@xmath295&@xmath296&@xmath297&@xmath298 + & & & @xmath297&@xmath299&@xmath300 + ngc 3454&type ii+iii&@xmath301&@xmath241&@xmath302&@xmath303 + & & & @xmath302&@xmath304&@xmath305 + & & & @xmath304&@xmath299&@xmath306 + ngc 3501&type ii&@xmath307&@xmath275&@xmath308&@xmath309 + & & & @xmath308&@xmath310&@xmath311 + ngc 3592&type ii&@xmath312&@xmath141&@xmath313&@xmath314 + & & & @xmath313&@xmath315&@xmath316 + ngc 3600&type iii&@xmath317&@xmath161&@xmath318&@xmath226 + & & & @xmath318&@xmath233&@xmath319 + ngc 3628&type ii+iii+ii&@xmath320&@xmath158&@xmath321&@xmath322 + & & & @xmath321&@xmath323&@xmath324 + & & & @xmath323&@xmath325&@xmath326 + & & & @xmath325&@xmath327&@xmath328 + ngc 4081&type ii+iii&@xmath329&@xmath296&@xmath330&@xmath331 + & & & @xmath330&@xmath332&@xmath333 + & & & @xmath332&@xmath334&@xmath335 + ngc 4111&type ii&@xmath336&@xmath337&@xmath338&@xmath339 + & & & @xmath338&@xmath340&@xmath341 + ngc 4330&type ii+iii+ii&@xmath342&@xmath343&@xmath344&@xmath345 + & & & @xmath344&@xmath346&@xmath347 + & & & @xmath346&@xmath348&@xmath349 + & & & @xmath348&@xmath166&@xmath350 + ngc 4359&type iii&@xmath351&@xmath296&@xmath352&@xmath353 + & & & @xmath352&@xmath257&@xmath354 + ngc 4437&type ii&@xmath355&@xmath356&@xmath357&@xmath358 + & & & @xmath357&@xmath359&@xmath360 + ngc 4565&type ii+ii&@xmath361&@xmath212&@xmath362&@xmath363 + & & & @xmath362&@xmath364&@xmath365 + & & & @xmath364&@xmath366&@xmath367 + ngc 4607&type ii+iii&@xmath368&@xmath226&@xmath369&@xmath370 + & & & @xmath369&@xmath371&@xmath372 + & & & @xmath371&@xmath310&@xmath373 + ngc 4747&type i&@xmath374&@xmath153&@xmath375&@xmath376 + ngc 5084&type iii&@xmath377&@xmath378&@xmath379&@xmath380 + & & & @xmath379&@xmath381&@xmath382 + ngc 5470&type ii&@xmath383&@xmath343&@xmath384&@xmath385 + & & & @xmath384&@xmath199&@xmath386 + + + ngc 5529&type ii+iii+ii&@xmath387&@xmath147&@xmath388&@xmath389 + & & & @xmath388&@xmath390&@xmath391 + & & & @xmath390&@xmath392&@xmath393 + & & & @xmath392&@xmath394&@xmath395 + ngc 5981&type ii&@xmath396&@xmath397&@xmath398&@xmath399 + & & & @xmath398&@xmath400&@xmath401 + ngc 6010&type ii&@xmath402&@xmath397&@xmath403&@xmath404 + & & & @xmath403&@xmath299&@xmath405 + ngc 7347&type ii&@xmath406&@xmath147&@xmath407&@xmath408 + & & & @xmath407&@xmath246&@xmath409 + pgc 013646&type ii&@xmath410&@xmath147&@xmath411&@xmath412 + & & & @xmath411&@xmath238&@xmath413 + pgc 028308&type ii&@xmath414&@xmath241&@xmath415&@xmath416 + & & & @xmath415&@xmath334&@xmath417 + pgc 030591&type i&@xmath418&@xmath147&@xmath419&@xmath420 + pgc 032548&type i&@xmath421&@xmath265&@xmath419&@xmath422 + pgc 052809&type ii&@xmath423&@xmath141&@xmath424&@xmath425 + & & & @xmath424&@xmath375&@xmath426 + ugc 00903&type ii+iii&@xmath427&@xmath241&@xmath428&@xmath429 + & & & @xmath428&@xmath430&@xmath431 + & & & @xmath430&@xmath334&@xmath432 + ugc 01970&type iii+ii&@xmath433&@xmath265&@xmath434&@xmath435 + & & & @xmath434&@xmath436&@xmath437 + & & & @xmath436&@xmath289&@xmath438 + ugc 05347&type ii&@xmath439&@xmath440&@xmath441&@xmath442 + & & & @xmath441&@xmath177&@xmath443 + ugc 05689&type ii&@xmath444&@xmath133&@xmath445&@xmath446 + & & & @xmath445&@xmath447&@xmath448 + ugc 05958&type i&@xmath449&@xmath153&@xmath334&@xmath432 + ugc 06526&type iii&@xmath450&@xmath440&@xmath451&@xmath452 + & & & @xmath451&@xmath150&@xmath453 + ugc 07086&type ii+iii&@xmath454&@xmath296&@xmath455&@xmath456 + & & & @xmath455&@xmath457&@xmath458 + & & & @xmath457&@xmath150&@xmath459 + + + ugc 08737&type ii+iii&@xmath460&@xmath343&@xmath461&@xmath462 + & & & @xmath461&@xmath463&@xmath464 + & & & @xmath463&@xmath238&@xmath465 + ugc 09448&type i&@xmath466&@xmath467&@xmath158&@xmath468 + ugc 09665&type ii&@xmath469&@xmath241&@xmath470&@xmath471 + & & & @xmath470&@xmath158&@xmath472 + ugc 10043&type ii&@xmath473&@xmath474&@xmath475&@xmath476 + & & & @xmath475&@xmath477&@xmath478 + ugc 10288&type ii+iii&@xmath479&@xmath147&@xmath480&@xmath481 + & & & @xmath480&@xmath482&@xmath483 + & & & @xmath482&@xmath484&@xmath485 + ugc 10297&type ii+iii&@xmath486&@xmath296&@xmath487&@xmath488 + & & & @xmath487&@xmath489&@xmath490 + & & & @xmath489&@xmath334&@xmath491 + ugc 12518&type ii+iii&@xmath492&@xmath153&@xmath493&@xmath494 + & & & @xmath493&@xmath495&@xmath172 + & & & @xmath495&@xmath138&@xmath496 + ugc 12692&type iii&@xmath497&@xmath153&@xmath498&@xmath499 + & & & @xmath498&@xmath419&@xmath500 + ugc 12857&type i&@xmath501&@xmath209&@xmath171&@xmath502 + note . @xmath128 and @xmath129 stand for the inner and the outer limit of each fitted section and @xmath130 corresponds to the scale length of that section . the lower limit of the innermost section and the upper limit of the outermost section have no error bar because they have been set manually . lccccc + i d & type & @xmath127 & @xmath128 & @xmath129 & @xmath130 + & & ( mag / as@xmath131 ) & ( ) & ( ) & ( ) + + i d & type & @xmath127 & @xmath128 & @xmath129 & @xmath130 + & & ( mag / as@xmath131 ) & ( ) & ( ) & ( ) + eso 157 - 049&type ii+iii&@xmath503&@xmath133&@xmath504&@xmath505 + & & & @xmath504&@xmath506&@xmath507 + & & & @xmath506&@xmath206&@xmath508 + eso 240 - 011&type ii&@xmath509&@xmath141&@xmath510&@xmath511 + & & & @xmath510&@xmath144&@xmath512 + + eso 292 - 014&type ii&@xmath513&@xmath147&@xmath514&@xmath515 + & & & @xmath514&@xmath150&@xmath516 + eso 346 - 001&type ii+ii&@xmath517&@xmath153&@xmath518&@xmath519 + & & & @xmath518&@xmath520&@xmath521 + & & & @xmath520&@xmath158&@xmath522 + eso 440 - 027&type ii+iii&@xmath523&@xmath161&@xmath524&@xmath525 + & & & @xmath524&@xmath526&@xmath527 + & & & @xmath526&@xmath166&@xmath528 + eso 443 - 021&type ii&@xmath529&@xmath141&@xmath530&@xmath531 + & & & @xmath530&@xmath171&@xmath532 + eso 466 - 014&type ii&@xmath533&@xmath174&@xmath534&@xmath535 + & & & @xmath534&@xmath177&@xmath536 + eso 469 - 015&type ii+iii&@xmath537&@xmath174&@xmath416&@xmath538 + & & & @xmath416&@xmath539&@xmath540 + & & & @xmath539&@xmath158&@xmath541 + eso 533 - 004&type ii&@xmath542&@xmath186&@xmath543&@xmath544 + & & & @xmath543&@xmath150&@xmath545 + eso 544 - 027&type ii&@xmath546&@xmath191&@xmath547&@xmath548 + & & & @xmath547&@xmath194&@xmath549 + ic 0217&type ii&@xmath550&@xmath153&@xmath551&@xmath552 + & & & @xmath551&@xmath199&@xmath553 + ic 0610&type ii+ii&@xmath554&@xmath141&@xmath555&@xmath556 + & & & @xmath555&@xmath557&@xmath558 + & & & @xmath557&@xmath206&@xmath559 + ic 1197&type ii&@xmath560&@xmath209&@xmath561&@xmath562 + & & & @xmath561&@xmath212&@xmath563 + ic 1553&type ii+iii&@xmath564&@xmath153&@xmath565&@xmath566 + & & & @xmath565&@xmath567&@xmath568 + & & & @xmath567&@xmath177&@xmath569 + ic 1711&type ii&@xmath570&@xmath221&@xmath571&@xmath572 + & & & @xmath571&@xmath199&@xmath573 + ic 1913&type ii&@xmath574&@xmath226&@xmath575&@xmath576 + & & & @xmath575&@xmath206&@xmath577 + ic 2058&type ii&@xmath578&@xmath174&@xmath579&@xmath580 + & & & @xmath579&@xmath233&@xmath581 + + ic 2135&type ii&@xmath582&@xmath147&@xmath583&@xmath584 + & & & @xmath583&@xmath238&@xmath585 + ic 5176&type iii&@xmath586&@xmath241&@xmath587&@xmath588 + & & & @xmath587&@xmath166&@xmath589 + ngc 0489&type i&@xmath590&@xmath241&@xmath246&@xmath591 + ngc 0522&type ii+iii&@xmath592&@xmath186&@xmath593&@xmath594 + & & & @xmath593&@xmath595&@xmath443 + & & & @xmath595&@xmath596&@xmath597 + ngc 0678&type ii&@xmath598&@xmath194&@xmath599&@xmath600 + & & & @xmath599&@xmath601&@xmath602 + ngc 1032&type ii&@xmath603&@xmath177&@xmath604&@xmath605 + & & & @xmath604&@xmath484&@xmath606 + ngc 1163&type ii+iii&@xmath607&@xmath265&@xmath608&@xmath609 + & & & @xmath608&@xmath610&@xmath611 + & & & @xmath610&@xmath596&@xmath612 + ngc 1422&type iii&@xmath613&@xmath133&@xmath614&@xmath615 + & & & @xmath614&@xmath375&@xmath616 + ngc 1495&type ii&@xmath617&@xmath275&@xmath618&@xmath619 + & & & @xmath618&@xmath278&@xmath620 + ngc 1596&type iii&@xmath621&@xmath275&@xmath622&@xmath623 + & & & @xmath622&@xmath283&@xmath624 + ngc 2732&type ii&@xmath625&@xmath286&@xmath626&@xmath627 + & & & @xmath626&@xmath628&@xmath629 + ngc 3098&type i&@xmath630&@xmath275&@xmath199&@xmath416 + ngc 3279&type ii&@xmath631&@xmath296&@xmath632&@xmath633 + & & & @xmath632&@xmath299&@xmath634 + ngc 3454&type ii+iii&@xmath635&@xmath241&@xmath636&@xmath637 + & & & @xmath636&@xmath638&@xmath639 + & & & @xmath638&@xmath212&@xmath640 + ngc 3501&type ii&@xmath641&@xmath275&@xmath642&@xmath643 + & & & @xmath642&@xmath340&@xmath644 + ngc 3592&type ii&@xmath645&@xmath141&@xmath646&@xmath647 + & & & @xmath646&@xmath289&@xmath648 + ngc 3600&type i&@xmath649&@xmath161&@xmath233&@xmath650 + + + ngc 3628&type ii+iii+ii&@xmath651&@xmath158&@xmath652&@xmath653 + & & & @xmath652&@xmath654&@xmath655 + & & & @xmath654&@xmath656&@xmath657 + & & & @xmath656&@xmath327&@xmath658 + ngc 4081&type ii+iii&@xmath659&@xmath296&@xmath660&@xmath661 + & & & @xmath660&@xmath662&@xmath663 + & & & @xmath662&@xmath334&@xmath664 + ngc 4111&type ii&@xmath665&@xmath337&@xmath666&@xmath667 + & & & @xmath666&@xmath668&@xmath669 + ngc 4330&type ii+iii+ii&@xmath670&@xmath343&@xmath671&@xmath672 + & & & @xmath671&@xmath673&@xmath674 + & & & @xmath673&@xmath675&@xmath393 + & & & @xmath675&@xmath166&@xmath676 + ngc 4359&type iii&@xmath677&@xmath296&@xmath678&@xmath679 + & & & @xmath678&@xmath257&@xmath680 + ngc 4437&type ii&@xmath681&@xmath356&@xmath682&@xmath683 + & & & @xmath682&@xmath684&@xmath685 + ngc 4565&type ii+ii&@xmath686&@xmath212&@xmath687&@xmath688 + & & & @xmath687&@xmath689&@xmath690 + & & & @xmath689&@xmath691&@xmath692 + ngc 4607&type ii+iii&@xmath693&@xmath226&@xmath694&@xmath695 + & & & @xmath694&@xmath696&@xmath697 + & & & @xmath696&@xmath310&@xmath353 + ngc 4747&type i&@xmath698&@xmath153&@xmath375&@xmath699 + ngc 5084&type iii&@xmath700&@xmath378&@xmath701&@xmath702 + & & & @xmath701&@xmath703&@xmath704 + ngc 5470&type ii&@xmath705&@xmath343&@xmath706&@xmath707 + & & & @xmath706&@xmath212&@xmath708 + ngc 5529&type ii+iii+ii&@xmath709&@xmath147&@xmath710&@xmath711 + & & & @xmath710&@xmath712&@xmath713 + & & & @xmath712&@xmath714&@xmath715 + & & & @xmath714&@xmath394&@xmath716 + ngc 5981&type ii&@xmath717&@xmath397&@xmath718&@xmath719 + & & & @xmath718&@xmath400&@xmath720 + ngc 6010&type ii&@xmath721&@xmath397&@xmath722&@xmath723 + & & & @xmath722&@xmath150&@xmath724 + ngc 7347&type ii&@xmath725&@xmath147&@xmath726&@xmath727 + & & & @xmath726&@xmath158&@xmath241 + pgc 013646&type ii&@xmath728&@xmath147&@xmath729&@xmath730 + & & & @xmath729&@xmath238&@xmath731 + pgc 028308&type ii&@xmath732&@xmath241&@xmath733&@xmath734 + & & & @xmath733&@xmath334&@xmath735 + pgc 030591&type i&@xmath736&@xmath737&@xmath738&@xmath532 + pgc 032548&type i&@xmath739&@xmath265&@xmath158&@xmath740 + pgc 052809&type ii&@xmath741&@xmath141&@xmath742&@xmath743 + & & & @xmath742&@xmath375&@xmath744 + ugc 00903&type ii+iii&@xmath745&@xmath241&@xmath746&@xmath747 + & & & @xmath746&@xmath748&@xmath749 + & & & @xmath748&@xmath334&@xmath750 + ugc 01970&type iii+ii&@xmath751&@xmath265&@xmath752&@xmath753 + & & & @xmath752&@xmath754&@xmath755 + & & & @xmath754&@xmath289&@xmath756 + ugc 05347&type ii&@xmath757&@xmath440&@xmath758&@xmath759 + & & & @xmath758&@xmath194&@xmath760 + ugc 05689&type ii&@xmath761&@xmath133&@xmath762&@xmath763 + & & & @xmath762&@xmath447&@xmath764 + ugc 05958&type i&@xmath765&@xmath153&@xmath158&@xmath766 + ugc 06526&type iii&@xmath767&@xmath440&@xmath768&@xmath769 + & & & @xmath768&@xmath150&@xmath770 + ugc 07086&type ii+iii&@xmath771&@xmath296&@xmath772&@xmath773 + & & & @xmath772&@xmath774&@xmath775 + & & & @xmath774&@xmath150&@xmath776 + ugc 08737&type ii+iii&@xmath777&@xmath343&@xmath778&@xmath779 + & & & @xmath778&@xmath780&@xmath781 + & & & @xmath780&@xmath233&@xmath782 + ugc 09448&type i&@xmath783&@xmath467&@xmath177&@xmath784 + ugc 09665&type ii&@xmath785&@xmath241&@xmath786&@xmath787 + & & & @xmath786&@xmath334&@xmath788 + ugc 10043&type ii&@xmath789&@xmath474&@xmath790&@xmath791 + & & & @xmath790&@xmath477&@xmath792 + + + ugc 10288&type ii+iii&@xmath793&@xmath147&@xmath794&@xmath795 + & & & @xmath794&@xmath796&@xmath797 + & & & @xmath796&@xmath144&@xmath798 + ugc 10297&type ii+iii&@xmath799&@xmath296&@xmath800&@xmath801 + & & & @xmath800&@xmath802&@xmath803 + & & & @xmath802&@xmath334&@xmath804 + ugc 12518&type ii+iii&@xmath805&@xmath153&@xmath806&@xmath807 + & & & @xmath806&@xmath808&@xmath809 + & & & @xmath808&@xmath138&@xmath810 + ugc 12692&type iii&@xmath811&@xmath153&@xmath812&@xmath813 + & & & @xmath812&@xmath419&@xmath814 + ugc 12857&type i&@xmath815&@xmath209&@xmath206&@xmath816 + note . @xmath128 and @xmath129 stand for the inner and the outer limit of each fitted section and @xmath130 corresponds to the scale length of that section . the lower limit of the innermost section and the upper limit of the outermost section have no error bar because they have been set manually . lccccc + i d & type & @xmath127 & @xmath128 & @xmath129 & @xmath130 + & & ( mag / as@xmath131 ) & ( ) & ( ) & ( ) + + i d & type & @xmath127 & @xmath128 & @xmath129 & @xmath130 + & & ( mag / as@xmath131 ) & ( ) & ( ) & ( ) + eso 157 - 049&type ii&@xmath817&@xmath174&@xmath818&@xmath819 + & & & @xmath818&@xmath138&@xmath820 + eso 240 - 011&type i&@xmath821&@xmath296&@xmath822&@xmath823 + eso 292 - 014&type ii&@xmath824&@xmath440&@xmath825&@xmath826 + & & & @xmath825&@xmath628&@xmath827 + eso 346 - 001&type ii&@xmath828&@xmath265&@xmath829&@xmath830 + & & & @xmath829&@xmath246&@xmath831 + eso 440 - 027&type i&@xmath832&@xmath209&@xmath166&@xmath833 + eso 443 - 021&type i&@xmath834&@xmath141&@xmath334&@xmath835 + eso 466 - 014&type i&@xmath836&@xmath153&@xmath726&@xmath837 + eso 469 - 015&type i&@xmath838&@xmath241&@xmath419&@xmath839 + eso 533 - 004&type ii&@xmath840&@xmath153&@xmath841&@xmath842 + & & & @xmath841&@xmath628&@xmath843 + + eso 544 - 027&type ii&@xmath844&@xmath191&@xmath845&@xmath846 + & & & @xmath845&@xmath847&@xmath848 + ic 0217&type ii&@xmath849&@xmath241&@xmath850&@xmath851 + & & & @xmath850&@xmath199&@xmath852 + ic 0610&type i&@xmath853&@xmath241&@xmath206&@xmath854 + ic 1197&type ii&@xmath855&@xmath226&@xmath856&@xmath857 + & & & @xmath856&@xmath299&@xmath858 + ic 1553&type i&@xmath859&@xmath737&@xmath194&@xmath860 + ic 1711&type i&@xmath861&@xmath862&@xmath150&@xmath863 + ic 1913&type i&@xmath864&@xmath296&@xmath206&@xmath865 + ic 2058&type ii&@xmath866&@xmath209&@xmath867&@xmath868 + & & & @xmath867&@xmath278&@xmath869 + ic 2135&type ii&@xmath870&@xmath871&@xmath872&@xmath873 + & & & @xmath872&@xmath238&@xmath874 + ic 5176&type iii&@xmath875&@xmath161&@xmath876&@xmath877 + & & & @xmath876&@xmath601&@xmath878 + ngc 0489&type i&@xmath879&@xmath296&@xmath194&@xmath880 + ngc 0522&type ii&@xmath881&@xmath186&@xmath882&@xmath883 + & & & @xmath882&@xmath212&@xmath884 + ngc 0678&type ii&@xmath885&@xmath356&@xmath886&@xmath887 + & & & @xmath886&@xmath340&@xmath888 + ngc 1032&type i&@xmath889&@xmath296&@xmath166&@xmath890 + ngc 1163&type i&@xmath891&@xmath209&@xmath150&@xmath892 + ngc 1422&type i&@xmath893&@xmath133&@xmath894&@xmath895 + ngc 1495&type i&@xmath896&@xmath209&@xmath212&@xmath897 + ngc 1596&type i&@xmath898&@xmath296&@xmath283&@xmath899 + ngc 2732&type ii&@xmath900&@xmath296&@xmath901&@xmath902 + & & & @xmath901&@xmath289&@xmath903 + ngc 3098&type i&@xmath904&@xmath241&@xmath150&@xmath905 + ngc 3279&type ii&@xmath906&@xmath241&@xmath907&@xmath908 + & & & @xmath907&@xmath628&@xmath909 + ngc 3454&type ii&@xmath910&@xmath296&@xmath911&@xmath393 + & & & @xmath911&@xmath199&@xmath912 + ngc 3501&type i&@xmath913&@xmath474&@xmath375&@xmath914 + + + ngc 3592&type ii&@xmath915&@xmath141&@xmath916&@xmath917 + & & & @xmath916&@xmath138&@xmath918 + ngc 3600&type iii&@xmath919&@xmath241&@xmath920&@xmath921 + & & & @xmath920&@xmath233&@xmath922 + ngc 3628&type ii&@xmath923&@xmath206&@xmath924&@xmath925 + & & & @xmath924&@xmath327&@xmath926 + ngc 4081&type i&@xmath927&@xmath209&@xmath334&@xmath928 + ngc 4111&type i&@xmath929&@xmath147&@xmath340&@xmath930 + ngc 4330&type ii&@xmath931&@xmath296&@xmath932&@xmath933 + & & & @xmath932&@xmath668&@xmath934 + ngc 4359&type i&@xmath935&@xmath147&@xmath310&@xmath936 + ngc 4437&type ii&@xmath937&@xmath356&@xmath938&@xmath939 + & & & @xmath938&@xmath940&@xmath941 + ngc 4565&type i&@xmath942&@xmath289&@xmath943&@xmath944 + ngc 4607&type i&@xmath945&@xmath174&@xmath310&@xmath946 + ngc 4747&type i&@xmath947&@xmath153&@xmath375&@xmath948 + ngc 5084&type iii&@xmath949&@xmath378&@xmath950&@xmath951 + & & & @xmath950&@xmath381&@xmath952 + ngc 5470&type ii&@xmath953&@xmath133&@xmath954&@xmath955 + & & & @xmath954&@xmath150&@xmath956 + ngc 5529&type iii+ii&@xmath957&@xmath275&@xmath958&@xmath959 + & & & @xmath958&@xmath960&@xmath393 + & & & @xmath960&@xmath394&@xmath961 + ngc 5981&type i&@xmath962&@xmath440&@xmath400&@xmath963 + ngc 6010&type i&@xmath964&@xmath296&@xmath199&@xmath965 + ngc 7347&type i&@xmath966&@xmath440&@xmath419&@xmath967 + pgc 013646&type i&@xmath968&@xmath153&@xmath278&@xmath969 + pgc 028308&type i&@xmath970&@xmath147&@xmath246&@xmath971 + pgc 030591&type i&@xmath972&@xmath209&@xmath847&@xmath973 + pgc 032548&type i&@xmath974&@xmath296&@xmath447&@xmath975 + pgc 052809&type ii&@xmath976&@xmath191&@xmath977&@xmath978 + & & & @xmath977&@xmath238&@xmath979 + ugc 00903&type i&@xmath980&@xmath174&@xmath158&@xmath981 + ugc 01970&type i&@xmath982&@xmath147&@xmath315&@xmath983 + ugc 05347&type i&@xmath984&@xmath209&@xmath985&@xmath986 + + ugc 05689&type ii&@xmath987&@xmath191&@xmath988&@xmath989 + & & & @xmath988&@xmath447&@xmath990 + ugc 05958&type i&@xmath991&@xmath153&@xmath334&@xmath992 + ugc 06526&type i&@xmath993&@xmath296&@xmath150&@xmath994 + ugc 07086&type i&@xmath995&@xmath241&@xmath150&@xmath996 + ugc 08737&type i&@xmath997&@xmath343&@xmath238&@xmath998 + ugc 09448&type i&@xmath999&@xmath467&@xmath378&@xmath1000 + ugc 09665&type i&@xmath1001&@xmath133&@xmath158&@xmath1002 + ugc 10043&type i&@xmath1003&@xmath862&@xmath289&@xmath1004 + ugc 10288&type i&@xmath1005&@xmath474&@xmath1006&@xmath1007 + ugc 10297&type i&@xmath1008&@xmath440&@xmath356&@xmath1009 + ugc 12518&type i&@xmath1010&@xmath191&@xmath138&@xmath1011 + ugc 12692&type i&@xmath1012&@xmath265&@xmath194&@xmath1013 + ugc 12857&type i&@xmath1014&@xmath209&@xmath194&@xmath1015 + note . @xmath128 and @xmath129 stand for the inner and the outer limit of each fitted section and @xmath130 corresponds to the scale length of that section . the lower limit of the innermost section and the upper limit of the outermost section have no error bar because they have been set manually . the following pages include the vertical and horizontal luminosity profile fits for all the galaxies in our final sample . they also include the image of each galaxy . the information is organized as follows : * the three images in the top row show the average of the 3.6 and 4.5@xmath13-band background - subtracted s@xmath4 g frames ( left ) , the same image after masking ( center ) , and the used mask ( right ) . the vertical red lines indicate the limits of the fitted vertical bins , the central one being ignored due to the possible presence of a bulge . * the next two rows show the fits to the luminosity profiles in these bins . the data points have @xmath1016 statistical error bars , the dashed curve represents the best fit , the dotted curves indicate the contributions of the thin and the thin discs . the dash - dotted vertical lines indicate the limits of the range in vertical distance above the mid - plane used for the fit . the vertical solid line indicates @xmath14 for each bin . the profiles in bins which have not been selected for the data processing in this paper ( section [ secdefrs ] ) are indicated with a `` not used '' label . * the bottom panel shows the horizontal luminosity profiles integrated from @xmath68 to @xmath1017 ( total ; yellow ) , from @xmath68 to @xmath1018 ( thin disc ; blue ) , and from @xmath1019 to @xmath1017 ( thick disc ; red ) . the black lines indicate fits to those profiles , with vertical intersecting lines indicating the fitting range and the truncation radii . the vertical solid gray lines represent @xmath1020 , @xmath1021 , and @xmath1022 . the vertical dashed line represents @xmath60 . the gray area represents the inferred average surface brightness of the thick disc in the range @xmath68 to @xmath1018 in the presence of a thin disc with a relative face - on surface brightness similar to which is found where the vertical luminosity profiles have been successfully fitted ( upper limit ) and in the absence of it ( lower limit ) extrapolated from the vertical luminosity profile fits . the bins for which valid vertical luminosity fits have been obtained can be seen in the bottom left of each panel . the errors in the fitted parameters are @xmath1016 fitting errors .

in 1957 , jaynes @xcite showed that maximizing statistical mechanic entropy for the purpose of revealing how gas molecules were distributed was simply the maximizing of the shannon information entropy @xcite with statistical mechanical information . this idea lead to maxent or his use of the method of maximum entropy for assigning probabilities . this method has recently evolved to a more general method , the method of maximum relative entropy ( mre ) @xcite which has the advantage of not only assigning probabilities but _ updating _ them when new information is given in the form of constraints on the family of allowed posteriors . one of the drawbacks of the maxent method was the inability to include data . when data was present , one used bayesian methods . the methods were combined in such a way that maxent was used for assigning a prior for bayesian methods , as bayesian methods could not deal with information in the form of constraints , such as expected values . previously it has been shown that one can use the mre method to reproduce a mean field solution for a simple fluid @xcite . the purpose of this was to illustrate that in addition to updating probabilities , mre can also be used for _ approximating _ probability distributions as an approximation tool . in a simple ferromagnetic material ( that is , a ferromagnetic material with a single domain ) , the electronic spins of the individual atoms are strong enough to affect one and other , and give rise to the so called exchange interaction @xcite . this effect , however , is temperature dependent . when the temperature is below a certain point ( the curie or critical temperature ) the spins tend to all point in the same direction due to their influence on each other . this establishes a permanent magnet as the individual atoms produce a net dipole effect . above this temperature , the atoms cease to have a significant effect on each other and the material behaves more like a paramagnetic substance . determining this net dipole effect can be difficult . first , the interactions are due to complicated quantum effects . second , since a given material has a very large number of atoms , computing the net dipole effect can be difficult in two dimensions and completely intractable in three dimensions . therefore , approximations such as using an ising model @xcite and/or the mean field approximation @xcite are made to facilitate computation . applications of the mre ( updating ) method together with information geometric methods used to characterize the complexity of dynamical systems described in terms of probabilistic tools are quite extensive @xcite . in ref . @xcite , using the mre method together with differential geometric techniques , we proposed an information - geometric characterization of chaotic energy level statistics of a quantum antiferromagnetic ising spin chain in a tilted magnetic field . in ref . @xcite , employing the very same aforementioned techniques , we were able to establish a connection between the behavior of the information - geometric complexity of a trivariate gaussian statistical model and the geometric frustration phenomena that appears in triangular ising models @xcite . however , the purpose of our article is to illustrate the use of the mre ( approximating ) method as a tool for attaining approximations for ferromagnetic materials that lie outside the ability of traditional methods . in doing so , we further the previous work done and show the versatility of the method . the layout of the remaining part of this manuscript is as follows . in section ii , we briefly outline the essential steps of the mre method in updating and approximating probability distributions . in section iii , we describe the basics of the ising model and mean field theory as approximate mathematical descriptions of ferromagnetic materials . in section iv , we compare magnetization properties of ferromagnets inferred by means of mre with those obtained via the ising model together with mean field theory . in section v , we further test the effectiveness of the mre methodology by considering ferromagnetic material in the presence of defects . our final remarks appear in section vi . in this section , we outline the essential elements of the mre method as a technique for updating and/or approximating probability distributions . the mre method is a technique for updating probabilities when new information is provided in the form of a constraint on the family of the allowed posteriors . the main feature of the mre method is the possibility of updating probabilities in the presence of both data and expected value constraints . this feature was first formally presented in @xcite where , in particular , it was shown that bayes updating can be regarded as a special case of the mre method . a first semi - quantitative analysis of the effective advantages of this powerful feature of the mre method appeared in @xcite . finally , the first fully quantitative investigation of the advantages of the mre method was carried out in @xcite where two toy problems were solved in detail . following these lines of investigation , we present here a novel application of the mre method to a real - world ferromagnetic problem . we use the mre method to update from a prior to a posterior probability distribution . specifically , we want to make inferences on some quantity @xmath0 given : i ) : : the prior information about @xmath1 ( the prior ) ; ii ) : : the known relationship between @xmath2 and @xmath3 ( the model ) ; iii ) : : the observed values of the variables ( data ) @xmath2 . the search space for the posterior probability distribution occurs in the product space @xmath4 , and the joint distribution is denoted as @xmath5 . the key idea is going from the old prior @xmath6 to the updated prior @xmath7,@xmath8 the joint probability @xmath9 maximizes the relative entropy functional @xmath10 $ ] , @xmath11 \overset{\text{def}}{=}-\int ddd\theta p\left ( d\text { , } \theta\right ) \log\left [ \frac{p\left ( d\text { , } \theta\right ) } { p_{\text{old}}\left ( d\text { , } \theta\right ) } \right ] \text { , } \label{ef}\ ] ] subject to the given information constraints . note that @xmath12,@xmath13 is called here the joint prior , while @xmath6 and @xmath14 denote the standard bayesian prior and the likelihood , respectively . we emphasize that both the joint prior and the standard bayesian prior encode prior information about @xmath0 . furthermore , despite the fact that the likelihood is not regarded as prior information in the conventional sense , it will be considered here as prior information since it represents the a priori established relation between @xmath0 and @xmath2 . let us specify now the relevant information constraints . first , we impose the standard normalization constraint,@xmath15 second , we consider an information constraint in the form of expected value of some smooth function @xmath16,@xmath17 finally , we consider the observed data @xmath18 . within the mre method , knowledge of this information leads to an infinite number of constraints,@xmath19 for any @xmath2 where @xmath20 denotes the dirac delta function . using the lagrange multipliers technique , we maximize the logarithmic relative entropy functional in eq . ( [ ef ] ) subject to the constraints in eqs . ( [ c1 ] ) , ( [ c2 ] ) , and ( [ c3 ] ) . we impose that the variation with respect to @xmath21 of the entropy functional @xmath10 $ ] is equal to zero,@xmath22{c}s\left [ p\left\vert p_{\text{old}}\right . \right ] + \alpha\left [ \int ddd\theta p\left ( d\text { , } \theta\right ) -1\right ] \\ + \beta\left [ \int ddd\theta g\left ( \theta\right ) p\left ( d\text { , } \theta\right ) -g\right ] \\ + \int dd\gamma\left ( d\right ) \left [ \int d\theta p\left ( d\text { , } \theta\right ) -\delta\left ( d - d^{\prime}\right ) \right ] \end{array } \right\ } = 0\text{. } \label{var}\ ] ] we note that eq . ( [ c3 ] ) describes the information in the data and represents an infinite number of constraints on the family @xmath23 . for this reason , there exist one constraint and one lagrange multiplier @xmath24 for each value of @xmath25 . after some simple algebra of variations , eq . ( [ var ] ) becomes@xmath26 \delta p\left ( d\text { , } \theta\right ) = 0\text { , } \label{get}\ ] ] for any @xmath27 . therefore , from eq . ( [ get ] ) we find@xmath28 where the lagrange multipliers @xmath29 , @xmath30 , and @xmath31 can be formally determined by substituting eq . ( [ newp ] ) into eqs . ( [ c1 ] ) , ( [ c2 ] ) and ( [ c3 ] ) , respectively . after some algebra , we obtain@xmath32 observe that the lagrange multiplier @xmath30 in eq . ( [ pnew ] ) can only be implicitly determined and depends on the observed data @xmath18 . finally , marginalizing @xmath9 over the variable @xmath25 , we obtain the updated prior probability distribution@xmath33 for the sake of notational simplicity , define@xmath34 finally , eq . ( [ gg ] ) becomes@xmath35 note that in the absence of constraints in the form of expected values , @xmath36 and eq . ( [ ggg ] ) reduces to the standard bayes updating relation@xmath37 for the sake of completeness , we point out that eq . ( [ bb ] ) can be obtained by combining bayes theorem,@xmath38 and bayes rule,@xmath39 in what follows , we shall regard the mre method as a tool for approximating probability distributions . for the sake of reasoning , assume that the microstates of a system are labeled by coordinates @xmath40 . furthermore , assume that the probability that the system is in a microstate within a given range @xmath41 is described by the canonical distribution @xcite,@xmath42 where the function @xmath43 is defined as,@xmath44 in statistical mechanical terms , the quantity @xmath45 in eq . ( [ ppp ] ) denotes the helmholtz free energy . in general , to describe a system in terms of its observables , one needs to compute expected values in terms of integrals over @xmath21 which , in turn , depends on the hamiltonian @xmath46 . calculating these integrals can be quite challenging especially when the hamiltonian exhibits nonlinear interaction terms . to make things more tractable , one can think of replacing the exact expression of @xmath21 with an approximate expression @xmath47 provided that the latter preserves the relevant information content essential for our specific inferences . more specifically , @xmath47 can be identified in two steps . first , identify a suitable family of trial distributions @xmath48 relevant for our specific problem at hand . second , select @xmath47 from @xmath49 . there is no recipe for the first step . the second step is mechanical . the distribution to be selected is the one that maximizes the entropy of @xmath50 relative to @xmath21,@xmath51 = -\int dxp_{a}\left ( x\right ) \log\left [ \frac{p_{a}\left ( x\right ) } { p\left ( x\right ) } \right ] \text{. } \label{re}\ ] ] a very convenient family of trial distributions is given by the canonical distributions with a modified hamiltonian @xmath52 where @xmath53 are parameters that label each distribution within a family,@xmath54 where the @xmath55 function is given by,@xmath56 we underline that in sec . ii , the symbol @xmath1 is used to denote one parameter or many parameters about which one wishes to make inferences . here , as mentioned earlier , the parameters @xmath53 are used to label a family of trial distributions that are canonical with the hamiltonian * * @xmath57 . substituting eqs . ( [ p ] ) , ( [ ppp ] ) , ( [ pp ] ) , and ( [ ppz ] ) into eq . ( [ re ] ) , we get@xmath58 = \beta\left ( f - f_{0}\right ) + \beta\left\langle \mathcal{h}_{0}-\mathcal{h}\right\rangle _ { 0}\text { , } \label{sp}\ ] ] where,@xmath59 note that the symbol * * @xmath60 * * in eq . ( [ sp ] ) denotes the average over the trial distribution * * @xmath61 * * and the subscript * * @xmath62 * * is simply used to recall that * * @xmath61 is canonical with the modified hamiltonian * * @xmath57 . since @xmath63 \leq0 $ ] , eq . ( [ sp ] ) implies@xmath64 loosely speaking , maximizing @xmath65 is equivalent to minimizing @xmath66 . this latter minimization procedure is also known as the bogoliubov variational principle @xcite which , as shown here , can be regarded as a special case of the mre method . [ c]|c|c|c|ferromagnet & symbol & critical temperature + iron & fe & 1043 + cobalt & co & 1388 + nickel & ni & 627 + gadolinium & gd & 293 + dysprosium & dy & 85 + one of the most interesting phenomena in solid state physics is represented by ferromagnetism @xcite . in spite of the extensive oversimplifications required to construct a tractable mathematical model describing such a phenomena , calculations concerning ferromagnetism are among the most impressive exhibition of brute force computation achieved by theoretical physicists @xcite . in what follows , we shall outline the basics of the so - called ising model of ferromagnetic materials @xcite . a few ferromagnetic materials together with their corresponding critical temperatures are listed in table i. one of the simplest systems that exhibits a transition from ordered to disordered states is that of a lattice composed of two different types of objects , @xmath67 and @xmath68 . assume that the objects interact with only their nearest neighbors . if one raises the temperature @xmath69 of the system at some critical point @xmath70 , the system will melt and become completely disordered . a relatively simple mathematical model of such a system was developed by ernst ising to describe ferromagnetism @xcite . the general expression for the ising hamiltonian is given by @xcite,@xmath71 where @xmath72 is the spin variable , @xmath73 is the cardinality of the spin variables , @xmath74 determines the spin configuration of the whole system , the symbol @xmath75 with @xmath76 denotes a pair of next neighbor spins , @xmath77 are the exchange coupling constants , and @xmath68 is the @xmath78-component of a uniform external magnetic field . ferromagnetic and antiferromagnetic materials are characterized by exchange coupling coefficients with @xmath79 and @xmath80 , respectively , for all pairs @xmath81 , @xmath82 . for instance , iron ( fe ) and iron oxide ( feo ) are examples of ferromagnetic and antiferromagnetic materials , respectively . their critical temperature is given by @xmath83 and @xmath84 @xcite , respectively . for the sake of reasoning , we shall assume in what follows that @xmath85 ( zero external magnetic field ) and @xmath86 for all pairs @xmath81 , @xmath82 ( isotropic interaction strength ) . the hamiltonian in eq . ( [ ham ] ) becomes,@xmath87 the sum over @xmath88 in eq . ( [ ham2 ] ) contains @xmath89 terms where @xmath90 is the number of next nearest neighbors of each lattice site and depends on the specific type of lattice structure being considered . for instance , for a two - dimensional square lattice , @xmath91 . the partition function @xmath92 corresponding to the hamiltonian in eq . ( [ ham2 ] ) is the sum of @xmath93 terms and is given by @xcite,@xmath94 where @xmath95 , with @xmath96 denoting the boltzmann constant . as a side remark , we recall that the thermodynamic functions ( internal energy and thermal capacity , for instance ) can be obtained in the usual manner from the helmholtz free energy * * @xmath97 @xcite,@xmath98 in the two - dimensional case with only nearest neighbor interactions and simple lattice configurations ( square and triangular , for instance ) , the exact free energy is known in the hypothesis of zero magnetic field @xcite . for the one dimensional case , the partition function @xmath99 can easily be calculated , @xmath100 ^{n-1}\overset{n\gg1}{\approx}\left [ 2\cosh\left ( \beta j\right ) \right ] ^{n}\text{. } \label{partition2}\ ] ] furthermore , the resulting average energy @xmath101 for @xmath102 would then be , @xmath103 where @xmath104 denotes the hyperbolic tangent function . for the sake of clarity , a rectangular two - dimensional ising model is depicted in fig . 1 . one of the most important starting points for more sophisticated calculations concerning ferromagnetic transitions is furnished by the so - called mean field theory ( or , molecular field theory @xcite ) . the mean field approximation for the ising model assumes that each and every spin is subject to a mean field generated by its next nearest neighbors only @xcite . such an approximation turns out to be especially important for three dimensional materials where each atom may have @xmath105 , @xmath106 or even @xmath107 close neighbors depending on the crystal geometry @xcite . we start with the same energy function as given in eq . ( [ ham2 ] ) but now examine only one atom ( sometimes called the _ central _ atom ) , @xmath108 where @xmath109 is the energy ascribed to one atom and @xmath110 is the number of nearest neighbors ( for instance , a face - centered cubic lattice would have @xmath111 ) with atoms at the edge of the material neglected . assuming that each neighbor contributes equally and averaging over the neighbor spins around this one atom , we obtain @xmath112 where,@xmath113 with @xmath114 . the boltzmann probability @xmath115 of finding this atom in the @xmath116 state is given by , @xmath117 and the partition function @xmath118 becomes , @xmath119 taking the average of @xmath116 for this one atom over its two possible states , using eqs . ( [ 1 ] ) and ( [ 2 ] ) , the mean magnetization becomes @xmath120 = \frac{2\sinh\left ( \beta jn\bar{s}_{j}\right ) } { 2\cosh\left ( \beta jn\bar{s}_{j}\right ) } = \tanh\left ( \beta jn\bar{s}_{j}\right ) ~\text{. } \label{expected_m}\ ] ] furthermore , assuming that _ all _ atoms will behave like this one labeled with the @xmath81 index and assuming that the average spin @xmath121 for the entire ferromagnetic material is formally defined as , @xmath122 from eqs . ( [ expected_m ] ) and ( [ as ] ) , we find @xmath123 vs. temperature @xmath69 in case of the mean field approximation applied to the ising model . we set @xmath124 and assume @xmath125 , @xmath126 , and the level curves @xmath127 with @xmath128,scaledwidth=50.0% ] observe that the inner and outer expectation values in the nested expectation value in eq . ( [ as ] ) used to define the average spin @xmath121 for the entire ferromagnetic material represent the averages over all possible states of an atom and over all atoms of the material , respectively . if one considers eq . ( [ ham ] ) instead of eq . ( [ ham2 ] ) , eq . ( [ avgm ] ) can be generalized and becomes @xcite,@xmath129 \text { , } \label{implicit}\ ] ] where * * @xmath68 * * denotes the intensity of the applied external magnetic field . ( [ implicit ] ) can be solved numerically in order to uncover the mean magnetization @xmath130 as a function of the temperature @xmath69 of the system . a contour plot of @xmath131 defined as,@xmath132 \text{,}\ ] ] as a function of @xmath133 and @xmath69 appears in fig . when an external magnetic field with strength @xmath68 is applied , the effective hamiltonian @xmath134 for the @xmath81-th atom becomes,@xmath135 where * * @xmath136 * * is the magnetic moment and * * @xmath137 * * is the net effective energy on the @xmath81-th atom . as a final remark , we also point out that if one were to write the full hamiltonian with the external field instead of @xmath138 as given in eq . ( [ ham2 ] ) , then one will not have as simple of a solution for @xmath139 as in eq . ( [ partition2 ] ) , i.e. @xmath140 ^{n}$ ] . as we have seen in the previous section , the ising model and mean field theory require various assumptions to be fulfilled . part of the strength of using the mre is that one does not need to justify some of the assumptions . stated otherwise , one simply supplies the information constraints that one has available and then allow the method to turn out the least biased solution based on the information given . this can be demonstrated by using the mre method to find an appropriate approximate description of a ferromagnetic material . following the procedure outlined in section ii , we begin by determining the full posterior solution for a ferromagnetic atomic system where we have both an expectation value constraint as well as observed data . for this example , the data observed will be the effect of the external magnetic field , @xmath141 on each atom . the appropriate entropy to consider becomes , * * @xmath142=-\int d^{n}b\sum_{\left\ { s\right\ } } p\left\ { s , b\right\ } \log\left [ \frac{p\left\ { s , b\right\ } } { p_{\text{old}}\left\ { s , b\right\ } } \right ] \label{h_entropy}\ ] ] where * * @xmath143** **@xmath144 * * , * * @xmath145 * * will be a flat prior ( constant in @xmath146 and uniform in @xmath68 ) , @xmath73 denotes the number of atoms , and@xmath147 second , we consider the energy constraint in the form of the expectation value of the hamiltonian,@xmath148 where * * @xmath149 * * denotes some general hamiltonian . for illustrative purposes , we adopt as our starting point the ising model so as to show how one can use the mre to arrive at a similar approximation as the one produced by mean field theory . let * * @xmath149 * * in eq . ( [ h_constraint ] ) be defined as , * * @xmath150 where * * @xmath151 equals * * @xmath152 * * in eq . ( [ ham2 ] ) and * * @xmath153 * * is the energy attributed to the observed external magnetic field * * @xmath141 * * acting on the individual atoms . next is to apply the data constraints . as mentioned in section ii , there are an infinite number of data constraints associated with a single data point . therefore , here we have * * @xmath73 * * sets of them , one set for each data point . for illustrative purposes , we will assume that the field is uniform and therefore the value * * @xmath154 * * for all * * @xmath81 , @xmath155 . * * we can then simply write the data constraint as,@xmath156 along with the normalization constraint , maximizing the logarithmic relative entropy with respect to the constraints yields the canonical distribution,@xmath157 where @xmath158 @xmath159 , @xmath160 is similar to the partition function from eq . ( [ partition ] ) with the exception of the data in the hamiltonian @xmath161 which is now,@xmath162 however , in order to determine the values of quantities such as the critical temperature , the total magnetic moment and the magnetic susceptibility , we are faced with the same dilemma as above : we can not compute the solutions explicitly . therefore , we must try to find an approximation that is computationally tractable . we now wish to use mre to find an approximation that is tractable . we accomplish this by first writing down the appropriate entropy functional @xmath163 $ ] , @xmath164=-\sum_{\left\ { s\right\ } } p_{\text{a}}\log\left [ \frac{p_{\text{a}}}{p}\right ] \text { , } \label{entropy_me}\ ] ] where @xmath165 is the canonical probability distribution as in eq . ( [ p ] ) with @xmath166 as given in eq . ( [ hamiltonian_nf ] ) and @xmath167 is the approximation that we seek . we proceed by rewriting the entropy functional @xmath163 $ ] as , @xmath164=-\sum_{\left\ { s\right\ } } p_{\text{a}}\log p_{\text{a}}+\sum_{\left\ { s\right\ } } p_{\text{a}}\log p\text{. } \label{sre}\ ] ] using eqs . ( [ p ] ) and ( [ hamiltonian_nf ] ) , eq . ( [ sre ] ) becomes @xmath164=\frac{1}{k_{b}}s_{\text{a}}-\sum_{\left\ { s\right\ } } p_{\text{a}}\beta\mathcal{h}^{\prime}\left\ { s\right\ } + \beta f\text { , } \label{sre1}\ ] ] where we have used the fact that the partition function @xmath160 can also be written in terms of the free energy @xmath45 as @xmath168 . ( [ sre1 ] ) can further be reduced to@xmath164=\frac{1}{k_{b}}s_{\text{a}}+\beta\left ( f-\left\langle \mathcal{h}^{\prime}\left\ { s\right\ } \right\rangle _ { \text{a}}\right ) \text { , } \label{aa}\ ] ] where @xmath169 can be regarded as the energy @xmath170 of the system and is formally defined as,@xmath171 since by definition @xmath163\leq0 $ ] , using eqs . ( [ aa ] ) and ( [ aaa ] ) leads to @xmath172 when using the mre method , we maximize the entropy functional in order to find the best posterior given the information provided . for this specific case , we have rewritten the entropy in terms of an inequality that compares the free energy of the system @xmath45 with the approximate values for the average energy @xmath170 and the entropy @xmath173 . the free energy is minimized to determine the best approximation for @xmath167 . this minimization problem is also known as the bogoliubov variational principle @xcite . therefore , from our discussion , we conclude that this variational principle is simply a special case of the mre method . to be more general , we proceed with using the mre method to find the best approximation . as stated earlier , when using the mre method , the goal is to search the family of possible posteriors in order to find the one that maximizes the entropy given the constraints . in addition to this , we need a solution that is tractable . therefore , we seek a posterior @xmath167 that has a form @xmath174 where @xmath175 is defined as , @xmath176 and @xmath177 is some effective energy similar to @xmath134 given in eq . ( [ hamiltonian_effm ] ) . for illustrative purposes , we assume that all atoms poses the same effective energy so that @xmath178 for any @xmath179 . the difference is that we do not yet know the form of @xmath180 . we continue by following a similar route to eq . ( [ expected_m ] ) by writing the expectation value for @xmath116 with respect to @xmath167 , @xmath181 except that here we are marginalizing over all atoms except the @xmath81-th . notice that because the effective energy @xmath180 is constant , the solution is independent of the index @xmath81 . next we maximize @xmath163 $ ] in eq . ( [ entropy_me ] ) or , in keeping with the current case , we minimize the free energy @xmath45 . from eq . ( [ aa ] ) , we have @xmath182 substituting eq . ( [ pa ] ) into eq . ( [ a4 ] ) , after some algebra we get@xmath183 note that @xmath184 is a constant and can be extracted from the summation in eq . ( [ pull ] ) . furthermore , since @xmath185 , eq . ( [ pull ] ) becomes @xmath186 or expressed otherwise,@xmath187 we now substitute the explicit expressions for the hamiltonians @xmath166 in eq . ( [ hamiltonian_nf ] ) and @xmath175 in eq . ( [ ha ] ) into the function @xmath188 in eq . ( [ fmin ] ) . after some algebra , we find @xmath189 noting that,@xmath190 @xmath188 becomes , @xmath191 substituting @xmath192 ^{n}$ ] into eq . ( [ fmin2 ] ) yields , @xmath193 ^{n}-j\sum_{i , j}^{n , n}\left\langle s_{i}s_{j}\right\rangle _ { \text{a}}-\mu b^{\prime}n\tanh\left ( \beta\epsilon_{\text{a}}\right ) \text{.}\ ] ] from eq . ( [ avga ] ) , @xmath194 and does not depend on the index @xmath81 , so that@xmath195 therefore , @xmath188 can be rewritten as @xmath196 -j\sum_{i , j}^{n , n}\left\langle s_{i}\right\rangle _ { \text{a}}\left\langle s_{j}\right\rangle _ { \text{a}}-\mu b^{\prime}n\tanh\left ( \beta\epsilon_{\text{a}}\right ) ~\text{. } \label{mmm}\ ] ] substituting eq . ( [ avga ] ) into eq . ( [ mmm ] ) yields , @xmath197 -j\frac{1}{2}nn\left [ \tanh\left ( \beta\epsilon_{\text{a}}\right ) \right ] ^{2}-\mu b^{\prime}n\tanh\left ( \beta\epsilon_{\text{a~}}\right ) \text { , } \label{minimi}\ ] ] where the factor @xmath110 specifies the number of nearest neighbors , the factor @xmath73 is from the total number of atoms and the @xmath198 appears in order to take into account double counting . we now minimize @xmath188 in eq . ( [ minimi ] ) with respect to @xmath180 , @xmath199 that is , @xmath200 after canceling terms in eq . ( [ simply ] ) , we have @xmath201 that is , @xmath202 eq . ( [ ea ] ) is our final result for this section . in fig . 2 , the numerical solution of the mean magnetization vs. temperature is reported . we note that for @xmath203 and @xmath204**,**@xmath205 the critical temperature @xmath70 is the temperature at which spontaneous magnetization in a lattice of magnetic material begins to appear , and this is where a phase transition occurs @xcite . roughly speaking , a phase transition in a lattice is a singularity in the limit of @xmath206 as @xmath25 , the size of the lattice , approaches infinity . for the sake of completeness , we remark here that while onsager was the first to obtain a closed - form solution to the ising two - dimensional ferromagnetic model in the absence of an external magnetic field @xcite , yang was the first to publish the exact calculation of the spontaneous magnetization for a two - dimensional ising model @xcite . returning to fig . 2 , we note that for @xmath207 , there is only one approximating probability distribution that is uniform over all states . for @xmath208 , there are two minima that correspond to approximating distributions that are symmetry - broken , with all spins more likely to be down , or all spins more likely to be up . furthermore , notice that if @xmath209 in eq . ( [ ea ] ) , we recover our solution using the mean field approximation above . as a matter of fact , letting @xmath210 , we rewrite eq . ( [ ea ] ) as , @xmath211 that is , we reattain eq . ( [ avgm]),@xmath212 however , from eq . ( [ ea ] ) alone we can formally solve for the critical temperature , the total magnetic moment and the magnetic susceptibility ( that is , the ratio between the magnetization of the material and the strength of the magnetic field applied to the material ) in the usual way @xcite . this is true even though we still do not know the explicit form for @xmath180 . using this mre approach , we did not need to assume that all atoms behave like a central atom and we did not need to know the explicit form of the effective energy @xmath180 , only that there was one . following the mre method we simply processed all of the information that we had available . notice that we also no longer need to assume ising conditions . we further explore such considerations in the next section . in this section , we employ mre techniques to infer the numerical estimates of effective energy levels of atoms in two cases : i ) atoms with three possible states ; ii ) defective atoms in a crystal lattice . to illustrate the use of mre in determining critical temperatures , we look at the ising model with three possible states : spin up , spin down , and no spin . in this example , we will let @xmath213 where an atom in the @xmath62-state would contribute no energy . we follow the same line of reasoning outlined in the previous section up until eq . ( [ avga ] ) where the new expected value for the @xmath81-th spin variable @xmath214 is now given by,@xmath215 after some tedious algebra , the new expression for @xmath188 becomes @xmath216 -j\frac{1}{2}nn\left [ \frac{2\sinh\left ( \beta\epsilon_{\text{a}}\right ) } { 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1}\right ] ^{2}-\mu b^{\prime}n\frac{2\sinh\left ( \beta\epsilon_{\text{a}}\right ) } { 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1}~\text{.}\ ] ] minimizing this function @xmath188 once again with respect to @xmath217 yields,@xmath218 ^{2}}\right ) + \nonumber\\ & \nonumber\\ & -\frac{n}{\beta}\frac{2\beta\sinh\left ( \beta\epsilon_{\text{a}}\right ) } { 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1}-j\frac{1}{2}nn2\left ( \frac{2\sinh\left ( \beta\epsilon_{\text{a}}\right ) } { 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1}\right ) \left ( \frac{2\beta\cosh\left ( \beta\epsilon_{\text{a}}\right ) } { 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1}-\frac{4\beta\sinh^{2}\left ( \beta\epsilon_{\text{a}}\right ) } { \left [ 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1\right ] ^{2}}\right ) + \nonumber\\ & \nonumber\\ & -\mu b^{\prime}n\left ( \frac{2\beta\cosh\left ( \beta\epsilon_{\text{a}}\right ) } { 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1}-\frac { 4\beta\sinh^{2}\left ( \beta\epsilon_{\text{a}}\right ) } { \left [ 2\cosh\left ( \beta\epsilon_{\text{a}}\right ) + 1\right ] ^{2}}\right ) ~\text{. } \label{mess}\ ] ] after canceling terms in eq . ( [ mess ] ) , we arrive at @xmath219 vs. temperature @xmath69 . we set @xmath124 and assume @xmath125 , @xmath126 , @xmath220 , @xmath221 and the level curves @xmath222 with @xmath223.,scaledwidth=50.0% ] eq . ( [ cc ] ) is our final result . a contour plot of @xmath224 defined as,@xmath225 as a function of @xmath180 and @xmath69 appears in fig . as stated earlier , this equation can then be used as above to solve for the critical temperature , the total magnetic moment and the magnetic susceptibility for this specific ferromagnetic material . as a side remark , we note that to make this result a little more general we can also write , @xmath226 where @xmath184 is once again the approximate partition function , @xmath227 is the number of nearest neighbors and @xmath228 is the approximate free energy . * * despite its lack of elegance , our entropic analysis leading to eq . ( [ cc ] ) together with fig . 3 seems to support the idea that the introduction of additional ( allowed ) states in the atomic ferromagnetic structure is consistent with both space shifts and nonhomogeneities of the critical temperature @xcite . a deeper understanding of these specific aspects require a deeper analysis that we leave to future investigations . as a side remark , although unnecessary and perhaps impractical , it should be noted that the * * _ spin state _ @xmath62 could be introduced as an observable quantity in a way that is analogous to how @xmath68 in eq . ( [ h_data ] ) was implemented . in @xcite , onsager s method was used to study the physical properties of a two - dimensional square lattice containing four kinds of atoms under the assumption that only the interaction of nearest neighbors was important and that only two distinct energies of interaction were permitted . here , following this line of investigation , we examine the case where one of the _ atoms _ is actually a missing one . in real cases , it would be very difficult to know how many atoms there are of each type . indeed , the point of the _ central atom _ idea is that we can not know the states of all the atoms in a lattice , so we must assume they are all the same . however , it can be that when a crystal is scanned , the scan can indicate where there might be impurities or defects . in solid state physics , a point defect ( or , vacancy ) in a monatomic bravais lattice occurs whenever a lattice site that would usually be occupied by an ion in the perfect crystal has not any ion associated with it . for the sake of clarity , a vacancy is illustrated in fig . 4 . in the previous application , we examined a three state atom . in this next example , we shall examine a material that has a known defect or defects . in this case , we need to use two effective energy terms . one for the atoms that are surrounded by non - defective atoms , @xmath229 and another for the ones affected by the defect , @xmath230 . for illustrative purposes , we will examine the case where we only have one defective atom . this means that there are @xmath231 atoms that are surrounded by non - defective atoms , @xmath110 atoms that have one defective atom next to it and @xmath232 atom which is the defective atom . for our purposes , let us think of the defect as an empty slot . given these conditions , the actual hamiltonian @xmath149 of the system is given by,@xmath233 where @xmath81 labels the non - defective atoms , @xmath82 labels the neighbors ( @xmath110 ) for these atoms , @xmath234 are the atoms affected by the defect , @xmath235 are the neighbors ( @xmath236 of the affected atoms and @xmath237 is the spin variable of the defective atom . since we are looking at this defect as an empty slot , we let @xmath238 . now we write our estimated hamiltonian @xmath175 for this case as , @xmath239 following the same procedures outlined above , we attain two expected values for the spin variables , one for each effective energy , @xmath240 and,@xmath241 once again , we write down the function @xmath188 we wish to minimize as@xmath242 where @xmath166 is arrived at in a similar way to eq . ( [ hamiltonian_nf ] ) . substituting eqs . ( [ himpo ] ) and ( [ himpa ] ) into eq . ( [ fimpo ] ) for @xmath188 yields , @xmath243 observe that,@xmath244 and , @xmath245 substituting eqs . ( [ asss ] ) and ( [ assss ] ) into eq . ( [ car ] ) , after some algebra , we obtain @xmath246 ~\text{,}\ ] ] where @xmath247 . substituting @xmath248 ^{\left ( n - n-1\right ) } $ ] and @xmath249 ^{\left ( n\right ) } $ ] yields , @xmath250 ^{n - n-1}+\nonumber\\ & \nonumber\\ -\frac{1}{\beta}\log\left [ 2\cosh\left ( \beta\epsilon_{\text{d}}\right ) \right ] ^{n-1 } & -j\sum_{i , j}^{n - n-1,n}\left\langle s_{i}s_{j}\right\rangle _ { \text{nd}}-j\sum_{k , l}^{n , n-1}\left\langle s_{k}s_{l}\right\rangle _ { _ { \text{d}}}-\mu b^{\prime}\left [ \epsilon_{\text{nd}}\left ( n - n-1\right ) \tanh\left ( \beta\epsilon_{\text{nd}}\right ) + \epsilon_{\text{d}}n\tanh\left ( \beta\epsilon_{\text{d}}\right ) \right ] ~\text{.}\ ] ] since @xmath214 in eq . ( [ s_nd ] ) and @xmath251 in eq . ( [ s_d ] ) are independent of @xmath81 and @xmath234 , respectively , we have @xmath252 and @xmath253 . therefore , we can write @xmath254 ^{n - n-1}-\frac { 1}{\beta}\log\left [ 2\cosh\left ( \beta\epsilon_{\text{d}}\right ) \right ] ^{n-1}+\nonumber\\ & \nonumber\\ & -j\sum_{i , j}^{n - n-1,n}\left\langle s_{i}\right\rangle _ { \text{nd}}\left\langle s_{j}\right\rangle _ { \text{nd}}-j\sum_{k , l}^{n , n-1}\left\langle s_{k}\right\rangle _ { \text{d}}\left\langle s_{l}\right\rangle _ { \text{d}}-\mu b^{\prime}\left [ \epsilon_{\text{nd}}\left ( n - n-1\right ) \tanh\left ( \beta\epsilon_{\text{nd}}\right ) + \epsilon_{\text{d}}\left ( n\right ) \tanh\left ( \beta\epsilon_{\text{d}}\right ) \right ] ~\text{. } \label{mm}\ ] ] substituting eqs . ( [ s_nd ] ) and ( [ s_d ] ) into eq . ( [ mm ] ) yields , @xmath255 ^{n - n-1}+\nonumber\\ & \nonumber\\ & -\frac{1}{\beta}\log\left [ 2\cosh\beta\epsilon_{\text{d}}\right ] ^{n-1}-\frac{1}{2}j\left ( n - n-1\right ) \left ( n\right ) \tanh^{2}\left ( \beta\epsilon_{\text{nd}}\right ) -\frac{1}{2}j\left ( n\right ) \left ( n-1\right ) \tanh^{2}\left ( \beta\epsilon_{\text{d}}\right ) + \nonumber\\ & \nonumber\\ & -\mu b^{\prime}\left [ \epsilon_{\text{nd}}\left ( n - n-1\right ) \tanh\left ( \beta\epsilon_{\text{nd}}\right ) + \epsilon_{\text{d}}\left ( n\right ) \tanh\left ( \beta\epsilon_{\text{d}}\right ) \right ] \text{.}\ ] ] after collecting @xmath229 and @xmath230 like terms , we have @xmath256 ^{n - n-1}+\nonumber\\ & \nonumber\\ & -\frac{1}{2}j\left ( n - n-1\right ) \left ( n\right ) \tanh^{2}\left ( \beta\epsilon_{\text{nd}}\right ) -\mu b^{\prime}\epsilon_{\text{nd}}\left ( n - n-1\right ) \tanh\left ( \beta\epsilon_{\text{nd}}\right ) + \nonumber\\ & \nonumber\\ & + \epsilon_{\text{d}}\left ( n\right ) \tanh\left ( \beta\epsilon_{\text{d}}\right ) -\frac{1}{\beta}\log\left [ 2\cosh\left ( \beta\epsilon_{\text{d}}\right ) \right ] ^{n-1}-\frac{1}{2}j\left ( n\right ) \left ( n-1\right ) \tanh^{2}\left ( \beta\epsilon_{\text{d}}\right ) -+\mu b^{\prime}\epsilon_{\text{d}}\left ( n\right ) \tanh\left ( \beta\epsilon_{\text{d}}\right ) ~\text{,}\ ] ] where the factor @xmath110 comes from the number of nearest neighbors , the factor @xmath73 is from the total number of atoms and the @xmath198 is due to double counting . we now choose the form that minimizes @xmath188 with respect to @xmath229 and @xmath230 . when minimizing with respect to @xmath229 , we obtain @xmath257 that is , after some simple algebra,@xmath258 similarly , when minimizing with respect to @xmath230 , one gets @xmath259 vs. temperature @xmath69 . we set @xmath124 and assume @xmath125 , @xmath126 , @xmath220 , @xmath221 and the level curves @xmath260 with @xmath261.,scaledwidth=50.0% ] eqs . ( [ fica1 ] ) and ( [ fica2 ] ) are final results . a contour plot of @xmath262 * * defined as,@xmath263 as a function of @xmath230 and @xmath69 appears in fig . these equations can then be used as above to solve for the critical temperature , the total magnetic moment and the magnetic susceptibility for this ferromagnetic material as usual . the difference here is that we will have two of each . for example , a critical temperature that applies to most of the atoms , and one that is local to the defect . therefore , the total magnetic moment will not simply be a result of just one temperature but a combination of each magnetic moment determined by these equations . notice that eq . ( [ ea_g ] ) still holds in general . in the two energy case , we have for one energy the number of nearest neighbors , @xmath264 and for the second energy , @xmath265 . as a final consideration , we point out that the outcomes of our entropic analysis in eqs . ( [ fica1 ] ) and ( [ fica2 ] ) together with the numerical representation in fig . 5 are consistent with the fact that imperfections in the lattice structure can be the cause of structural irregularities in the critical temperature @xmath70 @xcite . however , to obtain a deeper understanding of these aspects of ferromagnetic materials from an entropic viewpoint , a more thorough analysis would be required . just as in the previous example , the * * _ defect _ could be introduced here as an observable by means of a constraint like relation,@xmath266 where * * @xmath267 is a kronecker delta function . this makes much more sense here as this information may be obtained by common spectroscopic methods such as electron paramagnetic resonance , photoluminescence or cathodoluminescence . in the particular case considered here , we observed @xmath268 . in this article , we showed that not only can the mre method be used for updating probability distributions with both data and expectation values , but for determining effective approximations of such probability distributions equivalent to those provided by the bogoliubov variational principle and/or mean field theory . specifically , after describing the traditional ising model of a ferromagnetic material and the mean field approximation for a multidimensional ferromagnet , we applied the mre method to infer both magnetization and energetic features of ferromagnets characterized by simple lattice configurations . our main results can be summarized as follows : 1 . the standard mean field theory solution was recovered via mre methods for the ferromagnetic ising model . these findings are reported in eq . ( [ implicit ] ) and fig . the main advantage of our analysis is the fact that we do not require the knowledge of the exact form of the effective energy @xmath180 and we allow for departures from the centrality of the atom . 2 . moving past traditional methods and assumptions by using mre , we examined ferromagnets characterized by three state atoms . the outcomes of our entropic analysis are reported in eq . ( [ cc ] ) and fig . 3 . based on these results , we concluded that our analysis appears to support the idea that the introduction of additional states in the atomic ferromagnetic structure is consistent with both space shifts and nonhomogeneities of the critical temperature @xmath70 @xcite . a ferromagnetic material was considered * * that had known defects , which is beyond the scope of the mean field methodology . these findings are presented in eqs . ( [ fica1 ] ) , ( [ fica2 ] ) and fig . 5 . in each case we uncovered results independent of the explicit form of the effective energy in the hamiltonian . relying on these results , we concluded that our research appears to be consistent with the fact that imperfections in the lattice structure can be the cause of structural irregularities in the critical temperature @xmath70 @xcite . although it may seem trivial to introduce the observational data through an mre constraint , as opposed to simply _ putting it in _ as would be conventionally done , it should be noted that these toy examples are mostly for pedagogical purposes in order to demonstrate how the mechanism works . the power of this methodology enters when one does not know the expectation value of the hamiltonian , but may know how it relates to some observable , as done in @xcite . for instance , a more interesting example might be a case in which the joint prior equals * * @xmath269 * * where * * @xmath270 * * is the bayesian likelihood that represents a model that relates some observable * * @xmath141 * * with * * @xmath116 * * , * * like * * @xmath271 * * with * * @xmath153 * * described in eq . ( [ 45 ] ) . care must be taken when examining this scenario and such considerations will be explored in later works . a function @xmath272 can fail to be differentiable at a point @xmath273 in three scenarios : it has a vertical tangent line at @xmath273 , it has a discontinuity at @xmath273 , it exhibits a kink ( or , corner ) at @xmath273 . interestingly , we point out that figs . @xmath274 , @xmath275 , and @xmath276 show a vertical tangent , a jump discontinuity , and a kink , respectively . specifically , the arbitrary function @xmath277 is represented by the mean magnetization of the ferromagnetic material in fig . @xmath274 , by the effective energy of an atom in fig . @xmath275 , and finally , by the effective energy of a defective atom in fig . @xmath276 . although numerical methods are often ineffective near points of discontinuity , our qualitative entropic analysis seems to suggest that departures from the standard ising model ( with a vertical tangent type of singularity ) can lead to alternative types of non - differentiable singularities . while the findings uncovered in this manuscript are not conclusive ( about the nature of phase transitions in the presence of imperfections ) by any means , this work seems to suggest that using the mre method can be useful in inferring both thermal and electromagnetic properties of defective ferromagnetic materials in presence of available data from the substance being considered of course , a more detailed investigation would be required to deepen our understanding of these important phenomena in fundamental and applied science . indeed , it is our intention to extend our study to higher order spin configurations @xcite ( for example , xy spin chains , xx spin chains , heisenberg models , or , more generally , potts models @xcite ) . finally , we point out that mre methods employed in this article can be readily applied , in principle , to the investigation of more general dislocation , disclination , and dispiration defects in solid state physics @xcite . in particular , it would be very interesting to infer physical properties of materials exhibiting a non - uniform distribution of defects of different types . we leave these investigations to future efforts . a. giffin would like to acknowledge valuable discussions with ariel caticha regarding some of the material presented in this manuscript . finally , constructive criticisms from an anonymous referee leading to an improved version of this manuscript are sincerely acknowledged by the authors . f. jean , g. collin , m. andrieux , n. blanchard , j .- f . marucco , _ _ oxygen nonstoichiometry , point defects and critical temperature in superconducting oxide bi__@xmath278__sr__@xmath278__cacu__@xmath278__o__@xmath279 , physica * c339 * , 269 ( 2000 ) .

the need for a relativistic methodology that is broadly applicable to nuclear physics has never been more apparent . one of our most important sets of problems involves understanding the transition between the ( baryon , meson ) and the ( quark , gluon ) degrees of freedom . using a relativistic formulation of the hadronic degrees of freedom is necessary to avoid a misinterpretation of a kinematic effect as a signal for the transition . the goal of understanding future high momentum studies of nuclear targets using exclusive , nearly exclusive or inclusive processes can only be met through using relativistic techniques . the light front approach of dirac@xcite in which the time variable is taken as @xmath0 and the spatial variables are @xmath1 @xcite is one of the promising approaches because the momentum canonically conjugate to @xmath2 , @xmath3 , is directly related to the observables . it is worthwhile to begin with a qualitative explanation of the utility of these light cone variables and this light front approach in a qualitative fashion . consider lepton - nucleus deep inelastic scattering as a first example . the observed structure function depends on the bjorken variable @xmath4 which in the parton model is the ratio of the quark plus momentum to that of the target . if one regards the nucleus as a collection of nucleons , @xmath5 , where @xmath6 is the plus momentum of a nucleon bound in the nucleus . thus , a more direct relationship between the necessary nuclear theory and experiment occurs by using a theory in which @xmath6 is one of the canonical variables . since @xmath6 is conjugate to a spatial variable @xmath7 , it is natural to quantize the dynamical variables at the equal light cone time variable of @xmath8 . to use such a formalism is to use light front quantization , since the other three spatial coordinates ( @xmath9 ) are on a plane perpendicular to a light like vector@xcite . this use of light front quantization requires a new derivation of the nuclear wave function , because previous work used the equal time formalism . are these light cone variables useful only in nuclear deep inelastic scattering ? let s answer this by examining the origin of such coordinates . the four momentum of the incident virtual photon , q , can be said to have the components @xmath10 , with @xmath11 , and @xmath12 very large but @xmath13 finite ( the bjorken limit ) . then @xmath14 the condition that the reaction be elastic scattering from the quarks is that @xmath15 or @xmath16 . thus @xmath5 results from having only one large momentum in the problem , which can be taken in the negative z - direction , so that minus component is enhanced . more generally , one expects to be able to use light cone coordinates @xmath17 whenever there is such a large momentum in the problem as in any high energy scattering process . diverse applications are shown in the text by cheng and wu@xcite . examples of most relevance include high energy projectile nuclear scattering and high momentum transfer quasi - elastic reactions involving nuclear targets . light front techniques have previously been applied to systems of two hadrons . the two main approaches have been the relativistic quantum mechanics of directly interacting particles @xcite and relativistic field theory @xcite . we choose here to employ specific lagrangians which embody chiral and other symmetries , and thus use field theory . the light front quantization procedure necessary to treat nucleon interactions with scalar and vector mesons was derived by soper@xcite , and by yan and collaborators@xcite . here we combine the previous formalisms to obtain a light front treatment of a lagrangian which contains pions , vector and scalar mesons , and which respects the constraints of chiral symmetry . here is an outline . the bulk of the formalism is presented in sect . , a chiral lagrangian which includes pions , scalar mesons and neutral vector mesons is presented . the field equations are derived , and the quantization procedure for the free and interacting fields are quantized at the zero of light cone time @xmath18 . the energy momentum tensor , the light front hamiltonian @xmath19 and plus momentum operator @xmath20 are derived . the necessary contact interactions involving the exchange of instantaneous fermions and vector bosons are obtained . the principal purpose of the present work is to develop a technique that could have wide application in nuclear physics . thus we study and check the present formalism by applying it to three different examples- @xmath21 and @xmath22 scattering , and a mean field treatment of infinite nuclear matter- of relevance to nuclear physics . dealing successfully with each of these subjects is a prerequisite for making progress . 3 shows how light front field theory leads to a chiral treatment of low energy pion - nucleon scattering , which is consistent with the results of soft pion theorems . then nucleon - nucleon scattering is handled in a manifestly covariant manner , within the one - boson approximation , in sect . 4 . a discussion of the impact of chiral symmetry on the two - nucleon intermediate state contribution to the two pion exchange potential is also included . the mean field approximation is applied to infinite nuclear matter in sect 5 . glazek and shakin@xcite used a lagrangian containing nucleons and scalar mesons to study infinite nuclear matter . here vector mesons are included and the rotational - invariance arguments used in sect . 4 are used to derive the glazek - shakin @xmath6 variable . the energy of nuclear matter is computed and shown to be the same as found in the equal time formalism . the unique feature of the present formalism is the ability to obtain the nuclear and mesonic plus - momentum distributions from the energy momentum tensor . we find that that mesons can carry a significant fraction of the nuclear plus momentum , but have support only at 0 plus - momentum . some the results for nuclear matter have been presented in an earlier publication@xcite ; here the calculation is performed in two different ways and explained in more detail . sect . 6 summarizes the new results , presents a critique and discusses possible future applications . appendix a contains a summary of notation and some useful equations . we use a non - linear chiral model in which the nuclear constituents are nucleons @xmath23 ( or @xmath24 , pions @xmath25 scalar mesons @xmath26@xcite and vector mesons @xmath27 . the lagrangian @xmath28 is given by @xmath29 where the bare masses of the nucleon , scalar and vector mesons are given by @xmath30 @xmath31 , and @xmath32 . the unitary matrix @xmath33 can be chosen from amongst three forms @xmath34 : @xmath35 which correspond to different definitions of the fields . the pion - nucleon coupling here is chosen as that of linear representations of chiral symmetry used by gursey @xcite , with the the lagrangian approximately @xmath36 ) invariant under the chiral transformation @xmath37 one may transform the fermion fields , by taking @xmath38 as the nucleon field . one then gets lagrangians of the non - linear representation , as explained by weinberg @xcite . in this case the early soft pion theorems are manifest in the lagrangian , and the linear pion - fermion coupling is of the pseudovector type . however , the use of light front theory , requires that one find an easy way to solve the constraint equation that governs the fermion field . we shall show that the constraint can be handled in a simple fashion by using the linear representation . moreover , we shall see that the early soft pion theorems are indeed manifest from the form of the light front hamiltonian . the constant @xmath39 plays the role of the bare pion - nucleon coupling constant . if @xmath40 is chosen to be the pion decay constant , the goldberger - trieman relation yields the result that the axial vector coupling constant @xmath41 , which would be a problem for the lagrangian , unless loop effects can make up the needed 25% effect . corrections of that size are typical of order @xmath42 effects found in the cloudy bag model@xcite for many observables , including @xmath43 . there are no explicit @xmath44 in the above lagrangian . those will be handled in a future publication . for the moment we note that treating the higher order effects of the pion - nucleon inherent in this lagrangian is likely to lead to a resonance in the ( 3,3 ) channel of pion nucleon scattering . such effects can be included in the two - pion exchange contribution to nucleon - nucleon scattering . however , such an approach seems cumbersome . the choice of using an explicit @xmath45 instead of the iterated @xmath46 interaction is analogous to our use of a scalar meson even though the effects of @xmath47 interactions , which could lead to similar effects , are included in the lagrangian . we follow many authors ( see the review @xcite ) and include a scalar meson to simplify calculations . in this treatment , which follows that of refs.@xcite , the scalar meson @xmath26 is not a chiral partner of the pion- the chiral transformation is that of eq . ( [ chiral ] ) . the present lagrangian may be thought of as a low energy effective theory for nuclei under normal conditions . a more sophisticated lagrangian is reviewed in @xcite and used in @xcite ; the present one is used to show that light front techniques can be applied to hadronic theories relevant for nuclear physics . this hadronic model , when evaluated in mean field approximation , gives@xcite at least a qualitatively good description of many ( but not all ) nuclear properties and reactions . there are a variety of problems occuring when higher order terms are included@xcite . the aim here is to use a reasonably sophisticated lagrangian to study the effects that one might obtain by using a light front formulation . we could also have used the linear sigma model . the light front quantization for that model can be accomplished using a simple generalization of the work of refs . @xcite and @xcite , and is not shown here . according to the review @xcite the use of such a lagrangean precludes a successful description of nuclei at the mean - field level . the next step is to examine the field equations . the relevant dirac equation for the nucleons is @xmath48 the field equations for the mesons are @xmath49 @xmath50 @xmath51 = -m\bar\psi ' { \partial u\over \partial \pi_i}\psi ' , \label{pimeson}\end{aligned}\ ] ] where @xmath52 . the next step is obtain the light front hamiltonian ( @xmath19 ) as a sum of a free , non - interacting and a set of terms containing all of the interactions . this is accomplished by using the energy momentum tensor as @xmath53 the usual relations determine @xmath54 , with @xmath55 in which the degrees of freedom are labelled by @xmath56 . it is worthwhile to consider the limit in which the interactions between the fields are removed . this will allow us to define the free hamiltonian @xmath57 and to display the necessary commutation relations . the energy momentum tensors of the non - interacting fields are defined as as @xmath58 and @xmath59 the fermion fields are quantized in the next sub - section . then the use of eq.([tmunu ] ) leads to the result @xmath60,\ ] ] with @xmath61 the scalar field can be expressed in terms of creation and destruction operators : @xmath62,%\bbox{k}\cdot\bbox{x}}\right],\ ] ] where @xmath63 with @xmath64 , and @xmath65 . the @xmath66 function restricts @xmath6 to positive values . the commutation relations are @xmath67= \delta(\bbox{k}_\perp-\bbox{k}'_\perp ) \delta(k^+-k'^+)\label{comm}\ ] ] with @xmath68=0 $ ] . it is useful to define @xmath69 which will be used throughout this paper . the derivatives appearing in the quantity @xmath70 are evaluated and then one sets @xmath18 to 0 to obtain the result @xmath71 which has the interpretation of an operator the counts the light front energy @xmath64 of all of the particles . the pion field is treated in a similar manner , with the result @xmath72,\ ] ] and @xmath73 with commutation relations analogous to that of eq.([comm ] ) . the energy momentum tensor for the free vector meson field is obtained directly from the defining relation ( [ tmunu ] ) as @xmath74.\ ] ] it is desirable to obtain the symmetric energy momentum tensor . this is done by using @xmath75 , subtracting a total divergence and using the free field equations . the result is @xmath76.\ ] ] the component relevant for the light front hamiltonian can be shown to be @xmath77 the expression for the vector meson field operator is @xmath78,\label{vfield}\ ] ] where the polarization vectors are the usual ones : @xmath79 once again the four momenta are on - shell with @xmath80 the light front commutation relations : @xmath81=\delta_{\omega\omega ' } \delta^{(2,+)}(\bbox{k}-\bbox{k}'),\ ] ] with the others vanishing , lead to commutation relations amongst the field operators that are the same as in ref . the expression for @xmath82 can now be obtained from eqs.([t+- ] ) and ( [ vfield ] ) as @xmath83 this subject is complicated by the presence of massive vector meson fields . various difficulties were handled by soper@xcite and yan@xcite . the key features that we use are summarized here . in particular , the fields @xmath84 are chosen as the three independent fields , with the others expressible in terms of these . we shall need only one of these relationships in which the plus component of eq.([vmeson ] ) can be used to obtain @xmath85.\label{vpm}\ ] ] with @xmath86 and the inverse of @xmath87 is defined in refs . @xcite-@xcite . a more recent discussion is given by harindrinath and zhang@xcite , and the essentials are presented here in appendix a , we turn to the case of spin 1/2 fermions . although described by four - component spinors , these fields have only two independent degrees of freedom . the light front formalism allows a convenient separation of dependent and independent variables via the projection operators @xmath88@xcite , with @xmath89 . the independent fermion degree of freedom is chosen to be @xmath90 . the properties of the projection operators are discussed in appendix a. one gets two coupled equations for @xmath91 by multiplying eq.([dirac ] ) by @xmath92 and @xmath93 : @xmath94 the relation between @xmath95 and @xmath90 is very complicated unless one may set the plus component of the vector field to zero@xcite . this is a matter of a choice of gauge for qed and qcd , but the non - zero mass of the vector meson prevents such a choice here . instead , one simplifies the equation for @xmath95 by@xcite transforming the fermion field according to @xmath96 with @xmath97 this transformation leads to the result @xmath98 where @xmath99 note that all of the previously obtained fermionic sources of meson fields are unchanged by the transformation ( [ trans ] ) : @xmath100 the eigenmode expansion for @xmath101 is needed to compute the interaction between nucleons . ( [ vfield ] ) and ( [ vbar ] ) can be used to obtain @xmath102,\label{nvfield}\ ] ] where the polarization vectors @xmath103 are given by@xcite : @xmath104 with the properties @xmath105 the path towards the light front hamiltonian proceeds via the energy momentum tensor , which is given by @xmath106\psi'\nonumber\\ + \partial^\mu\phi\partial^\nu\phi + \partial^\mu\bbox{\pi}\cdot \partial^\nu \bbox{\pi } + \bbox{\pi}\cdot\partial^\nu\bbox{\pi } { \bbox{\pi}\cdot\partial^\mu\bbox{\pi}\over\pi^2}(1-{f^2\over \pi^2 } \mbox{sin}^2{\pi\over f}).\end{aligned}\ ] ] the use of the fermion field equation allows one to obtain the light front hamiltonian density @xmath107 it is now worthwhile to discuss a subtle feature regarding chiral symmetry in light front formalisms . chiral invariance is defined as invariance under the transformation defined by eq.([chiral ] ) if the equal time formalism is used . now the independent fermion variable is @xmath108 and @xmath109 is a functional of this . thus chiral invariance is the invariance under the transformation @xmath110 which is not the same as eq.([chiral ] ) because eq.([newchiral ] ) produces a change in @xmath109 that is different than using @xmath111 @xcite . the @xmath70 ( or equivalently the light front hamiltonian ) of eq.([tpm1 ] ) is invariant under the transformation ( [ newchiral ] ) if the pion mass is neglected so the usual chiral properties are obtained in these light front dynamics . the expression ( [ tpm1 ] ) is useful for situations , such as in the mean field approximation case for infinite nuclear matter examined below , for which a simple expression for @xmath108 is known . this is not always the case , so it is worthwhile to use the dirac equation to express @xmath70 in an alternate form : @xmath112 it is convenient to consider @xmath109 as a sum of terms , one @xmath113 whose relation with @xmath108 is free of interactions@xcite , the other @xmath114 containing the interactions . that is , rewrite the second of eq . ( [ yan ] ) as @xcite @xmath115 furthermore , define @xmath116 , so that @xmath117 where @xmath118 . the purpose of the above decomposition is to separate the dependent and independent parts of @xmath23 and to allow one to expand @xmath119 in terms of eigenstates of momentum . one may make a similar treatment for the vector meson fields . the operator @xmath120 , determined by eq ( [ vpm ] ) , is relevant for the hamiltonian . part of this operator is determined by a constraint equation . to see this examine eq ( [ vpm ] ) , and make a definition : @xmath121 where @xmath122 next use equations ( [ fcon ] ) and ( [ vcon ] ) to rewrite the hamiltonian as a sum of a free and interacting terms . the sum of the last term of eq . ( [ tpm2 ] ) and the terms involving @xmath123 is the density of the interaction hamiltonian , @xmath124 , plus the free fermion term , @xmath125 . use equations ( [ fcon ] ) and ( [ vcon ] ) in the expression ( [ tpm2 ] ) for @xmath70 along with the field equations and integration by parts to find : @xmath126 and @xmath127 with @xmath128 @xmath129 and @xmath130 the term @xmath131 accounts the emission or absorption of a single vector or scalar meson , as well as the emission or absorption of any number of pions through the operator @xmath132 . the term @xmath133 includes contact terms in which there is propagation of an instantaneous fermion . the term @xmath134 accounts for the propagation of an instantaneous vector meson . we may now quantize the fermion fields using @xmath135 , \label{dq}\ ] ] where again the momenta are on shell : @xmath136 , and the anti - commutation relations are given by @xmath137 the properties of the dirac spinors are described in appendix a. the term @xmath125 of eq . ( [ freef ] ) can now be expressed as @xmath138 the component that is related to the plus momentum is @xmath139 . the necessary expression is given by @xmath140 we begin by showing that , if one starts with a non - linear representation of chiral symmetry , the requirement of solving the constraint equation for the @xmath141 component of the fermion field leads one to a lagrangian of the gursey - type linear representation . the focus is on chiral properties and pion - nucleon scattering , so we dispense with the vector and non - chiral @xmath26 meson fields for this section , and it is sufficient to examine only the following fermion - pion term of a non - linear representation@xcite : @xmath142n.\ ] ] next obtain the fermion field equation and make the usual decomposition : @xmath143 with @xmath144n_-\label{nn } \nonumber\\ \left(i\partial^+-o^+\right)n_-= [ \bbox{\alpha}_\perp\cdot\bbox{(p}_\perp-\bbox{o}_\perp)+\beta m]n_+,\end{aligned}\ ] ] where the operator @xmath145 has been defined as @xmath146 we wish to remove the @xmath147 term from the left hand side of the equation for @xmath148 . this can be done by defining a unitary operator @xmath149 and fermion field @xmath150 such that @xmath151 with @xmath152 the identity@xcite @xmath153 where @xmath154 is given in eq . ( [ us ] ) , helps a good deal . its use in eq . ( [ feq ] ) , combined with the condition @xmath155 , leads to the result @xmath156 so that using eqs . ( [ ff ] ) and ( [ defchi ] ) in ( [ nn ] ) yields @xmath157\chi_- \nonumber\\ i\partial^+\chi_-=\left [ \bbox{\alpha}_\perp\cdot\bbox { p}_\perp+\beta mu_2\right]\chi_+.\end{aligned}\ ] ] this is of the desired form in which no interactions appear on the left - hand - side of the equation for @xmath158 . thus the use of light front quantization mandates that the pion - nucleon interactions be of the form of eq . ( [ lag ] ) . the first test for any chiral formalism is to reproduce the early soft pion theorems@xcite . here we concentrate on low energy pion nucleon scattering because of its relation to the nucleon - nucleon force . we work to second order in @xmath159 in this first application . in this case , each of the @xmath34 takes the same form : @xmath160 this expression is to be used in the potentials @xmath131 and @xmath133 of eqs . ( [ v1 ] ) and ( [ v2 ] ) . the second order scattering graphs are of three types and are shown as time @xmath18 ordered perturbation theory diagrams in fig . 1 . the kinematics are such that @xmath161 , with @xmath162 and @xmath163 . the iteration of @xmath131 to second order yields the direct and crossed graphs of fig . 1a . in this formalism @xmath131 is proportional to the matrix element of @xmath164 between @xmath165 spinors , so it is proportional to the momentum of the absorbed or emitted pion . thus the terms of fig . 1a vanish near threshold . the terms of fig . 1b are generated by the @xmath166 terms of @xmath131 . using the various field expansions in the expression ( [ v1 ] ) for @xmath131 leads to the result that plus - momentum is conserved and the plus momentum of every particle is greater than zero . this means that the first of fig.1b vanishes identically and the second vanishes for values of the initial pion plus momentum that are less than twice the nucleon mass . the net result is that only the instantaneous term of @xmath133 and the @xmath167 term of @xmath131 ( shown in fig . 1c ) remain to be evaluated . proceeding more formally , we evaluate the s - matrix given by @xmath168 where @xmath169 is the @xmath18 ( light - front time ) ordering operator and @xmath170 is the interaction representation light front hamiltonian . then @xmath171 with @xmath172 the evaluation proceeds by using the field expansions in the expressions for @xmath131 and @xmath133 . integrating over @xmath173 and evaluating the result between the relevant initial and final pion - nucleon states leads to the result that each contribution to the s - matrix is proportional to a common factor , @xmath174 which combines with the result of the required integration over the light cone time ( @xmath18 ) to provide the necessary momentum conservation and flux factors . the remaining factor of each term is its contribution to the invariant amplitude @xmath175 . the result is @xmath176 where the three terms here correspond to the three terms of fig . the role of cancellations in the reduction of the term proportional to @xmath177 is already apparent . to understand the threshold physics take @xmath178 and @xmath179 . then one finds @xmath180 to leading order in @xmath181 . the weak nature of the @xmath177 term and the presence of the second weinberg - tomazowa term is the hallmark of chiral symmetry@xcite . the same results could be obtained using the linear sigma model , with @xmath182 exchange playing the role of the @xmath167 term of eq.([um1 ] ) . the ultimate aim is to derive the nuclear wave function including correlation effects . the first step is to understand nucleon - nucleon scattering using our light front formalism . we start with the one boson exchange approximation , discuss the light front wave equation and show that this procedure gives the same scattering amplitude as the usual procedure of computing the one - boson exchange contributions to the invariant amplitudes and using the blankenbeckler - sugar reduction of the bethe - salpeter equation@xcite . this usual procedure is covariant , so that our construction shows that the light front wave procedure respects rotational invariance . this invariance is the result of strikman and frankfurt@xcite and others . the present treatment explicitly includes the effects of nucleon spin and the nucleon - nucleon interaction can derived from an underlying chiral lagrangian . the starting point is the s matrix of eqs . ( [ smat1 ] ) and ( [ smat2 ] ) . here the initial state @xmath183 consists of nucleons with quantum numbers labeled by 1 and 2 , and the state @xmath40 consists of nucleons 3 and 4 . to be definite , we take the plus momentum of nucleon 1 to be greater than that of nucleon 3 , and the momentum transfer @xmath184 to be @xmath185 so that @xmath186 . the lowest order contributions to the invariant amplitude are represented by the light - front - time ordered graphs shown in fig . the graphs of fig . 2a represent terms of the form @xmath187 , and that of fig . 2b accounts for the instantaneous massive vector boson exchange term of @xmath134 . these terms may be evaluated by using the field expansions and doing the relevant integrals over the @xmath188 coordinate space . each term has a common factor of @xmath189 where the factor @xmath190 in the numerator is compensated by dividing the invariant amplitude by @xmath190 . it is simplest to consider the effects of scalar @xmath26 and pseudoscalar @xmath191 exchanges at the same time . the scattering amplitudes @xmath192 take the form @xmath193 where the notation is that @xmath194 is the spinor for a nucleon of quantum numbers @xmath183 , and @xmath195 is either of the form @xmath196 or @xmath197 . the momentum of the exchanged meson is @xmath198 , and it is necessary to realize that @xmath199 but @xmath200 where @xmath201 is the mass of the exchanged scalar meson or pion . the factor @xmath202 arises from the denominators of the field expansions and @xmath203 is the result of evaluating the light front energy denominator @xmath204 . define the energy denominator of eq.([gen ] ) to be @xmath205 so that @xmath206 using eqs . ( [ off ] ) and ( [ off1 ] ) immediately yields @xmath207 so the amplitude @xmath208 takes the form @xmath209 this is the usual @xcite expression for a one - boson exchange potential , if no form factor effects are included . note that the klein - gordon propagator is obtained using only a single time - ordered graph . the calculation with the equal - time formulation requires the summation of two time - ordered graphs . the derivation of the contribution of vector meson exchange proceeds by adding the terms of fig 2a and 2b . the term of fig . 2a can immediately seen to be @xmath210.\ ] ] the factor in brackets arises from the polarization sum , recall eq . ( [ ebar ] ) . it is worthwhile to define the contribution of the second two terms in the bracket , which result from the difference between @xmath101 and @xmath27 fields , as @xmath211 , with @xmath212 next use the relations @xmath213 and the equality of the @xmath214 and @xmath215 components of @xmath198 with those of @xmath184 to obtain the results @xmath216 but @xmath217 , which leads to a compact rewriting of eq . ( [ bar ] ) as @xmath218 the term of fig . 2b is obtained by using the field expansion in the equation for @xmath134 , ( [ v3 ] ) , integrating over coordinate space and removing the common factor . the result is @xmath219 which exactly cancels the term @xmath211 . the net result is that the amplitude for vector meson exchange , @xmath220 , takes the familiar form : @xmath221 the sum of the amplitudes arising from each of the individual one boson exchange terms : @xmath222 gives the invariant amplitude to second order in each of the coupling constants . these amplitudes are strong , so computing the nucleon - nucleon scattering amplitude and phase shifts requires including higher order terms . one may include a sum which gives unitarity by including all iterations of the scattering operator @xmath223 through intermediate two - nucleon states : @xmath224 where @xmath225 is the negative - momentum in the initial state and @xmath226 projects on to two - nucleon intermediate states . more explicitly , eq . ( [ iter ] ) is given by @xmath227 after removing the common factor and accounting for the momentum conserving delta functions . one realizes that this is of the form of the weinberg equation@xcite by expressing the plus - momentum variable in terms of a light - front momentum fraction @xmath228 such that @xmath229 and using the relative and total momentum variables : @xmath230 then @xmath231 where @xmath232 is square of the total initial four - momentum , otherwise known as the invariant energy @xmath233 and @xmath234 is the corresponding quantity for the intermediate state . because the kernal @xmath223 is itself an invariant amplitude the procedure of solving this equation to determine observables is manifestly covariant . equation ( [ 419 ] ) can in turn can be re - expressed as the blankenbecler - sugar ( bbs ) equation @xcite by using the variable transformation@xcite : @xmath235 with @xmath236 . the result is : @xmath237 which is the desired equation . the three - dimensional propagator is exactly that of the bbs equation ; there is one difference . our one boson exchange potentials depend on the square of the four momentum @xmath238 transferred when a meson is absorbed or emitted by a nucleon . thus the energy difference between the initial and final on - shell nucleons is included and @xmath239 . the derivation of the bbs equation from the bethe - salpeter equation specifies that @xmath240 is used in the meson propagator . including @xmath241 instead of @xmath240 increases the range of the potential . such an effect can be hidden in phenomenological potentials by changing the pion - nucleon coupling constant or form factor . one can easily convert eq . ( [ bsbs ] ) into the lippman - schwinger equation of non - relativistic scattering theory by removing the factor @xmath242 with a simple transformation@xcite . the present results are that one can use the light front technique to derive nucleon - nucleon potentials in the one - boson exchange obe approximation and use these in an appropriate wave equation . therefore our procedure is directly comparable to the one used in constructing the realistic bonn one - boson exchange potentials used in momentum space . those potentials also have a close connection with an underlying lagrangian . our purpose here is to argue that the present procedure can yield potentials essentially identical to the bonn obep potentials and therefore would lead to a good description of the nn data . the bonn one - boson exchange potentials employ six different mesons @xmath243 and the ( isovector scalar ) @xmath244 meson . the present techniques can be used to handle all of these mesons and their couplings , with the possible exception of the tensor @xmath245 part of the @xmath246-nucleon interaction . the presence of such a tensor interaction makes it difficult ( or impossible ) to write the equation for @xmath109 as @xmath247 this is relevant because the standard value of the ratio of the tensor to vector @xmath246-nucleon coupling @xmath248 is 6.1 , based upon ref . @xcite and subsequent papers . reproducing the observed values of @xmath249 and p - wave wave phase shifts requires a large value @xmath248 ; see ref . however the lagrangian compensates for its lack of a @xmath246-n interaction with tensor coupling by generating such a term via vertex correction diagrams ( which are the origin of the anomalous magnetic moment of the electron in qed ) . such diagrams probably do not generate the phenomenologically required values of the coupling constants , but all that is needed here is that terms of the correct form be produced . this is because the standard procedure is to choose the values of the coupling constants so as to yield a good description of the nn scattering data . indeed the potentials a , b , and c are defined by the parameters which account for the mesonic masses , coupling constants and form factors . thus we end up with the same procedure that is used in the bonn one boson exchange potentials . this brings us to the treatment of divergent terms in our procedure . the definition of any effective lagrangian requires the specification of such a procedure . for the present , it is sufficient to say that we introduce form factors , @xmath250 which reduce the strength of the @xmath228 meson - nucleon coupling for large values of @xmath251 . this is also the procedure of ref @xcite . the net result is that the one - boson exchange treatment of the nucleon - nucleon potential and the t - matrix resulting from its use in the bbs equation is essentially the same as the one - boson exchange procedure of ref . thus our light front treatment is guaranteed to be consistent with nucleon - nucleon nucleon scattering data measured in the standard energy range . such a similarity has also been obtained using relativistic hamiltonian dynamics@xcite . the dominant contribution to the two pion exchange potential arises from contributions to intermediate states that include one or two @xmath45s@xcite , and a treatment of such effects based on chiral symmetry has been provided by van kolck and collaborators@xcite . including the effects of @xmath45 s is beyond the scope of the present work , but we are able to discuss the two pion exchange contribution ( of order @xmath252 ) to the nucleon nucleon potential . the property that a sum of light cone time - ordered diagrams is equal a single feynman graph can be used to simplify the calculation . the relevant feynman graphs are displayed in fig . 3 ; the terms originating from the linear @xmath253 coupling ( a , b ) , from the quadratic @xmath254 coupling ( c ) and from a combination of the linear and quadratic interactions ( d ) are indicated . the line through the two - nucleon intermediate state of fig . 3a is meant to indicate that the contribution arising from iterating the one pion exchange interaction is removed . this has been a standard procedure since the work of ref . @xcite , and will not be discussed further . the sum of the terms of fig . 3a and 3b is equal to the partovi - lomon two pion exchange potential , as they used the pseudoscalar pion - nucleon interaction . this interaction certainly simplifies the calculation ; in particular the diagrams of fig . 3a , b and d are convergent ( whereas they would be strongly divergent if pseudovector coupling were to be used . one can use such a pseudoscalar coupling , and include the effects of chiral symmetry , provided one also includes the effects of the @xmath254 coupling shown in fig . 3c , and the combined effects of the linear and quadratic interactions , fig . the quadratic interaction term cancels the large pair terms in pion - nucleon scattering and should also play a significant role here in reducing the size of the computed potential . thus we expect that the partovi - lomon potential contains too large an attraction . next turn to the procedure used in constructing the full bonn potential . this potential is constructed by ignoring all of the z - graphs and including the the effects of the two - nucleon intermediate states which arise from the crossed graph , fig . 3b , as well as the parts of fig . 3a arising from time ordered terms in which two pions exist at the same time . ( for such contributions to the tpep the linear pseudoscalar and pseudovector interactions are are evaluated between on shell positive energy nucleon spinors , and are therefore equivalent . the resulting contribution to the tpep is small , but is comparable to that of the iterated opep . the neglect of the z graphs goes a long way towards including the effects of chiral symmetry . however , terms involving the weinberg - tomazowa interaction at one or two vertices are ignored . the computation of the graphs of fig . 3 would include such effects implicitly as well as that of pair suppression . thus a detailed comparison would be useful . however , the small nature of the effects that we discuss now indicate that the dominance of the tpep by effects of intermediate @xmath45 s will remain unchallenged . the nucleon - nucleon interaction of the previous section can be used as the basis for a light front brueckner theory of nuclei . we study the mean field approximation for infinite nuclear matter as a first step . the nuclear mean field model- the shell model- occupies pre - eminence in understanding nuclear structure . we need to see if our formalism can describe this physics . in the mean field approximation@xcite , the coupling constants are considered strong and the fermion density large . then the meson fields can be approximated as classical- the sources of the meson fields are replaced by their expectation values . in this case , the nucleon mode functions will be plane waves and the nuclear matter ground state can be assumed to be a normal fermi gas , of fermi momentum @xmath255 , and of large volume @xmath256 in its rest frame . we consider the case that there is an equal number of protons and neutrons . first we examine the mesonic field equations ( [ vmeson])-([pimeson ] ) . the baryon source of the pion field is a pseudoscalar operator , so its expectation value vanishes in the ground state . thus this mean field approximation leads to the result that @xmath257 . the other meson fields are constants , independent of space and time , given by @xmath258 where the brackets denote expectation values of the nuclear ground state in its rest frame and the baryon density is @xmath259 this result that @xmath27 is a constant , along with eqs . ( [ vbar ] ) and ( [ mfa ] ) , can be used to determine @xmath101 . in particular , @xmath260 by construction . furthermore , the conditions that @xmath261 and @xmath262 tell us that @xmath263 . finally @xmath264 , so that @xmath265 , so the net result is that the only non - vanishing component of @xmath101 is @xmath266 . with this mean field approximation , the fermionic field equations ( [ yan ] ) can be rewritten as @xmath267 now @xmath26 and @xmath268 are constants so we expect the mode functions for the field expansion of @xmath23 to be of the plane wave form @xmath269 and can be obtained from eq . ( [ nyan ] ) as@xcite @xmath270 the light front eigenenergy @xmath271 is the sum of a kinetic energy term in which the mass is shifted by the presence of the scalar field , and an energy arising from the vector field . comparing this equation with the one for free nucleons , @xmath272 , shows that the nucleons have a mass @xmath273 and move in plane wave states . the nucleon field operator is constructed using the solutions of eq . ( [ sol ] ) as the plane wave basis states . this means that the nuclear matter ground state , defined by operators that create and destroy baryons in eigenstates of eq . ( [ sol ] ) , is the correct wave function and that equations ( [ mfa ] ) and ( [ sol ] ) represent the solution of the approximate field equations , and the diagonalization of the hamiltonian . one question remains . we are going to fill up a fermi sea , but @xmath255 is the magnitude of a three vector . how is this three vector defined ? this was answered in the paper of glazek and shakin @xcite who showed that rotational invariance is manifest if one uses the definition : @xmath274 which implicitly defines @xmath275 . using eq . ( [ kplus ] ) allows one to maintain the equivalence between energies computed in the light front and equal time formulations of scalar field theories@xcite . a similar equation has been used to restore manifest rotational invariance in light - front qed@xcite . we shall show that this same expression also restores rotational invariance in this mean field problem when vector mesons are included . equation ( [ kplus ] ) has the correct form in the limit of non - interacting nucleons and therefore seems natural@xcite . we attempt a heuristic derivation of this equation using the requirement that manifest rotational invariance be restored . the starting point is the observation that eq . ( [ alpha ] ) , with its definition of @xmath228 as the plus momentum fraction carried by a nucleon , restores manifest rotational invariance in the two - nucleon system . let s consider the mean field approximation as involving an interaction between a nucleon and a very heavy particle containing a-1 nucleons ( with @xmath276 ) . then the variable @xmath277 , which is the fraction of the nuclear plus - momentum carried by a nucleon , is given by @xmath278 and is a suitable generalization of the variable @xmath228 . the nucleon mass is taken to be @xmath273 , because it is this mass that appears in the nucleon field equations . the mass of the @xmath279 body system is dominated by the mass of the ( a-1 ) nucleons mass ( the binding energy per particle is 16 mev @xmath280 compared with 940 mev ) , so that we may write @xmath281 in which the last line results from the limit @xmath282 . the key feature is that the variable @xmath277 is defined as a momentum fraction , so that @xmath283 comparing eq . ( [ approx ] ) and eq . ( [ begin ] ) leads to eq . ( [ kplus ] ) . the computation of the energy and plus momentum distribution proceeds from taking the appropriate expectation values of the energy momentum tensor @xmath284 discussed in sect . 2 and @xmath285 we are concerned with the light front energy @xmath19 and momentum @xmath20 . the relevant components of @xmath284 are presented in eqs . ( [ tpm1 ] ) and ( [ tpp ] ) . within the mean field approximation(mfa ) , the derivatives of the meson fields are zero so that one finds @xmath286 taking the nuclear matter expectation value of @xmath70 and @xmath139 and performing the spatial integral of eq . ( [ pmu ] ) leads to the result @xmath287 the subscript f denotes that @xmath288 with @xmath275 defined by the relation ( [ kplus ] ) . equations ( [ pminus ] ) and ( [ pplus ] ) along with the expression for @xmath6 , ( [ kplus ] ) allow an evaluation of @xmath19 and @xmath20 . this shall be done in two different ways . in the first method we evaluate the energy of the a - nucleon system @xmath289@xcite , which turns out to be the same as in the usual equal - time treatment@xcite . this can be seen by summing equations ( [ pminus ] ) and ( [ pplus ] ) to obtain @xmath290 then replace the integration over @xmath6 by one over @xmath275 , using eq . ( [ kplus ] ) so that @xmath291 where @xmath292 then eq . ( [ ei ] ) takes the form : @xmath293 which is the expression familiar from the walecka model , this confluence of energies is a nice check on the present result because a manifestly covariant solution of the present problem , with the usual energy , has been obtained@xcite . we consider the system to be at a fixed large volume , @xmath256 , so that @xmath294 depends on @xmath26 and @xmath255 . the ground state energy is determined by minimizing @xmath294 with respect to those two parameters . setting @xmath295@xcite to zero reproduces the field equation for @xmath26 ( [ smf ] ) as is also the case in the equal - time formalism . the next step is to minimize the energy per particle @xmath296 at fixed volume with respect to @xmath255 . ( one may also minimize the energy with respect to the volume@xcite . ) start this calculation by using @xmath297 to obtain : @xmath298 using eq . ( [ e ] ) followed by eqs . ( [ smf ] ) and ( [ mfa ] ) leads to the result @xmath299 where @xmath300 . this is a transcendental equation which determines @xmath255 , so that the calculation of @xmath301 is complete . it is useful to note that the relation @xmath302 ( which must hold for a system in its rest frame ) also emerges as a result of this minimization . to see this rewrite the left hand side of eq . ( [ mid ] ) as @xmath303 using this in eq . ( [ mid ] ) leads to @xmath304 which is just the relation that one obtains by setting @xmath302 with the versions of eqs . ( [ pminus ] ) and ( [ pplus ] ) obtained by replacing the variable @xmath6 by @xmath275 . another way to obtain the energy of the ground state is to minimize the value of @xmath305 subject to the constraint that @xmath306 , or to minimize the quantity @xmath307 with @xmath308 where @xmath309 is a lagrange multiplier . setting @xmath310 leads to @xcite @xmath311 but the field equation ( [ smf ] ) for @xmath26 can be restated as @xmath312 combining eqs.([mini ] ) and ( [ rew ] ) leads to the result that @xmath313 so that the minimization of @xmath314 with respect to @xmath255 is the same as minimizing @xmath294 with respect to @xmath255 . this ends the discussion of how the expressions for @xmath315 are used to determine the energy of the system . solving the field equations and minimizing the energy density determines the properties of nuclear matter , once the meson - nucleon coupling constants and masses are chosen . one can now discuss the properties of the resulting system . of course the necessary calculations have been done long ago . in particular , the parameters @xmath316 and @xmath317 have been chosen @xcite so as to give the binding energy per particle of nuclear matter as 15.75 mev with @xmath255=1.42 fm@xmath318 . in this case , solving the equation for @xmath26 gives @xmath319 . the light front formalism embodies the use of @xmath6 as a canonical variable which allows us to study the nucleonic and mesonic contributions to the nuclear plus momentum . the study of the plus momentum content is motivated by the desire to obtain a better understanding of lepton - nucleus deep inelastic scattering . the emc effect@xcite that the structure function of a bound nucleon differs from that of a free one , showed a principal effect that the plus momentum carried by the valence quarks is less for a bound nucleon than for a free one . many different interesting interpretations and related experiments@xcite were stimulated by these experiments . but a correct interpretation requires that the role of conventional effects , such as nuclear binding , be assessed and understood . our formalism employs plus component of the momentum so that it s use in assessing the nucleon s ( and therefore the valence quark s ) plus - momentum is necessary . we therefore examine the @xmath320 component of the energy momentum tensor eq . ( [ pplus ] ) to determine how much momentum is carried by nucleons and how much by mesons . rewrite eq . ( [ pplus ] ) as a sum of mesonic @xmath321 and nucleonic @xmath322 terms @xmath323 with @xmath324 and @xmath325 the parameters of eqs . ( [ cv ] ) and ( [ cc ] ) leads to @xmath326 and the use of eq . ( [ cv ] ) in eq . ( [ pm ] ) gives : @xmath327 while performing the integral involved in eq . ( [ nuc ] ) leads to @xmath328 the result is that 65% of the nuclear plus momentum must be carried by the nucleons and the remainder of 35% is carried by the mesons . how do these numbers relate to experiments ? to answer we need to recall that the nuclear structure function @xmath329 can be obtained from the light front distribution function @xmath330 ( which gives the probability for a nucleon to have a plus momentum fraction @xmath331 ) and the nucleon structure function @xmath332 using the relation@xcite : @xmath333 where @xmath331 is the @xmath334 times the fraction of the nuclear plus - momentum carried by the nucleon , and @xmath335 is the bjorken variable computed using the nuclear mass divided by @xmath334 ( @xmath336 ) : @xmath337 . this formula is the expression of the convolution model in which one means to assess , via @xmath330 , only the influence of nuclear binding . other effects such as the nuclear modification of the nucleon structure function ( if @xmath332 is obtained from deep inelastic scattering on the free nucleon ) and any influence of the final state interaction between the debris of the struck nucleon and the residual nucleus@xcite are neglected . our formalism enables us to calculate the function @xmath330 from the integrand of eq.([pplus ] ) . since the integral gives the total plus - momentum carried by nucleons , the integrand which multiplies the factor @xmath6 can be interpreted as the necessary probability distribution . thus : @xmath338 @xmath339 the function @xmath330 can be obtained by replacing @xmath6 by the dimensionless variable @xmath331 using @xmath340 with @xmath341 mev . then using eq.([fdef ] ) leads to the result @xmath342 , \label{fy}\ ] ] where @xmath343 and @xmath344 . knowing the following integrals is useful : @xmath345 with the 0.65 representing the earlier 65% result . we may now assess the implications of the statement that nucleons carry only 65% of the nuclear plus - momentum . this number is to be compared with the value obtained by frankfurt and strikman@xcite . they used data for @xmath329 and @xmath332 along with eq . ( [ deep ] ) to determine the average value of @xmath331 required by experiments with the result that @xmath346 this means that nucleons carrying 95% of the nuclear plus momentum ( a 5% depletion effect ) is sufficient to explain the 10 - 15% depletion effect observed for the fe nucleus . our 35% depletion seems to be rather large , but one must remember that it results from nuclear matter . a result that compares more closely with experiment could be obtained in a version of the present model for which the value of @xmath273 is closer to @xmath347 . however , determining specific features of the present model is not the goal of the present work . instead we wish to demonstrate that the light front formalism can be used to obtain a nuclear wave function expressed in terms of the plus - momentum variable which is closely related to experiment . indeed we can verify that the ability to obtain the nucleonic plus momentum , feature that requires the use of a light front front formalism , instead of an equal time formalism . to do this compare the 0.65 fraction with the result of a relativistic calculation using the equal time ( et ) formalism@xcite . in this calculation , which uses eq . ( [ lag ] ) ) and for which the scalar and vector fields are the same as here , the plus momentum of a nucleon was chosen as the sum of the dirac eigenenergy and @xmath275 : @xmath348 using this leads to an average nucleon plus momentum fraction @xmath349 , which when evaluated with our parameters for @xmath350 and @xmath268 , leads to @xmath351 ! the big difference between our result and the earlier equal time result -compare eqs . ( [ kplus ] ) and ( [ ident])- arises from our use of the plus momentum as a canonical momentum variable and the consequent use of @xmath54 to construct the light front momentum and energy density . in particular , the first line of eq.([ident ] ) is only a reasonable guess . we note also that the baryon number distribution @xmath352 ( number of baryons per @xmath331 , normalized to unity ) can be determined from the expectation value of @xmath353 . the result is @xmath354 -{1\over 2y^2}({k_f^4\over \bar { m}^4}-({e_f\over \bar{m}}-y)^4 ) \right].\label{fby}\ ] ] some phenomenological models treat the two distributions @xmath330 and @xmath352 as identical . the distributions have the same normalization , but they are different as shown by eqs . ( [ fy ] ) and ( [ fby ] ) . consider that the average value of @xmath331 equal to 0.65 represents a very strong binding effect on lepton - nucleus deep inelastic scattering . one might think that the mesons , which cause this binding , would also have huge effects on deep inelastic scattering . it is therefore certainly necessary to determine the momentum distributions of the mesons . the mesons contribute 0.35 of the total nuclear plus momentum , but we need to know how this is distributed over different individual values . the paramount feature is that @xmath26 and @xmath27 are the same constants for any and all values of the spatial coordinates @xmath355 ( and also @xmath18 ) . this means that the related momentum distribution can only be proportional to a delta function setting both the plus and @xmath215 components of the momentum to zero . this result is attributed to the mean field approximation for infinite nuclear matter , in which the meson fields are treated as classical quantitates . thus the finite plus momentum can be thought of as coming from an infinite number of quanta , each carrying an infinitesimal amount of plus momentum . a plus momentum of 0 can only be accessed experimentally at @xmath356 , which requires an infinite amount of energy . thus , in the mean field approximation , the scalar and vector mesons can not contribute to deep inelastic scattering . the usual term for a field that is constant over space is a zero mode , and the present lagrangian provides a simple example . for finite nuclei , in the mean field approximation , the mesons would carry a very small momentum of scale given by the inverse of the nuclear radius , under the mean field approximation . if fluctuations were to be included , the relevant momentum scale would be of the order of the inverse of the average distance between nucleons ( about 2 fm ) . we can understand the significance of the presence of components of a wave function that carry plus momentum but do not participate deep inelastic processes by reviewing a bit of history . the nuclear binding effect is that the plus momentum of a bound nucleon is reduced by the binding energy , and so is that of its confined quarks . conservation of momentum implies that if nucleons lose momentum , other constituents such as nuclear pions@xcite , must gain momentum . this partitioning of the total plus momentum amongst the various constituents is called the momentum sum rule . pions are quark anti - quark pairs so that a specific enhancement of the nuclear antiquark momentum distribution , mandated by momentum conservation , is a testable @xcite consequence of this idea . a nuclear drell yan experiment @xcite , in which a quark from a beam proton annihilates with a nuclear antiquark to form a @xmath357 pair , was performed . no influence of nuclear pion enhancement was seen , leading bertsch et al.@xcite to question the idea that the pion is a dominant carrier of the nuclear force . in the present situation , we have a huge depletion effect of 35% , but with no consequence for either the nuclear deep inelastic scattering or drell - yan experiments . we hasten to add that the lagrangian of eq . ( [ lag ] ) and its evaluation in mean field approximation for nuclear matter have been used to provide a simple but semi - realistic example . it would be premature to compare the present results with data before obtaining light front dynamics for a model in which the correlational corrections to the mean field approximation are included , and which treats finite nuclei . thus the specific numerical results of the present work are far less relevant than the central feature that the mesons responsible for nuclear binding need not be accessible in deep inelastic scattering . however , the present model may be regarded as being one of a class of models in which the mean field plays an important role@xcite . for such models nuclei would have constituents that contribute to the momentum sum rule but do not contribute to deep inelastic scattering . in particular , a model can have a large binding effect , nucleons can carry a significantly less fraction of @xmath20 than unity , and it is not necessary to include the influence of mesons that could be ruled out in a drell - yan experiment . the present paper shows how the light front quantization of a chiral lagrangian can be accomplished . the resulting formalism can be applied to many problems of interest to nuclear physics . in particular , pion - nucleon , nucleon - nucleon scattering , and infinite nuclear matter ( in the mean field approximation ) are presented here . soft pion theorems for pion - nucleon scattering are reproduced . the treatment of nucleon - nucleon scattering is shown to be manifestly covariant in the one boson exchange approximation . the implications of chiral symmetry for the two - nucleon intermediate state contribution to the two pion exchange potential are discussed . the present results mainly constitute a feasibility study , in that the emphasis here is on checking the formalism by reproducing known results . but this light front treatment does allow the effects of chiral symmetry to be incorporated within a relativistic formalism , and therefore should have a broad applicability in the future . one remaining technical problem is to provide a light front quantization of a lagrangian for spin 3/2 particles . the special feature of the light front formalism is its use of the plus momentum as one of the canonical variables . this enables a close contact with the experimental variables used to analyse deep inelastic scattering and any experiment in which there is one large momentum . this feature is exploited here in the derivation ( within the mean field approximation ) of the nucleonic and mesonic distribution functions for infinite nuclear matter . the mesons are shown to carry a significant fraction of the nuclear plus momentum , but only a zero plus - momentum ( a zero mode ) , and therefore do not participate in nuclear deep inelastic scattering or drell - yan experiments . the ultimate validity of the above ( perhaps startling ) statement depends on whether or not the dominance of mesonic zero modes survives calculations performed for finite nuclei and calculations which include the effects of nucleon - nucleon correlations . there seems to be no technical barrier precluding such calculations . this work is partially supported by the usdoe . i thank the slac theory group and the national int for their hospitality . i thank p. blunden , s.j . brodsky , l. frankfurt , s. glazek , r. machleidt , r.j . perry , c.m . shakin , m. strikman and u. van kolck for useful discussions . this is patterned after the review of harindrinath@xcite the light - front variables are defined by @xmath358 so the four - vector @xmath359 is denoted @xmath360 with this notation the scalar product is denoted by @xmath361 the metric tensor @xmath362 with @xmath363 is obtained from the usual one by using ( [ def ] ) ( i.e. @xmath364 ) . then @xmath365 , with the other elements vanishing . the term @xmath366 is obtained from the condition that @xmath367 . its elements are the same as those of @xmath368 except for @xmath369 . thus @xmath370 and the partial derivatives are similarly given by @xmath371 the step function is defined as @xmath372 for @xmath373 , and @xmath374 for @xmath375 . the antisymmetric step function is given by @xmath376 with @xmath377 in this notation , @xmath378 . the above few definitions allow us to express the inverse operators appearing in the text in terms of integrals : @xmath379 @xmath380 the hermitian projection operators @xmath383 are given by @xmath384 and obey the following relations @xmath385 @xmath386 @xmath387 @xmath388 and @xmath389 the dirac spinors are given by @xmath390 \ ; \chi_{\lambda } , \end{aligned}\ ] ] where @xmath391 the anti - particle spinors are given by @xmath392 where @xmath393 is the charge conjugation operator , so that @xmath394 \ ; \eta_{\lambda } \end{aligned}\ ] ] with @xmath395 other useful relations are @xmath396 @xmath397 and @xmath398 note that in the above three equations @xmath399 is an on - shell four vector with @xmath400 and @xmath401 . 99 p. a. m. dirac , rev . * 21 * , 392 ( 1949 ) . s. j. brodsky , g. mccartor , h. c. pauli , s. s. pinsky part . world * 3 * , 109 ( 1993 ) ; s. j. brodsky , h - c pauli , s.s . pinsky , slac - pub-7474 , hep - ph/9705477 m. burkardt adv . * 23 * , 1 ( 1993 ) . j. m. namyslowski(1985 ) : progress in particle and nuclear physics , vol . * 14 * , faessler a. ( ed . ) ( pergamon press ) ; s. j. brodsky and g. p. lepage in _ perturbative quantum chromodynamics _ , a. mueller ( ed . ) . ( world scientific , singapore 1989 ) ; x. d. ji comments nucl . * 21 * , 123 ( 1992 ) w .- m . zhang , chinese j. phys . * 32 * , 717 ( 1994 ) . theory of hadrons and light - front qcd " edited by s.d . glazek , world scientific , ( singapore ) 1994 . a. harindranath , an introduction to light front dynamics for pedestrians , hep - ph/9612244 our notation is that a four vector @xmath402 is defined by the plus , minus and perpendicular components as @xmath403 and @xmath404 . see the appendix for more on this . h. cheng and t. wu expanding protons : scattering at high energies " , mit press cambridge ( 1987 ) f. coester , helv . phys . acta . * 38 * , 7 ( 1965 ) ; and in constraint s theory and relativistic dynamics , edited by g. longhi and l.lusanna ( world scientific singapore , 1987 ) , p. 159 and in the three body force in the three - nucleon system " edited by b.l.berman and bf gibson ( springer new york , 1986 ) b.d . keister and w. n.polyzou relativistic hamiltonian dynamics in nuclear and particle physics in _ advances in nuclear physics_. j.w . negele and e. vogt ( ed . ) . * 20*,1 ( 1991 ) ( plenum press , new york ) m. g. fuda and y.zhang , phys.rev . * c51 * , 23 ( 1995 ) ; phys . * c54 * , 495 ( 1996 ) . l. l. frankfurt and m. i.strikman phys . 76 * , 215 ( 1981 ) . frankfurt and m.i . strikman , phys . rep . * 160 * , 235 ( 1988 ) . soper , `` field theories in the infinite momentum frame '' , slac pub-137 ( 1971 ) ; d.e . soper , phys . rev d4 , 1620 ( 1971 ) ; j.b . kogut and d.e . soper , phys . rev . * d1 * , 2901 ( 1971 ) . s - j chang , r.g . root , and t - m yan , phys . rev . * d7 * , 1133 , 1147 ( 1973 ) . t - m yan phys . rev . * d7 * , 1760 , 1780 ( 1974 ) . miller , a light front treatment of the nucleus- implications for deep inelastic scattering " , to be published phys . rev . c. rapid communications , nucl - th/9702036 the use of scalar mesons is meant as a simple representation of part of the two pion exchange potential which causes much of the the medium range attraction between nucleons , as well as the effects of a possible fundamental scalar meson . f. gursey , nuovo cimento * 16 * 230 ( 196 ) ; ann . phys . ( n.y.)*12 * 91 ( 1961 ) ; p. chang and f. gursey , phys . rev . * 164 * , 1752 ( 1967 ) . s. weinberg , phys . rev . * 166 * , 1568 ( 1968 ) . s. thberge , a. w. thomas and g. a. miller , phys . d22 * ( 1980 ) 2838 ; * d23 * ( 1981 ) 2106(e ) . a. w. thomas , s. thberge , and g. a. miller , phys . rev . * d24 * , 216 ( 1981 ) ; a. w. thomas , adv . * 13 * ( 1984 ) 1 ; g. a. miller , p 190 in int . * 2 * ( 1984 ) . b.d . serot and j.d . walecka , adv . . phys . * 16*,1 ( 1986 ) . furnstahl , b.d . serot , h - b tang , nucl - th/9608035 . nucl.phys . * a615 * , 441 ( 1997 ) . serot and j.d . walecka , iu - ntc-96 - 17 , jan . 1997 nucl - th/9701058 the determination of the inverse of the operator @xmath405 is discussed in the original literature @xcite and in other places . the equations in appendix a are sufficient for our calculations . for a more recent discussion see a. harindrinath and w - m zhang , phys . rev . * d48 * , 4861 , 4881 , 4903 ( 1993 ) . the most general non - linear realization does not specify the ratio of the two pion - nucleon couplings . b.h . brandsden and r.g . moorhouse , `` the pion - nucleon system '' ( princeton university press , princeton , 1973 ) ; s.l . adler and r.f . dashen , `` current algebras '' , ( w.a . benjamin , new york , 1968 ) . r. machleidt , k. holinde , ch . elster , phys . rep . * 149 * , 1 ( 1987 ) ; r. machleidt , adv . * 19 * , 189 ( 1989 ) . r. brockmann and r. machleidt , `` the dirac - brueckner approach '' , prepared for `` open problems in nuclear matter '' , m. baldo , ed ( world scientific , singapore , to be published ) , chapt . nucl - th/9612004 . nucleon - nucleon interaction , g.e . brown and a.d . jackson north - holland ( amsterdam ) 1976 ; `` pions and nuclei '' , t. ericson and w. weise oxford science publications , ( new york ) 1987 . s. weinberg phys . rev . * 150 * , 1313 ( 1966 ) r. blankenbecler and r. sugar , phys.rev . * 142 * , 1051 ( 1966 ) . terentev , sov . * 24 * , 106 ( 1976 ) . m - h partovi , and e.l . lomon , phys . rev . * d2 * , 1999 ( 1970 ) . g. hoehler and h. pietarinen , nucl . phys . * b95 * , 210 ( 1975 ) ; g.e . brown and r. machleidt , phys . rev . * c50 * , 1731 ( 1994 ) . c. ordonez , l. ray , and u. van kolck , phys . rev . * c53 * , 2086 ( 1996 ) ; phys . rev . lett . * 72 * , 1982 ( 1994 ) ; c. ordonez , and u. van kolck phys.lett . * b291*,459 ( 1992 ) . the symbol for the nucleon field operator and the mode functions of that field is taken to be the same -@xmath23 to reduce the amount of notation . bartnik and st . glazek , phys . * d39 * , 1249 ( 1989 ) . m. brisudov and r. perry , phys . rev . * d54 * , 6453 ( 1996 ) . minimal substitution would require that the plus component of the vector field would enter into the equation ( [ kplus ] ) for @xmath6 , but we are using a formalism for which that component vanishes . serot , and r.j . furnstahl , phys . rev . * c43 * , 105 ( 1991 ) . minimizing @xmath301 with respect to @xmath26 ( for fixed @xmath255 ) is the same as minimizing @xmath294 with respect to @xmath26 ( for fixed @xmath255 ) because the volume @xmath256 is held fixed . s. a. chin , and j.d . walecka , phys . lett . * b52 * , 24 ( 1974 ) j. aubert _ et al . , b123 * , 275 ( 1982 ) . et al.,_phys . * 52 * 727 ( 1984 ) . a. bodek , _ et al.,_phys . * 51 , * 534 ( 1983 ) . see e.g. the reviews : r.l . jaffe , in relativistic dynamics and quark - nuclear physics . ed . by m. b. johnson and a. picklesimer ; m. arneodo , phys . 240 , 301 ( 1994 ) . geesaman , k. saito , a.w . thomas , ann . nucl . part . sci . * 45 * , 337 ( 1995 ) . ( [ deep ] ) is essentially eq . ( 5.2 ) of frankfurt and strikman , @xcite , with a correspondence between our @xmath406 and their @xmath407 . k. saito , and a.w . thomas , nucl . phys . * a574 * , 659 ( 1994 ) . birse , phys . lett . * b299 * , 186 ( 1930 ) . the standard nuclear shell model is one model of finite nuclei in which the mean field is dominant . the quark meson coupling model , see e.g. p.a.m . guichon , k. saito , e. rodionov , and a.w . thomas , nucl . * a601 * , 349 ( 1996 ) ; and p.g . blunden and g.a . miller , phys.rev . * c54 * , 359 ( 1996 ) is another . bjorkenand s.d . relativistic quantum fields _ ( mcgraw - hill , new york ) ( 1965 ) . fig . 1 @xmath18-ordered graphs for low energy pion - nucleon scattering . ( a ) second - order effects of the @xmath408 term @xmath131 . ( b ) second - order effects of the @xmath166 and @xmath409 terms of @xmath131 . ( c ) effects of the instantaneous fermion propagation terms of @xmath133 , and of the @xmath167 term of @xmath131 . the terms @xmath410 are defined in eqs , ( 2.46)-(2.48 ) . @xmath18-ordered graphs for one boson exchange contributions to nucleon - nucleon scattering . the numbers 1 - 4 represent the momentum , spin and charge states of the nucleons . here ( a ) meson propagation terms ( b ) instantaneous vector meson exchange of @xmath134 , eq . ( 2.48 ) fig . 3 feynman graphs for the two - pion exchange potential ( a ) uncrossed box diagram- the horizontal line represents the subtraction of the contribution arising from the iterated one pion exchange potential . ( b ) crossed box diagram ( c ) second order effect of the @xmath167 term of @xmath131 , eq . ( 2.46 ) ( d ) terms with one @xmath167 term .

the coalescence of binary black holes is one the most promising sources of gravitational waves for interferometric gravitational - wave detectors , such as ligo , virgo and geo600 @xcite . the first - generation ligo detectors have achieved their design sensitivity and recorded over one year of coincident data @xcite . this data , together with data from the virgo detector , are currently being searched for gravitational waves from compact binary coalescence @xcite . upgrades to improve the sensitivity of these detectors by a factor of two , and ultimately 10 , are underway . optimal searches using the enhanced detectors in 2009 will be sensitive to black - hole coalescence out to hundreds of megaparsecs @xcite . the advanced detectors , operational next decade , could detect black - hole binaries at distances of over @xcite . optimal searches for gravitational waves use matched filtering , which requires accurate knowledge of the waveform @xcite . previous searches in ligo data have used post - newtonian and phenomenological templates to search for the coalescence of black - hole binaries @xcite . over the last several years numerical relativity has made remarkable breakthroughs in simulating the late inspiral , merger and ringdown of black - hole binaries . the computational cost of these simulations is high , however , making it impractical to use them directly as template waveforms for use in a matched - filter search . it has been shown that there is good agreement between the waveforms generated by numerical relativity with analytic post - newtonian waveforms to within just a few orbits of merger @xcite . this paper uses the high - accuracy caltech cornell numerical - relativity waveforms to suggest improvements to the analytic waveforms currently used in gravitational - wave searches by ligo and virgo . a similar study has been performed by pan _ et al._using numerical data from pretorius and the goddard groups @xcite . our main results are in agreement with their conclusion that a simple extension of the existing stationary - phase approximation to the adiabatic post - newtonian waveforms ( called _ taylorf2 _ in ref . @xcite ) yields high overlaps with numerical waveforms . in sec . [ sec : searches ] , we review the current techniques used for searching for gravitational waves in gravitational - wave detector data . we discuss the construction of the waveform a pn nr hybrid in sec . [ sec : pnnrhybridwaveform ] . in sec . [ sec : efficiency ] we employ the hybrid waveform in a comparison of the detection efficiency of gravitational - wave templates that may be used in upcoming searches of ligo and virgo data . finally , in sec . [ sec : recommendations ] , we discuss improvements that may be made to the current data - analysis techniques to optimize overlaps . throughout this paper , we use only the @xmath5 component of the waveform @xmath6 ( as defined , e.g. , in @xcite ) . for convenience , we drop the superscript . whenever possible , we use dimensionless quantities , like @xmath7 , where @xmath8 is the areal radius of the observation sphere , and @xmath9 is the total apparent - horizon mass of the holes in the initial data . however , for any calculation involving the ligo noise curve , we have a physical scale , and thus use standard mks units . current searches for gravitational waves from binary black - hole coalescence use matched filtering to search for a waveform buried in noise . the matched filter is the optimal filter for detecting a signal in stationary gaussian noise . suppose that @xmath10 is a stationary gaussian noise process with one - sided power spectral density @xmath11 given by @xmath12 . for long integration times , the data stream @xmath13 output by the detector will always be dominated by the noise . thus , we can simply approximate @xmath14 to calculate @xmath15 . using this power spectral density ( psd ) , we can define the inner product between two real - valued signals the data stream @xmath16 and the filter template @xmath17by @xmath18 then , given data @xmath16 which may contain either noise @xmath19 or noise and a gravitational wave signal @xmath17 , @xmath20 the matched - filter signal - to - noise ratio ( snr ) is defined as @xmath21 the snr can then be used to construct a detection statistic ( directly or in combination with other statistics ) . it is therefore important to ensure that the templates used in searches accurately model the expected waveforms to avoid reduction in the value of @xmath22 . the _ overlap _ between two templates @xmath17 and @xmath23 is defined as @xmath24 the overlap encodes the fractional loss in snr that results from using the template @xmath23 rather than the true waveform @xmath17 . in a search that uses @xmath22 as a detection statistic this corresponds to the fractional loss in distance to which the search is sensitive . the filter template includes arbitrary time and phase offsets , encoded by the arrival time and phase , @xmath25 and @xmath26 . under a change of these quantities , the fourier transform behaves as @xmath27 we maximize over these two variables by calculating the inner product as @xmath28 note that this integral is just the ( inverse ) fourier transform of the quantity @xmath29 evaluated at @xmath25 . thus finding the maximum over @xmath25 involves taking the fourier transform and selecting the largest element of the finite set that results from discretization . in this paper we are concerned with overlaps between pn waveforms and nr signals ; to weight the inner product we use the following psds for initial and advanced ligo : for initial ligo we use an analytic approximation to the ligo design psd given by @xmath30\end{aligned}\ ] ] all integrals start from 40 hz . as shown in fig . [ fig : stildesandinitialpsd ] , at this frequency the noise is an order of magnitude higher than its lowest value , and below this frequency it rises rapidly as @xmath31 . the region below 40 hz therefore contributes very little signal power to the snr @xcite . the psd for enhanced ligo , which will begin operation in mid 2009 , has a similar shape to that for initial ligo although it has a factor of @xmath32 increase in strain sensitivity . our results using the initial - ligo psd are therefore valid for enhanced ligo , as the sensitivity factor cancels in eq . ( [ eq : overlapdefinition ] ) ; overlaps depend on the _ shape _ of the psd . for advanced ligo we use the gwinc program @xcite to generate the psd . bench reports the psd in increments of 0.0124 hz . when calculating discrete integrals against signals sampled at other frequencies we obtain values for the psd by linearly interpolating between the values provided by bench . we start integrals at 10 hz as that is the point where the noise has increased by two orders of magnitude above its minimum , as also shown in fig . [ fig : stildesandinitialpsd ] . searches for gravitational waves in ligo and virgo use a post - newtonian waveform known as__taylorf2 _ _ @xcite . this is a frequency - domain waveform obtained via the stationary - phase approximation @xcite , which assumes that the frequency - domain amplitude is simply proportional to @xmath33 ( the lowest - order behavior ) , while its phasing is given by the phase of the time - domain waveform , as a function of frequency . for a binary consisting of masses @xmath34 and @xmath35 , located at an `` effective '' distance @xmath36 , we have @xmath37{mpc}}{{\ensuremath{d_{\mathrm{eff}}}}}\right)\ , { \mathcal{a}}_{\unit[1]{mpc}}(m,\eta)\ , f^{-7/6}\ , \e^{i\varphi(f;m,\eta)}\ , \label{e : waveform}\ ] ] where @xmath38{mpc}}(m,\eta ) & = & \left(\frac{5\pi}{24}\right)^{1/2 } \left(\frac{{g}{\ensuremath{m_{\odot}}}/{c}^2}{\unit[1]{mpc}}\right ) \times \nonumber \\ & \quad & \times \left(\frac{\pi { g}{\ensuremath{m_{\odot}}}}{{c}^3}\right)^{-1/6 } \left(\frac{\eta}{{\ensuremath{m_{\odot}}}}\right)^{1/2 } \left(\frac{m}{{\ensuremath{m_{\odot}}}}\right)^{1/3}\ , \end{aligned}\ ] ] and the phasing @xmath39 of the frequency - domain waveform is given to 3.5pn accuracy by the formula @xcite @xmath40 \left[1 + 3 \ln\left(\frac{v}{v_0}\right ) \right ] \nonumber \\ & & + \left[\frac{11\,583\,231\,236\,531}{4\,694\,215\,680 } - \frac{640}{3 } \pi^2 - \frac{6\,848}{21 } \gamma \right ] v \nonumber \\ & & + \bigg[\left(-\frac{15\,335\,597\,827}{3\,048\,192 } + \frac{2\,255}{12 } \pi^2 - \frac{47\,324.0}{63.0 } - \frac{794\,8}{9}\right ) \eta \nonumber \\ & & \qquad + \frac{76\,055}{1\,728 } \eta^2 - \frac{127\,825}{1\,296 } \eta^3 \bigg ] v \nonumber \\ & & + \pi \left [ \frac{77\,096\,675}{254\,016 } + \frac{378\,515}{1\,512 } \eta - \frac{74\,045}{756 } \eta^2 \right ] v^2\bigg]\ .\end{aligned}\ ] ] where @xmath41 , @xmath42 and @xmath43 . the overall frequency scale is set by the total mass @xmath9 , as can be seen by observing that each occurrence of @xmath44 is accompanied by a factor of @xmath9 . might be rewritten as @xmath45 . this term accounts for a time offset altering the phase of the fourier transform . ] thus , going to a higher - mass system shifts the waveform to lower frequencies . on the other hand , to first order , the timescale for the rate of change of the frequency is given by the _ chirp mass _ : @xmath46 clearly , the total mass and chirp mass give us two very different handles on the behavior of the waveform . these two handles will be important when we try to match the template to our waveform in regions where the post - newtonian and stationary - phase approximations are poor . this is typically the case for high - mass systems , which only enter the detector band late in the inspiral . in this case , we can still obtain a high match , at the cost of using templates with the wrong values of @xmath9 and @xmath1 . we also note that physical binary systems are restricted to @xmath47 . however , for higher values of @xmath1 , the formulas shown above still give plausible waveforms ; in fact , in some cases these templates match the true waveform better than any physical template . we will explore the implications of allowing unphysical values for @xmath1 in searches over the templates in sec . [ sec : unrestrictedeta ] . note the heaviside function in eq . ( [ e : waveform ] ) . this contains a cutoff frequency @xmath48 which is used to ensure that the template does not extend to frequencies much greater than the frequencies contained in the expected signal . this is essentially a third parameter for the template waveform , and will be searched over . see sec . [ sec : effectofupperfreqcutoff ] for a discussion of strategies for optimizing detection by changing this cutoff . frequencies which are often used to characterize coalescing binary black holes are : ( i ) the frequency at the innermost stable circular orbit ( isco ) , @xmath49 , around a single schwarzschild black hole with the total mass of the binary system ; ( ii ) the frequency at the light ring , @xmath50 , around a single schwarzschild black hole with the total mass of the binary system ; ( iii ) the ringdown frequency of the final black hole , which depends on both its spin and mass ; and ( iv ) an `` effective ringdown , '' @xmath51 defined in @xcite . current searches use 2pn stationary - phase approximation ( spa ) _ taylorf2 _ templates @xcite . it has previously been shown that such waveforms provide acceptable detection templates for binary neutron stars and sub - solar mass black holes @xcite , but not necessarily for higher - mass black holes . this is an issue we will investigate below by testing 2pn and 3.5pn templates . in order to perform our comparison we need to construct a `` true '' black - hole binary waveform , which we might expect to observe with detectors . a numerical simulation will provide the data for the crucial nonlinear merger phase . we carefully extract the data and extrapolate it to large radius , and investigate the effects of numerical error on the final result . because this waveform is very computationally expensive to produce , it covers only about 32 cycles , which is not sufficient for a thorough investigation of the possibility of detecting it in searches of data from gravitational - wave detectors . thus , we match the numerical waveform to a post - newtonian waveform , producing a hybrid which extends for many thousands of cycles , covering the entire band of interest . the numerical simulation is the same as that described in refs . @xcite : an equal - mass , non - spinning , black - hole binary with reduced eccentricity @xcite , beginning roughly 16 orbits before merger , continuing through merger and ringdown @xcite . it is performed with the caltech cornell pseudospectral code , using boundary conditions designed to prevent constraint violations and gravitational radiation from entering the domain @xcite . data is extracted from the simulation in the form of the newman penrose scalar @xmath52 where @xmath53 and the complex vector @xmath54 are constructed with reference to the coordinate basis . along the positive @xmath55 axis , we have @xmath56 here , @xmath57 is the timelike unit normal to the spatial hypersurface , and @xmath58 is the unit vector in the positive @xmath55 direction . the vectors @xmath59 and @xmath60 are the standard coordinate vectors , which are not normalized . @xmath61 is extracted as a function of time , at various radii along the positive @xmath55 axis . this is then extrapolated to large radii , as described in ref . @xcite , and in greater detail in ref . @xcite . the measured ( instantaneous ) frequency at the beginning of the simulation is @xmath62{hz}\ , \frac{{\ensuremath{m_{\odot}}}}{m}\ .\ ] ] the measured ringdown frequency is @xmath63{hz}\ , \frac{{\ensuremath{m_{\odot}}}}{m}\ .\ ] ] the measured _ christodoulou _ mass and spin of the final black hole are @xmath64 using this value for the spin , a quasi - analytic formula due to echeverria @xcite predicts a value of @xmath65{hz}\ , \frac{{\ensuremath{m_{\odot}}}}{m}$ ] , for the ringdown frequency , in close agreement with the measured frequency . the numerical waveform will be the standard against which we will judge the _ taylorf2 _ waveforms used in ligo data analysis . to understand how precisely we should trust our final results , we need to understand the accuracy of the waveform itself . the most obvious measure of the error in this fiducial waveform is its convergence with increasing numerical resolution . [ f : accuracy ] shows the overlap ( eq . ( [ eq : overlapdefinition ] ) ) between waveforms computed at different resolutions . the data used here are the extrapolated @xmath66 waveforms , integrated in time twice . because of the short extent of the numerical waveforms , we need to be careful when using their fourier transforms . the signal can be corrupted easily by the non - periodicity of the waveforms , and the discontinuous jumps that result . for fig . [ f : accuracy ] we mitigate this problem by increasing the sampling frequency of the input data , and restricting the fourier transform to frequencies corresponding to instantaneous frequencies contained in the data . the input data can easily be upsampled in the time domain by interpolating the phase and amplitude of the complex data to a finer time grid . we then perform the transform , and explicitly set the data to zero at frequencies below @xmath67 and above @xmath68 , as given in eqs . [ eq : measuredstartingfreq ] and [ eq : measuredringdownfreq ] . while the results do depend on whether or not we impose these cutoffs , they do not depend sensitively on the actual cutoff frequencies . the overlap between the lowest- and highest - resolution simulations ( dashed lines ) actually passes through zero , as shown in the upper panel . presumably , this is because of loss of phase accuracy over the course of the simulation . all three simulations begin with the same initial data , so the waveforms are most similar at the beginning . masses for which this is the most important segment ( the lowest masses ) will naturally have the highest overlap between resolutions . as the simulation progresses , numerical error accumulates notably in the phase so the overlap decreases with masses for which later segments dominate the overlap ( higher masses ) . when the overlap is optimized over arrival time and phase , we can see that the overlap becomes much better , as shown in the lower panel , indicating sufficient accuracy within any frequency band for which phase coherence is required . in either case , the medium and highest resolutions are much more nearly the same . without optimization , their overlap is within a few tenths of a percent of 1 ; after optimization , the overlap is within @xmath69 of 1 . in the rest of our analysis we use the highest - resolution waveform . because we always optimize over arrival time and phase , the lower panel of fig . [ f : accuracy ] is the most relevant , and shows that the waveform has converged to very high accuracy . the overlaps we quote below will only be given to three decimal places at most , because this is roughly the accuracy of the single - precision numerical methods used in the rest of the paper . this accuracy is also sufficient for searches of gravitational - wave data . thus , the truncation error of the simulated waveform is irrelevant for those purposes . other sources of error include residual eccentricity and spin , the influence of the outer boundary of the simulation , extrapolation errors , and coordinate effects , as discussed in ref . the eccentricity had a disproportionately large effect on the error quoted in that paper because of the matching technique , which is not used here . restricting attention to the other effects of eccentricity , the uncertainty falls below that due to numerical error . similarly , using the techniques of ref . @xcite , the initial spins of the black holes have been measured more reliably , and found to be more than an order of magnitude smaller than previously determined , allowing us to reduce the estimate for that error to less than the numerical truncation error . the various coordinate effects were all estimated to be of roughly the same magnitude as the numerical error . with the numerical error being many times more accurate than needed for this analysis , and the other sources of uncertainty being of roughly the same size , these considerations indicate that the overall error in our fiducial waveform is substantially less than the precision needed for this analysis . numerical simulations can not simulate a very large portion of the inspiral of a black - hole binary system . indeed , the longest such simulation currently in the literature is the one used here which extends over just 32 gravitational wave cycles before merger . fortunately , this is the only stage in which simulations are needed . it has been shown previously @xcite that the _ taylort4 _ waveform with 3.5-pn phase and 3.0-pn amplitude matches the early part of this simulation to very high accuracy . we generate a _ taylort4 _ waveform of over 8000 gravitational wave cycles ( @xmath70 , starting at @xmath71 ) , and transition between the two to create a hybrid . this long waveform is sufficient to ensure that even for the lowest - mass systems we will consider the waveform begins well before it enters the frequency band of interest to ligo . we begin with @xmath66 data , which will later be integrated to obtain @xmath17 . following ref . @xcite , we match the numerical waveform to the pn waveform by adjusting the time and phase offsets of the pn waveform to minimize the quantity @xmath72^{2}\ , d t \ .\ ] ] here , we choose @xmath73 and @xmath74 , which is closer to the beginning of the waveform than in the previous paper . this particular interval is chosen to begin and end at troughs of the small oscillations due to the residual eccentricity @xmath75 in our numerical waveform . taking a range from trough to trough or peak to peak rather than node to node , for example of the eccentricity effects minimizes their influence on the matching . the eccentricity oscillations can be seen more easily after low - pass filtering the waveform , though we find filtering to be unnecessary for this paper . the junk radiation apparent in the waveform as shown here has no effect on the resulting match as we have verified by filtering , and redoing the match . because the final waveform will incorporate no numerical data before @xmath76 and very little immediately thereafter ( as explained below ) , the junk radiation will have no effect on any of our results as we have also explicitly verified . in particular , by integrating @xmath61 to obtain @xmath17 , we will effectively smooth the junk radiation . , that are blended to create the hybrid waveform . the vertical lines at @xmath77 and @xmath78 denote the region over which matching and hybridization occur . note that the agreement is well within the numerical accuracy of the simulation , represented by the horizontal bands , throughout the matching region . also note that the phase difference is fairly flat for a significant period of time after the matching range , which indicates that the match is not sensitive to the particular range chosen for matching . ] in fig . [ fig : matchingphasecomparison ] we compare the phase of the numerical and pn waveforms . the quantities plotted are @xmath79 shown over the interval on which both data sets exist . the vertical bars denote the matching region . note that the phase difference is well within the accuracy of the simulation ( about 0.01 radians , represented by the horizontal band ) over a range extending later than the matching region . also , the difference between the two is fairly flat , which implies that the match is not very sensitive to the region chosen for matching . because of this , we expect that the phase coherence between the early pn data and the late nr data will be physically accurate to high precision . the hybrid waveform is then constructed by blending the two matched waveforms together according to @xmath80\ , a_{{\mathrm{pn}}}(t)\ , \\ \phi_{{\mathrm{hyb}}}(t ) & = & \tau(t)\ , \phi_{{\mathrm{nr } } } + \left [ 1 - \tau(t ) \right]\ , \phi_{{\mathrm{pn}}}(t)\ .\end{aligned}\ ] ] the blending function @xmath81 is defined by @xmath82 the values of @xmath76 and @xmath83 are the same as those used for the matching . the amplitude discrepancy between the pn waveform and the nr waveform over this interval is within numerical uncertainty roughly @xmath84 . as with the matching technique ( eq . ( [ eq : matchingchisquared ] ) ) , this method is similar to that of ref . @xcite , but distinct , in that we blend the phase and amplitude , rather than the real and imaginary parts . this leads to a smoothly blended waveform , shown in fig [ fig : waveformsnapshot ] . of the hybrid waveform used in this analysis : the @xmath85 waveform seen by an observer on the positive @xmath55 axis . the vertical lines denote the matching and hybridization region . the @xmath86 on the time axis corresponds to the beginning of data from the numerical simulation . ] up to this point , the waveform has been @xmath61 data . with the long waveform in hand , we numerically integrate twice to obtain @xmath17 , and set the four integration constants so that the final waveform has zero average and first moment @xcite . because of the very long duration of the waveforms , this gives a reasonable result , which is only incorrect at very low frequencies lower than any frequency of interest to us . we have also checked that our results do not change when we effectively integrate in the frequency domain by taking @xmath87 which is the frequency - domain analog of the equation @xmath88 . we now compare the signal described in the previous section to restricted , stationary phase _ taylorf2 _ post - newtonian templates with terms up to order 2.0 , order 3.5 , and a `` pseudo-4.0 pn - order '' term recommended in ref . overlaps are calculated using the techniques of sec . [ sec : matchedfiltering ] , with the signal @xmath16 being the hybrid waveform described in sec . [ sec : pnnrhybridwaveform ] scaled to a range of masses . we consider both the initial- and advanced - ligo noise curves . plots of the hybrid waveforms in comparison to the initial - ligo noise curve are shown in fig . [ fig : stildesandinitialpsd ] . use the initial - ligo noise curve , with sources at a distance of 100mpc . ] the masses are chosen so that various frequencies of interest ( the final stitching frequency , the isco , and the ringdown ) occur at the `` seismic wall '' for initial ligo : . the waveforms @xmath89 are scaled to depict the detectability of the signal , typically quantified by the snr introduced in ( [ eq : innerproductsnr ] ) , which may be written as @xmath90 in the final expression , the numerator and denominator have the same units , and are directly comparable . because the square root of the denominator is familiar , we plot that along with the square root of the numerator . plotting these two quantities together gives a graphical impression of the detectability of the waveform , and the relative importance of each part of the waveform , by its height above the noise curve . in ref . @xcite , brady and creighton define a slightly different quantity , the characteristic strain @xmath91 the relative factor of @xmath92 they use is present so that they can plot @xmath93 against @xmath94 . cutler and thorne @xcite define still another quantity , the signal strength @xmath95 , which is related to the fourier transform by @xmath96 the factor of @xmath97 comes from averaging over the orientation of the binary , which we do not do . @xmath98 is the ratio of the threshold to the rms noise at the endpoint of signal processing . , with randomized initial conditions . note the logarithmic scale on the vertical axis . the majority of instances produced a lower overlap than the optimum . we interpret this as pointing to the existence of a broad local maximum which did not coincide with the global maximum . ] for each template family we initially optimize over signal mass @xmath9 , symmetric mass ratio @xmath99 , and upper cutoff frequency @xmath100 . the optimization is performed using a nelder mead ( `` amoeba '' ) algorithm @xcite . the amoeba starts with a simplex in the parameter space , and proceeds through a series of steps , each of which will improve the value of the function at at least one vertex . the algorithm terminates when all vertices have converged to the same point to within a specified tolerance . this process is deterministic , and amounts to an enhanced steepest - ascent algorithm . it is therefore only guaranteed to find a local maximum , and indeed we find that an amoeba instance started at a random point in the parameter space is most likely to converge to a point that does not give the highest possible overlap . we interpret this as being due to a large region in parameter space containing a local maximum and a relatively smaller region containing the global maximum . we therefore supplement the basic amoeba by running 300 instances with random starting values , and taking the best match obtained over all instances . in repeated runs the same optimal parameters were found by at least some of the amoebas , which supports the claim that this is the true maximum . the results of optimizing over all of @xmath101 and @xmath100 for selected masses for initial ligo are given in table [ tab : threeparamoverlapdetailinitial ] and summarized in fig . [ fig : threeparamoverlapsummaries ] . for initial ligo , in the range covered by the current compact binary coalescence ( cbc ) low - mass search @xmath102 @xcite , the pseudo-4.0 pn _ taylorf2 _ waveforms achieve the highest overlaps , exceeding those obtained with 3.5 pn waveforms by @xmath103 . above @xmath104 the 3.5 pn waveforms produce overlaps as much as 4% greater than those obtained with pseudo-4.0 pn waveforms over a range from @xmath105@xmath106 . with the advanced - ligo noise curve , in the cbc low - mass range , the 3.5 pn and pseudo-4.0 pn waveforms produce overlaps within 2% of each other , with 3.5 pn producing higher overlaps below 20 @xmath107 and pseudo-4.0 pn producing higher overlaps in the range @xmath108@xmath104 . pseudo-4.0 pn continues to give the highest overlaps up to @xmath109 , producing overlaps as much as 4% greater than those obtained with 3.5 pn waveforms . above @xmath109 3.5 pn waveforms again yield the best overlaps , by as much as 6% around 90 @xmath107 . [ cols="<,^,^,^,^",options="header " , ] a significant feature of tables [ tab : threeparamoverlapdetailinitial ] and [ tab : threeparamoverlapdetail ] is the size of the error bars on the cutoff frequencies . for @xmath110 the cutoff frequency can vary as much as 128% above and 28% below the optimal value while losing no more than 1% of overlap this leads us to consider the range of possible template parameters which may give high overlaps . in the next section we consider the reduction in overlap as the parameters @xmath48 and @xmath1 are independently varied from the optimal value . and @xmath1 , which the cutoff frequency @xmath48 is prescribed by the weighted average described below . the mass ratio @xmath1 is allowed to range over unphysical values . the best - fit values found for the pseudo-4.0 pn templates are always physical in this case . see sec . [ sec : unrestrictedeta ] . right : the same , for the advanced - ligo psd , title="fig : " ] and @xmath1 , which the cutoff frequency @xmath48 is prescribed by the weighted average described below . the mass ratio @xmath1 is allowed to range over unphysical values . the best - fit values found for the pseudo-4.0 pn templates are always physical in this case . see sec . [ sec : unrestrictedeta ] . right : the same , for the advanced - ligo psd , title="fig : " ] as shown in fig . [ fig : stildesandinitialpsd ] the amplitude of the nr waveforms drops sharply at around the lightring frequency , which depends on the total mass of the binary . the _ taylorf2 _ waveforms do not model the late inspiral , merger or ringdown and hence will continue to evolve as @xmath33 at all frequencies , increasingly deviating from the nr waveform . this suggests that the upper frequency cutoff of the _ taylorf2 _ waveform should be chosen to be below the frequency at which the two diverge . however , the effect of the divergence is mitigated by the psd . the denominator of the overlap , eq . ( [ eq : overlapdefinition ] ) , depends on @xmath111 which is a constant , and @xmath112 which would increase without limit if not for the psd . [ fig : filterintegrand ] shows @xmath113the integrand of @xmath112for the initial - ligo noise curve for an example _ taylorf2 _ waveform for an equal - mass @xmath114 binary . we see that above about 450 hz there is very little contribution to the integrand , and so extending the cutoff frequency above this will not impact the overlap . ) for a _ taylorf2 _ , 3.5 pn waveform with @xmath115 and @xmath116 , at a distance of 100mpc , using the initial - ligo noise curve . note that the shape of this curve does not change as we change @xmath9 and @xmath1 ; only the vertical scale changes . right : overlap between caltech cornell waveform scaled to @xmath117 and restricted _ taylorf2 _ , 3.5 pn waveform using the best - match values for @xmath9 and @xmath1 , as a function of the cutoff frequency @xmath100 , with the initial - ligo noise curve . the vertical bars are meant to delineate 1% loss . note that the upper bound extends to higher frequencies indefinitely . , title="fig : " ] ) for a _ taylorf2 _ , 3.5 pn waveform with @xmath115 and @xmath116 , at a distance of 100mpc , using the initial - ligo noise curve . note that the shape of this curve does not change as we change @xmath9 and @xmath1 ; only the vertical scale changes . right : overlap between caltech cornell waveform scaled to @xmath117 and restricted _ taylorf2 _ , 3.5 pn waveform using the best - match values for @xmath9 and @xmath1 , as a function of the cutoff frequency @xmath100 , with the initial - ligo noise curve . the vertical bars are meant to delineate 1% loss . note that the upper bound extends to higher frequencies indefinitely . , title="fig : " ] the numerator of the overlap , @xmath118 , can only increase as the cutoff frequency is raised , however frequencies above the lightring where the waveforms have diverged will contribute very little . the effect of including higher frequencies on the overlap is therefore determined by the @xmath112 term in the denominator . for systems with ringdown frequencies well above the peak of the integrand in fig . [ fig : filterintegrand ] , this term will not significantly reduce the overlap . for example , binaries of total mass roughly @xmath119 have ringdown frequencies at roughly 450hz . only a small fraction of the snr comes from higher frequencies . thus , we expect that systems with lower masses should not suffer great loss in overlap if the cutoff frequency is higher than ringdown . however for higher - mass systems the overlap can be significantly reduced if the upper frequency cutoff is too large . this is indeed what we find , as shown by a representative example on the right in fig . [ fig : filterintegrand ] . for this @xmath119 system , using the initial - ligo noise curve , the optimal cutoff frequency is around 450hz roughly the ringdown frequency . decreasing the cutoff quickly decreases the overlap . the cutoff may be increased almost indefinitely , however , with only 0.5% loss in overlap . this , of course , changes when using the advanced - ligo noise curve . we revisit this issue in sec . [ sec : recommendations ] . the physical symmetric mass ratio is restricted to the range @xmath120 , values above this imply complex - valued masses . however the pn waveforms are well - behaved for @xmath121 , and as seen from tables [ tab : threeparamoverlapdetailinitial ] and [ tab : threeparamoverlapdetail ] , the highest overlaps are often obtained at unphysical values of @xmath1 . in fig . [ fig : physicaleta ] we show the effect of limiting the optimization to physical @xmath1 . at high masses , the limitation reduces the optimal overlap by up to 12% . _ taylorf2 _ waveforms with @xmath122 would not be expected to accurately model the late - inspiral and merger part of the waveform , as non - newtonian effects are increasingly significant in this region . we find that allowing unphysical @xmath1 broadens the space of waveforms covered by the _ approximation sufficiently to capture more of the late - inspiral and merger . to range over unphysical values , compared to those obtained by restricting the range of @xmath1 . these overlaps are generated using 3.5 pn taylorf2 templates , searching over values of the total mass and mass ratio . extending to unphysical values of @xmath1 improves the match by up to 11% . ] based on the analysis of the previous sections we propose a series of adjustments to searches using _ taylorf2 _ template waveforms to enhance the efficiency of those searches . first , as seen in fig . [ fig : threeparamoverlapsummaries ] for initial ligo , adding terms up to 3.5 pn order produces overlaps as large or larger than the current 2.0 pn templates over most of the mass range , while the pseudo-4.0 pn templates recommended in ref . @xcite produce slightly larger overlaps at masses near @xmath123 . thus , we recommend pseudo-4.0 pn templates for the low mass range , @xmath124 , and 3.5 pn templates for higher masses . the improvement due to 3.5 pn templates over 2.0 pn generally holds for advanced ligo as well . the 3.5 pn templates produce larger overlaps than 2.0 pn templates above @xmath125 without a significant loss ( within 1% ) at lower masses . however , there is a large region for which the pseudo-4.0 pn term does significantly better . when using an advanced - ligo noise curve , we recommend 3.5 pn templates generally , 2.0 pn templates in the range @xmath126@xmath127 and pseudo-4.0 pn templates for masses in the range @xmath128@xmath4 . as a second improvement , we note from fig . [ fig : physicaleta ] that allowing @xmath1 to range over unphysical values significantly improves matches with 3.5 pn templates above @xmath129 . in preliminary studies we have found that extending to @xmath130 roughly doubles the size of the template bank , and the advantages must therefore be weighed against the increase in false alarm rate . values for 3.5 pn templates with initial ligo . the dark gray band contains cutoff frequencies with matches within 1% of the value at which the best overlap was obtained . the light gray band contains frequencies with matches within 3% . right : candidate @xmath100 values for 3.5 pn templates with advanced ligo . the dark gray band contains cutoff frequencies with matches within 1% of the value at which the best overlap was obtained . the light gray band contains frequencies with matches within 3% . note that the weighted - average cutoff extends past the 1% error bars for @xmath131 . however , in that same region , the 3.5 pn templates do poorly overall , and we recommend pseudo-4.0 pn templates . the optimal cutoff frequency for pseudo-4.0 pn templates is much closer to the weighted - average cutoff in this mass range.,title="fig : " ] values for 3.5 pn templates with initial ligo . the dark gray band contains cutoff frequencies with matches within 1% of the value at which the best overlap was obtained . the light gray band contains frequencies with matches within 3% . right : candidate @xmath100 values for 3.5 pn templates with advanced ligo . the dark gray band contains cutoff frequencies with matches within 1% of the value at which the best overlap was obtained . the light gray band contains frequencies with matches within 3% . note that the weighted - average cutoff extends past the 1% error bars for @xmath131 . however , in that same region , the 3.5 pn templates do poorly overall , and we recommend pseudo-4.0 pn templates . the optimal cutoff frequency for pseudo-4.0 pn templates is much closer to the weighted - average cutoff in this mass range.,title="fig : " ] our third recommendation involves the cutoff frequency used for the template waveform . optimization over the cutoff frequency is too computationally intensive to be done in searches . currently , the cutoff frequency is typically taken to be the schwarzschild isco frequency . to examine the effect of this choice we vary @xmath100 while keeping the mass and @xmath1 at their optimal values , for each of the signal masses in our range . the result of one such variation is shown in fig . [ fig : filterintegrand ] ( right ) . figs . [ fig : fcrecomendations ] shows the variations for all masses , highlighting the regions within which the overlap drops by less than 1% ( dark gray ) and 3% ( light gray ) of the optimal value . this figure also shows the isco and erd frequencies , neither of which stays within the 1% band for both initial and advanced ligo . in particular , the isco is a poor choice for both initial and advanced ligo except at very low masses , where the precise value of the cutoff is almost irrelevant . the isco is often pointed to somewhat arbitrarily as a good estimate of the breakdown of post - newtonian approximations @xcite . so , for instance , if we were to match a pn template to a physical waveform , beginning at some point in the distant past , we might expect them to separate quite badly near the isco . of course , for realistic black - hole binaries , the gravitational waves will only enter the ligo band late in the inspiral just before the isco for low - mass systems , or after the isco for high - mass systems . we can see from fig . [ fig : stildesandinitialpsd ] that , for masses below about @xmath129 , the isco is high enough that lower - frequency parts of the waveform contribute the most to the snr . for very high masses , however , this basically cuts the waveform down to nothing . in initial ligo , the isco is completely buried in seismic noise for masses above about @xmath132 . thus , we must move the cutoff frequency up . we can not push the cutoff far above ringdown , because the physical waveform simply ceases to exist ( see fig . [ fig : stildesandinitialpsd ] ) . it has been suggested that an `` effective ringdown '' ( erd ) frequency @xmath51 is a useful upper limit @xcite . for intermediate masses , we would like to interpolate somehow between these two extremes of isco and erd . we suggest setting the cutoff frequency to a weighted average of the two , where the weights are the contributions to the snr below the given frequency . if we assume coherent phasing between the template and the physical waveform , we can simply take the amplitudes of the two waveforms . also , note that the restricted spa approximation for the amplitude is reasonable . thus , define @xmath133 we have already dropped constant factors in the expressions for @xmath22 that will cancel out . note that these expressions only depend on the total mass by way of the limits of integrations which are very simple , known functions of the mass so these integrals could be done just once for a given noise curve , storing the intermediate values . when the cutoff needs to be calculated , the cumulative integral could be evaluated at the given isco and ringdown frequencies . hence , this would be a fast way of calculating the cutoff , with no need to do the integrals each time the cutoff is needed . we can test this recommended frequency by comparing it to the optimal cutoff frequency found by the amoeba search described in sec . [ sec : efficiency ] . for 3.5 pn templates in initial ligo , we find that it is an excellent match to the optimal frequency . [ fig : fcrecomendations ] shows these two values , along with dark and light bands showing the regions in which changing @xmath48 results in a loss of overlap of 1% and 3% , respectively . of course , the same figure shows that using the erd recommendation would stay within the 1% error bounds . nonetheless , the close match between this recommendation and the true optimum suggests that it is sound . thus , our final recommendation is to use the weighted - average frequency cutoff throughout the entire mass range . while our analysis has been restricted to equal - mass systems , the cutoff frequency we have defined here could be applied to unequal - mass systems as well . it will be interesting to see how this cutoff fares in those situations . similar results hold for advanced ligo , when using our recommended template for each mass . that is , in regions where 3.5 pn templates do poorly ( see fig . [ fig : threeparamoverlapsummaries ] ) , the weighted average is a poor predictor of the optimal cutoff frequency using those templates , as shown in fig . [ fig : fcrecomendations ] . however , in those same regions where pseudo-4.0 pn templates do well the weighted average is a good predictor of the optimal cutoff frequency for 4.0 pn templates . thus , again , we recommend using the weighted - average frequency cutoff throughout the entire mass range with advanced ligo . by prescribing a cutoff frequency , the search does not need to extend over that parameter . similarly , by prescribing a post - newtonian order , we need use only one template for a given total mass . on the other hand , if these recommendations decrease the overlap found by too much when using them compared to the overlap found by an unconstrained search , it may be better to search the larger parameter space . we can evaluate the loss in overlap by comparing the results found using our recommendations to the results found when searching over the set of all three template families , and all masses , mass ratios , and cutoff frequencies . we have determined that this loss in overlap when using our recommendations is always less than 0.0025 for initial ligo , and less than 0.007 for advanced ligo . we have compared high - accuracy nr waveforms for equal - mass binary black holes from the caltech cornell group to stationary phase post - newtonian waveforms . we examined a number of factors that influence the matches between the two , with the goal of optimizing the matches and hence improving the efficiency of templated searches in initial and advanced ligo . we first considered the effect of the post - newtonian order to which the phase evolution is taken , and found that adding terms up to 3.5 pn or pseudo-4.0 pn to the currently - used 2.0 pn templates significantly improves the matches over a large range of masses , as shown in fig . [ fig : threeparamoverlapsummaries ] . we then studied the effect of varying the upper cutoff frequency of the templates . the frequency that achieves the optimal match is a function of mass , and we find this function is well - approximated by an average between isco and erd , weighted by contribution to the snr , as shown in fig . [ fig : fcrecomendations ] . finally , we allow the symmetric mass ratio @xmath1 to range over unphysical values up to @xmath134 , and find that this dramatically improves matches , as shown in fig . [ fig : physicaleta ] . based on the results we recommend adjusting the searches using _ taylorf2 _ template waveforms by going up to 3.5 pn or 4.0 pn over most of the mass range , integrating up to our recommended cutoff , and allowing allowing @xmath1 to extend up to 1 . for initial ligo , the overlaps obtained using these parameters is always within 0.0025 of overlaps achievable by optimizing over all three parameters . in future work we plan to extend this analysis to unequal - mass and spinning black - hole systems . we have found that allowing unphysical values of @xmath1 roughly doubles the size of the template bank , and we also plan to study the impact of this on the false alarm rate . we thank luisa buchman , yanbei chen , curt cutler , lisa goggin , larry kidder , luis lehner , lee lindblom , harald pfeiffer , mark scheel and kip thorne for useful discussions . this work was supported in part by grants from the sherman fairchild foundation and by nsf grants phy-0601459 , phy-0652995 , and dms-0553302 to caltech . 10 url # 1#1urlprefix[2][]#2 thorne k s 1987 _ three hundred years of gravitation _ ed hawking s w and israel w ( cambridge : cambridge university press ) ajith p , babak s , chen y , hewitson m , krishnan b , sintes a m , whelan j t , brgmann b , diener p , dorband n , gonzalez j , hannam m , husa s , pollney d , rezzolla l , santamara l , sperhake u and thornburg j 2008 _ phys . rev . d _ * 77 * 104017

the phenomenon of shear localization during high strain - rate deformations of metals @xcite is a striking instance of material instability in mechanics , that has attracted considerable attention in the mechanics literature ( _ e.g. _ @xcite ) as well as a number of mathematical studies ( @xcite and references therein ) . material instability is typically associated with ill - posedness of an underlying initial value problem , what has coined the term _ hadamard instability _ for its description in the mechanics literature . it should however be emphasized that while hadamard instability indicates the catastrophic growth of oscillations around a mean state , what is observed in localization is the orderly albeit extremely fast development of coherent structures : the shear bands . the latter is a nonlinear phenomenon and a linearized analysis can only serve as a precursor to such behavior . we work with the simplest model capturing the mechanism of shear band formation , a simple shear motion of a non - newtonian fluid with temperature decreasing viscosity @xmath0 , @xmath1 where @xmath2 describes the strain - rate sensitivity which typically in these problems satisfies @xmath3 . mathematical studies of , for power law viscosities @xmath4 , @xmath5 , show that the uniform shear is asymptotically stable for @xmath6 @xcite , but indicate unstable response in the complementary region @xmath7 , @xcite . the mathematical analysis has not so far characterized the precise behavior in the instability region ; however , detailed and resolved numerical studies ( _ e.g. _ @xcite ) show that instability is followed by formation of shear bands . the difficulty is that in order to analyze the unstable regime one needs to account for the combined effect of parabolic regularizations in conjunction to ill - posed initial value problems . an asymptotic theory for calculating an effective equation for the compound effect is proposed in @xcite . it is based on the theory of relaxation systems and the chapman - enskog expansion and produces an effective asymptotic equation . for a power law the effective equation changes type from forward to backward parabolic , across the threshold @xmath8 of the parameter space . the criteria of @xcite are based on asymptotic analysis and as such are formal in nature . our objective , here , is to develop a mathematical theory for the onset of localisation and the emergence of a nonlinear localized state , thus providing a rigorous justification to the behavior conjectured by the effective equation in @xcite . we employ the constitutive hypothesis of an _ exponential law _ for the viscosity coefficient , @xmath9 the model for the exponential law admits a special class of solutions describing uniform shearing @xmath10 the graph @xmath11 describing the average stress vs. average strain response is monotonically decreasing , and thus the deformation of uniform shear is characterized by strain softening response . we set the following goals : 1 . to study the linearized problem around the uniform shearing solution and to assess the effects of the various parameters of the problem 2 . to inquire whether there are special solutions of the nonlinear problem that indicate the emergence of localized deformations and formation of shear bands . since the base solution around which we linearize is time - dependent the linearized analysis necessarily involves analysis of non - autonomous systems and also perturbations have to be compared against the base solution . we perform a linearization using ideas of relative perturbations @xmath12 leading to linearized system @xmath13 with @xmath14 computed by @xmath15 . this problem turns out to have the following behavior : 1 . [ en : introa]for the case without thermal diffusion @xmath16 : + when @xmath17 the high frequency modes grow exponentially fast with the exponent of the order of the frequency what is characteristic of ill - posedness and _ hadamard instability_. + when @xmath2 the modes are still unstable but the rate of growth is bounded independently of the frequency . in fact , the behavior in that range is characteristic of _ turing instability_. 2 . the effect of thermal diffusion ( @xmath18 ) for the non - autonomous linearized model can also be assessed : perturbations do grow initially in the early stages of deformation , but over time the effect of the diffusion intensifies and eventually the system gets stabilized . for a quantitative description of the above qualitative statement we refer to lemmas [ lemdecay ] and [ lembound ] . in addition , this behavior can also occur for the full solution of the nonlinear problem subject to simple shear boundary conditions as indicated in numerical simulations presented in figure [ mtstb ] . this type of metastable response was already conjectured ( for a somewhat different model ) in numerical experiments of @xcite . the linearized analysis presented here offers a theoretical explanation of this interesting behavior . in the second part of this article we consider the adiabatic variant of the model with exponential constitutive law . the system is expressed for @xmath19 in the form @xmath20 and is now considered on the whole space @xmath21 . for the system , we construct a class of localizing solutions @xmath22 \sigma ( x , t ) & = \sigma_s ( t ) \frac{1}{\phi_\lambda ( t ) } \ , \sigma_\lambda \big ( \sqrt{\lambda } x \phi_\lambda ( t ) \big ) , \\[2pt ] \theta(x , t ) & = \theta_s(t ) + \lambda \tfrac{n+1}{\alpha } ( \theta_s(t ) - \theta_0 ) + \theta_\lambda \big ( \sqrt{\lambda } x \phi_\lambda ( t ) \big ) , \end{aligned}\ ] ] where @xmath23 depends on the parameter @xmath24 , and is determined via the solution @xmath25 of the singular initial value problem , with @xmath26 . the existence of the solution @xmath25 is established in theorem [ connection ] where most importantly its behavior at the origin and at infinity is made precise . the emerging solution depends on three parameters : @xmath27 linked to the uniform shear in , a parameter @xmath28 which can be viewed as a length scale of initial data , and a parameter @xmath29 determining the size of the initial profile . to our knowledge this is the first instance that the compound effect of hadamard instability and parabolic regularizations is mathematically analyzed in a nonlinear context . the reader should contrast the nonlinear response captured by the exact solution to the response of the associated linearized problem . the solutions indicate that the competition among hadamard instability and momentum diffusion in a nonlinear context can lead to a nonlinear coherent structure that localizes in a narrow band ( see theorem [ shearband ] and remark [ rmkbeh ] ) . by contrast , for a linear model , the combined effect of hadamard instability and parabolicity leads to unstable oscillatory modes growing in amplitude ( see ) . this justifies the scenario that nonlinearity can kill the oscillations caused by hadamard instability so that the non - uniformity of deformation produced by the instability merges into a single localized zone . it would be interesting to explore this type of response for other models related to material instability in solid mechanics . the exposition is organized as follows : section [ secdescription ] presents a description of the problem and the formulation via relative perturbations . the linearized analysis is presented in section [ stabar ] , first the part that can be done via spectral analysis for autonomous systems , and then the part concerning the effect of thermal diffusion , which is based on energy estimates for a non - autonomous linear system . detailed asymptotics for the eigenvalues and the connection with hadamard and turing instabilities are presented there . the presentation of the localized solutions is split in sections [ seclocal]-[seclocalii ] : section [ seclocal ] presents the invariance properties , the _ ansatz _ of localized solutions , and also introduces the relevant auxiliary problem , . the construction of solutions to the singular initial - value problem , and the proof of their properties is the centerpiece of our analysis and forms the objective of section [ secss ] . the construction is based on a series of steps , involving desingularization , a nonlinear transformation , and eventually the existence of a heteroclinic orbit via the poincar - bendixson theorem . the construction of the localized solutions is then effected in section [ seclocalii ] . in this work we study the onset of localization and formation of shear bands for the model describing shear motions of a non - newtonian fluid with viscosity @xmath30 decreasing with temperature . for concreteness we take a viscosity satisfying the exponential law and are interested in the range where the rate sensitivity @xmath31 is positive and small . the model then becomes @xmath32 and describes shear motions between two parallel plates located at @xmath33 and @xmath34 with @xmath35 the velocity in the shear direction , @xmath36 the shear stress , @xmath37 the temperature , and @xmath38 depicting the strain rate . the boundary conditions are taken @xmath39 equation reflects prescribed velocities at the boundaries , while manifests that the bounding plates are adiabatic . note that implies that the stress satisfies @xmath40 and that @xmath41 the initial conditions are @xmath42 throughout the study we will consider two cases : when @xmath43 the process is adiabatic and the resulting model will be the main object of study , as it is the simplest model proposed for studying the phenomenon of formation of shear bands . the case @xmath18 will also be studied in section [ stabar ] as a paradigm to assess the effect of thermal diffusion . the problem , , admits a class of special solutions describing uniform shear @xmath44 for the special case of the exponential law , @xmath45 and @xmath46 read @xmath47 these are universal solutions , _ i.e. _ they hold for all values of the parameters @xmath31 , @xmath48 , @xmath49 in the model including the limiting elliptic initial value problem when @xmath50 , @xmath43 . the graph @xmath51 is viewed as describing the stress vs. average strain response . as the graph is decreasing , it means that there is always softening response in the course of the deformation . the system has served as a paradigm for the mathematical study of the phenomenon of shear bands ( _ e.g. _ @xcite ) occurring during the high strain - rate deformations of metals . formation of shear bands is a well known material instability that usually leads to rupture , and has received considerable attention in the mechanics literature ( _ e.g. _ @xcite ) . the reader is referred to @xcite for a description of the problem intended for a mathematically oriented reader that is briefly also outlined below . it was recognized by zener and hollomon @xcite ( see also @xcite ) that the effect of the deformation speed is twofold : ( a ) to change the deformation conditions from isothermal to nearly adiabatic ; under nearly adiabatic conditions the combined effect of thermal softening and strain hardening tends to produce net softening response . ( b ) strain rate has an effect _ per se _ and induces momentum diffusion that needs to be included in the constitutive modeling . a theory of thermoviscoplasticity is commonly used to model shear bands and we refer to @xcite for various accounts of the modeling aspects . the basic mechanism for localization is encoded in the behavior of the following simple system of differential equations @xmath52 ( which is the adiabatic case @xmath43 of ) . this model is appropriate for a non - newtonian fluid with temperature dependent viscosity , and is also a simplification of the models proposed in @xcite to capture the mechanism for shear band formation in high strain - rate plastic deformations of metals . the argument - adapted to the language of a temperature dependent fluid - goes as follows : as the fluid gets sheared the dissipation produces heat and elevates the temperature of the fluid . the viscosity decreases with the elevation of the temperature and the fluid becomes softer and easier to shear . one eventuality is that the shear deformation distributes uniformly across the material in the same manner as it does for the usual newtonian fluids . this arrangement is described by the uniform shearing solution . a second mode of response is suggested when comparing the behavior of the fluid in two spots : one hot and one cold . since the viscosity is weaker in the hot spot and stronger in the cold spots , the amount of shear generated in the hot spot will be larger than the shear generated in the cold spot and in turn the temperature difference will be intensified . it is then conceivable that the deformation localizes into a narrow band where the material is considerably hotter ( and weaker ) than the surrounding environment . the two possibilities are depicted in fig . [ shearflow ] . the outcome of the competition of thermal softening and momentum diffusion is captured by the behavior of solutions to . a series of works analyze the system mostly for power law viscosities @xmath53 these studies show stable response @xcite when @xmath54 , and indicate unstable response @xcite but without characterizing the behavior in the complementary region @xmath55 . more precisely , in the unstable regime tzavaras @xcite shows that development of shear bands can be induced by stress boundary conditions leading to finite - time blow - up , and that their occurrence is associated with a collapse of stress diffusion across the band . on the other hand , bertsch , peletier and verduyn - lunel @xcite show that if the problem has global solutions and there is a sufficiently large initial temperature perturbation , then a shear band appears as the asymptotic in time state . careful and detailed numerical investigations @xcite indicate that instability is widespread in the parameter regime @xmath55 and is typically followed by the development of shear bands . the behavior in the instability domain " is at present poorly understood , especially regarding the precise mechanism of formation of the shear band . the reason is the response in this regime results from the competition between an ill - posed problem and a parabolic regularization , and little mathematical theory is available to study such effects . nevertheless , an asymptotic theory similar to the chapman - enskog expansion is used by katsaounis - tzavaras @xcite to calculate an effective equation for the compound effect of thermal softening and strain - rate sensitivity . the effective equation changes type from forward to backward parabolic across the instability threshold , and thus provides an asymptotic criterion for the onset of localization . in the present work , we will develop a mathematical framework for studying the competition and for examining the conjecture of @xcite . our analysis is facilitated by the use of the exponential law inducing special scaling properties on the resulting system of partial differential equations . in the mechanics literature shear localization is typically associated with hadamard - instability . this is easily put on display for , by considering the limiting case @xmath50 and @xmath43 . in the parlance of mechanics , this corresponds to rate - insensitive deformations and absence of heat conduction , and leads to the system of conservation laws @xmath56 for @xmath57 the wave speeds are both imaginary , @xmath58 , and the linearized initial value problem for is ill - posed . two questions then emerge : 1 . to assess the combined effect of the competition between hadamard instability and momentum diffusion and/or heat conduction ( as manifested by small @xmath2 and @xmath18 respectively ) . while hadamard instability indicates the catastrophic growth of oscillations around a mean state , it does not by itself explain the formation of coherent structures typically observed in localization . the latter is an outcome of the nonlinear effects and has to be assessed at the level of the nonlinear problem . the first objective is to develop quantitative criteria for the stability of the uniform shearing solutions . since the basic solutions are time dependent an issue arises how to define their stability . following @xcite , we call the uniform shearing solution _ asymptotically stable _ if perturbations of the solution decay faster than the basic solution , and we call it _ unstable _ if they grow faster than the uniform shearing solution . it is expedient to introduce a relative perturbation formulation of the problem : for the _ exponential law _ , we introduce the state @xmath59 describing the relative perturbation and defined through the transformation @xmath60 where @xmath61 is given by and satisfies in particular that @xmath62 using and we derive the system satisfied by the relative perturbation @xmath59 in the form @xmath63 \theta_t - \kappa \theta_{xx } & = \sigma_s ( t ) \ , \big ( \sigma \ , u -1 \big ) , \\[2pt ] \sigma & = e^{-\alpha \theta } u^n . \end{aligned}\ ] ] note that the uniform shearing solution is mapped to the state @xmath64 , @xmath65 and @xmath66 , and the latter is an equilibrium for the system . the problem then becomes to study the stability of the equilibrium @xmath67 for the non - autonomous system . in this section we present an analysis of the linearized system . since the uniform shearing solution is time - dependent the linearized analysis will seek to classify the precise growth in - time of perturbations . we compute the linearized problem , using the ansatz @xmath68 where @xmath69 is a small parameter measuring the size of the perturbation and @xmath70 . expansion of the equation @xmath71 gives to the leading order of @xmath69 @xmath72 while linearizing the other two equations gives @xmath73 collecting the equations together and neglecting the terms of @xmath74 we obtain the linearized system : @xmath75 { \bar \theta}_t - \kappa { \bar \theta}_{xx } & = \sigma_s ( t ) \ , \big ( ( n+1 ) { \bar u}- \alpha { \bar \theta}\big ) , \\[2pt ] { \bar \sigma}&= n { \bar u}- \alpha { \bar \theta}. \end{aligned } \right .\ ] ] the boundary conditions and and the transformations and , imply that the induced boundary conditions for the linearized system are @xmath76 in view of the transformations and , the relation between the original solution @xmath77 and the solution of the linear problem is as follows @xmath78 our analysis of the linearized equation is facilitated by introducing the transformation of variables @xmath79 where @xmath80 is a rescaling of time , defined by @xmath81 the map @xmath80 is invertible and its inverse is given by @xmath82 one computes that @xmath83 satisfy the linear system @xmath84 \frac{{\partial}{\hat \theta}}{{\partial}\tau } - \kappa c_0 e^{\alpha \tau } { \hat \theta}_{xx } & = ( n+1 ) { \hat u}- \alpha { \hat \theta } , \\[2pt ] { \hat \sigma}&= n { \hat u}- \alpha { \hat \theta } , \end{aligned } \right .\ ] ] with boundary conditions @xmath85 for further reference , we record that the original variables @xmath86 and the solution @xmath87 of the rescaled problem - are related through the formulas @xmath88 we next consider the system @xmath89 \frac{{\partial}{\hat \theta}}{{\partial}\tau } - k { \hat \theta}_{xx } & = ( n+1 ) { \hat u}- \alpha { \hat \theta } , \end{aligned } \right .\ ] ] with boundary conditions . note that in the thermal diffusion coefficient @xmath90 is time dependent , while in this coefficient is frozen in time . we analyze the solutions of via an eigenvalue analysis . this provides the following information : 1 . for @xmath91 it provides the stability or instability properties of the linearized system . 2 . for @xmath92 the eigenvalue analysis will only offer an indication of the effect of thermal diffusion on the linearized analysis . this will be complemented by an energy estimate analysis of the exact system in the following section . the solutions of - are expressed as a cosine fourier series @xmath93 where the fourier coefficients @xmath94 satisfy the ordinary differential equations @xmath95 the eigenvalues of the matrix @xmath96 are the roots of the binomial equation @xmath97 * the mode @xmath98*. for @xmath99 the equation becomes @xmath100 and has two eigenvalues @xmath101 and @xmath102 . this corresponds to marginal stability for the linearized system . however , any instability associated with the eigenvalue @xmath101 is eliminated by , which imply for the perturbation @xmath103 in the condition @xmath104 . indeed , the system for @xmath105 takes the form @xmath106 this implies , using , , and the boundary condition , that @xmath107 and clearly @xmath108 decays . thus , the @xmath109-th mode decays . * the modes @xmath110*. for these modes the eigenvalues are @xmath111 where the discriminant @xmath112 since @xmath113 we conclude that @xmath114 in summary : 1 . if @xmath115 then all modes @xmath116 are unstable . 2 . if @xmath92 the first few modes might be unstable , that is the modes @xmath117 for which @xmath118 , but the high frequency modes are stable . 3 . the higher @xmath119 the fewer the number of unstable modes . for @xmath119 larger than a certain threshold ( here equal to @xmath120 ) all modes are asymptotically stable . we analyze the behavior of the linearized modes as a function of @xmath117 and discuss the nature of the instability in the various special cases of interest . the analysis reveals the role of the parameters @xmath121 regarding the nature of the instability . in this regime @xmath122 one eigenvalue is negative and the other strictly positive and increasing with @xmath117 . using the taylor series development @xmath123 we obtain the following asymptotic developments for the eigenvalues @xmath124 referring to the formulas , and , we see that the @xmath117-th mode of the perturbation grows to the leading order like @xmath125 clearly the perturbation grows much faster than the uniform shearing solution @xmath126 given by where the growth is logarithmic , and the behavior is characteristic of _ hadamard instability _ , that is the high frequency oscillations grow at a catastrophic rate . in this case the eigenvalues of the @xmath117-th mode are given by the formulas @xmath127 again we have one negative and one positive eigenvalue and thus unstable response in this regime as well . using once again the taylor expansion we obtain the asymptotic expansions @xmath128 for the positive and negative eigenvalues respectively . the positive eigenvalue @xmath129 satisfies the following properties : @xmath130 set @xmath131 and recall that @xmath132 satisfies the identity @xmath133 differentiating , we derive @xmath134 next , using the elementary bound @xmath135 for @xmath136 we obtain @xmath137 _ + ( x ) & = - + ( n x + ) < < . @xmath137 this shows that @xmath138 and provides and . then follows from . the following remarks are in order : observe first that , by virtue of , and , the @xmath117-th mode of the perturbation grows like @xmath139 clearly the perturbation grows much faster than the uniform shearing solution @xmath126 , but the rate of growth is bounded in @xmath117 . on the other hand , the growth becomes very fast as the rate sensitivity @xmath140 . the behavior in this regime is characterized by _ turing instability_. recall , that turing instability @xcite corresponds to the following scenario that occurs in some instances of dynamical systems . there are two dynamical systems @xmath141 where the spectrum of both matrices @xmath142 and @xmath143 lies in the left - hand plane , but the trotter product of the two systems @xmath144 exhibits unstable response . indeed , the dynamical system with @xmath115 may be visualized as the trotter product of the two systems @xmath145 and @xmath146 each of and has spectrum in the left half space but the eigenvalues of have unstable response . in this regime the eigenvalues are given by and they are real and have the properties @xmath147 therefore , the high modes are stable , but the low modes might be unstable if @xmath119 is sufficiently small . the most unstable " mode is the first mode . one can again carry out the large @xmath117 asymptotics of the eigenvalues . the result , in the case that @xmath148 , reads @xmath149 a similar formula holds when @xmath150 . consider now the linearized system with boundary condition @xmath151 . since we are interested in perturbations of the uniform shearing solution we will consider initial data that satisfy @xmath152 and thus focus on solutions with @xmath153 the objective is now to analyze the effect of the time - dependent diffusion on the dynamics of the system . the linearized analysis of the system with frozen coefficients suggests to expect that diffusion stabilizes the system and its effect is intensified with the passage of time . indeed , here we provide a rigorous proof of this statement via energy estimates . [ lemdecay ] there exist @xmath154 and @xmath155 such that @xmath156 and @xmath157 we multiply @xmath158 by @xmath159 and @xmath160 by @xmath161 and obtain the identities @xmath162 using cauchy - schwarz inequality , gives @xmath163 in view of , we may use the poincare inequality @xmath164 combining the above , we note that for @xmath142 a positive constant we have @xmath165 next we select @xmath142 such that @xmath166 , and deduce that for @xmath142 thus selected there is a choice @xmath167 such that @xmath168 select now @xmath169 such that @xmath170 for @xmath171 ( recall from that @xmath172 when @xmath173 ) . on the interval @xmath171 , we have the differential inequality @xmath174 the latter implies and yields , upon using gronwall s inequality and , @xmath175 and concludes the proof the next lemma provides control of the intermediate times . [ lembound ] there exist @xmath176 and a constant @xmath177 such that @xmath178 using again and we have for any @xmath179 @xmath180 select now @xmath143 large so that @xmath181 for all @xmath182 , and then we obtain @xmath183 from where follows via gronwall s inequality . we conclude now the analysis of stability . suppose that for some @xmath184 the initial data satisfy @xmath185 and are controlled in @xmath186 by @xmath187 then , imply @xmath188 together they imply that the state @xmath189 is stable in @xmath186 , and in fact on account of it is asymptotically stable . we state the result in the following : for the initial - boundary value problem , with @xmath18 , if the initial perturbation @xmath190 satisfies @xmath191 then the equilibrium @xmath189 is asymptotically stable in @xmath186 . the proof of the above lemmas and the analysis of the linearized problem with frozen coefficients indicate that the perturbation of the base solution may drift far from the origin until eventually it is recaptured as the effect of the diffusion intensifies with time . this phenomenon of _ metastability _ also appears in numerical experiments for the nonlinear problem ; an instance of these experiments appears in figure [ mtstb ] . we solve numerically the rescaled system for @xmath192 . the initial conditions are perturbations of the uniform shearing solutions and they are : @xmath193 and @xmath194 , where @xmath195 is a small gaussian perturbation centered at the middle of the interval . the numerical method is based on adaptivity in space and time . an adaptive finite element method is used for the spatial discretisation while a runge - kutta method with variable time step is used for the discretisation in time . further details concerning the numerical scheme and the adaptivity criteria can be found in @xcite . in figure [ mtstb ] the evolution in time ( in natural logarithmic scale ) is presented for the original variables @xmath35 and @xmath37 in ; @xmath37 is sketched in natural logarithmic scale . we notice that , after a transient period of localization , thermal diffusion dominates the process , inhomogeneities of the state variables dissipate , and eventually the solution approaches a uniform shearing solution . ( left ) , @xmath196 ( right).,title="fig:",width=302,height=226 ] ( left ) , @xmath196 ( right).,title="fig:",width=302,height=226 ] we consider next the problem on the whole real line and put aside the boundary conditions . the objective is to construct special solutions that capture a process of localization . this is done over three sections . the present one is preliminary in nature and discusses various properties that make the construction feasible and the setting of the problem . the main step is performed in section [ secss ] where a special class of self - similar solutions are constructed . the localized solutions are then presented in section [ seclocalii ] . we consider on @xmath197 and recall the notation @xmath198 . the uniform shearing solutions are defined by solving . introduce the transformation @xmath199 which involves a passage to relative perturbations ( as in section [ secrelpert ] ) and a change of time scale from the original to a rescaled time @xmath80 . the scaling @xmath80 is precisely the one used in the linearized analysis of sections [ secchatime ] and [ seclinfroz ] ; it is defined by @xmath200 it satisfies @xmath201 where @xmath202 is as in , @xmath203 . the scaling @xmath204 satisfies @xmath205 as @xmath206 , and @xmath80 is invertible with inverse map given by @xmath82 the functions @xmath207 satisfy the system of partial differential equations @xmath208 { \hat \theta}_\tau - \kappa c_0 e^{\alpha \tau } { \hat \theta}_{xx } & = \big ( { \hat \sigma}\ , { \hat u}-1 \big ) , \\[2pt ] { \hat \sigma}&= e^{-\alpha { \hat \theta } } { \hat u}^n . \end{aligned}\ ] ] we emphasize that * for @xmath18 the system is non - autonomous , which is natural in view of the nature of the transformation . * notably , for @xmath43 the system becomes autonomous . next , we consider the adiabatic case @xmath43 where takes the form @xmath209 { \hat \theta}_\tau & = \big ( { \hat \sigma}\ , { \hat u}-1 \big ) , \\[2pt ] { \hat \sigma}&= e^{-\alpha { \hat \theta } } { \hat u}^n \ , . \end{aligned}\ ] ] to understand the long time response of we perform a formal asymptotic analysis following ideas from @xcite . fix an observational time scale @xmath169 and introduce the change of variables @xmath210 one easily checks that @xmath211 satisfies the system @xmath212 \tfrac{1}{t } \ , { \tilde \theta}_s & = \big ( { \tilde \sigma}\ , { \tilde u}-1 \big ) , \\[2pt ] { \tilde \sigma}&= e^{-\alpha { \tilde \theta } } { \tilde u}^n . \end{aligned}\ ] ] we are interested in calculating the equation describing the effective response of for @xmath169 sufficiently large . this asymptotic analysis involves a parabolic equation @xmath158 , which can be thought as a moment equation , and equations @xmath160 and @xmath71 , which may be thought as describing a relaxation process towards the equilibrium curve described by @xmath213 note that the equilibria are a one - dimensional curve parametrized by @xmath214 . to calculate the effective equation we perform a procedure analogous to the chapman - enskog expansion of kinetic theory . we refer to ( * ? ? ? * sec 5 ) for the details of the asymptotic analysis . the procedure produces the following effective equation @xmath215 which captures the effective response up to order @xmath216 . the leading term is backward parabolic and the first order correction is fourth order . one checks that the effect of the fourth order term is to stabilize the equation , in the sense that the linearized equation around the equilibrium @xmath217 is stable ( see @xcite for the details ) . it was further shown in @xcite that admits special solutions that exhibit localization , which is not surprising as the leading response in is backward parabolic . localized solutions of a similar nature will be constructed here directly for the system . this is a more unexpected " behavior , because is at least _ pro - forma _ a hyperbolic - parabolic " system , and the explanation for this behavior lies precisely in the asymptotic relation between and . it is easy to check that for @xmath43 the system is invariant under the one - parameter scaling transformation @xmath218 for any @xmath219 . it is also easy to check that this is the only scaling transformation under which the problem is invariant . this transformation has the distinctive feature that the scaling of space is independent of the scaling of time . finally , the system is not scaling invariant when the thermal diffusion @xmath49 is positive . the scaling invariance motivates a change of variables and a related _ ansatz _ of solutions . introduce a variable in time scaling @xmath220 , the variable @xmath221 , and the change of variables @xmath222 we denote by @xmath223 and by @xmath224 . introducing to , we obtain that the function @xmath225 satisfies the system of partial differential equations @xmath226 { \bar \theta}_\tau - \frac{\dot r}{r } \left ( \tfrac{n+1}{\alpha } + \xi { \bar \theta}_{\xi } \right ) , - \kappa c_0 \frac{e^{\alpha \tau}}{r^2 ( \tau ) } { \bar \theta}_{\xi \xi } & = \big ( { \bar \sigma}\ , { \bar u}-1 \big ) , \\[2pt ] { \bar \sigma}&= e^{-\alpha { \bar \theta } } { \bar u}^n . \end{aligned}\ ] ] the resulting system depends on the choice of the scaling function @xmath227 . we will be interested in solutions that are steady states in the new variables . of course such solutions are dynamically evolving in the original variables . there are two cases that produce simplified results of interest : 1 . consider the case @xmath18 , and select @xmath228 . then the system is consistent with the ansatz of steady state solutions @xmath229 and the latter are constructed by solving the system of ordinary differential equations @xmath230 -\frac{\alpha}{2 } \left ( \tfrac{n+1}{\alpha } + \xi { \bar \theta } ' \right ) - \kappa c_0 { \bar \theta } '' & = { \bar \sigma}\ , { \bar u}-1 , \\[2pt ] { \bar \sigma}&= e^{-\alpha { \bar \theta } } { \bar u}^n . \end{aligned}\ ] ] + since @xmath231 as @xmath232 increases , solutions of will produce defocusing solutions for . a study of is expected to give information about the convergence to the uniform shearing state in the non adiabatic setting . this is not the focus of the present work and we will not deal further with the system . 2 . instead we consider the case @xmath43 . we may now select any exponential function for @xmath227 but are most interested in @xmath233 with @xmath24 as this produces focusing solutions . with this selection , @xmath229 is a steady solution for with @xmath43 provided it satisfies @xmath234 \lambda \left ( \tfrac{n+1}{\alpha } + \xi { \bar \theta } ' \right ) & = { \bar \sigma}\ , { \bar u}-1 , \\[2pt ] { \bar \sigma}&= e^{-\alpha { \bar \theta } } { \bar u}^n , \end{aligned}\ ] ] where @xmath24 is a parameter . note that : * the form of the transformation is motivated by the scaling invariance . * the scaling connects the two variables and produces interesting systems even in the case @xmath49 positive when the system is not scale invariant . we are ready to state and prove the main result of the paper . this concerns the adiabatic case @xmath378 of system , taken now on the whole real line @xmath379 and expressed in terms of the variable @xmath19 in the form : @xmath380 our analysis is based on the transformations of section [ seclocal ] and the self - similar solution constructed in section [ secss ] . [ shearband ] let @xmath381 and @xmath247 be given . for any @xmath24 , the system admits a special class of solutions describing localization of the form : @xmath382 \label{locsigma } \sigma ( x , t ) & = \sigma_s ( t ) \left ( \tfrac{\alpha}{c_0 } t + 1 \right ) ^ { - \tfrac{\lambda}{\alpha } } \sigma_\lambda \left ( \sqrt{\lambda } x \left ( \tfrac{\alpha}{c_0 } t + 1 \right ) ^{\tfrac{\lambda}{\alpha } } \right ) , \\[2pt ] \label{loctheta } \theta(x , t ) & = \left ( 1 + \lambda \tfrac{n+1}{\alpha } \right ) \theta_s ( t ) - \lambda \tfrac{n+1}{\alpha } \theta_0 + \theta_\lambda \left ( \sqrt{\lambda } x \left ( \tfrac{\alpha}{c_0 } t + 1 \right ) ^{\tfrac{\lambda}{\alpha } } \right ) , \end{aligned}\ ] ] where @xmath46 , @xmath45 are the uniform shear solutions in , @xmath203 , while @xmath383 is an even function defined on @xmath384 , which solves the initial value problem , with @xmath385 for @xmath386 . [ rmkbeh ] 1 . 2 . because @xmath387 goes to infinity in time , the level curves of @xmath388 accumulate along the @xmath389 axis in the @xmath390 plane . the functions , , thus describe a localizing solution where the flow tends to concentrate around @xmath33 as the time proceeds . the solution , , is defined on @xmath391 and emanates from initial data @xmath392 \sigma ( x,0 ) & = \sigma_s ( 0 ) \sigma_\lambda \big ( \sqrt{\lambda } x \big ) , \\[2pt ] \theta(x,0 ) & = \theta_s ( 0 ) + \theta_\lambda \big ( \sqrt{\lambda } x \big ) \ , . \end{aligned}\ ] ] the initial data depend on two parameters : the parameter @xmath28 which can be thought as a length scale in initial data , and @xmath29 in which describes the amplitude of the initial perturbation . this parametric family of solutions is valid for any @xmath24 . if @xmath28 is large then the length scale of the initial perturbation is short while the growth in time and the rate of localization for the solution become fast . we introduce the changes of variables and the stationary variant of , which once combined together reads @xmath393 we select @xmath394 with @xmath24 . according to the analysis in section [ secinvariance ] the function @xmath395 is sought as a solution of defined on the interval @xmath396 . the desired @xmath395 is constructed as follows . first , observe the relation between the system and the system . let @xmath29 be fixed , and let @xmath397 be the solution of with @xmath26 and , defined for @xmath398 . this solution is well defined by theorem [ connection ] . using a direct computation , the smoothness properties in ( iii ) of theorem [ connection ] , and we conclude that 1 . the function @xmath399 defined by @xmath400 satisfies on @xmath401 and the condition @xmath402 . if we extend @xmath399 on @xmath403 by setting @xmath404 for @xmath405 , the resulting function @xmath406 , @xmath407 is a solution of defined on the whole real line that is even . noting that @xmath408 we combine with and to arrive at , , . in figure [ uqsol ] the profiles of @xmath409 and @xmath410 in and , respectively , are drawn at various instances of time from @xmath411 to @xmath412 . to achieve these profiles , we construct numerically the heteroclinic orbit for whose existence is justified in proposition [ thode ] . the construction is simple , if one exploits the special properties of system , and proceeds as follows : initial conditions @xmath413 are selected close to the equilibrium @xmath303 inside the region @xmath311 and in the direction of the stable manifold of @xmath284 for the linearized problem . then is solved backward in time , using an ode solver , and since @xmath414 is an unstable node ( for the forward problem ) the resulting orbit is a good approximation of the heteroclinic . for the integration of we use an explicit runge kutta predictor - corrector method of order ( 4,5 ) with adaptive time stepping . then @xmath29 is selected by choosing an appropriate parametrisation @xmath344 following the construction in . , and @xmath410 , .,title="fig:",width=302,height=226 ] , and @xmath410 , .,title="fig:",width=302,height=226 ] for the numerical runs the constitutive parameters are @xmath415 , @xmath416 . the parameters related to the solution are @xmath417 , @xmath418 while for the selected reparametrization of the heteroclinic orbit the value of @xmath419 . once the heteroclinic has been constructed , the profiles @xmath420 are computed via the changes of variables and . finally , @xmath409 and @xmath410 are computed by , respectively with @xmath417 , and are presented in fig [ uqsol ] . in this section we consider the system of ordinary differential equations @xmath235 \nu \left ( \tfrac{n+1}{\alpha } + \xi \theta ' \right ) & = \sigma \ , u -1 , \\[2pt ] \sigma & = e^{-\alpha \theta } u^n \ , , \end{aligned}\ ] ] on the domain @xmath236 with @xmath237 , subject to the initial condition @xmath238 this is an auxiliary problem that will be studied in detail in this section . we emphasize that it does not coincide with the integrated form of , however it will be used together with additional properties in order to construct solutions of and localizing solutions for in the following section . in this section , we are interested in constructing solutions @xmath239 for the singular initial value problem , for any @xmath237 . the following elementary remarks are in order : 1 . when @xmath240 , simplifies to @xmath241 \sigma ( 0 ) & = \sigma_0 \end{aligned } \right . \left \ { \begin{aligned } \sigma \ , u & = 1 \\[2pt ] \theta & = \frac{n+1}{\alpha } \log u \end{aligned } \right . .\ ] ] the solution of is computed explicitly by @xmath242 2 . when @xmath243 , the system is invariant under the family of scaling transformations @xmath244 for @xmath245 . there is a special solution of that is self - similar with respect to the scaling invariance and reads @xmath246 this solution does not however satisfy the initial condition . we next construct a function @xmath239 which satisfies , for all values of @xmath237 and @xmath247 . we will show that the constructed solution will have the inner behavior of and the outer behavior of . [ connection ] let @xmath247 be fixed . let us note @xmath248 . given @xmath237 there exists a solution @xmath249 of , defined for @xmath250 and satisfying the properties : 1 . it achieves the initial data @xmath251 where @xmath29 as in and @xmath252 , @xmath253 are defined via @xmath254 2 . as @xmath255 it has the limiting behavior @xmath256 \theta ( \xi ) & = - \tfrac{n+1}{\alpha } \log \xi + o(1 ) . \end{aligned}\ ] ] 3 . the solution has the regularity @xmath257 , @xmath258 , @xmath259 . it satisfies @xmath260 and has the taylor expansion as @xmath261 , @xmath262 where as usual @xmath263 ( but the rate may degenerate as @xmath264 tends to @xmath109 ) . _ step 1 . _ motivated by the special solution we introduce a change of variables in two steps : @xmath265 \theta ( \xi ) & = - \tfrac{n+1}{\alpha } \log \xi + { \bar \theta } ( \xi ) & & = - \tfrac{n+1}{\alpha } \log \xi + \theta ( \log \xi ) , \nonumber\end{aligned}\ ] ] indeed , we easily check that @xmath266 satisfies the system of equations @xmath267 \nu \xi \ , { \bar \theta } ' & = { \bar \sigma } \ , { \bar u } -1 , \\[2pt ] { \bar \sigma } & = e^{-\alpha { \bar \theta } } { \bar u}^n \ , . \end{aligned}\ ] ] in the second step , we introduce in the change of independent variable @xmath268 and derive that @xmath269 satisfy @xmath270 \nu \frac{d\theta}{d\eta } & = \sigma \ , u -1 , \\[2pt ] \sigma & = e^{-\alpha \theta } u^n \ , . \end{aligned}\ ] ] the system is far simpler than the original in that it is autonomous and non singular . we next consider and perform a further simplification , via the change of dependent variables @xmath271 note that @xmath272 . we now compute @xmath273 and @xmath274 in summary , we conclude that @xmath275 satisfies the autonomous system of ordinary differential equations @xmath276 where @xmath277 there are two advantages in relative to : 1 . is expressed via only two independent variables @xmath278 . the equilibrium at infinity in is pulled at the axis for via the transformation . we perform a phase space analysis of and establish the existence of a heteroclinic connection . we show : [ thode ] consider the autonomous planar system of differential equations with @xmath279 defined by . for any @xmath237 the system has the following properties : * @xmath280 has two equilibria @xmath281 , @xmath282 in the first quadrant . both @xmath283 and @xmath284 are hyperbolic points , @xmath283 is a repelling node while @xmath284 is a saddle point . * there exists a heteroclinic orbit connecting @xmath283 to @xmath284 . _ proof of proposition [ thode ] . _ 1 . the system has the special solution @xmath285 this solution splits the phase portrait into the left and right plane and acts as a barrier from crossing from the one side to the other . the equilibria of are computed by solving @xmath286 and they are @xmath287 and @xmath288 and @xmath289 . focusing on the behavior in the first quadrant we neglect the equilibrium @xmath290 . a computation shows @xmath291 -2 \tfrac{n+1}{n } a & \tfrac{\alpha}{n\nu } c_\nu \end{pmatrix } .\ ] ] at the equilibrium @xmath283 we have @xmath292 and we have eigenvalues @xmath293 with eigenvector @xmath294 , and @xmath295 with eigenvector @xmath296 . in view of , we have @xmath297 . the local behavior of solutions of around the equilibrium @xmath283 is that of a repelling node . standard phase plane theory for second order systems ( _ e.g. _ @xcite ) states that there is a small neighbourhood of the equilibrium @xmath283 such that for any given solution @xmath298 of there are constants @xmath299 , @xmath300 such that @xmath301 as @xmath302 . 4 . at the point @xmath303 equation gives @xmath304 the eigenvalues @xmath305 , computed by @xmath306 are @xmath307 with eigenvector @xmath308 and @xmath309 with eigenvector @xmath310 . @xmath284 is a saddle point . 5 . consider the region @xmath311 bounded by the curves * @xmath312 is the @xmath313-axis defined by the equation @xmath314 , * @xmath315 is the horizontal line @xmath316 , * @xmath317 is the parabola @xmath318 . + the region @xmath311 is expressed as @xmath319 see figure [ fig : phase2 ] . + + we will show that @xmath311 is negatively invariant . + indeed , the flow of is parallel to @xmath312 as indicated in ( b ) . on the horizontal line @xmath320 the system gives @xmath321 on the curve @xmath322 the system gives @xmath323 therefore @xmath311 is negatively invariant . 6 . for completeness we also show that the vector @xmath324 associated to the eigenvalue @xmath325 at @xmath284 points towards the interior of the region @xmath311 . recall that @xmath284 is the upper right corner of the region @xmath311 and that * the tangent vector to @xmath315 is @xmath326 , * the tangent vector to @xmath317 at @xmath303 is @xmath327 , * the eigenvector @xmath328 . + we will show that @xmath329 this then implies that the vector @xmath330 when placed at the corner @xmath284 points strictly towards the interior of @xmath311 . to show recall that @xmath307 and use and to rewrite @xmath331 where @xmath332 this shows @xmath333 and concludes . it is easy to check that there can be no closed limit cycle in @xmath311 because in that region @xmath334 the poincar - bendixson theorem implies that there exists a heteroclinic orbit joining @xmath335 and @xmath336 which emanates along the direction of the stable manifold at @xmath336 and progresses backwards in time towards the node at @xmath335 . we turn now to defining the solution of , . let @xmath337 stand for a parametrization of the heteroclinic orbit . in view of and the fact that @xmath338 , there exists a constant @xmath299 such that @xmath339 due to the fact that the vertical axis is a trajectory of its own and thus distinct from the trajectory we are considering , and due to the form of the phase portrait we can assert that @xmath340 . for any @xmath341 the function @xmath342 offers an alternative parametrization of the same heteroclinic orbit . applying to this parametrization yields @xmath343 choose @xmath344 such that @xmath345 and let @xmath298 denote the corresponding parametrization of the heteroclinic : @xmath346 the selected parametrization satisfies as @xmath302 @xmath347 where as usual @xmath348 as @xmath302 . recalling the changes of variables and we define @xmath349 the function clearly satisfies the system and , since @xmath350 as @xmath351 , it satisfies @xmath352 \theta ( \xi ) & = - \tfrac{n+1}{\alpha } \log \xi + o(1 ) , \end{aligned}\ ] ] as @xmath353 goes to infinity , which shows . in the last step , we study the behavior near the ( apparent ) singular point at @xmath354 and prove that the solution is in fact regular there and satisfies , and . clearly and @xmath160 imply that @xmath355 as @xmath261 . let @xmath252 and @xmath253 be defined by . since @xmath356 , gives @xmath357 now , , , imply @xmath358 and together with and provide . observe now that @xmath359 using , and , we obtain @xmath360 \\ & = \tfrac{1}{n } u^3 b - u \tfrac{\alpha}{\nu n } \frac{c_\nu}{b } \big ( b - \frac{1}{c_\nu } \big ) \\ & = \tfrac{1}{n } \tfrac{u_0 ^ 3}{c_\nu } e^{3 \eta}(1 + o(1 ) ) - \tfrac{\alpha}{\nu n } u_0 o(1 ) e^{2\eta } \qquad \mbox{as $ \eta \to -\infty$ } \end{aligned}\ ] ] and thus @xmath361 since @xmath362 \cap c^1(0,1)$ ] the mean value theorem implies @xmath363 and implies @xmath364 exists and @xmath365 . we conclude that @xmath366 has the taylor expansion @xmath158 . in turn , @xmath367 , @xmath368 , @xmath369 and @xmath370 achieves the taylor expansion @xmath160 . turning to @xmath371 , observe that by and @xmath71 @xmath372 while @xmath373 in addition , @xmath374 combined with the formula @xmath375 gives @xmath376 this shows @xmath377 has the taylor expansion @xmath71 at the origin . we considered the system as a paradigm to study the onset of localization and formation of shear bands and to examine the role of thermal softening , strain - rate hardening and thermal diffusion in their development . the stability of the uniform shear solution leads , in general , to the analysis of a non - autonomous system . the present model has the remarkable property that the system of relative perturbations is autonomous in the adiabatic case ( @xmath421 ) . a study of linearized stability for the case with no thermal diffusion @xmath378 shows that : ( i ) for @xmath17 the high frequency modes grow exponentially fast , with the exponent proportional to the frequency , what is characteristic of ill - posedness and hadamard instability . ( ii ) when @xmath2 the modes are still unstable but the rate of growth is bounded independently of the frequency . the behavior in that range is that characteristic of turing instability . the effect of thermal diffusion ( @xmath18 ) for the non - autonomous linearized model was also assessed ; it was shown that perturbations grow in the early stages of deformation , but over time the compound effect of diffusion stabilizes the response of the system . then we turned to the nonlinear system and constructed a class of self - similar solutions on the real line , leading to the explicit solutions , , . their construction is based on the self - similar profiles of , , in turn determined by a suitable heteroclinic connection . they depend on three parameters : @xmath27 linked to the uniform shear in , a parameter @xmath28 which can be viewed as a length scale of initial data , and @xmath29 determining the size of the initial profile . the solutions presented in fig [ uqsol ] evidently exhibit localisation : as time increases a coherent structure forms with the deformation localizing in a narrow band . in this self - similar solution there is no trace of the oscillations . a comparison with the modes of the linearized problem indicates that the nonlinearity suppresses the oscillations and a seamless coherent structure emerges as the net outcome of hadamard instability , rate dependence , and the nonlinear structure of the problem . , _ localization and shear bands in high strain - rate plasticity_. in nonlinear conservation laws and applications , a. bressan , g .- q . chen , m. lewiscka and d. wang , eds ; i m a vol . math . appl . , 153 , springer , new york , 2011 , pp 365 - 378 .

given some radiation field , a prediction can be made about the outcome of an experiment where a suitable detector is placed at some position @xmath0 and probes the radiation field for some time @xmath1 . even when the radiation field is known as precisely as possible , there is still a finite uncertainty in the outcome . this statement can also be formulated in the inverse way : it is not possible to build an apparatus that provides a `` better '' ( = more defined ) illumination at the point @xmath0 than this limit . we will demonstrate in this paper that the uncertainty of the photocount , quantified by its variance , depends on both the number of modes , including the energy distribution among the modes , and on the spectral properties and quantum correlations of the radiation field . another contribution to the uncertainty will appear if the radiation field is known at the source but the mode structure is then perturbed in some uncontrolled way before it reaches the location of the detector . the latter effect is frequently referred to as static speckle @xcite named after the pattern it creates on a camera . the opposing term `` dynamic speckle '' unfortunately is used to denote two completely different concepts in the literature . frequently it refers to a static speckle pattern that is changing over time because the source of the perturbation is moving @xcite but in this paper we refer to `` dynamic laser speckle '' @xcite an effect that is due to the quantum dynamics of the radiation field . quantum effects are most prominent on small length scales . modern optical lithography operates precisely in this regime , printing structures smaller than the wavelength of the light with a precision of a few nanometres @xcite . to this end , photo resists have been developed that are more sensitive to changes of the light intensity than most technical sensors , thereby acting as efficient ( albeit unintentional ) detectors for uncertainties of the photocount . later in this paper we will show that the effects of multi - mode multi - frequency radiation are most pronounced neither at the limit of large or short times but in the intermediate regime . recent studies have confirmed that this is precisely the regime used in modern optical lithography @xcite . classically a radiation field is described by its electric field @xmath2 as function of position @xmath3 and time @xmath4 . in a quantum treatment , the electric field @xmath5 is replaced by a suitable operator . in both treatments , the electric field is not a perfectly determined quantity , and the maximum knowledge possible is contained in the mutual - coherence function @xcite @xmath6 we restrict ourselves to stationary fields such that @xmath7 does not depend on @xmath8 and @xmath9 but only on the time - difference @xmath10 . furthermore , for ease of writing we will only treat one component @xmath11 of the electric field @xmath5 but the extension to cover the other components is straight - forward . the fourier transform @xmath12 of @xmath13 with respect to the time difference is a nonnegative hermitian operator , and thus possesses an eigenrepresentation @xcite @xmath14 the eigenvectors @xmath15 are called the modes of the electric field and form a complete orthonormal set . we assume that the frequency spectrum is small enough such that @xmath15 is independent of @xmath16 . whenever the frequency spectrum is equal to the natural linewidth of the light source , i.e. , the spectrum is due to the finite lifetime of some excited light - emitting medium such as in a laser , this assumption is always fulfilled . for other light sources , this assumption can sometimes be problematic for cavity - like systems but for open system or in a waveguide geometry , this is less of an issue . if the condition that @xmath15 is independent of @xmath16 , should be violated , one can split the mode into several discrete modes ( one for each frequency interval in which the shape of the mode can assumed to be constant ) and the remainder of this paper be applied nonetheless . the electric field can then be written as @xmath17 since @xmath18 is independent of @xmath4 , the @xmath18 are the modes instead of being some arbitrary base of the electric field . comparison of eqs . ( [ eqgamma2 ] ) and ( [ eqgamma3 ] ) shows that the magnitude of @xmath19 can be computed from @xmath20 but not its phase . a semiclassical treatment of the photocount statistics would assume that @xmath20 is fluctuating in time whereas within quantum optics , which we will apply in this paper , fluctuations are inherent to the operator description . this paper is organised as follows . in sec . [ seczeitunschaerfe ] we compute the variance of the photocount on a quantum - optical level when the radiation field is completely known . we will find that there is a shot noise term , a quantum correlation term and a term describing the beating of different frequencies . frequently , by design or unintentionally , the mode structure emitted by a known light source is completely changed and unknown when it reaches the point of the detector . as will be demonstrated in sec . [ secstaticspeckle ] , random - matrix theory allows a compact and exact treatment of this problem . the resulting `` static speckle '' becomes the smaller the more modes of the radiation field are excited , and the more uniform the energy distribution among the modes is . depending on the `` size '' of the radiation field , there is thus a minimum amount of static speckle , which is computed in sec . [ secetendue ] . in sec . [ secstatickorrelation ] we extend this question to computing the most likely amount of static speckle . since we will demonstrate this quantity to be self - averaging , the computed average is more than just an average in that it is characteristic for almost all individual speckle values . while in `` general '' electrodynamics the electric field @xmath2 can be probed directly , this is not possible in the realm of optics as @xmath2 changes too quickly in time and space to allow a direct measurement . rather , the radiation field is probed by means of photodetection , i.e. , by absorbing photons inside some device and counting the number of photons absorbed . this principle applies to technical machines as well as to the human eye or to photo resists . on small length and time scales , quantum effects become important . for this reason , and because all effects can then be treated in a more compact way , we will use the quantum theory of photodetection in the following . this theory @xcite is usually formulated for single - mode detection . since multi - mode fields @xcite are at the core of this paper , we will present an extension of the theory to multi - mode photodetection here . we will treat the multi - frequency aspect explicitly and not express the different frequencies by different modes as is frequently done in textbooks . we label the modes of the electromagnetic field by @xmath21 . the annihilation operator associated with this mode is @xmath19 . using this notation , the quantum operator for the electric field at some position @xmath0 is given by @xcite @xmath22 later it will prove helpful to switch from the time representation @xmath23 to the spectral representation @xmath24 , @xmath25 since in time representation , operators taken at different times do not necessarily commute whereas they do in the spectral representation , @xmath26=\delta_{nm}\delta(\omega-\omega ' ) \;. \label{eqvertauschungsrelation}\ ] ] when we label the number of photons in the @xmath27-the mode with @xmath28 , which basically amounts to the total intensity of that mode , this gives the expectation value @xmath29 where @xmath30 is the ( normalised ) spectrum of the radiation in the @xmath27-th mode , and the prefactor @xmath31 has been introduced for later convenience . photodetection within some time interval @xmath1 is then described by the quantity @xcite @xmath32 where @xmath33 marks the detection efficiency and includes information on the size of the detector . this amounts to a perturbative description of the interaction between the detector and the radiation field . the perturbative approach neglects the effect that every detected photon decreases the number of photons remaining in the radiation field @xcite but this is mainly an issue for microcavities where only a small number of photons are excited at one point in time and the dynamics are then studied . for the purpose of this paper , this is no relevant restriction . the factorial moment @xmath34 of the photodetection count distribution is given by @xmath35 where the colons denote normal - ordering of the operator within , and the brackets @xmath36 denote the average which has to be taken over both the quantum fluctuations of the operators @xmath19 and the frequency distribution @xmath30 . mean and variance of the photo count then follow from @xmath37 ^ 2\;. \label{eqdefvar}\ ] ] for the first moment , this yields the result @xmath38 this is the same result as expected by classical theory since @xmath39 corresponds to the intensity of the @xmath27-th mode at position @xmath0 . the second factorial moment is @xmath40 } \;.\end{gathered}\ ] ] taking the average over the quantum mechanical expectation operators gives nonzero contributions for three cases of the indices @xmath27 : @xmath41 , @xmath42 , and @xmath43 , thus @xmath44 ^ 2 [ a_{n}(\omega)]^2 \rangle \\ + \frac{\eta^2}{4\pi^2}\int\!\!\!\int_0^t\mathrm{d}t_{1,2 } \int\mathrm{d}\omega_{1,3 } \sum_{n_1\ne n_3 } { \lvert\phi_{n_1}(\vec{r})\rvert}^2{\lvert\phi_{n_3}(\vec{r})\rvert}^2 \langle a^{\dagger}_{n_1}(\omega_1 ) a_{n_1}(\omega_1)a^{\dagger}_{n_3}(\omega_3)a_{n_3}(\omega_3 ) \rangle \\ + \frac{\eta^2}{4\pi^2}\int\!\!\!\int \mathrm{d}\omega_{1,2 } \int\!\!\!\int_0^t\mathrm{d}t_{1,2 } \sum_{n_1\ne n_2 } { \lvert\phi_{n_1}(\vec{r})\rvert}^2{\lvert\phi_{n_2}(\vec{r})\rvert}^2 \langle a^{\dagger}_{n_1}(\omega_1 ) a_{n_1}(\omega_1)a^{\dagger}_{n_2}(\omega_2)a_{n_2}(\omega_2 ) \rangle \mathrm{e}^{\mathrm{i}\bigl[(\omega_1-\omega_2)t_1 + ( \omega_2-\omega_1)t_2\bigr ] } \;. \label{eqlang1}\end{gathered}\ ] ] this equation can be simplified by inserting the expectation values from eq . ( [ eqerwart1 ] ) . the double integral over @xmath8 and @xmath9 in the last line of eq . ( [ eqlang1 ] ) can be reduced to a single integral via @xmath45\;. \label{eqttransform}\ ] ] the remaining two integrations over @xmath46 and @xmath47 in that line each yield the fourier transform @xmath48 of @xmath49 , @xmath50 this thus gives @xmath51 ^ 2 [ a_{n}(\omega)]^2 \rangle \\ + \eta^2 t^2 \sum_{n_1\ne n_3 } { \lvert\phi_{n_1}(\vec{r})\rvert}^2{\lvert\phi_{n_3}(\vec{r})\rvert}^2 i_{n_1 } i_{n_3 } \\ + \frac{\eta^2}{2\pi } \int_{-t}^t\mathrm{d}t \sum_{n_1\ne n_2 } { \lvert\phi_{n_1}(\vec{r})\rvert}^2{\lvert\phi_{n_2}(\vec{r})\rvert}^2 \\ \times i_{n_1 } i_{n_2 } \hat{g}_{n_1}(t ) \hat{g}^*_{n_2}(t ) [ t-{\lvertt\rvert } ] \;. \label{eqlang4}\end{gathered}\ ] ] the summation over @xmath52 is rather inconvenient . thus , we include the missing terms @xmath53 in that summation , yielding the value @xmath54 ^ 2 $ ] , and correct for this by subtracting in the first summation the terms just added . there now appears the quantity @xmath55 ^ 2 [ a_{n}(\omega)]^2 \rangle - \langle a^{\dagger}_{n}(\omega ) a_{n}(\omega ) \rangle^2 \ ; , \label{eqdeff}\ ] ] that quantifies the deviation of the radiation in the @xmath27-th mode from a coherent state . for coherent radiation , @xmath56 since then @xmath57 ^ 2 [ a_{n}(\omega)]^2 \rangle=\langle a^{\dagger}_{n}(\omega ) a_{n}(\omega ) \rangle^2 $ ] whereas for any gaussian light source , in particular any object emitting thermal radiation , @xmath57 ^ 2 [ a_{n}(\omega)]^2 \rangle=2 \langle a^{\dagger}_{n}(\omega ) a_{n}(\omega ) \rangle^2 $ ] , and @xmath58 . for classical radiation , @xmath59 whereas for certain nonclassical radiation @xmath60 is possible . collecting results , using eq . ( [ eqdefvar ] ) , gives the variance of the photocount @xmath61 \;. \label{eqlang3}\end{gathered}\ ] ] equation ( [ eqlang3 ] ) constitutes a core result of this paper . it gives the general expression for the noise that is intrinsic to any measurement of a multi - mode multi - frequency radiation field over some time @xmath1 . the first term is the well - known shot noise term that appears naturally in the quantum treatment : electric field and photocount are quantum - optically related via the factorial moment , see eq . ( [ eqfactorialmoment1 ] ) , whereas in classical optics the relation is via the `` plain '' moment , with the difference between factorial and plain moments being precisely the shot - noise term . the second term , quantified by @xmath62 from eq . ( [ eqdeff ] ) , marks the excess noise when the radiation field is not in a coherent state . as mentioned above , @xmath56 for coherent radiation , and @xmath63 for thermal radiation . even when @xmath62 is small , any nonzero @xmath62 will eventually become the dominating effect for the photonoise as its contribution increases quadratically with time , in contrast to all other contributions . the cross - over point , where this term becomes larger than the shot - noise term , is easily estimated from eq . ( [ eqlang3 ] ) as @xmath64 . for any thermal radiation this regime thus is already entered once the measurement time is long enough for one photon per mode to be counted on average . @xmath28 can be computed from the bose - einstein factor , evaluated at the temperature @xmath65 of the source , @xmath66 this value depends on the ratio of frequency and temperature , and thus has the same value @xmath67 for both a light bulb emitting visible light and an euv plasma source operating at up to @xmath68 to emit radiation with @xmath69 . the third and final contribution in eq . ( [ eqlang3 ] ) quantifies the beating of different frequency components when the measurement time is finite . it is sometimes referred to as dynamic laser speckle since it is most relevant for measurement times @xmath1 not much larger than the coherence time @xmath70 of the radiation , @xmath71 hence the name `` temporal degree of coherence '' frequently given to @xmath48 . however , the factor @xmath72 $ ] is also relevant , and it would be wrong to replace the integral simply by @xmath73 , as we will now demonstrate . the extra factor @xmath72 $ ] , which seems to be largely ignored in phenomenological literature , gives less weight to the frequently surprisingly large tails of @xmath48 . two frequently encountered spectral distributions are gaussian and lorentzian , @xmath74 the curves above are already normalised to @xmath75 , i.e. , their integral ( not their square integral ) is @xmath75 . the widths of the spectral distributions have been chosen such that the coherence time , computed from eq . ( [ eqdeftau ] ) , is the same value @xmath70 , allowing for a direct comparison of these two curves . their fourier transforms are @xmath76 \;,\\ \hat{g}_{\mathrm{lorentz}}(t ) & = \frac{1}{\sqrt{2\pi } } \exp\bigl [ -\frac{{\lvertt\rvert}}{2\pi\tau } \bigr ] \;. \label{eqkurvenformen}\end{aligned}\ ] ] the results for entering these expressions into the integral in eq . ( [ eqlang3 ] ) are shown in fig . [ figkurvenformen ] . there is a significant dependence on the spectral shape , with the longer tails of the lorentzian leading to a slower approach to saturation . please note that the graph corresponds to the variance of the photocount . when the square root of the variance is scaled by the mean intensity this quantity is called the contrast the saturating curve shown in the figure turns into a curve decreasing to zero as @xmath1 becomes larger . @xmath77 is a plain exponential and thus conforms best to the simple model of how to model finite temporal coherence . we thus have used this curve for contrasting eq . ( [ eqlang3 ] ) by an approximation where the factor @xmath78 is replaced by @xmath1 , thereby ignoring the additional dependence on @xmath79 . for all values of @xmath1 , the integral increases by about @xmath80 , making the factor @xmath78 important for correct modelling of the photon counting statistics . ) that quantifies the strength of the dynamic speckle . the two curves at the bottom follow by inserting the expressions for a gaussian and a lorentzian spectrum from eq . ( [ eqkurvenformen ] ) . when the term @xmath81 in eq . ( [ eqlang3 ] ) is replaced by @xmath1 , and an exponentially decreasing @xmath48 is assumed , the third curve follows . ] a final word about the cross - over from `` dynamics '' to `` statics '' seems to be in order . since the temporal effect begins to saturate at time @xmath70 it is frequently assumed that the nontemporal regime is entered once @xmath82 . while this statement is correct for single - mode radiation fields , it is incorrect here since the shot - noise term @xmath83 , setting the reference value , scales linearly with the number @xmath84 of modes whereas the prefactor of the temporal term scales as @xmath85 such that the condition @xmath82 has to be replaced by @xmath86 . incidentally , since it is the regime @xmath87 that current high - power excimer lasers are operating in @xcite , this distinction is of actual importance . similar to classical mechanics , theoretical optics knows the concept of phase space @xcite . the optical phase space is spanned by position and spatial frequency , informally called @xmath88-vector , and the radiation field is completely described by the ( pseudo)-density function @xmath89 known as the wigner function . any radiation field then occupies a certain volume in phase space . unfortunately , there is no good mathematical metric to actually measure or at least define such a volume . in experimental and technological areas of optics , in contrast , the volume of phase space occupied is well - defined and is referred to as etendue . the energy density @xmath90 of the light , given as a function of position and angle , is assumed to be larger than zero only inside a finite and well - defined region . this allows for an easy and direct definition of the etendue of the radiation field . etendue is an important quantity since , at least within geometrical optics , it can only be changed by vignetting the radiation field . the etendue supplied by some radiation field incident on some optical apparatus should thus be no larger than the etendue that this apparatus can accept . convolving the wigner function @xmath89 with a small kernel , its size given by an uncertainty - relation @xcite , yields @xmath90 when the relation @xmath91 between angle and @xmath88-vector is utilised . thus on first sight it might seem that the relation between microscopic and macroscopic definition would be straight - forward . the essential difference is that @xmath89 has no finite support whereas @xmath90 is assumed to have . it can be shown that due to the uncertainty principle , every mode @xmath18 needs to have infinite tails in at least either real space ( position ) or @xmath88-space ( angle ) @xcite , and this is reflected in @xmath89 . of course , also @xmath90 has the same infinite tails but in the macroscopic world , the function values inside those tails drop down so quickly that they are `` assumed '' to be zero once they are so low that they have become irrelevant for practical purposes . we thus need a definition for phase space volume that is based on a microscopic radiation field but reduces to the macroscopic etendue for `` large '' radiation fields . for the purpose of this paper , the relevant way of quantifying phase space volume is by the number of modes that can be fit into it . we will tackle the problem in the opposite way : we pick a certain number of `` sensible '' base functions and compute all possible radiation fields that can be formed from them . we will then identify for the intensity profiles @xmath92 as function of position and @xmath93 as a function of direction the properties that allow to deduce the number of base functions used . to have shorter mathematical expressions , we will express directions in terms of the @xmath88-vector instead of the angle @xmath94 . furthermore we make two assumptions . first , we restrict ourselves to one spatial dimension but the extension is straight - forward and becomes trivial when the radiation field factorises in its spatial dimensions . second , we assume that the intensity distribution @xmath95 factorises into a spatial term @xmath96 and an angular term @xmath97 . since the etendue accepted by practically any real - life apparatus factorises , these assumptions do not limit the applicability of this approach . we start with @xmath98 orthonormal base functions @xmath99 , the orthogonality is essential since this prevents the base functions from being too similar and thus collapsing into a very small volume of phase space . furthermore , all modes of the radiation field have to be mutually orthogonal anyhow as they are eigenfunctions of a hermitian operator , cf . eq . ( [ eqgamma2 ] ) we assume them to be normalised for the ease of the calculation . for every @xmath99 , there also exists its fourier transform @xmath101 to progress , we need to choose a certain set of base functions . the aim is to pack the base functions as tightly as possible in phase space , and without loss of generality we try to pack them around the origin . if all @xmath18 , @xmath100 , are assumed to fulfil the conditions [ eqhardy1 ] @xmath102 then it can be shown @xcite that @xmath103 is the most condensed situation for which there are solutions . furthermore , all solutions can then be written as @xmath104 where @xmath105 , @xmath106 are polynomials of order not higher than @xmath84 . only the product @xmath107 is fixed by this but not @xmath108 or @xmath109 on their own . this freedom is equivalent to changing @xmath110 with the simultaneous change @xmath111 . this also reflects that , while phase space volume is well - defined and preserved as volume , its projection onto a particular axis is allowed to change . the set of solutions @xmath99 of eq . ( [ eqhardy2 ] ) is @xmath98 dimensional , and a convenient orthonormal set consists of the hermite functions @xmath112 , defined as @xmath113 h_n(x / r)\ ; , \label{eqdefhermite1}\ ] ] where @xmath114 are the hermite polynomials , @xmath115 the prefactor in eq . ( [ eqdefhermite1 ] ) ensures normalisation of @xmath112 . the hermite functions are eigenfunctions of the fourier transform , @xmath116 such that the derivation presented in the following for real space also applies to @xmath88-space . all normalised functions @xmath117 inside the solution space of eq . ( [ eqhardy1 ] ) can be written as @xmath118 the maximum intensity @xmath119 possible for such a state can be computed using the method of lagrange multipliers . finding an extremum of @xmath117 is equivalent to finding one of @xmath119 , and we will pick either of these quantities depending on which one is more convenient . the condition @xmath120 of the function @xmath121 yields the intermediary result @xmath122 . by multiplying this with @xmath123 respectively @xmath124 and summing , this gives the two conditions [ eqlagrange3 ] @xmath125 solving for @xmath126 by eliminating @xmath117 and inserting @xmath126 into eq . ( [ eqlagrange3b ] ) , remembering that @xmath127 , one arrives at the result for the maximum intensity , namely @xmath128 the method of lagrange multipliers only gives necessary but not sufficient conditions for extrema , meaning that there can not be any additional extrema different from eq . ( [ eqlagrange4 ] ) but this equation might not describe an extremum in the first place . equation ( [ eqlagrange4 ] ) actually describes two solutions , @xmath129 and @xmath130 . it is obvious from physics reasons that at least two extrema , one being the most positive and one the most negative , exist . hence , eq . ( [ eqlagrange4 ] ) uniquely describes these two extrema . and @xmath131 , computed from eq . ( [ eqlagrange4 ] ) , as a function of @xmath84 where @xmath84 is the number of basis functions minus 1 . depicted are the curves for @xmath132 . if the corresponding curves for a normalised radiation field are below a given line , the radiation field can not have more modes than indicated by the number @xmath84 on the curve . ] a graphical display of the solutions @xmath133 and @xmath134 can be found in fig . [ figlagrange1 ] as a function of @xmath84 . the right edge of @xmath133 becomes progressively steeper as @xmath84 increases . this means that in the macroscopic limit @xmath135 there exist a sharp edge that can be associated with the macroscopic etendue . to define the position of this edge , one could pick the @xmath136-coordinate where @xmath137 or where it is equal some other particular value but this would lead to an arbitrary result . it can be shown that the number of basis functions is then proportional to the square of the width at which the `` cut '' is done @xcite such that the result depends more on this arbitrary decision than on the radiation field . also the method of `` almost bandwidth - limited functions '' ( which is usually formulated for time - dependent electric signals , hence the term `` bandwidth '' instead of `` angular range '' ) suffers from a similar problem @xcite . luckily , there exists a strong link between the microscopic and macroscopic worlds because the hermite functions @xmath112 are the solutions of the eigenvalue equation @xmath138 that can be found in any quantum mechanics textbook as it describes the harmonic oscillator . the eigenfunction @xmath139 has eigenvalue ( = energy ) of @xmath140 , and a classical oscillator with the same energy is restricted to the interval @xmath141 . the extremal curve @xmath117 computed above thus is the extremal superposition of harmonic - oscillator solutions with energy level @xmath27 up to @xmath84 . hence , the classical edge beyond which the oscillator can not be found is given by @xmath142 . summarising , a microscopic radiation field has at most @xmath98 modes when for all of its modes @xmath143 the conditions @xmath144 are fulfilled for some value @xmath145 identical for all modes . since @xmath146 and @xmath147 are the intensities as a function of position and angle , respectively , these are quantities that can in principle be measured . a macroscopic radiation field with a half - diameter @xmath148 in real space and a half - diameter @xmath149 in angle space is thus spanned by modes @xmath139 with @xmath100 determined by @xmath150 the scale factor @xmath145 drops out when expressing results in terms of the macroscopic etendue . for given etendue @xmath151 , the number @xmath84 of modes ( minus 1 ) is thus @xmath152 if the etendue is specified in angle ( radians ) instead of a @xmath88-vector , this becomes @xmath153 the variance of the photocount computed in sec . [ seczeitunschaerfe ] describes the uncertainty of the photocount if the radiation field at the detector is completely known . frequently , the mode structure is known at the light source but not at the detector . the standard undergraduate example is a hene - laser pointed at the wall of the lecture room where the observer can see a speckle pattern . when the eyes are then moved , the direction of motion of the observed speckle pattern depends on whether the viewer is near - sighted or far - sighted @xcite . the hene - laser emits a very regular mode structure but the modes are then changed by scattering at the rough wall . the influence of the eye movement demonstrates that scattering at the wall does not result in local intensity changes at the place of scattering but rather in a deformation of the mode structure that translates into intensity changes only by propagation to and focusing in the eye . the change of the mode structure on its way from the radiation source to the detector can be unintentional , just as in the example above , or it can be intentional due to some apparatus designed to mix the modes of the radiation field . unless optical systems are manufactured to the highest standards , they will have surface roughness in excess of the wavelengths . this results in uncontrolled thus in a certain sense random mode deformations . in addition , any apparatus designed to provide a uniform and stable illumination at its exit can only do this by means of `` mixing '' the incoming light . illumination systems used in optical lithography are the supreme example of such an apparatus as the local intensity at the output varies only by about @xmath154 even when the form of the illumination of its entrance is completely changed . while the light paths inside such an apparatus might be very controlled , for an outside observer there ( intentionally ) is no recognisable connection between the light at the entrance and the exit . in other words , an apparently random change of the mode structure occurs in this case as well . in sec . [ seczeitunschaerfe ] we have used the modes @xmath21 of the radiation field to expand the operator for the electric field , cf . ( [ eqfexpansion ] ) . any orthonormal basis could have been used instead but using the modes offered the advantage that the associated annihilation operators @xmath24 become uncorrelated , cf . ( [ eqerwart1 ] ) . we will now switch to a different basis set , at the expense of replacing eq . ( [ eqerwart1 ] ) by something more complicated . , one switches from the modes @xmath18 that are optimal to describe the light source , to new base functions @xmath155 that are optimal to describe the photodetector . along with the transformation @xmath156 , the annihilation operators also need to be transformed , @xmath157 . the additional action of scattering or mixing is then described by the matrix @xmath158 , transforming @xmath159 into @xmath109 . ] considering only the radiation source and the detector , hence ignoring the scattering between them for the moment , we pick new basis functions @xmath160 and associated annihilation operators @xmath161 such that at the location @xmath0 of the detector all @xmath155 except for @xmath162 vanish . such a basis set can always be found and offers the advantage that photodetection at the point @xmath0 has been reduced to the photocount of the operator @xmath163 . since both bases are orthonormal , the transformation between @xmath164 and @xmath165 can be described by a unitary matrix @xmath166 , @xmath167 with @xmath168 . computing @xmath166 explicitly would be a very formidable task but it will turn out that knowledge of @xmath166 is not needed . we will fix the number @xmath169 of new basis functions @xmath155 shortly but need to allow for the case that @xmath169 is larger than the number @xmath84 of excited modes of the radiation field . this is easily achieved by simply adding @xmath170 vacuum states to the input for eq . ( [ secsmatrix ] ) . we now assume that the radiation field is perturbed between the light source and the place of the detector by means of some `` virtual device '' either intentionally by a properly designed apparatus or , usually unintentionally , by quasi - random scattering . this can be described in two opposite but physically equivalent approaches . either , one assumes that the mode structure is perturbed , thus changing @xmath171 while keeping @xmath172 unchanged , or one assumes that the mode structure @xmath171 is unchanged while energy is transferred between modes , thus modifying @xmath172 . we use the second approach , also known as the method of input - output relations @xcite , since the annihilation operators @xmath173 describing the radiation leaving the device can then be expressed in terms of the annihilation operators @xmath172 entering device . in the absence of nonlinear media , this relation is linear , and can be described by a matrix @xmath158 @xmath174 if there is no absorption , @xmath158 is unitary . apart from energy conservation , already the commutation relation ( [ eqvertauschungsrelation ] ) demands the unitarity of @xmath175 . in the presence of absorption , eq . ( [ secsmatrix2 ] ) would need to be supplemented by noise sources @xcite to ensure these commutation relations . in the absence of further information , the unitary matrix @xmath166 describing the distortion of the mode structure is uniformly distributed in the space of unitary matrices @xcite . this concept of `` maximum uncertainty '' is known from other areas of physics as well , most notably from thermodynamics where the basic lemma is that every microstate compatible with the macroscopic boundary conditions is equally likely . the photocount distribution at some point @xmath0 has thus been reduced to the photocount distribution for the mode @xmath176 , related to the original incident mode structure of the radiation source by @xmath177 since for a given position of detector and light source @xmath0 , @xmath166 is a fixed unitary matrix , the product @xmath178 is also distributed uniformly @xcite , and so no knowledge of @xmath166 is needed , as promised above . the only information about what happens between light source and detector enters in the form of the number @xmath169 of basis functions . this number is not arbitrary but is determined by the physics of the mode deformation process described by the matrix @xmath158 . please remember that @xmath169 is the number of base functions that are mixed by the transformation @xmath158 . if the mode deformation is due to some intentional mixing in an apparatus accepting and emitting some etendue @xmath11 , the number @xmath169 follows from the results of sec . [ secetendue ] and has a finite , well - defined value . if the mode deformation is due to scattering or a diffuser , @xmath169 is determined by the number of basis functions that could in principle end up in the detector due to scattering or diffusing . depending on maximum scattering angle , this can be a very large number and is always larger than @xmath84 . computing the first factorial moment analogue to eq . ( [ eqafrequenz2 ] ) yields @xmath179 where in the final equality sign all averages over the fluctuations of electromagnetic field have been taken , just as described in sec . [ seczeitunschaerfe ] , and an average over the ensemble of random unitary matrices remains . the average is easily computed from @xmath168 , hence the average in eq . ( [ eqafrequenz3 ] ) is equal to @xmath180 where @xmath169 is the order of the matrix , yielding @xmath181 this is the expected result , namely that the mean intensity at the detector is equal to the properly scaled total intensity of the original light field . computing the first moment demonstrated that to translate the equations from sec . [ seczeitunschaerfe ] , all terms of the form @xmath182 have to be replaced by @xmath183 , and all terms @xmath184 by @xmath185 . the correct starting point is eq . ( [ eqlang4 ] ) , which transforms into @xmath186 ^ 2 [ a_{n}(\omega)]^2 \rangle \\ + \eta^2 t^2 \sum_{n_1\ne n_3 } \langle { \lverts_{1n_1}\rvert}^2 { \lverts_{1n_3}\rvert}^2 \rangle i_{n_1 } i_{n_3 } \\ + \frac{\eta^2}{2\pi } \int_{-t}^t\mathrm{d}t \sum_{n_1\ne n_2 } \langle { \lverts_{1n_1}\rvert}^2 { \lverts_{1n_2}\rvert}^2 \rangle \\ \times i_{n_1 } i_{n_2 } \hat{g}_{n_1}(t ) \hat{g}^*_{n_2}(t ) [ t-{\lvertt\rvert } ] \;. \label{eqlang6}\end{gathered}\ ] ] it only remains to compute the necessary averages over the unitary group in a similar spirit to above where we found that @xmath187 . the average of the square of this expression is frequently needed and can be found in many papers , e.g. in ref . , with the result @xmath188 ^ 2 \rangle = \frac{2}{m(m+1 ) } \;. \label{eqsquadratmittel}\ ] ] the average @xmath189 for @xmath190 is not identical to @xmath191 as the unitary condition @xmath192 introduces correlations among the elements of @xmath175 . starting from the exact relation @xmath193 allows us to write @xmath194 which , together with eq . ( [ eqsquadratmittel ] ) , gives us the desired average , @xmath195 this allows us to transform eq . ( [ eqlang6 ] ) into @xmath196 ^ 2 [ a_{n}(\omega)]^2 \rangle \\ + \frac{\eta^2 t^2}{m(m+1 ) } \sum_{n_1\ne n_3 } i_{n_1 } i_{n_3 } \\ + \frac{\eta^2}{2\pi m(m+1 ) } \int_{-t}^t\mathrm{d}t \sum_{n_1\ne n_2 } i_{n_1 } i_{n_2 } \hat{g}_{n_1}(t ) \hat{g}^*_{n_2}(t ) [ t-{\lvertt\rvert } ] \;. \label{eqtemp1exakt}\end{gathered}\ ] ] when computing the variance in sec . [ seczeitunschaerfe ] , the term with @xmath197 dropped out when @xmath54 ^ 2 $ ] was subtracted from @xmath198 . this is no longer the case here , and one arrives at @xmath199 \;. \label{eqtemp2exakt}\end{gathered}\ ] ] before discussing this result , we will analyse an approximation that is frequently done : the exact averages from eq . ( [ eqmittelwertes ] ) are replaced by their gaussian approximation , equivalent to a phasor representation @xcite , @xmath200 the denominator in eq . ( [ eqmittelwertes ] ) was @xmath201 so that the mixing of the modes would conserve energy . equation ( [ eqmittelwertesnaeherung ] ) neglects this constraint . while this error might seem to be negligible for large @xmath169 , we will demonstrate that its effect is surprisingly large . first , we compute @xmath202 ^ 2 [ a_{n}(\omega)]^2 \rangle \\ + \frac{\eta^2 t^2}{m^2 } \sum_{n_1\ne n_3 } i_{n_1 } i_{n_3 } \\ + \frac{\eta^2}{2\pi m^2 } \int_{-t}^t\mathrm{d}t \sum_{n_1\ne n_2 } i_{n_1 } i_{n_2 } \hat{g}_{n_1}(t ) \hat{g}^*_{n_2}(t ) [ t-{\lvertt\rvert } ] \;.\end{gathered}\ ] ] in contrast to eq . ( [ eqtemp1exakt ] ) , the term with @xmath203 now cancels against @xmath54 ^ 2 $ ] . the variance then becomes @xmath204 \;.\end{gathered}\ ] ] this is basically eq . ( [ eqlang3 ] ) with an additional contribution @xmath205 that is called `` static '' as its scaling with @xmath206 implies that its strength , when scaled by the mean squared intensity , is independent of measurement time @xmath1 . we arrived at the ensemble of random @xmath175 for calculating the variance @xmath207 by assuming a fixed position of the detector while the mode deformations between radiation source and detector is different for a every member of the ensemble . however , from eq . ( [ eqsstatischgesamt ] ) it follows that this variance can equally well be computed or measured by keeping the mode formations constant while changing the position of the detector . the static speckle contrast @xmath208 , @xmath209^{1/2}}{\sum_n i_n } \;,\ ] ] can thus equally well be determined by doing a ( long ) measurement at different detector positions @xmath0 . in practise , one uses a camera instead of a point - like photodetector , and @xmath208 is calculated from the observed contrast of the image taken by the camera . the image resembles the speckle pattern found on the fur or skin of many animals , hence the name `` speckle '' . the static speckle contrast can , in this approximation , conveniently be expressed in terms of the effective number of modes @xmath210 , @xmath211 ^ 2}{\sum_n i_n^2 } \;,\ ] ] that is equal to the actual number of modes if all intensities are equal , and smaller ( implying more static speckle ) otherwise . for the exact result ( [ eqtemp2exakt ] ) , static speckle is not so easy to define , as the difference between eq . ( [ eqtemp2exakt ] ) and the results from sec . [ seczeitunschaerfe ] is not a simple term proportional to @xmath206 . however , taking the limit @xmath212 , hence living up to the name `` static '' , the difference between these two terms gives @xmath213^{1/2 } } { \overline{n } } \nonumber\\ & = \left [ \frac{m}{m+1 } \frac{\sum_n i_n^2}{\bigl[\sum_n i_n\bigr]^2 } + \frac{m-1}{m+1 } \frac{\sum_n f_n i_n^2}{\bigl[\sum_n i_n\bigr]^2 } -\frac{1}{m+1}\right]^{1/2 } \nonumber\\ & = \left [ \frac{m}{m+1 } \bigl(\frac{1 } { n_{\mathrm{eff } } } -\frac{1}{m } \bigr ) + \frac{m-1}{m+1 } \frac{\sum_n f_n i_n^2}{\bigl[\sum_n i_n\bigr]^2 } \right]^{1/2 } \;. \label{eqcexakt}\end{aligned}\ ] ] there thus is a very significant difference between the exact result and the approximation . in the latter , the contribution ( [ eqvarnstatisch ] ) is minimal but still nonzero if all modes carry the same intensity @xmath28 whereas for the exact result eq . ( [ eqtemp2exakt ] ) , assuming coherent radiation , this contribution vanishes if all @xmath28 are identical since then @xmath214 . the approximation becomes valid in the limit that there are many modes but only a small number of them are actually excited : equation ( [ eqvarnstatisch ] ) depends on the average squared intensity @xmath215 of the modes of the light source , where for a fixed light source the average is taken over its modes . this kind of average is indicated by the notation @xmath216 . in contrast , the exact result eq . ( [ eqtemp2exakt ] ) depends on @xmath217 $ ] but the term @xmath218 scales with the square of the fraction of excited modes and may thus be neglected in the limit stated above . this important conclusion can also be formulated in terms of the etendue introduced in sec . [ secetendue ] . the approximation is valid in the limit that the light source fills only a small fraction of the etendue that the `` device '' used for mixing can accept . it also means that mixing will not introduce static speckle as long as the entire etendue of the mixer is uniformly filled by the light source ( assuming coherent radiation ) . this might seem to contradict phenomenological theories that claim that there should aways be some pattern in the detector plane because light from the same point of the light source can reach the detector along different paths , thereby resulting in uncontrolled interference that inevitably creates intensity fluctuations . however , the unitary condition @xmath168 ensures that , whenever a mode of the light source is brighter at the detector , another mode has to be dimmer . this property is fulfilled for any individual @xmath175 since @xmath219 , and no averaging over @xmath175 is necessary to arrive at this conclusion . technologically , mixing apparatuses have a well - defined etendue , hence a well - defined value of @xmath169 . when the etendue that they can accept , is filled to a large extend by the incident radiation , the effect of the unitary condition is strong , and the exact formula ( [ eqtemp2exakt ] ) has to be used . for random scattering , on the other hand , in particular by a diffuser , frequently @xmath220 , and therefore the approximate formula ( [ eqvarnstatisch ] ) might suffice . in sec . [ secstaticspeckle ] we demonstrated that the static speckle depends on the mean squared intensity in each mode of the radiation source , hence on @xmath221 , cf . ( [ eqtemp2exakt ] ) and ( [ eqvarnstatisch ] ) . the effect is minimal if all modes carry equal weight , and the maximum number of modes possible follows from the etendue via sec . [ secetendue ] . the minimum amount of static speckle possible can then be computed it is zero if the etendue is not increased by the scattering and mixing processes , and finite otherwise . this raises the following question : is it possible not only to compute the minimum amount of static speckle for given etendue of the radiation field but also the `` typical '' amount of speckle ? we will demonstrate that static speckle contrast is a self - averaging quantity , i.e. , in the limit of many modes @xmath169 , the relative fluctuation while taking the average over the unitary group becomes small . this implies that the ensemble average is characteristic for almost all realisation of the ensemble . when @xmath21 are the modes of the electric field , the electric field @xmath222 is uniquely described by coefficients @xmath19 , @xmath223 and the static speckle contrast is determined by the effective number @xmath210 of modes in a more or less complicated way depending on whether the exact result or the approximation is used , @xmath224 ^ 2 } \;. \label{eqneff1}\ ] ] in sec . [ secetendue ] the linear span corresponding to some given etendue @xmath11 was computed . in the absence of dynamic information , one can not select the modes out of all possible sets of basis vectors , and any set of basis vectors could thus represent the modes . we assume that the decomposition ( [ eqfeld1 ] ) represents the actual modes @xmath99 of the radiation field . since , as just explained , we can not know @xmath99 , the equations for the static speckle contrast are instead evaluated using some other orthonormal basis @xmath160 . the two bases are related by a unitary transformation , @xmath225 , and consequently @xmath226 , but @xmath227 is unknown for the reasons just explained . if no additional information is available , the best available estimate is the average of @xmath228 ^ 2 } \nonumber \\ & = \frac{\sum_{n , i_1,\ldots , i_4 } u^\dagger_{n , i_1 } u_{n , i_2 } u^\dagger_{n , i_3 } u_{n , i_4 } a^*_{i_1}a_{i_2}a^*_{i_3}a_{i_4 } } { \bigl[\sum_n { \lverta_n\rvert}^2\bigr]^2 } \ ; , \label{eqneffmittel}\end{aligned}\ ] ] after integrating over the unitary group , i.e. , over all possible unitary matrices . we study the average of @xmath229 instead of @xmath230 since it both is easier to compute and is the relevant quantity for the speckle contrast . the variance @xmath231 ^ 2\rangle -\langle 1/n'_{\mathrm{eff}}\rangle^2 $ ] then specifies how well the average is representative for all possible members of the ensemble . averaging over the unitary group , the nonvanishing terms are of the form @xmath232 where @xmath233 and @xmath234 are permutations of the numbers @xmath235 , and @xmath236 denotes the cycle structure of a permutation . the coefficients @xmath237 depend only on the cycle structure of @xmath238 and have been tabulated @xcite . labelling the numerator in eq . ( [ eqneffmittel ] ) as @xmath239 , @xmath240 and counting all the permutations of the indices in it yields @xmath241 for the square , @xmath242 one finds @xmath243 the terms @xmath244 appearing in the averages above cancel against the denominator of eq . ( [ eqneffmittel ] ) . this is as expected since , as explained , these coefficients depend on the base that was used to determine them , i.e. , they depend on the choice of one particular unitary matrix . averaging over the unitary group , this choice must be irrelevant , thus the coefficients drop out , and the results depends only on the number @xmath169 of modes . inserting eq . ( [ equmittel2 ] ) into eq . ( [ eqneffmittel ] ) and using the tabulated coefficients from ref . gives the result @xmath245 which has an easy interpretation . the term @xmath246 instead of the naive term @xmath169 in the denominator is due to the correlations imposed on @xmath227 as it is unitary , cf . the factor @xmath246 in eq . ( [ eqcexakt ] ) . the @xmath247 in the numerator implies that every mode is occupied with an effective intensity of @xmath248 relative to the maximum intensity of any mode a very plausible result since the intensity is a random quantity between @xmath249 and that maximum value . a similar calculation using eq . ( [ equmittel4 ] ) gives for the square @xmath250 so that the variance becomes @xmath251 the variance decreases much faster with @xmath169 than @xmath229 or @xmath252 as the lowest - orders in @xmath180 cancel when computing the variance . this implies that @xmath229 is a self - averaging quantity , and its value computed from the number @xmath169 and hence from the etendue of the radiation field gives a good estimation of the actual value of @xmath229 . one should be aware , however , that this is a statistical statement : in `` almost all cases '' this statement is correct . by use of particular light sources it still is possible to have a very different value for @xmath229 . this is also reflected in the calculation presented here . equation ( [ eqneff1 ] ) gives the actual value of @xmath253 which , of course , depends on the values of the @xmath19 . the average value from eq . ( [ eqeffmittelwert ] ) no longer depends on the @xmath19 and can thus deviate . the modes @xmath21 of the electromagnetic field are mutually orthogonal and stationary , cf . ( [ eqgamma2 ] ) , and all the time - dependence of the electromagnetic field is included in the frequency - dependence of the annihilation operators . however , frequently it is rather inconvenient to describe time - dependence in this way . an extreme example are pulsed lasers where it is more appropriate to describe the mode structure of each pulse separately . this comes at a price , though , as the modes of one pulse are then not necessarily orthogonal to the modes of another pulse . this is primarily an issue for the static speckle computed in sec . [ secstaticspeckle ] as the mutual orthogonality of the modes was essential for the derivation presented there . the static speckle contrast in eq . ( [ eqcexakt ] ) has an easy interpretation : except for a minor correction due to the unitary of the scattering matrix , every mode creates a mutually uncorrelated speckle pattern in the detector plane . the effect of a superposition of different modes is easily understood from this , and for the static part it is irrelevant if all the modes are excited simultaneously or rather one after the other . in this section , we will show that the speckle patterns of two nonorthogonal modes ( i.e. , necessarily at two different times ) are correlated . we will , without loss of generality , refer to these two times as first and second pulse , respectively . the first pulse consists of modes @xmath18 and annihilation operators @xmath19 . for a fixed arrangement of mode structure of the light source , scattering between source and detector , and detector position , thus only averaging over the fluctuations of the radiation field , the mean photocount is , completely analogue to eq . ( [ eqafrequenz3 ] ) , @xmath254 we now consider a second set @xmath155 of modes of the light source but keep the scattering between source and detector and the detector position fixed . the transformation from the @xmath18 to the @xmath155 is linear and can thus be described by a matrix @xmath166 , @xmath255 where the notation @xmath256 denotes the overlap integral . the annihilation operators transform as @xmath257 and the mean photocount thus becomes @xmath258 the essential quantity is the cross - correlation between @xmath83 and @xmath259 after averaging over all possible scattering configurations , hence over all @xmath175 , and only the matrix @xmath166 may remain in the result . a positive correlation means that the light patterns caused by the two radiation fields are similar . when the photocounts of two such fields are subsequently integrated on the same detector , the static speckle contrast will be higher than naively expected by the inverse square - root of number - of - modes law . the cross - correlation between @xmath83 and @xmath259 is given by @xmath260 and the average has to be taken over the matrix @xmath175 keeping the matrix @xmath166 fixed . writing out the first term explicitly , @xmath261 the average yields nonzero terms only for @xmath262 , and the two cases @xmath263 and @xmath264 have to be distinguished . the necessary averages have already been given in eq . ( [ eqmittelwertes ] ) , and the result is @xmath265 this can be simplified by noting that @xmath266 . the cross - correlator then becomes @xmath267 i_n i'_m \;. \label{eqcrosscorrelator}\ ] ] this equation gives the expected result in the two extreme situations . if all intensities are equal , the cross - correlation becomes zero since , as explained earlier in the context of eq . ( [ eqcexakt ] ) , the static speckle itself vanishes , and thus also the cross - correlation does . if only a single mode @xmath27 is excited , the cross - correlation becomes proportional to @xmath268 . the average value of @xmath269 when assuming a random @xmath166 is equal to @xmath180 , and the cross - correlator is again zero . if @xmath18 and @xmath270 are identical , the cross - correlator becomes maximal . the cross - correlator thus correctly quantifies if @xmath18 and @xmath155 are stronger correlated than in a random configuration , and the overlap integral @xmath271 is directly related to the static speckle pattern . when a photodetector sums over @xmath84 pulses , the variance @xmath272 of the integrated photocount follows from the variances @xmath273 of the @xmath88-th pulse , given by eq . ( [ eqtemp2exakt ] ) , and from the cross - correlator @xmath274 , given by eq . ( [ eqcrosscorrelator ] ) , between the @xmath88-th and the @xmath275-th pulse , @xmath276 apart from the rather obvious relation ( [ eqvarmehrere ] ) , the results from this section have another application . frequently , radiation fields are expanded not in their modes but rather in a set of `` convenient '' functions , such as gaussian beams or plane waves . drawback of this approach is that this set then usually has to be overcomplete , hence it has more elements than there had been modes , and not all its elements are mutually orthogonal . for example , two gaussian beams have an overlap that decreases exponentially with separation but is always larger than zero . two plane waves restricted to a finite interval have an overlap that decreases exponentially with the angle between the two waves but is always finite . depending on the way that this expansion is done , some coherence information might be lost , namely when doing the expansion such that only the time - averaged intensities at all field - points are matched , and each @xmath18 is then assigned an intensity @xmath28 . this might sound like a bad approximation but effectively amounts to using the assumptions from sec . [ secstatickorrelation ] . inserting the @xmath28 into eq . ( [ eqtemp2exakt ] ) would yield a low static speckle contrast since the number of @xmath28 s is large . however , the correct equation to use is eq . ( [ eqvarmehrere ] ) where every term is evaluated for a single function @xmath18 , and the nonorthogonality of the @xmath18 implies that the cross - correlation terms are large . this is as expected since introducing additional functions @xmath18 to describe the same system should not change its computed properties . the purpose of this paper has been to give a consistent and concise description of multi - frequency and multi - mode effects on photodetection on a quantum level trying to avoid the use of uncontrolled approximations . its focus lies on exactness , not on ease of presentation or application to some particular device . it thus complements previous work , for example by joseph w. goodman or the group around christer rydberg @xcite . in sec . [ seczeitunschaerfe ] we derived how the uncertainty of the photocount depends on the measurement time and the temporal coherence function . among others , it was shown that a description via the coherence time of the radiation is insufficient to arrive at an exact result . similarly , in sec . [ secstaticspeckle ] not only previously - known approximations of the static speckle contrast were retrieved , namely an inverse square - root law with the effective number of modes , but in addition also effects of finite etendue accepted by the `` mixing device '' or of noncoherent , e.g. thermal , radiation followed from the chosen mathematical formalism without additional effort . some of the results of this paper can be explained , albeit only on a qualitative level , by the `` brick wall '' model : the radiation field is like a brick wall ( horizontally : time , vertically : position ) , and only bricks in the same column or row can cause interference effects . counting the fraction of such bricks gives an intuitive explanation of the scaling of dynamic and static speckle contrast upon changing system parameters . if this is sufficient , the methods and results presented in this paper are a bit of an overkill . if , on the other hand , exact agreement between predication and performance of some device is essential , this paper offers the advantage of giving results using only well - defined input parameters , thus not relying on effective parameters that need to be tuned until the desired result is retrieved . main function of any theory is to predict whether some device or experimental setup will work as expected and/or required . for state - of - the - art technological applications , the performance might improve by about @xmath277 from one generation of a device or setup to the next but rarely more . to predict if something actually is an improvement , the error due to approximations thus has to be much smaller this number . as an example , fig . [ figkurvenformen ] gives an indication of the error made when using the coherence time instead of the exact shape of the frequency spectrum . if improvement of dynamic speckle is a topic , one should thus use the exact formulae and refrain from approximations . depending on the desired application , either the present paper or one of the previous works by other authors will thus be more appropriate . furthermore , we restrict us to universal problems . for example , we study in sec . [ secstaticspeckle ] only the variance of the photocount , hence the contrast of the dark and bright `` spots '' in the plane of the detector , but not their spatial correlations , i.e. , the size of these spots . reason is that , while the contrast is universal , the size of the spots is not and depends on the setup . the book of goodman @xcite is to a large extend devoted to studying this question ( on the classical level ) for a large number of setups of technological or scientific importance . in sec . [ seczeitunschaerfe ] the uncertainty of the photocount for known radiation field has been computed , yielding in addition to the well - known shot noise term two terms depending on the frequency spectrum and on the photon correlations of the radiation source , respectively . the importance of the frequency spectrum is obvious , in particular in view of the two examples presented in this paper . for the importance of the effects of noncoherent radiation , the situation is less clear . for traditional thermal light sources , the necessary information can be found in this paper ; for traditional lasers , the emitted radiation is coherent , and the question becomes trivial . for modern light sources , such as excimer lasers @xcite that are increasingly used for high - power applications , insufficient data is available . technological advances are fast , and no good characterisation of the photon correlations for modern devices seems to have been published . since this contribution to the noise inevitably becomes the dominant term once the measurement time is sufficiently long , this lack of knowledge is an actual issue . the static speckle , computed in sec . [ secstaticspeckle ] , depends on the number of excited modes of the radiation field ( and the energy distribution among them ) but , in addition , also on the number of modes that could in principle be excited . this is closely related to the concept of etendue , cf . [ secetendue ] . even though etendue is a very basic quantity , its microscopic definition does not seem to have been addressed before . the relation presented between number of basis function and macroscopic etendue has application beyond this paper as any modelling of a radiation field for the purpose of simulation or for an analytic study needs to start with a given number of modes , and knowing this number a priori makes this process much more efficient . while the applicability of the relation between etendue and maximum number of modes is obvious , this is less the case for sec . [ secstatickorrelation ] discussing the relation between etendue and average effective number of modes . we demonstrated that this average is identical to the actual number for almost all radiation fields . however , that is `` only '' a statistical statement : it assumes that all possible radiation fields with given etendue are equally likely , and it is still allowed that a few cases ( albeit of measure zero ) deviate strongly . different light sources have different properties , and for any technological application , one will choose the most appropriate light source . selecting a single - mode laser that operates in a high - order hermite - gauss mode would give the most extreme disagreement between the actual number of modes ( = one ) and the prediction from sec . [ secstatickorrelation ] . the results from this section can thus only be applied when the choice for a particular light source does not adversely impair the freedom of the mode structure of the generated light field .

the yuan - tseh lee array for microwave background anisotropy ( amiba ) is a platform - mounted 7-element interferometer operating at 3-mm wavelength to study the structure of the cosmic microwave background ( cmb ) radiation . it is being constructed as part of the cosmology and particle astrophysics ( cospa ) project , funded by the taiwan ministry of education initiative on academic excellence . this excellence initiative was aimed at stimulating interdisciplinary research and large scale integration of independent research programs . cospa is designed to jump start a program of research in cosmology , with both theory and experimental projects , while incorporating research in high energy physics , development of infrastructure for optical astronomy in taiwan , as well as accessing observing time on a 4-m class optical telescope . amiba is a collaboration between principally the academia sinica institute of astronomy and astrophysics ( asiaa ) , the national taiwan university ( ntu ) physics and electrical engineering departments , and the australia telescope national facility ( atnf ) . the project was started in 2000 . a two - element prototype was deployed in 2002 to mauna loa ( elevation 3396@xmath0 m ) in hawaii for testing of design concepts . site development was completed in 2004 . the amiba mount was delivered and installed in 2004 , while the platform was delivered and integrated in 2005 . with the integration of the first seven elements of the array and successful first light , the amiba was dedicated in october 2006 , and named after then academia sinica president yuan tseh lee for his important contributions in promoting the growth of astronomy in taiwan . figure 1 shows the amiba at the dedication ceremony . the aim of amiba is to study spatial structure in the cmb radiation ( ho et al . 2008 ; wu et al . 2008a ) , which carries imprints of various physical processes in early epochs of the universe . since its initial detection by penzias & wilson ( 1965 ) , the cmb has been recognized as the definitive signature of the big bang which began the expansion of the universe . subsequent studies have established the properties of this relic radiation after its decoupling from the matter in the early universe around @xmath1 : a mean temperature of @xmath2k ( present ) with minute fractional anisotropies at the level of @xmath3 ( cobe , mather et al . 1990 ; smoot et al . 1992 ) , and polarization at the level of a few to 10% of temperature fluctuations ( dasi , kovac et al . 2002 : wmap , kogut et al . 2003 ; page et al . 2007 : nolta et al . 2008 : cbi , readhead et al . 2006 : quad , pryke et al . in particular , the cmb structures seen on various angular scales by cobe and then wmap ( bennett et al . 2003 ; spergel et al . 2007 ; komatsu et al . 2008 ) demonstrated that the angular power spectrum of cmb anisotropies is a powerful probe of our cosmological model of the universe . amiba is built to sample the angular range from @xmath4 to @xmath5 , corresponding to spherical harmonic multipoles @xmath6@xmath7 , at a wavelength of 3@xmath0 mm , with full polarization . these capabilities complement existing , on - going , and planned experiments . the angular scales sampled by amiba address the higher - order acoustic peaks of the cmb structures to further constrain cosmological models . amiba also aims to search for , and study , distant high - redshift clusters of galaxies whose hot intracluster gas will distort the cmb spectrum via the thermal sunyaev zeldovich effect ( hereafter sze , sunyaev & zeldovich 1970 , 1972 ; rephaeli 1995 ; birkinshaw 1999 ; carlstrom , holder , & reese 2002 ) . the optical and x - ray surface brightnesses of clusters of galaxies decrease rapidly with increasing redshift due to cosmological redshift dimming , while the detectable sze is close to being independent of redshift because it is a spectral distortion of the cmb radiation which itself increases in intensity with increasing redshift as @xmath8 in a standard cosmological model . thus sze measurements are potentially more sensitive than x - ray observations for finding clusters of galaxies beyond a redshift @xmath9 , and will be an important probe for the matter distribution in the high - redshift universe . in this paper , we describe the design and construction of amiba , the first observational results , and the scientific potential of this instrument . throughout this paper , we adopt a concordance @xmath10cdm cosmology with @xmath11 , @xmath12 , and @xmath13 . the basic characteristics of amiba are summarized in table 1 . with the funding of the cospa / amiba projects in 2000 , a workshop was held to define the scientific objectives and the design of amiba ( lo et al . 2001 ; liang 2002 ) . _ the first design criterion was to operate at 3-mm wavelength . _ this was to take advantage of the sweet spot at 3@xmath0 mm where the fractional sze decrement with respect to the primary cmb is close to its maximum ( see figure 1 of zhang et al . 2002 ) and the sze signal is minimally - contaminated by the galactic synchrotron emission , dust foregrounds , and the population of cluster and background radio sources . operations at 3@xmath0 mm also complement the wavelength coverage of other existing and planned cmb instruments : interferometers such as cbi at 30@xmath0ghz ( padin et al . 2002 ) , ami at 15@xmath0ghz ( scaife et al . 2008 ) , sza ( mroczkowski et al . 2008 ) at 30 and 90@xmath0ghz , and vsa at 30@xmath0ghz ( watson et al . 2002 ) ; bolometer arrays such as act , apex - sz ( halverson et al . 2008 ) , and spt . _ the second design criterion was to choose interferometry . _ this was a somewhat difficult choice as many new cmb projects were then planning to use bolometer arrays , which held the promise of a greater inherent sensitivity because of the broad wavelength coverage and a greater speed because of the multiple elements of the detector arrays . the choice of interferometry was based on the desire to utilize cross correlations to suppress systematic effects , since the ability for bolometers to integrate down to theoretical noise was unknown at that point . furthermore , asiaa had the experience of working in millimeter wavelength interferometry from being a partner on the submillimeter array ( sma ) project ( ho et al . amiba was seen as an extension and application of the technical capabilities within asiaa . interferometry is also a natural way to sample simultaneously the spatial structures on various scales and to construct a map by fourier inversion . imaging the entire primary beam of the individual elements of an interferometer is equivalent to a multi - element detector array . _ the third design criterion concerns the angular sizes to be pursued . _ amiba was specifically designed to sample structures in the cmb on small angular scales at multipoles @xmath6@xmath7 , complementing in angular scales earlier large - sky cmb experiments ( e.g. , cobe , maxima , boomerang ) and the dasi experiment on degree / sub - degree angular scales ( @xmath14@xmath15 ; leitch et al . 2002a , 2002b ) . in the left panel of figure 2 , we show the initial compact configuration of seven 0.6@xmath0 m antennas ( _ small solid circles _ ) on the 6@xmath0 m platform ( _ solid outer circle _ ) , along with the distribution of holes ( _ crosses _ ) for mounting receiver packages . at each of the frequency channels centered at about 90 and 98@xmath0ghz , this compact configuration provides 21 simultaneous baselines with three baseline lengths of @xmath16 , 1.05 , and 1.21@xmath0 m , corresponding to angular multipoles @xmath17 of @xmath18 at @xmath19 . also shown in the right panel of figure 2 is the sensitivity of the 7-element compact array as a function of @xmath20 at two frequency channels , represented by window functions ( see eq . [ 14 ] of white et al . 1999 ) for the three different baseline - lengths . for the initial configuration of amiba at 3@xmath0 mm ( see figures 2 and 3 ) , we are sensitive to the multipole range of @xmath21 , which roughly overlaps with the cbi experiment at 1 cm ( padin et al . 2002 ; mason et al . 2003 ; pearson et al . 2003 ) but is complementary in observing frequency . hence amiba aimed at angular scales from @xmath22 to @xmath23arcmin , in order to extend the coverage of angular sizescales by one more order of magnitude . the choice of 3-mm wavelength meant the required baselines were 0.6@xmath0 m to 6@xmath0 m . a maximum baseline of 6@xmath0 m immediately suggested that this interferometer is small enough to be mounted on a platform , as pioneered by cbi and dasi . _ the fourth design criterion was therefore to choose a carbon fiber platform for weight and stiffness considerations . _ the platform was made with multiple holes for mounting the receiver packages , to accommodate multiple baseline configurations . the layout of the receiver ports has a hexagonal pattern in order to allow close packing , and also to utilize the triangular patterns used by small interferometers such as the sma , cbi , and dasi , in order to achieve the most uniform @xmath24-coverage . an asiaa team led by robert n. martin and philippe raffin designed the platform , which was manufactured by composite mirrors applications . a single rigid platform has the advantages of stable differential pointing , a stable baseline solution without differential delay tracking , no mutual shadowing by individual reflectors , and a single drive system . however , a platform also means the interferometer does not have different projected baselines because of earth rotation . _ the fifth design criterion was the choice of a hexapod mount , which provides six degrees of freedom ( which are tightly constrained ) in driving the telescope and rotating the platform . _ by rotating the reflectors with respect to the celestial source , additional @xmath24-spacings are sampled . the rotation of the platform , described by the so - called polarization angle of the platform , also allows us to set different orientations of the receivers relative to the ground , which is useful for discriminating between various environmental effects and checking for ground pick - up . these capabilities were introduced both by cbi and dasi , and we have incorporated them . the hexapod mount was designed and built by vertex antennentechnik , duisburg , germany . to protect the telescope from the elements , we use a retractable shelter with seven steel trusses covered with a pvc fabric , which is manufactured by american spaceframe fabricator inc . the individual parabolic reflectors are built in carbon fiber in order to minimize their weight . the choice of 0.6 m ( @xmath25 m , to be precise ) for the reflector diameter was driven by the desire to cover angular sizescales from @xmath22@xmath23arcmin in order to extend and overlap with existing cmb data . the choice of 1.2@xmath0 m for the diameter of the second - generation reflectors was to increase the collecting area and to improve the spatial dynamic range and angular resolution for sze observations while utilizing the entire platform . the reflectors were designed by asiaa , and manufactured by cotech , taichung , taiwan . the design includes baffles to shield against crosstalk between individual elements , and gortex covers to shield against direct solar irradiation . the height of the baffles is approximately 30% of the diameter of the reflectors so that the secondary mirrors are well shielded . the supports for the secondary are attached to the baffles which are attached to the edges of the primary . replicating an accurate carbon fiber surface against a steel mold was not simple , and some hand polishing was required . depositing the actual reflecting surface and a protective coating was also not simple especially for the 1.2@xmath0 m size . nevertheless , a collaborative effort between asiaa and cotech was successful in delivering the reflectors with surface accuracy better than 50 microns . this is 2% of the operating wavelength of 3@xmath0 mm , and would allow efficient operation up to 1 mm . laboratory measurements and outdoor beam pattern measurements showed that the performances of the reflectors met the specifications ( for details see koch et al . 2006 ) . _ the sixth design criterion was to choose heterodyne receiver systems , which operate between 86 and 102@xmath0ghz . _ this choice of frequency covers an excellent region of atmospheric transparency . the design , construction , and integration of the receiver systems were performed by asiaa staff led by ming - tang chen . dual polarization capabilities are provided by waveguide orthomode transducers , which follow the circular corrugated feedhorns and the circular - to - linear polarizers . each polarization is then fed into a jpl monolithic - microwave - integrated - circuit ( mmic ) inp hemt low noise amplifier ( lna ) cooled to 15@xmath0k with standard cti22 refrigerators . these lnas performed well with measured noise temperatures of 3550@xmath0k across a 20@xmath0ghz bandwidth ( weinreb et al . we used subharmonically pumped mixers ( shm ) using the same mmic technology . the lo / if systems are designed and built by tah - hsiung chu and his group at the ntu ee - department . the phase - locked lo signal at 21@xmath0ghz is doubled to 42@xmath0ghz and then phase switched with walsh functions before being combined with the sky signal at the shm . variable attenuators and amplifiers , before and after mixing , control the if levels before correlation . slope equalizers , phase stabilized cables , and adjustable delays are used to further adjust the if signals . we measure the effective receiver temperatures across the @xmath22@xmath26ghz if window to be @xmath27@xmath28k . figure 3 shows the amiba fully loaded with all the receivers and correlator modules . more details on the receiver system are given by chen et al . ( 2008 ) . _ the seventh design criterion was to choose a wide band correlator with analog technology . _ this is a joint development effort between asiaa and atnf , with chao - te li , derek kubo , and warwick wilson leading the effort ( li et al . a digital correlator would have required sampling at too high a rate to be practical . the amiba correlator uses balanced diode mixers to multiply the signals from each pair of antennas . after application of the 4 different lags , the cross - correlated signal is then digitized and read out with integrated circuits designed by tzi - dar chiueh of the ntu ee department . the readout ics demodulate the phase - switched signals and accumulate the counts for specified times . the choice of a small number of lags is to reduce bandwidth smearing effects while maintaining the maximum bandwidth response through the cascade of electronics . the bandwidth smearing effect is described by the product of the fractional bandwidth and the displacement of the source from the field center . a @xmath29 bandwidth will keep the distortion below @xmath30 ( @xmath31 ) at the fwhm of the primary beam of the @xmath32 m ( @xmath33 m ) reflectors . because of the small number of lags to cover a large bandwidth , the conversion to complex visibility is strongly affected by gain variations over the passband , differential delays between lags , and non - linear phase response for each lag . the lag - to - visibility transformation is calibrated with a noise source . this is discussed further by lin et al . ( 2008 ) . the hexapod drive system is more complex than the conventional azimuth - elevation drive systems used in radio astronomy . the difficulty lies in the necessity of driving all six hexapod jacks without over - extension or collisions . vertex antennentechnik provided the control software . the performance of the drive system was checked via pointing and tracking tests ( koch et al . pointing of the amiba utilized an optical telescope mounted on the platform . we identified a misalignment of the anchor cone and the mount , tilt of the optical telescope with respect to the mount , flexure in the platform as a function of the platform polarization angle , and local platform deformation . we also measured the repeatability of pointing on both short and long timescales , and we performed photogrammetry to measure the stability of the platform . pointing was found to be repeatable at the @xmath34 level on a timescale of several hours . the absolute rms pointing error appears to be about @xmath35 for the platform set at zero polarization angle , and up to 3@xmath36 if the platform is allowed to rotate to different polarization angles . however , with the implementation of an interpolation table , absolute rms pointing can be reduced to @xmath37 over all sky . this is less than @xmath29 of the primary beams of the 0.6@xmath0 m reflectors , and is within specifications for the operation of the 7-element compact array ( koch et al . however , when the 1.2@xmath0 m reflectors are deployed , pointing needs to be improved by a factor of 2 . the segmented approach instead of a monolithic design for our carbon fiber platform resulted in unforeseen difficulties . the bolted joints between the six outer segments were not stiff enough . in spite of efforts to strengthen the joints with additional plates and brackets , the platform still deforms under operational loads to a level beyond our specifications . a saddle - shaped deformation pattern is present with an amplitude dependent on the hexapod azimuth , elevation and polarization position . fortunately , the deformation appears repeatable when measured with photogrammetry which means that it can be modeled . maximum deformation at the edge of the platform under simulated full loading is 0.38@xmath0 mm . the @xmath0maximum deformation in the inner 3@xmath0 m of the platform is more modest at 0.120@xmath0 mm , and within specifications for the operation of the 7-element compact array . the deformation is due principally to the platform not being stiff enough , and also because the stresses from the drive system are transmitted to the platform in spite of a steel interface ring ( koch et al . the immediate ramification is that the individual primary beams of the interferometer elements will be mis - pointed relative to each other . this is equivalent to the primary mirror of a single dish telescope being deformed under gravity or atmospheric distortion . the equivalent adaptive optics approach for an interferometer involves adjusting the gains and phases of the individual elements either through a look - up table or via self calibration if the signals are strong enough . the advantage of an interferometer is that we can adjust the signals during the postprocessing phase before adding or multiplying the signals from different baselines . for our current configuration , such correction schemes for the platform deformation do not need to be implemented . calibration of the amiba system was led by kai - yang lin and chao - te li . the receiver temperatures were measured by the standard hot / cold load method to be 55 to @xmath28k . the contribution from the sky and ground pickup to the total system temperatures is about @xmath38k . since the correlator only has 4 lags , only 2 frequency channels can be extracted . the response of the correlator is therefore not easy to measure , and we use planets to calibrate the correlator response periodically . we find the complex gain to be stable to 5% in gain and 0.1 rad in phase on time scales of a few hours . using the detected fluxes for saturn and jupiter , the overall system efficiency was estimated to be @xmath39@xmath40 for each of the baselines of the interferometer . this efficiency accounts for all losses due to illumination , blockage , spillover , alignment of the reflectors with respect to the radio axis of the platform , deformation of the platform , and pointing , as well as the narrower correlator response ( lin et al . 2008 ) . the losses in front of the receiver account for the significant contributions from the sky and ground pickup to the system temperatures . in two - patch differencing observations a sensitivity of about @xmath41mjy is achieved in 1@xmath0hour of on - source integration under good sky conditions ( chen et al . 2008 ; lin et al . 2008 ) . since the sze signals and the cmb anisotropy are quite faint , the behavior of the noise is very important . this analysis has been led by hiroaki nishioka . we conducted tests of blank sky data as well as data where the inputs are terminated by absorbers . statistical analysis of different samples of such data showed no significant differences . time variable signals from the correlator due to electronics and ground pickups can be seen in the various datasets . however , a power spectrum analysis demonstrated white noise behavior for frequencies between 10@xmath42 and @xmath43hz . hence two - patch differencing of the sky signal at 600 sec intervals is adequate to remove the slowly varying contaminations . by cross correlating the outputs from the different lags of the correlator , we find only a weak correlation at less than 10% . by dividing the datasets into smaller samples , we verify that the sample variance is consistent with gaussian random noise which would integrate down over time . we also applied the kolmogorov - smirnov ( k - s ) test to our cluster data , and we find that 90% of the data are consistent with gaussian behavior at the 5% significance level . the data which fail the test at 5% level are often found to be associated with hardware problems , and we use this criterion to discard questionable samples . more details on the gaussianity tests are given in nishioka et al . ( 2008 ) . since the sze signals are faint , they can be confused with the cmb anisotropies and foreground emission structures . the analysis of possible contaminations has been led by guo - chin liu . examining the wmap data , we can see that cmb structures on the scale of a degree are suppressed by our two - patch observing technique , while contamination from galactic emission is typically fainter than the cmb emission by an order of magnitude . from wmap data , we also estimate possible contamination from primary cmb anisotropies , by analyzing 500 simulated cmb fields , for the sizescales which we sample . we find contamination at the level of @xmath44 to @xmath45mjy for the sampled multipoles of @xmath46 to @xmath47 ( at @xmath48ghz ) . we also estimated potential contamination from foreground point sources . the best way to correct for this issue is to have high angular resolution interferometer data at the observed frequencies . lacking such data , we extrapolate the contributions from known point sources in low frequency radio surveys . since spectral indices are not known accurately to @xmath49 mm , a statistical approach is used to estimate the contamination in each cluster from known point sources . in all clusters , a net positive contribution of the point sources was found in our main - trail / lead differencing amiba observations , indicating that there are more radio sources towards clusters than in the background ( as expected ; see coble et al . the corrections for contamination are described more fully in liu et al . ( 2008 ) . first light with the 7-element amiba was achieved in september 2006 . the array of 0.6@xmath0 m reflectors was in the close - packed configuration on the platform ( figures 2 and 3 ) . the image of jupiter shown in figure 4 is the first end - to - end test of the system based on drift - scan data , including the pipeline for data analysis . the image was constructed from drift scans at four equally - divided platform polarization angles to provide better @xmath24-coverage . only the transit data of about 0.23@xmath0s exposure time from each drift scan were used , yielding an overall noise level in the dirty map of about 3@xmath0jy , much lower than the jupiter flux density of 844@xmath0jy . successful images were obtained for saturn , venus , and the crab nebula , all with fluxes on the order of 200@xmath0jy . in particular , the measured angular size of the crab nebula is consistent with existing optical and radio images . in december 2006 , using saturn as the calibrator , uranus was imaged with a flux density of @xmath50jy ( at @xmath51 confidence level ; in 16 @xmath43s exposures ) , consistent with the expected level of @xmath52jy . at the achieved sensitivity , the close - packed 7-element configuration of amiba is suitable for imaging galaxy clusters via the sze . during 2007 , we imaged 6 clusters , chosen to be relatively nearby ( a median redshift of @xmath53 ) , already detected in sze , and resolvable with amiba . target selection and data analysis are described in wu et al . ( 2008b ) . to detect the much fainter sze signals from clusters of galaxies , we need to integrate longer . at this level of sensitivity , emission from local terrestrial sources such as buildings and the ground will enter through the system either through the sidelobes of the antenna response or via scattering into the primary beam . the interferometer has the advantage over a bolometer array in that the cross correlation will suppress extended low level emission which enters the system through the sidelobe response of the reflectors . however , structures on angular scales corresponding to the interferometer baselines will still persist . the design of the baffles around the edges of the reflectors helps further suppression , but we did not exercise the option of installing ground shields . instead we employ a two- or three - patch observing technique where the target source is preceded and/or followed by tracking over the same range of azimuth and elevation just traversed by the source . this allows us to subtract and cancel the terrestrial ground and sky emission . this is the amiba equivalent of the position - switching technique used in single - dish radio astronomy , where we track specifically the same part of the atmosphere . as shown in figure 5 , and in wu et al . ( 2008b ) , this procedure is quite successful and we have mapped the sze decrement towards a number of clusters . these detections serve to demonstrate the scientific potentials of amiba . amiba will be sensitive to structures as large as @xmath54 with resolution as high as @xmath55 so that sze structures can be resolved . even at the current angular resolution of @xmath56 fwhm , an extended , elliptical shape can be seen for some of the clusters . these shapes can be compared to the results from weak gravitational lensing or x - ray imaging . the sze and x - ray images are sensitive to the distribution of the hot intracluster gas , while the weak lensing images probe the distribution of all the gravitational mass including the dominant `` invisible '' dark matter . furthermore , the x - ray emission is sensitive to the square of the electron density while the sze measures the thermal electron pressure projected along the line - of - sight , linearly tracing the electron column density . hence , the complementary x - ray , the sze , and the weak lensing data , probe progressively further out in the mass distribution . for four of the massive clusters in our sample ( a1689 , a2142 , a2261 , a2390 ) , for which high - quality subaru weak lensing data are available , we have compared the sze results with the subaru weak lensing measurements ( umetsu et al . 2008 ) . in the case of abell cluster a2390 , the elliptical shape seen in sze is consistent with the shape of the dark matter distribution as deduced from weak lensing data . for this comparison , a2142 is of particular interest since this is our brightest , most - nearby ( resolvable ) sze cluster at @xmath57 , known as a merging cluster with two x - ray _ cold fronts _ , which are sharp edges in x - ray dense cores ( markevitch et al . 2000 ) . in figure 6 we compare for a2142 our amiba sze map with the projected mass distribution ( _ white contours _ ) as deduced from our weak lensing analysis . here the mass map is smoothed to a resolution of @xmath55 fwhm . our amiba sze map shows an elliptical structure extending in the northwest ( nw ) - southeast ( sw ) direction , similar to x - ray and weak lensing distribution shapes ( figure 6 ; see also figure 10 of okabe & umetsu 2008 ) . the almost @xmath54 angular extent of the cluster sze justifies the use of observing with antennas as small as @xmath32 m , so that the observations are sensitive to large scale structures . further , relative positions ( between pixels ) are important in such maps of extended structures , where the interferometer has the inherent advantage of relying on the phase to determine relative positions , while single dish studies might be affected more strongly by pointing of the telescopes . in the cluster a2142 , a mass subclump located @xmath58 ( @xmath59kpc@xmath60 at the cluster redshift ) to the nw of the cluster center has been detected from the subaru weak lensing analysis of okabe & umetsu ( 2008 ) and umetsu et al . this nw mass subclump , lying @xmath61 ahead of the nw- edge of the x - ray dense core , has no x - ray counterpart in chandra and xmm - newton observations , but is associated with a slight excess of cluster sequence galaxies ( see figure 10 of okabe & umetsu 2008 ) . our amiba map exhibits slight excess sze signals in the nw region ( figure 6 ) at @xmath62 significance levels . we note this slight excess sze appears extended over a couple of synthesized beams , although the per - beam significance level is marginal . good consistency found between the sze and weak lensing maps is encouraging , and may suggest that the nw excess sze is a pressure increase in the icm associated with the moving nw substructure . clearly further improvements in both sensitivity and resolution of the sze mapping ( see [ subsec : expansion ] ) are needed to better constrain the merger geometry and physical properties of the merging substructure . this however demonstrates the potential of sze observations as a powerful tool for measuring the distribution of icm in cluster outskirts where the x - ray emission measure ( @xmath63 ) is less sensitive . this also demonstrates the potential and reliability of amiba , and the power of multiwavelength cluster analysis for probing the distribution of mass and baryons in clusters . finally , for our small sample with a mean virial mass of @xmath64 , the hot gas mass fraction is about @xmath65 ( umetsu et al . 2008 ) . as compared to the cosmic baryon fraction @xmath66 of @xmath67 , as deduced from the wmap 5-year data ( dunkley et al . 2008 ) , possibly @xmath68 of the baryons are missing from the hot cluster environment ( umetsu et al . 2008 ) . this missing cluster baryon fraction is partially made up by observed stellar and cold gas fractions of @xmath69 several @xmath70 in our x - ray temperature range ( @xmath71kev ) . by comparing the sze and x - ray imaging results , we can deduce a value for the hubble constant @xmath72 . since the sze is induced by inverse - compton scattering of the cmb photons by hot electrons in the cluster environment , it is sensitive to the total column of electrons . the x - ray emission is due to the bremstrahlung process between protons and electrons , and is therefore sensitive to the square of the electron density . by combining these two relations , the angular diameter distance can be deduced . however this derivation , and the subsequent measurement of @xmath72 depends on the assumed geometry of the emission region . we have derived a value of @xmath72 from our sample of sze results combined with x - ray data . the uncertainty is high because of the small size of our sample , and the bias from the assumed geometry . however , the deduced value is consistent with other sze constraints , and is an indicator of integrity of the amiba data so far . better limits will come with a much larger sample to be obtained in the future . while the initial 7-element @xmath32 m reflectors have been commissioned , we are proceeding with the expansion of the amiba to its 13-element configuration . in this configuration , we will upgrade from @xmath32 m to @xmath33 m antennas . this will increase the collecting area by a factor of @xmath73 , and the speed of the interferometer by a factor of almost 60 in single pointed observations . we will place the 13 elements over the platform to generate the longest possible baselines , which will result in angular resolutions up to @xmath55 . the correlator is also being expanded in order to handle the larger number of cross correlations . in this second phase of amiba operations , the first science target will be to measure the angular power spectrum of cmb temperature anisotropies to the higher multipole numbers in order to examine the shape at and beyond the third acoustic peak ( @xmath74 ; nolta et al . 2008 ) , where data taken with the 7-element close - pack 0.6@xmath0 m configuration will be combined with upcoming 13-element 1.2@xmath0 m data . accurate measurements of the angular power spectrum of cmb temperature anisotropies ( park et al . 2003 ) through @xmath75 up to @xmath76 will allow us to see secondary effects such as the sze ( lin et al . 2004 ) and possible cosmic string structures ( wu 2004 ) . the second science target will be to resolve cluster sze structures on the sky in order to compare with dark matter structures as deduced from weak gravitational lensing studies ( umetsu & futamase 2000 ; broadhurst et al . 2005 , 2008 ; umetsu & broadhurst 2008 ; okabe & umetsu 2008 ) . the third science target will be to survey for the distribution of galaxy clusters via the sze ( zhang et al . 2002 ; umetsu et al . 2004 ) . to obtain redshifts of cluster candidates , optical follow up observations will be conducted with ground - based telescopes . at the time of the publication of this paper , the expansion is already in progress . we anticipate first operations during 2009 . in this introductory paper , we have presented the design and construction of the yuan - tseh lee amiba project . the telescope is now in scientific operation , and the first science results are the detection and imaging of six galaxy clusters via the sunyaev zeldovich effect . our companion papers elaborate on various aspects of our results . chen et al . ( 2008 ) describe the technical design of the instruments in detail . wu et al . ( 2008b ) describe the observations and data analysis for the first year of amiba sze data . lin et al . ( 2008 ) examine the system performance , and discuss the astrophysical properties of planetary calibrators . koch et al . ( 2008a ) describe the performance of the hexapod mount and the carbon fiber platform . nishioka et al . ( 2008 ) examine the integrity of the amiba data and its statistical properties . liu et al . ( 2008 ) address the issues of foreground and primary - cmb contamination in amiba sze observations . umetsu et al . ( 2008 ) discuss the combination of amiba sze and subaru weak lensing data in order to probe the structure of dark matter and to derive the cluster gas mass fractions . koch et al . ( 2008b ) derive the value of the hubble constant @xmath72 from amiba and x - ray data on the clusters . huang et al . ( 2008 ) derive cluster scaling relations between the sze and x - ray observables . finally , molnar et al . ( 2008 ) discuss the potential of amiba data in its 13-element configuration to constrain the intra - cluster gas distributions . we thank the ministry of education , the national science council , and the academia sinica for their support of this project . we thank the smithsonian astrophysical observatory for hosting the amiba project staff at the sma hilo base facility . we thank the noaa for locating the amiba project on their site on mauna loa . we thank the hawaiian people for allowing astronomers to work on their mountains in order to study the universe . bennett , c. et al . 2003 , apjs , 148 , 1 birkinshaw , m. 1999 , phys rep , 310 , 97 , t. , takada , m. , umetsu , k. , kong , x. , arimoto , n. , chiba , m. , & futamase , t. 2005 , , 619 , l143 , t. , umetsu , k. , medezinski , e. , oguri , m. , & rephaeli , y. 2008 , apj , 685 , l9 carlstrom , j.d . , holder , g.p . , and reese , e.d . 2002 , araa . , 40 , 643 chen , m .- 2008 , , submitted coble , k. , bonamente , m. , carlstrom , j. e. , dawson , k. , hasler , n. , holzapfel , w. , joy , m. , la roque , s. , marrone , d. p. , & reese , e. d. 2007 , , 134 , 897 dunkley et al . 2008 , submitted to apjs , preprint ( arxiv:0803.0586 ) halverson , n.w . 2008 , , submitted ( arxiv : 0807.4208 ) ho , p.t.p . , moran , j.m . , & lo , k.y . 2004 , , 616 , l1 ho , p.t.p . et al . 2008 , modern physics letter a , 23 , 1243 huang , c .- w . l. et al . 2008 , in preparation ( apj companion paper ) koch , p. et al . 2008a , , submitted koch , p. et al . 2008b , , in preparation koch , p. et al . 2006 in _ proceedings of the european conference on antennas and propagation _ : eucap 2006 ( esa sp-626 ) , eds . : h. lacoste & l. ouwehand . published on cdrom . , p.668.1 kogut , a. et al . 2003 , apjs , 148 , 161 komatsu , e. et al . 2008 , apjs submitted ( arxiv : astro - ph/08030547 ) kovac , j.m . , et al . 2002 , nature , 420 , 772 leitch , e.m . et al . 2002a , , 568 , l28 leitch , e.m . et al . 2002b , nature , 420 , 763 li , c .- t . 2004 in _ millimeter and submillimeter detectors for astronomy ii _ , ed . j. zmuidzinas , w. holland , and s. withington , spie , bellingham , wa liang , h. 2002 , in _ amiba 2001 : high - z clusters , missing baryons , and cmb polarization , asp conference proceedings , vol . chen et al . ( astronomical society of pacific ) liao , y .- w . 2008 , ( lett . ) , submitted ( this issue ) lin , k .- y . , p . , tseng , y .- h . , lin , l. , & chiueh , t. 2004 , , 608 , l1 lin , k .- y . 2008 , , submitted liu , g .- c . 2008 , , in preparation lo , k.y . 2001 , in _ aip conference proceedings _ , 586 , 172 markevitch , m. et al . 2000 , , 541 , 542 mason , b.s . , et al 2003 , , 591 , 540 mather , j.c . et al . 1990 , , 354 , l37 molnar , s. et al . 2008 , , in preparation mroczkowski , t.et al . 2008 , apj , submitted ( arxiv:0809.5077 ) nishioka , h. et al . 2008 , , in press ( arxiv:0811.1675 ) nolta , m. r. et al . 2008 , apjs submitted ( arxiv : astro - ph/08030593 ) okabe , n. & umetsu , k. 2008 , , 60 , 345 padin , s. et al . 2002 , , 114 , 83 page , l. et al . 2007 , apjs , 170 , 335 park , c .- g . , ng . , k .- w . , park , c. , liu , g .- c . , & umetsu , k. 2003 , , 589 , 67 pearson , t.j . , et al . 2003 , , 591 , 556 penzias , a. a. & wilson , r. w. 1965 , , 142 , 419 pryke , c. et al . 2008 , apj submitted ( arxiv : astro - ph/08051944 ) readhead , a. c. s. et al . 2004 , science , 306 , 836 rephaeli , y. 1995 , araa , 33 , 541 scaife , a. m. m. et al . 2008 , , 385 , 809 smoot , g.f . 1992 , , 396 , l1 spergel , d.n . et al . 2003 , apjs , 148 , 175 spergel , d.n . et al . 2007 , apjs , 170 , 377 sunyaev , r.a . & zeldovich , y.b . 1970 , comments astrophys . space phys . , 2 , 66 sunyaev , r.a . & zeldovich , y.b . 1972 , comments astrophys . space phys . , 4 , 173 umetsu , k. & futamase , t. 2000 , , 539 , l5 umetsu , k. , chiueh , t. , lin , k .- y . , wu , j .- m . , & tsheng , y .- h . 2004 , monder physics letters a , 19 , 1027 umetsu , k. & broadhurst , t. 2008 , apj , 684 , 177 umetsu , k. et al . 2008 , , in press ( arxiv:0810.0969 ) weinreb , s. , lai , r. , erickson , n. , gaier , t. , wielgus , j. 1999 , _ microwave symposium digest _ , 1999 ieee mtt - s international volume 1 , 101 white , m. , carlstrom , j.e . , dragovan , m. , & holzapfel , w.l . 1999 , apj , 514 , 12 wu , j. h. p. 2004 , modern physics letters a , 19 , 1019 wu , j. h. p. et al . 2008a , modern physics letters a , 23 , 1675 wu , j. h. p. et al . 2008b , , in press ( arxiv:0810.1015 ) zhang , p. , pen , , wang , b. 2002 , , 577 , 555 ll interferometer elements & 7 0.6-meter ( 13 1.2-meter ) , f/2.0 cassegrain + telescope mount & hexapod + telescope backup structure & 6-meter carbon fiber platform + primary reflector & monolithic carbon fiber + surface accuracy & 30 microns rms + secondary reflector & carbon fiber , fixed + array configuration & rings at 0.6@xmath0 m spacings + available baselines & 0.6 - 6.0 meters + operating frequencies & 94@xmath0ghz + maximum angular resolution & @xmath4 + primary beam field of view & 23@xmath77 ( 11@xmath77 ) + receiver band & 86102@xmath0ghz + number of receivers & 7 ( 13 ) dual polarization mmic ip hemt + correlator & 4-lag analog 2 @xmath78 21 ( 78 ) baselines + point source sensitivity & 63 ( 8) mjy in 1 hour on - source integration + m antennas ( _ small solid circles _ ) on a 6@xmath0 m single platform ( _ outer solid circle _ ) . _ right _ : sensitivity of the 7-element amiba as a function of spherical harmonic multipole @xmath79 at two frequency channels @xmath80 ( _ solid _ ) and 98 ( _ dashed _ ) ghz , shown by window functions for three baselines with different lengths ( see eq . [ 14 ] of white et al . [ fig2],width=529 ] m antennas co - mounted on a 6@xmath0 m platform . shown in the right panel are the receiver packages mounted on the platform together with various electronics such as the correlator and lo / if systems . the reflectors and receivers can be deployed at various locations on the platform in order to achieve different projected baselines . [ fig3],title="fig:",width=264 ] m antennas co - mounted on a 6@xmath0 m platform . shown in the right panel are the receiver packages mounted on the platform together with various electronics such as the correlator and lo / if systems . the reflectors and receivers can be deployed at various locations on the platform in order to achieve different projected baselines . [ fig3],title="fig:",width=264 ] ) , revealing a strong sze decrement of about @xmath81 in the cluster center ( wu et al . the field size shown is @xmath82 , corresponding to @xmath83mpc@xmath60 at the cluster redshift . the black circle indicates the size of the amiba field - of - view ( @xmath84 fwhm ) , and the black filled circle at the bottom - right corner shows the size of the amiba synthesized beam ( @xmath56 fwhm ) . the residual rms noise level in the cleaned map is @xmath85mjy ( wu et al . overlaid are the contours ( _ white _ ) of the projected mass distribution reconstructed from subaru weak lensing data ( see umetsu et al . 2008 ; okabe & umetsu 2008 ) . the contours are spaced from the @xmath51 noise level at intervals of @xmath86 . shown at the bottom - left corner is the @xmath55 fwhm of the gaussian smoothing kernel ( _ white filled circle _ ) used for mass reconstruction . the weak lensing map shows a mass subclump in the northwest region located about @xmath87 northwest from the cluster center . a slight excess of the sze signal , extending in the northwest direction , is seen in the northwest region at @xmath62 levels . [ fig6],width=491 ]

in the quest for new coating materials with outperforming corrosion and wear resistance properties , chromium carbide thin films are of high interest in applications for electrical contacts and decorative purposes @xcite . metal carbide films produced by physical vapor deposition ( pvd ) , show different microstructures , ranging from epitaxial single - crystal materials to nanocomposites with crystallites of the carbide in an amorphous carbon / amorphous carbide matrix @xcite . completely amorphous metal carbide coatings are also found , in both binary and ternary systems such as cr - c , fe - c , ni - c , and w - fe - c @xcite . it is an important methodological challenge to theoretically model such complex non - crystalline environments , in order to understand their structure and bonding properties . in the case of _ crystalline _ materials , there is a multitude of successful combined experimental and theoretical studies in the literature , see for example refs . @xcite . on the contrary , it is far from straightforward to model _ amorphous _ structures for studying materials properties . the lack of long - range translational symmetry provides a formidable challenge for obtaining suitable model structures of the systems . this is also the case experimentally , as the samples become _ x - ray amorphous _ with no usable diffraction peaks to determine the structure . furthermore , in the computation of xanes and exafs spectra , the absence of symmetry means that the excitation response of each atom is unique and can be very distinct . as spectral measurements are effective samplings of individual excitations of atoms in different local chemical environments , an average over a large number of sites must be considered in a theoretical approach . while detailed explanation for the formation of non - crystalline phases during pvd in certain carbide systems remains unclear , it involves a significant quenching for conditions with neutral atoms and substrate temperatures below the melting point of the carbide , giving rise to limited surface mobility where atoms only find locally favorable sites . it is known that amorphous films often occur in metallic carbide systems where crystalline phases exhibit structural units that have carbon atoms in a mixture of both octahedral and prismatic sites @xcite . previous experimental studies of amorphous cr@xmath0c@xmath1 films @xcite suggest that the films contain two components ; an amorphous metal - rich carbide phase , and an amorphous carbon - rich phase . with increasing total carbon content , the amorphous carbon - rich contribution is believed to grow and incorporate the additional carbon , leaving the metal - rich phase more or less constant in carbon content @xcite . from c @xmath3 x - ray photoelectron spectroscopy ( xps ) measurements on the same samples as in this work , it was determined , based on quantitative estimations of carbide and a - c contributions in the spectra , that the amorphous metal carbide phase contained approximately 20 - 30 at.% of carbon . this is in the same compositional range as e.g. the crystalline cr@xmath7c@xmath8 and cr@xmath5c@xmath9 phases . on the other hand , the amorphous carbon phase contained approximately pure carbon with a high amount of @xmath10-hybridized graphene - like carbon for large amounts of total carbon contents @xcite . however , to identify the detailed structure and the bonding of the carbide phase in these complex non - equilibrium materials @xcite , more detailed experimental and theoretical studies are needed , in particular focusing on the surrounding of the metal atoms . in this work , we investigate and compare experiments for nanocomposite cr@xmath0c@xmath1 thin films with theory for _ completely amorphous _ systems as a function of carbon concentration , together with _ ideal crystal _ structures . x - ray absorption near - edge structure ( xanes ) and extended x - ray absorption fine structure ( exafs ) spectroscopies are ideal tools for characterizing the structure and chemical bonding in crystalline materials @xcite . since these techniques are element specific , we anticipate that they can also be applied as efficient probes of the local environment in more complex non - crystalline materials . in particular , we investigate the cr @xmath2 near - edge ( @xmath3 ) and exafs signals of the crc@xmath11 carbide phase in amorphous cr@xmath0c@xmath1 thin films by experiment and _ ab initio _ calculations , including metal cr for comparison . in order to model _ completely _ amorphous systems , we apply the so - called _ stochastic quenching _ ( sq ) method @xcite . all structural models were obtained using density functional theory ( dft ) @xcite utilizing the projector augmented wave ( paw ) @xcite method as implemented in the vienna ab - initio simulation package ( vasp ) @xcite . the generalized gradient approximation ( gga ) @xcite for electronic exchange and correlation effects was used in these simulations . the atoms are first placed randomly in a supercell and then relaxed using dft until the force on each atom is negligible @xcite . in this way a large set of structures is obtained , which together represent a specific concentration , at a comparatively low computational cost . finally , to produce the theoretical x - ray absorption spectra , an average is taken over spectra computed at each cr - atom site in the amorphous structure . in order to consider possible microstructures and phases within an amorphous matrix , we also consider crystalline crc@xmath1 structures that exist experimentally : _ _ cf__116-cr@xmath7c@xmath8 , _ _ op__16-cr@xmath9c , _ _ hp__20-cr@xmath5c@xmath9 , _ _ hp__80-cr@xmath5c@xmath9 , _ _ op__40-cr@xmath5c@xmath9 , _ _ op__20-cr@xmath9c@xmath12 , as well as the cubic b1 ( octahedral ) and hypothetical hexagonal hp2 ( trigonal prismatic ) type of crc . all spectra are calculated by employing the computationally efficient real - space green s function code feff9 @xcite . in this way it is possible to produce the many individual spectra which are needed to represent amorphous structures . different magnetic structures , ferromagnetic and antiferromagnetic were considered as initial configurations , although , with the exception of b1 crc , magnetic moments relaxed to zero or almost zero during the electronic structure calculation . the structural models used for simulations of the spectra of amorphous cr@xmath13c@xmath14 , cr@xmath15c@xmath15 , and cr@xmath14c@xmath13 were obtained using the stochastic quenching approach @xcite in combination with dft . cubic supercells containing 250 atoms ( cr@xmath15c@xmath15 ) or 252 atoms ( cr@xmath13c@xmath14 and cr@xmath14c@xmath13 ) were used , allowing for the selected concentrations to be modeled accurately . the stochastic quenching procedure was carried out in a similar procedure as in ref . the atomic positions were initiated with a random distribution in the simulation box , avoiding , for numerical stability concerns in the electronic structure calculations , pair distributions shorter than 0.8 , which is about 0.6 times the calculated c - c dimer distance in vacuum . this random structure is then quenched by gradually relaxing the internal atomic positions inside the supercell , until they reach locally stable positions with forces on each ion being smaller or in the order of 0.02 ev / . for each composition , five or six volumes , with 19 - 25 sq structures per volume , were used to obtain the equilibrium volume with a cubic polynomial fit . a similar number of sq structures were then used to obtain average pair distribution functions g(r ) for each bond type and to simulate spectroscopic properties . the calculations were initiated with individual cr atoms having either spin up or spin down magnetic polarization with equal probability and a spatially random distribution in order to simulate a paramagnetic configuration . some cr atoms , in particular in the compositions cr@xmath14c@xmath13 , retained a magnetic moment after the sq procedure with a mean magnitude of about 0.8 @xmath16 with minimum and maximum magnitudes 0.1 and 2.2 @xmath16 , respectively . for the composition cr@xmath15c@xmath15 the corresponding mean value is about 0.3 @xmath16 with a few of the cr moments being around 2 @xmath16 . for the cr@xmath13c@xmath14 composition most of the local moments disappear with a mean magnitude of around 0.02 @xmath16 . overall , the local magnetic moments are clearly less robust as compared to similar simulations for the related amorphous crn - system @xcite . a gaussian broadening of the calculated pair distances of 0.2 was used in the simulations . figure 1 shows one of the supercells ( sq1 ) generated by the procedure of the sq method in order to model amorphous cr@xmath15c@xmath15 . large blue spheres correspond to cr and smaller brown spheres are c atoms . in order to obtain the theoretical x - ray absorption spectrum for an amorphous system , an average of all the spectra for the different sq supercells and cr sites was performed . the cr - atoms are represented by large blue spheres , while c - atoms are shown as smaller brown spheres . the supercell was drawn using vesta @xcite.,width=340 ] in order to produce theoretical xanes and exafs spectra for the structures described above , we employ the single particle green s function technique within dft @xcite . all calculations were performed utilizing the computationally efficient _ ab initio _ real - space green s function ( rsgf ) approach of the feff9 code @xcite . for example , in the case of cr@xmath15c@xmath15 the resulting cr @xmath2-edge spectrum is an average over 22 sq structures , each in turn an average over 125 cr sites , amounting to 2750 spectra in total to represent the amorphous system . all the potentials are calculated self - consistently and the default hedin - lundquist exchange correlation potential is used . in more detail , to obtain accurate near - edge structures considering xanes , we utilized the real - space full multiple scattering ( rsfms ) approach within feff9 . here , the potentials were calculated for the central scattering cr atom and for the other cr and c atoms . to take the final state effect into account , a fully screened core - hole was used for the central scattering atom , i.e. applying the final - state rule , or core - hole approximation . clusters with more than 550 atoms were employed in order to converge the spectral features in the different systems . for a brief overview of different methodologies for calculating xanes see ref . @xcite and references within . for exafs , which considers the energy region far above the near - edge , it is sufficient to use multiple scattering paths with no approximation for the core - hole @xcite . finally , the theoretical spectra were aligned with experiment by applying a rigid shift . amorphous chromium carbide ( cr@xmath0c@xmath1 ) samples with different carbon content were made using direct current ( dc ) magnetron sputtering as described elsewhere @xcite . it was also found @xcite that the samples contain about 2 - 5 at.% oxygen present . the xanes and exafs spectra were measured at the undulator beamline i811 on the max ii ring of the max iv laboratory , lund university , sweden @xcite . the energy resolution at the cr @xmath3 edge of the beamline monochromator was better than 0.5 ev . the x - ray absorption spectra were recorded in reflection mode by collecting the fluorescence yield using a pips detector from -150 ev below to 1000 ev above the cr @xmath3 absorption edge ( 5989 ev ) with 0.5 ev energy steps . to avoid self - absorption effects in the sample and bragg peaks from the substrate , the incidence angle on the sample was varied in 0.25 degrees steps in a maximum range of @xmath173 degrees from normal incidence with a step motor . based on the xanes fitting results ( described in the next section ) , two cr - c and two cr - cr scattering paths obtained from feff9 @xcite were included in the exafs fitting procedure with _ _ hp__20-cr@xmath18c@xmath19 as a model system using the viper software package @xcite . the reference energy @xmath20 was set to the pre - peak of each cr @xmath3 x - ray absorption spectrum as determined from the first peak of the derivative of each spectrum relative to the pure bcc - cr reference sample . in the fitting procedure , @xmath21 was set to 5988.6 ev ( 25% c ) , + 0.1 ev ( 42 and 47% , as compared to the 25% c sample ) , + 0.4 ev ( 67% c ) and + 0.6 ev ( 85% c ) , respectively . the exafs functions @xmath22 were extracted from the raw absorption data after removing remaining glitches originating from substrate diffraction peaks , subsequent background subtraction , and summarizing and averaging of 15 absorption spectra . the bond distances ( r ) , number of neighbors ( n ) , debye - waller factors ( @xmath23 , representing the amount of disorder ) and @xmath24=reduced @xmath25 as the squared area of the residual , were determined by fitting the back - fourier - transform signal between @xmath26=0 - 12 @xmath27 obtained from the fourier - transform within @xmath28=0.7 - 3 of the first coordination shell using a hanning window function ) @xcite and a global electron reduction factor of @xmath29=0.8 . the calculated pairs of cr - c and cr - cr 1st and 2nd scattering paths in the model were obtained using the feff9 code @xcite . in this section , we start by considering theoretical xanes and exafs cr @xmath2-edge spectra for the amorphous and crystal chromium carbide structures , with a comparison to measurements carried out for the thin film systems and pure cr metal . an analysis of the xanes data was done by fitting the measured spectra with the calculations using a superposition of an amorphous phase and a smaller contribution of different crystalline phases . using the results of the xanes fitting , an analysis of the exafs data was carried out for the experimental data using a standard fitting procedure for crystalline materials , in order to obtain reduced ( radial ) pair distribution functions ( rdfs ) , that are compared to theoretical rdf results . finally , the trend in the structural parameters of bond lengths and coordination numbers extracted from the exafs analysis are compared and discussed with theoretical values . -edge , a ) xanes and b ) exafs , 125 cr - sites ( gray lines ) , are exemplified for the structure sq1 in figure 1 , together with the spectra for cr@xmath15c@xmath15 over all 22 sq structures . the average of the spectra are shown as black lines.,width=340 ] in figure 2 , the large differences of the theoretical a ) xanes and b ) exafs spectra in the case of cr@xmath15c@xmath15 are shown . the upper parts of the two panels a ) and b ) give the cr @xmath2-edge spectra calculated over the 125 different cr - sites ( gray lines ) for the sq1 structure illustrated in fig . 1 . the averaged spectra are shown as solid black lines . the lower parts of the panels correspond to the averaged results obtained for each of the 22 sq structures ( gray lines ) . finally , the resulting xanes and exafs spectra over the whole amorphous cr@xmath15c@xmath15 structure are given by the solid black lines in the lower part of each panel . as observed in fig . 2 , the x - ray absorption fine structures are highly sensitive to the atom local environment in the amorphous matrix . similar behavior was found at the other concentrations . our result in fig . 2 contains an important message to theoreticians working with spectroscopical simulations of disordered materials , and in particular amorphous substances . it is clear that in order to give even a qualitatively reasonable description of the cr xanes or exafs spectra , averaging over hundreds of atomic sites is needed . however , for a quantitative simulation of the average signal , which is what is measured experimentally , in particular for the exafs case it might not be enough to consider just one amorphous supercell . it can be seen in the lower curve of fig . 2 that the computed average signals for some of the sq supercells deviates considerably from the average line of all the samples . thus , it might not be reliable to simulate exafs spectra with just one supercell , say generated with molecular dynamics and containing one- or two hundred atoms , even if the average spectra of all atoms in the supercell is considered . -edge , a ) xanes and b ) exafs , are compared for amorphous cr@xmath13c@xmath14 ( i ) , cr@xmath15c@xmath15 ( ii ) , cr@xmath14c@xmath13 ( iii ) and with the different crystal structures : bcc - cr ( 1 ) , _ _ cf__116-cr@xmath7c@xmath8 ( 2 ) , _ _ op__40-cr@xmath5c@xmath9 ( 3 ) , _ _ hp__80-cr@xmath5c@xmath9 ( 4 ) , _ _ hp__20-cr@xmath5c@xmath9 ( 5 ) , _ _ op__20-cr@xmath9c@xmath12 ( 6 ) , _ _ op__16-cr@xmath9c ( 7 ) , phases and cubic b1 [ nacl ] ( 8) and hexagonal hp2 [ wc ] ( 9 ) type crc.,width=340 ] in figure 3 , a comparison of theoretical a ) xanes and b ) exafs spectra is made between amorphous cr@xmath0c@xmath1 and crystalline crc@xmath11 structures . in the upper parts of each panel , results for three different amorphous cr@xmath0c@xmath1 structures with concentrations @xmath30 ( blue dashed lines ) , @xmath31 ( green line ) and @xmath32 ( red dotted line ) are shown . in the lower part of each panel , the corresponding spectra for the crystals are presented : bcc - cr , _ _ cf__116-cr@xmath7c@xmath8 , _ _ op__16-cr@xmath9c , _ _ hp__20-cr@xmath5c@xmath9 , _ _ hp__80-cr@xmath5c@xmath9 , _ _ op__40-cr@xmath5c@xmath9 , _ _ op__20-cr@xmath9c@xmath12 , structures from experiment and cubic b1 ( octahedral nacl ) and hypothetical hexagonal _ _ hp__2 ( trigonal prismatic wc ) type of crc . xanes spectra of amorphous cr@xmath0c@xmath1 and crystal pure cr are shown . the spectra were normalized below the edge and 100 ev above the absorption edge . the inset shows the experimental intensities at the pre - edge peak . at the bottom , the first derivative of the measured spectra are shown . the inset is a close - up of the first peak ( marked with arrow ) . the edge energies @xmath20 and the chemical shifts were determined by curve fitting of the first peak of the derivative to 5989.0 ev ( pure cr ) , 5988.6 ev ( 25% c ) , + 0.1 ev ( 42 and 47% , as compared with 25% ) , + 0.4 ev ( 67 % ) and + 0.6 ev ( 85 % ) .,width=340 ] in figure 4 , a direct comparison is made between the cr @xmath2-edge xanes spectra from experimental measurements for different compositions , @xmath33=0.25 , 0.42 , 0.47 , 0.67 and 0.85 ( top ) , and the structures obtained by our theoretical modeling @xmath33=0.33 , 0.5 and 0.67 ( middle ) , including results for pure crystal cr . the theoretical concentrations are chosen to capture the overall trend in the system with the variation of carbon concentration . at the bottom , the first derivatives of the experimental spectra are shown . the chemical shift to higher energy is a signature of higher ionization of the chemical state of the absorbing cr atoms as the carbon content increses . for the exafs spectra , theory reproduces the general shape of the measurements . as in the case of xanes , the amorphous materials yield softer spectral features compared to crystalline materials . xanes spectra at 25@xmath34 c ( red circles ) with theory for amorphous cr@xmath13c@xmath14 ( green line ) and _ _ hp__20-cr@xmath5c@xmath9 ( blue line ) . sum ( black line ) shows the best fit , compare with table 1.,width=340 ] @xmath35 typically , it is most fruitful to analyze the measured exafs data in the form of reduced pair distribution functions , which provide information about the average cr - cr and cr - c bond lengths , which in turn can be compared with corresponding theoretical pair distribution functions . in order to evaluate the probability of different compositions , we have fitted each measured xanes spectrum with the theoretical fully amorphous 33 @xmath34 c structure together with each one of the crystal structure candidates as listed in table 1 . the curve fitting analysis was done using the spectral analysis by curve fitting macro package ( spancf ) software @xcite package . in figure 5 , an example is shown for the fit ( black line ) of the 25 @xmath34 c sample ( red dots ) with amorphous cr@xmath36c@xmath37 ( green line ) and _ _ hp__20-cr@xmath5c@xmath9 ( blue line ) . the background was taken into account both by an integrated and a linear background ( dashed lines ) and the residual is shown by a thin black curve . the square of the residual was used to compare the quality of different fittings . from the fitting analysis of the xanes data in table 1 , the results show that the contribution of the amorphous cr@xmath0c@xmath1 phase increases with carbon content from 40% to 90% . in all simulations the cr@xmath5c@xmath9 , cr@xmath7c@xmath8 and cr@xmath9c@xmath12 are the most likely crystalline - like coordination while wc , pure cr , cr@xmath19c and b1 are unlikely . by combining experimental and theoretical xanes and exafs data at the cr @xmath2-edge in figs . 2 - 5 we make several interesting observations . both in the xanes ( fig . 4 ) and the exafs spectra , we note that the main features observed in experiment are also captured in the calculated spectra . concerning the xanes data , it is known that the energy position and the shape of the main peak and the pre - edge depends on the chemical state of the absorbing atom . the pre - peak is due to a transition of a @xmath3 electron to hybridized @xmath38 and @xmath39 ( cr @xmath40 - c @xmath41 ) states . on the contrary , the main peak is a pure cr @xmath3 @xmath42 @xmath43 dipole transition . the first derivative of the xanes in the bottom of fig . 4 highlights the differences between the samples . the intensity of the pre - peak is a signature of the amount of @xmath44-@xmath45 hybridization @xcite , which here decreases as a function of carbon content . generally , the chemical state of the absorbing cr atom is different in cr metal , cr@xmath9c@xmath12 and cr@xmath12o@xmath9 @xcite and the spectra shift to higher energies with increasing average oxidation states that are higher than in cr metal . note that this could also be affected by the presence of small amounts of oxygen in the samples , up to 5 at.% @xcite , though it is difficult to precisely estimate the effect . although the shape of the peaks and intensities are similar in all the studied cr@xmath0c@xmath1 samples ( fig . 4 ) there is a noticeable chemical energy shift towards higher energy with increasing c content . it is also possible to distinguish these differences in a close study of the spectral fine structure . for instance , the samples containing 42 and 47% c show sharper resolved pre - edge peaks ( fig . 4 ) . in the case of the calculated spectra for the amorphous systems ( fig . 2a ) , the individual spectrum of each and every atom is unique with a detailed fine structure of the pre - edge peak and the main line . the crystallite systems ( fig . 3a ) present very distinct and sharp spectral features both at the main peak and the pre - edge . however , the spread of the different crystal spectra is smaller than the rich variation found within an amorphous structure . for the amorphous cr@xmath0c@xmath1 materials , the averaged spectra over thousands of cr atoms show more smooth features as compared with the crystallites , in agreement with measurements . still , the overall shape of the spectra with pre - edge and main peak is similar . these spectral features are also seen for the pure cr metal , similar to previous results @xcite . c@xmath1 films are shown for the different c concentrations , with comparison to crystal pure cr . the fit ( black curve ) is compared to the back - fourier - transform of the first coordination shell within @xmath28=0.7 - 3 and @xmath26=0 - 12 @xmath27 . , width=340 ] figure 6 shows experimental exafs data in @xmath46-space of bcc - cr metal and the five amorphous nanocomposite cr@xmath0c@xmath1 films obtained after normalizing and background removal using the viper software package @xcite . for bcc - cr , the main sharp oscillations occur in the 4 - 11 @xmath27 region . on the contrary , for the amorphous cr@xmath0c@xmath1 films , the main oscillations occur in the region @xmath26=2 - 7 @xmath27 , whereas for higher @xmath26-values the oscillations are damped out in increasing noise from about @xmath26=12 @xmath27 . this behavior is characteristic of backscattering from light c atoms @xcite . c@xmath1 extracted from the fourier transform of the exafs s(q ) data at the cr @xmath3 absorption edge . the vertical dotted lines indicate the relative positions of the cr - c and cr - cr peaks but the g(r ) functions represent raw data that have not been phase shifted by @xmath47 + 0.5 ev ( compare with table 2).,width=347 ] figure 7 shows the radial distribution functions , obtained directly by fourier transforming the measured exafs structure factors in fig . 6 by the standard exafs procedure @xcite . the quantitative analysis was made using @xmath48 weight as is normal for conductive samples with a reduced @xmath26-range between 0 and 12 @xmath27 . the main peaks are dominated by the short - range order of the cr - c and cr - cr orbital overlaps observed at @xmath471.5 and @xmath472.2 , in the first coordination shell , corresponding to @xmath26 @xmath49 2 - 6 @xmath27 in fig . 6 . above 2.5 , corresponding to @xmath26 @xmath4 6 @xmath27 , a smaller peak is observed at 3 - 3.5 due to the second coordination shell . note that the radius - values on the abscissa shown in fig . 7 represent raw data and are shown non - phase shifted to absolute values . @xmath50 table 2 shows the results of the exafs fitting using scattering paths from the most probable structure ( _ _ hp__20-cr@xmath5c@xmath9 with a large unit cell with lattice parameters @xmath51=@xmath52=6.96 and @xmath53=4.45 ) from the xanes results in table 1 . for amorphous materials , it is not possible to choose a perfect model structure due to the inherent disorder , however , selecting a simpler ( and less probable ) model such as b1-crc results in less accurate fitting with larger residuals . the phase shift ( @xmath47 + 0.5 ) is included in the fitting model for the bond lengths for easier comparison with the theoretical values . the details of the fitting procedure are described in the experimental section iii . the first line in table 2 presents the fitting results of the reference sample of pure bcc - cr with known coordination numbers for the first ( 8) and second ( 6 ) scattering paths . these nearest neighbor scattering paths represent the bond lengths space diagonal from the corner of the unit cell to the center ( 2.49 ) and the edge length equal to the lattice parameter ( 2.88 ) . as expected , the amorphous samples on the following lines exhibit larger debye - waller factors than crystal bcc - cr due to a greater amount of disorder . for comparison of the coordination numbers , the debye - waller factors were kept fixed for the amorphous samples . generally , we observe a trend of decreasing cr - cr coordination numbers ( n@xmath54+n@xmath55 ) from 9.5 , 7.6 , 6.7 , 5.9 to 4.6 with increasing carbon content . with exception of the 25% c sample , we also observe that the cr - cr and cr - c radii slightly decrease as a function of carbon content . for the cr - c coordination numbers , the decreasing trend is less obvious although as it first increases from 1.7 to 4.1 as the carbon content increases from 25% c to 42% c , probably associated with the formation of domain structures with interfaces . thereafter , the coordination number decreases to 3.6 at 47% c , followed by a small increase to 3.7 and 3.8 as a function of carbon content . the dashed curves in fig . 7 show the individual scattering path components resulting from fitting the pairs of cr - c ( 1,2 ) and cr - cr ( 3,4 ) contributions in the first coordination shell obtained from feff9 of the _ _ h__p20-cr@xmath5c@xmath9 structure . as observed in fig . 7 ( and table 2 ) , the cr - cr peaks in the first coordination shell are not experimentally resolved from the cr - c peaks for the lowest carbon content ( 25 at% ) . as will be discussed below , the most likely explanation is that the difference between the cr - c and cr - cr bonds becomes smaller at low c - content due to carbon vacancies in the structure . the cr - cr bond distance is shorter in the carbide compared to the more open reference bcc - cr structure . as the carbon content increases to 42 at% , the cr - cr peak shifts to larger radius values due to the expansion of the structure . at the same time , the cr - c peak becomes clearly distinguished at lower radius . for carbon contents above 42 at% the cr - c and cr - cr peaks stays approximately at the same values as indicated by the vertical dotted lines . this signifies that the carbide structure remains the same , independent of carbon content . for the second coordination shell corresponding to larger cr - cr and cr - c distances between 3.7 - 4.2 , it is interesting to notice that the relative intensity is rather independent of carbon content while the shape changes . figure 8 shows theoretical ( radial ) pair distribution functions g(r ) obtained directly from the sq structures for amorphous cr@xmath14c@xmath13 , cr@xmath15c@xmath15 , and cr@xmath13c@xmath14 . from top to bottom , the graphs display g(r ) for cr - cr , cr - c and c - c , respectively . the inset is zoomed in on the first cr - cr peak , illustrating the small but distinct shift to larger cr - cr distances upon increasing carbon content . the pair distribution function g(r ) is related to the reduced pair distribution function g(r ) by g(r)=g(r)/4@xmath56@xmath57 + 1 , where @xmath57 is the average number density of the material . the ( radial ) pair - distribution function is normalized so that g(r ) @xmath42 1 when r @xmath42 @xmath58 @xcite . note that while the theoretical g(r ) is split into separate parts , g(r ) obtained from the measurements includes all the cr bond lengths , in particular , the cr - cr and cr - c peaks in the first coordination shell . figure 9 shows theoretical cr - cr , cr - c and c - c g(r ) obtained from the optimized geometries of crystalline crc@xmath11 phases : _ _ cf__116-cr@xmath7c@xmath8 , _ _ op__16-cr@xmath9c , _ _ hp__20-cr@xmath5c@xmath9 , _ _ hp__80-cr@xmath5c@xmath9 , _ _ op__40-cr@xmath5c@xmath9 , _ _ op__20-cr@xmath9c@xmath12 , the cubic b1 ( octahedral ) and hp2 ( trigonal prismatic , wc ) type crc and bcc - cr . these are the same phases whose xanes and exafs spectra are shown previously in fig . 3 . calculated average bond lengths and nearest neighbors are summarized in table 3 for the amorphous systems and crystalline phases , also including c - c bonds . as shown by the exafs results in table 3 , we generally observe that the crystalline bond lengths are similar to both the theoretical fully amorphous structures and measurements ( table 2 ) , and the coordination numbers are closer to the measurements . considering the bond lengths , cr@xmath5c@xmath9 phases give the closest fit with the xanes measurements ( table 1 ) . from the fitting of the xanes measurements to the theoretical results , we find that the dominant contribution is amorphous crc@xmath1 carbide with a coordination that is most similar to cr@xmath5c@xmath6 carbide in all the samples , ranging from @xmath47 30 at% . this is followed by less possible contributions of coordination similar to cr@xmath7c@xmath8 , cr@xmath9c@xmath12 and cr@xmath9c . note that these results are in a general good agreement with the previous experimental findings by maury _ et al . _ @xcite and nygren _ et al . _ @xcite . @xmath59 in table 3 , a very small increase of both the cr - c and cr - cr bond lengths as a function of c content is predicted in the theoretical purely amorphous structures . this is consistent with the observed increase in crystal carbide lattice parameters when they are present in nanocrystal form in a carbon matrix @xcite . this trend is possibly due to the fact that the carbon - rich environment induces charge - transfer that reduces the electron concentration around the cr atoms with a following reduction of the strength of cr - cr bonding . in addition , our calculations of pure amorphous materials predict a reduction of the coordination numbers of cr and an increase of coordination numbers of c as a function of carbon content . the predicted reduction of the coordination number of cr is consistent with our measurements . however , an increase in the coordination number of c is only experimentally observed between the 25% and 42% samples while at higher c contents , a weak decrease is observed . this could be due to the fact that the size of the amorphous crc@xmath1 carbide and amorphous carbon domains changes as a function of total carbon content . as observed in transmission electron microscopy ( tem ) @xcite , domain sizes are less than 1 nm for x=0.25 , about 3 nm for x=0.47 , and about 2 nm for the higher carbon contents . thus , the situation in amorphous nanocomposites is more complex than in pure amorphous materials as there is also a contribution from the interface grain boundaries that changes as a function of total carbon content . furthermore , for the crc@xmath1 carbide phase , there are at least two different competing effects that can give rise to a change in bond length : i ) at low c - content , the cr - rich phase has vacancies that cause an increase or an decrease of the lattice constant that has a maximum for under - stoichiometric carbide . ii ) more carbon atoms filling into the vacancies of the structure gives rise to a larger lattice parameter @xmath51 , see e.g. , ref . @xcite . in a previous experimental study on the same samples @xcite , @xmath33=0.25 was observed to have the most amorphous character , in high resolution transmission electron microscopy ( hrtem ) . a possible interpretation is that the low carbon concentration sample can be described with a cr@xmath5c@xmath6 coordination with @xmath60 -0.5 , prohibited from forming the optimal structure by carbon vacancies . the fitting with theoretical xanes spectra ( table 1 ) does not indicate a change in structure or a possible concentration of bcc - cr , while having the lowest amorphous character as compared with the other samples . we find that in general , exafs modeling can give useful information about the local structure in amorphous materials but largely depends on the chosen model . therefore , it is important to use _ ab initio _ theory as a complement . while the present work focus on new theoretical and experimental results for the cr @xmath2-edge , the corresponding results for the _ carbon _ @xmath2-edge can provide valuable complementary information . therefore , it is also of interest to compare the theoretical c - c bond for the amorphous structures and crystallites ( fig . 8 , 9 and table 3 ) with measurements . generally , the calculated cr - c and cr - cr bond lengths were found to be rather similar between the crystallites and amorphous structures . however , the theoretical c - c bond distances are strikingly different in amorphous carbides ( @xmath471.5 ) compared to crystalline carbide materials ( @xmath473 ) , which are closer to the second c - c coordination shell . as discussed in a previous work @xcite , these results qualitatively agree well with the relative distribution of carbon in the carbidic , crc@xmath1 phase and the amorphous carbon phase as reported in ref . @xcite . for comparison , the c - c bond lengths are @xmath471.5 in a - c , graphite and diamond @xcite . the experimentally observed c - c bond lengths ( @xmath472 ) @xcite are larger than the theoretical , but significantly shorter than the bonds for the crc crystallites in table 3 , which are similar to previous calculations @xcite . note that in this study , the cr - c bond length ( @xmath47 2.0 ) was extracted from exafs modeling , while in ref . @xcite the bond length ( @xmath47 2.2 ) was only estimated directly from the rdf obtained from electron energy loss spectroscopy with lower energy resolution . with the combination of experiment and theory for xanes and exafs , we have shown that it is possible to obtain detailed information about the local structure in amorphous nanocomposite materials . using the computationally efficient _ ab initio _ stochastic quenching method , we modeled fully amorphous structures , taking into account local cr magnetic moments in the derivation . experimentally , amorphous structures require advanced fitting procedures where the result is very sensitive to the structure of the model . amorphous two - phase cr@xmath0c@xmath1 nanocomposite films were investigated , consisting of a chromium - rich amorphous carbide phase and an amorphous carbon - rich phase . we find that the short - range coordination in the cr - rich amorphous phase is similar to that of a hexagonal cr@xmath5c@xmath6 structure . at total carbon contents below 30 at% , the cr - rich phase is clearly under - stoichiometric with carbon vacancies that causes shorter cr - cr bond lengths and directly depend on the packing and orbital overlap with the cr - c bonds . furthermore , the cr - cr bond length in the amorphous carbide has shorter bond lengths than the more open structure of bcc - cr metal ( space diagonal=2.49 , lattice parameter @xmath51=2.88 ) that indicates stronger bonds . above 30 at% total carbon content , the structure of the amorphous cr@xmath5c@xmath6-like domains remains , while the excessive carbons assemble in the amorphous carbon phase . the coordination numbers in the first coordination shell of cr decreases as a function of total carbon content from 9.5 to 4.6 , while the coordination number of c only weakly increases . this is associated with a smaller number and size of the cr - rich domains causing a larger fraction of interface bonding that lacks cr - nearest neighbors on one side . theoretically , we also predict that the c - c bond lengths are very sensitive to the local structure , with considerably shorter bond lengths in amorphous materials ( @xmath471.5 ) compared to various crystals ( @xmath472.8 - 3.2 ) . generally , it is important to use theory as an assessment to investigate the trends in complex amorphous materials . we would like to thank the staff at max - lab for experimental support and u. jansson and m. andersson for providing the samples . this work was supported by the swedish research council ( vr ) linnaeus grant lili - nfm , the funcase project supported swedish strategic research foundation ( ssf ) . acknowledge financial support from vr grant no . 621 - 2011 - 4426 , the swedish government strategic research area in materials science on functional materials at linkping university ( faculty grant sfo - mat - liu no 2009 00971 ) , knut and alice wallenbergs foundation project strong field physics and new states of matter 2014 - 2019 ( cotxs ) . b.a . would like to thank e. holmstrm and r. lizrraga for support with the sq method and acknowledges financial support by the swedish research council ( vr ) through the young researcher grant no . 621 - 2011 - 4417 and the international career grant no . 330 - 2014 - 6336 and marie sklodowska curie actions , cofund , project inca 600398 . the calculations were performed using supercomputer resources provided by the swedish national infrastructure for computing ( snic ) at the national supercomputer centre ( nsc ) and center for parallel computing ( pdc ) . cavaleiro , a. ; de hosson , j. t. m. _ nanostructured coatings _ ; springer : new york , 2007 . cheng , y. q. ; ma , e. atomic - level structure and structure - property relationship in metallic glasses . _ prog . mat . science _ * 2011 * , _ 56 _ , 379 . furlan , a. ; lu , j. ; hultman , l. ; jansson , u. ; magnusson , m. crystallization characteristics and chemical bonding properties of nickel carbide thin film nanocomposites . _ j. phys . : matt . _ * 2014 * , _ 26 _ , 415501 . magnuson , m. ; wilhelmsson , o. ; mattesini , m. ; li , s. ; ahuja , r. ; eriksson , o. ; hgberg , h. ; hultman , l. ; jansson , u. anisotropy in the electronic structure of v2gec investigated by soft x - ray emission spectroscopy and first - principles theory . b _ * 2008 * , _ 78 _ , 035117 . olovsson , w. ; tanaka , i. ; mizoguchi , t. ; radtke , g. ; puschnig , p. ; ambrosch - draxl , c. al l2,3 edge x - ray absorption spectra in iii - v semiconductors : many - body perturbation theory in comparison with experiment . b _ * 2011 * , _ 83 _ , 195206 . olovsson , w. ; weinhardt , l. ; fuchs , o. ; tanaka , i. ; puschnig , p. ; umbach , e. ; heske , c. ; draxl , c. the be k - edge in beryllium oxide and chalcogenides : soft x - ray absorption spectra from first - principles theory and experiment . _ j. phys . : matt . _ * 2013 * , _ 25 _ , 315501 . kdas , k. ; andersson , m. ; holmstrm , e. ; wende , h. ; karis , o. ; urbonaite , s. ; butorin , s. m. ; nikitenko , s. ; kvashnina , k. o. ; jansson , u. ; eriksson , o. structural properties of amorphous metal carbides : theory and experiment . _ acta materialia _ , 2012 , * 2012 * , _ 60 _ , 4720 - 4728 . holmstrm , e. ; bock , n. ; peery , t. b. ; lizrraga , r. ; delorenzi - venneri , g. ; chisolm , e. d. ; wallace , d. c. ab initio method for locating characteristic potential - energy minima of liquids . e _ , 2009 , * 2009 * , _ 80 _ , 051111 . holmstrm , e. ; bock , n. ; peery , t. ; chisolm , e. ; lizrraga , r. ; delorenzi - venneri , g. ; wallace , d. structure discovery for metallic glasses using stochastic quenching . _ phys . rev . b _ * 2010 * , _ 82 _ , 024203 . lindmaa , a. ; lizrraga , r. ; holmstrm , e. ; abrikosov , i. a. ; alling , b. exchange interactions in paramagnetic amorphous and disordered crystalline crn - based systems . b _ * 2013 * , _ 88 _ , 054414 . pantelouris , a. ; modrow , h. ; pantelouris , m. ; hormes , j. ; reinen , d. the influence of coordination geometry and valency on the k - edge absorption near edge spectra of selected chromium compounds . phys . _ * 2004 * , _ 300 _ , 13 - 22 . arcon , i. ; mirtic , b. ; kodre , a. determination of valence states of chromium in calcium chromates by using x - ray absorption near - edge structure ( xanes ) spectroscopy . ceram . _ * 1998 * , _ 81 _ , 222 - 224 . maury , f. ; oquab , d. ; manse , j. c. ; morancho , r. structural characterization of chromium carbide coatings deposited at low temperature by low pressure chemical vapour decomposition using dicumene chromium . coat . tech . _ * 1990 * , _ 41 _ , 51 - 61 . nygren , k ; andersson , m. ; hgstrm , j. ; fredriksson , w. ; edstrm , k. ; nyholm , l. ; jansson , u. influence of deposition temperature and amorphous carbon on microstructure and oxidation resistance of magnetron sputtered nanocomposite cr - c films . sci . _ * 2014 * , _ 305 _ , 143 - 153 . darmstadt , h. ; roy , c. ; kaliaguine , s. ; xu , g. ; auger , m. ; tuel , a. ; ramaswamy , v. solid state 13c - nmr spectroscopy and xrd studies of commercial and pyrolytic carbon blacks . _ carbon _ * 2000 * , _ 38 _ , 1279 - 1287 . li , y. ; gao , j. ; li , q.l . ; peng , m. f. ; sun , x. h. ; li , y. y. ; yuan , g. ; wen , w. j. the electronic , mechanical properties and theoretical hardness of chromium carbides by first - principles calculations . _ j. alloys comp . _ * 2011 * , _ 509 _ , 5242 - 5249 .

our understanding of stellar structure and stability has been built upon classical analytical studies of equilibrium configurations of incompressible self - gravitating systems by maclaurin , jacobi , dedekind , darwin , riemann , and especially @xcite . with the help of modern computers and numerical methods , such as the self - consistent - field ( scf ) technique @xcite and hydrodynamics techniques , these studies have been extended to compressible configurations by a variety of researchers @xcite . however , most three - dimensional ( 3d ) hydrodynamical studies have been confined to a limited parameter space , i.e. , starting from two - dimensional ( 2d ) axisymmetric equilibrium models ( compressible analogues of maclaurin spheroids ) because we have not been able to build 3d compressible equilibrium models with complicated flows . to the author s knowledge , techniques have only been developed to build compressible equilibrium models of nonaxisymmetric configurations for a few systems with simplified rotational profiles , e.g. , rigidly rotating systems @xcite , irrotational systems @xcite , and configurations that are stationary in the inertial frame @xcite . our aim is to find a method by which a much wider class of nonaxisymmetric configurations can be readily constructed . this will permit us to more clearly understand the viability of evolutionary sequences of triaxial configurations and will provide models whose stability properties can be readily analyzed . previous linear stability analyses @xcite have revealed a number of hydrodynamical instabilities that might arise in incompressible riemann ellipsoids , but no corresponding numerical work using state - of - the - art 3d hydrodynamic techniques has been carried out . in particular , riemann s - type ellipsoids have been found to be subject to a hydrodynamic strain instability associated with elliptical stream lines @xcite , which raises concerns about the stability of certain types of geophysical flows and leads to suspicions about the evolutionary path of stars that are driven by gravitational - radiation - reaction ( grr ) forces toward the dedekind - sequence . in a recent nonlinear study of the secular bar - mode instability induced by grr @xcite , a uniformly rotating , axisymmetric neutron star , which was secularly unstable and driven by artifically enhanced grr , evolved into a bar - like configuration with a very slow pattern speed . however , this special bar configuration became unstable to some kind of turbulence - like instability while evolving toward the dedekind - sequence , that is , toward a stationary ellipsoidal structure in the inertial frame maintained purely by the internal motion of the fluid . it is suspected that this turbulence - like instability may be the elliptical strain instability identified in the earlier linear stability analyses . but no definite conclusion can be drawn because the stability of only one dedekind - like configuration was studied . following @xcite , one might consider generating additional ellipsoidal structures hydrodynamically , but the generation of each additional ellipsoidal model would be expensive in the sense that one has to evolve an initially axisymmetric model for a very long time , even with artificially enhanced grr @xcite . this approach would make it impractical to undertake a full stability investigation through the entire parameter space of ellipsoidal models . here we present a new 3d scf technique that is capable of building the full range of incompressible riemann s - type ellipsoids with nontrivial internal flows ( i.e. , not just the uniformly rotating jacobi configurations , which are thought to be the end point of secularly unstable stars driven by viscosity , and stationary dedekind configurations ) as well as compressible triaxial models that share the same divergence - free velocity field as riemann s - type ellipsoids . our compressible models satisfy the steady - state euler s equation exactly , but only satisfy the steady - state continuity equation approximately , hence they are only in quasi - equilibrium . however , the violation of the steady - state continuity equation is very small for nearly incompressible models , hence these models can be used effectively as initial configurations for a hydrodynamical study of the stability of compressible analogues of riemann s - type ellipsoids . we note that , to date , hydrodynamical techniques have not been used to examine the stability of riemann ellipsoids because the sound speed is formally infinite for incompressible models , while the time step in hydrodynamic codes is usually inversely proportional to the sound speed . on the other hand , since the elliptical strain instability associated with elliptical stream lines would set in on a dynamical time scale , we believe that our new `` quasi - equilibrium '' models will prove to be good starting points for hydrodynamical studies of this instability . our numerical method inherits one central idea of the traditional scf method in that it makes use of the properties of the velocity field of riemann s - type ellipsoids to turn the vector form of the steady - state euler s equation into a scalar form , thus yielding a uniform bernoulli s function . this enables us to follow the procedures of the traditional scf method to use several boundary conditions to determine quantities elsewhere . to be concise , we will refer readers to @xcite and references therein for details of the traditional scf technique , and focus on the new aspects of our new method . figure [ triaxial ] reproduces key elements of a plot that ( * ? ? ? * see his chapter 7 , figure 15 ) used to discuss sequences of riemann s - type ellipsoids . in this ( @xmath0,@xmath1 ) parameter plot , @xmath2 , @xmath3 , and @xmath4 are the three principal semiaxes of an ellipsoid , and each riemann s - type ellipsoid with angular velocity @xmath5 and uniform vorticity @xmath6 in the rotating frame resides between the two self - adjoint sequences ( curves lp and lk ) with @xmath7 , where @xmath8 the velocity field of a riemann s - type ellipsoid as viewed from a frame rotating with angular velocity @xmath5 takes the following form : @xmath9 where @xmath10 is a constant that determines the magnitude of the internal motion of the fluid , and the origin of the x - y coordinate system is at the center of the ellipsoid . this velocity field @xmath11 is designed so that velocity vectors everywhere are always aligned with elliptical stream lines by demanding that they be tangent to the equipotential contours , which are concentric ellipses . now , let s turn to a general self - gravitating system with some compressibility . if we adopt a polytropic equation of state , @xmath12 ( @xmath13 is pressure , @xmath14 is the polytropic constant and @xmath15 is the polytropic index ) , euler s equation for steady - state flows in the rotating frame is , @xmath16 - 2 \vec{\omega } \times \vec{v } \ , , \label{euler}\ ] ] where h is enthalpy , and @xmath17 is the gravitational potential . with the velocity field specified by eq . ( [ vel ] ) , one can show that eq . ( [ euler ] ) is equivalent to the following equation , @xmath18 = -\nabla [ h+\phi-\frac{1}{2}\omega^2 ( x^2+y^2 ) ] \nonumber\\ -\nabla [ \omega \lambda(\frac{b}{a}x^2 + \frac{a}{b}y^2 ) ] \,.\end{aligned}\ ] ] hence , within the configuration the following bernoulli s function must be uniform in space : @xmath19 where @xmath20 is a constant . it should be mentioned that in this bernoulli s function , @xmath5 and @xmath10 are interchangeable . this reflects the fact that , for any direct configuration ( @xmath21 ) in which fluid motion is dominated by figure rotation , the adjoint configuration ( @xmath22 ) in which fluid motion is dominated by internal motions can be obtained by simply interchanging @xmath5 and @xmath10 . we find it is useful to define an effective potential @xmath23 as : @xmath24 hence , the bernoulli s function can also be written as : @xmath25 in our later discussion of equi - potential ( iso - density ) surfaces , we will often use the phrase `` equi - potential '' referring to @xmath23 . on the other hand , the steady - state continuity equation in the rotating frame requires , @xmath26 this is trivially satisfied for incompressible riemann s - type ellipsoids because @xmath27 is zero everywhere , and the velocity vectors in the rotating frame are always tangent to the concentric equi - potential contours and perpendicular to the unit vector that is normal to the ellipsoidal surface . in our compressible models , the divergence - free velocity field given by eq . ( [ vel ] ) will continue to be adopted , so if the equi - potential ( and iso - density ) contours continue to be concentric ellipses ( as assumed by * ? ? ? * ) , the velocity vectors will always be orthogonal to the density gradient everywhere ; this would satisfy the steady - state continuity equation exactly . however , as we demand that euler s equation be satisfied via the solution of eq . ( [ bernoulli ] ) , the iso - density contours for our compressible models are not guaranteed to be self - similar ellipses . therefore , in our current numerical method , the extent to which the steady - state continuity equation is satisfied will depend on how far the density shells deviate from self - similar ellipses . for the remainder of this paper , we will only focus on obtaining a solution of bernoulli s function given by eq . ( [ bernoulli ] ) . we will return to issues related to the steady - state continuity equation in section [ discussion ] . for a given choice of axes ( @xmath2 , @xmath3 , @xmath4 ) and a mass distribution @xmath28 ( hence @xmath29 is determined by poisson s equation ) , there are three unknown constants in eq . ( [ bernoulli ] ) : @xmath20 , @xmath5 and @xmath10 . on the other hand , we have three boundary conditions on the three axes at the surface of the star , i.e. , points a , b and c as illustrated in figure [ boundary ] , where @xmath30 drops to zero . ( note that points a , b and c are located on @xmath31 , @xmath32 , and @xmath33 axes at @xmath34 , @xmath35 , and @xmath36 , respectively . ) these boundary conditions give the following equations : @xmath37 subtracting equation ( [ bc2 ] ) from ( [ bc1 ] ) , we obtain @xmath38 adding equation ( [ bc2 ] ) and ( [ bc1 ] ) together , we obtain @xmath39 \label{ch2 } \,.\ ] ] since the right - hand - side of this last expression is non - negative , it follows that @xmath40 . it can be shown that equations ( [ ch1 ] ) and ( [ ch2 ] ) are equivalent to equations ( 29 ) and ( 30 ) in chapter 7 of @xcite . the solutions of equations ( [ ch1 ] ) and ( [ ch2 ] ) for a direct configuration are : @xmath41 where @xmath42 for adjoint configurations , we only need to interchange @xmath43 and @xmath44 . the signs of @xmath5 and @xmath10 are determined by the sign of @xmath45 ( which is predetermined by the specified axis ratios ) , through eq . ( [ eqn : f ] ) and the relationship , @xmath46 with these constants and the bernoulli s function in hand , we can solve for @xmath30 throughout the interior of the configuration . in the incompressible case ( @xmath47 ) , we know from the classical analytical results that the adopted velocity field generates equilibrium structures that are uniform density ellipsoids . hence , we can obtain riemann s - type ellipsoids directly by setting up a mass distribution that is confined by the ellipsoidal surface defined by three axes @xmath2 , @xmath3 and @xmath4 ; then we solve poisson s equation to get the potential everywhere ; finally , @xmath5 and @xmath10 can be determined by equations ( [ omega ] ) and ( [ lambda ] ) . compressible polytropic models ( @xmath48 ) having the same flow pattern as that of riemann ellipsoids ( given by eq . [ vel ] ) , can also be constructed by an iterative procedure that is very similar to hachisu s scf method @xcite . the main steps of this method include : ( i ) set up a trial ellipsoidal mass distribution defined by the choice of @xmath2 , @xmath3 , and @xmath4 , then solve poisson s equation to obtain the gravitational potential everywhere ( the details of our poisson solver are discussed in * ? ? ? * ) ; ( ii ) calculate @xmath20 , @xmath5 and @xmath10 using the three boundary conditions as discussed above ; ( iii ) calculate the enthalpy everywhere inside the configuration using eq . ( [ bernoulli ] ) ; ( iv ) calculate the new trial " density distribution according to the relationship between the density and enthalpy for a polytrope . these steps are repeated until the model converges . in this section , we show our results for incompressible riemann s - type ellipsoids and their quasi - equilibrium , compressible counterparts . since the available parameter space occupies a large portion of figure [ triaxial ] , we choose to build models only along the vertical dashed lines , which largely represent most of the configurations . our models are constructed on a cylindrical grid with a resolution of @xmath49 in the @xmath50 ( cylindrical radius ) , @xmath51 ( vertical ) , and @xmath52 ( azimuthal ) directions , respectively . following @xcite , we adopt a set of polytropic units in which the gravitational constant @xmath53 , the radius of the entire grid @xmath54 , and maximum density @xmath55 are all set to 1 . for all of our models , we set the semiaxis @xmath2 to @xmath56 . we choose to present our results in such a way that readers can readily normalize our data to any unit system they are familiar with for the purpose of comparison . table [ n0 ] shows the results for direct configurations of incompressible ( @xmath47 ) riemann s - type ellipsoids . tables [ n05 ] , [ n10 ] , and [ n15 ] show the same results but for compressible counterparts of riemann ellipsoids with @xmath57 , @xmath58 , and @xmath59 , respectively . tables [ n0a ] , [ n05a ] , and [ n10a ] show results for adjoint configurations with @xmath60 , @xmath57 and @xmath58 , respectively . in these tables , @xmath61 is the rotational energy of the system , @xmath62 is the total mass , @xmath63 is the total angular momentum , @xmath64 is the mean density , @xmath65 is the gravitational potential energy , and @xmath66 is the thermal energy . the quantity @xmath67 measures how well our models are in virial equilibrium ; ideally , this ratio should have a value of 0.5 . all of our models have values that are very close to 0.5 , especially for compressible configurations . in table [ n0 ] , the parameters @xmath68 and @xmath69 are analytically computed results for @xmath5 and @xmath10 , respectively . several properties of our models are worth discussing . first , for direct configurations with a fixed value of @xmath0 , as one proceeds from larger values of @xmath1 to smaller values , @xmath10 starts from a positive value , then continuously decreases ; it reaches zero when the rigidly rotating jacobi sequence is hit , and then becomes negative after passing the jacobi sequence . the parameter @xmath45 also switches sign when crossing the jacobi / dedekind sequences . for adjoint configurations , a similar behavior is observed for @xmath5 , while @xmath45 approaches @xmath70 around the dedekind sequence because @xmath71 . second , in the parameter space above the jacobi / dedekind sequence where @xmath72 , @xmath5 is positive for direct configurations , and negative for adjoint configurations , which reflects the fact that the jacobi mode in maclaurin spheroids is the `` forward moving '' wave but the dedekind mode is the `` backward moving '' wave . this also implies that in the adjoint configurations with @xmath72 , the ellipsoidal pattern is moving in a direction that is opposite to the overall rotation ! furthermore , when the critical jacobi / dedekind sequence is passed , @xmath45 switches sign ; so does @xmath10 in direct configurations and @xmath5 in adjoint configurations . in maclaurin spheroids , this is where the frequency of the backward moving wave becomes positive in the inertial frame ; hence , it is where the secular bar - mode instability sets in through gravitational radiation ( chandrasekhar - friedman - schutz instability ) or viscosity @xcite . finally , for softer equations of state , the scf technique failed to converge to models with relatively small values of @xmath1 because , in these cases , the centrifugal force at the boundary point a ( see figure [ boundary ] ) would exceed the gravitational force . in fact , for @xmath73 and @xmath59 , even the jacobi sequence is not reached . this is consistent with previous studies that have indicated that the jacobi sequence can not exist for @xmath74 @xcite . furthermore , as was pointed out by @xcite in their analysis of uniformly rotating polytropes , the mass - shedding limit never extends much beyond the bifurcation point ( where the jacobi / dedekind sequence branches off the maclaurin spheriod sequence ) even for very small @xmath75 : for @xmath76 , this mass - shedding limit corresponds to @xmath77 . our @xmath78 models are consistent with their results . therefore , the available parameter space for compressible counterparts of riemann ellipsoids becomes smaller as the equation of state becomes softer . figures [ direct05 ] and [ adjoint05 ] show equatorial - plane iso - density contours and velocity fields for direct and adjoint configurations of our @xmath57 model with @xmath79 and @xmath80 . both the rotating frame velocity field @xmath11 and inertial frame velocity field @xmath81 are plotted for comparison . the flow patterns in these two models are different from each other although they have identical geometrical shapes . it is also clear that the fluid motion in the adjoint configuration is dominated by internal motion rather than rotation . as another illustration , figures [ direct ] and [ adjoint ] illustrate the structure of our @xmath73 model with @xmath79 and @xmath82 . notice that @xmath5 , which is also the pattern speed of the ellipsoid , is negative in both fig . [ adjoint05 ] and fig . [ adjoint ] , so the ellipsoidal figures in both plots are spinning retrograde ( clockwise ) relative to the overall fluid rotation . comparisons between previous published results and our models yield very good agreement . for example , from table iv in chapter 6 of @xcite , a jacobi ellipsoid with @xmath83 and @xmath0=0.60 , @xmath84 , has @xmath85 and @xmath86 , where @xmath87 . the closest match among our models shown in table [ n0 ] , with @xmath79 , @xmath88 , has @xmath89 , @xmath90 , and an almost vanishing @xmath91 , which shows that this model from our table is very close to the jacobi sequence . it should be noticed that , due to the discrete nature of our grid , we can not construct models with axis ratios that will place the models precisely on the jacobi / dedekind sequence , but we can construct models that are very close to this sequence ( with very small values of @xmath10 or @xmath5 ) . as another comparison , the last four columns in table [ n0 ] give analytically computed results of @xmath92 and @xmath93 for riemann s - type ellipsoids having the specified axes ratios @xmath0 and @xmath1 , as well as the ratio between numerical and analytical results ( only models with @xmath94 have been computed ) . to obtain these analytical results , we used standard incomplete elliptic integrals to evaluate the potential field of a homogeneous ellipsoid ( * ? ? ? * see their table 2 - 2 ) . in virtually all cases , the error in @xmath5 is @xmath95 a few percent . on the other hand , for the direct configurations the fractional error in @xmath10 can be as large as 50% ; the error is largest for models whose @xmath10 almost vanishes . this is understandable because a small variation in the magnitude of @xmath10 will result in a large percentage difference when @xmath10 is approaching zero . also some mismatch between numerical and analytical results may arise because the discrete nature of our grids makes it difficult for us to define an exact axis ratio for our models . for compressible equations of state , we can compare our results with those presented by @xcite , who used an energy variational method to determine approximate equilibrium configurations for ellipsoidal self - gravitating systems . @xcite assumed that the iso - density contours of each configuration are self - similar concentric ellipses , which as we have already emphasized is not the case . in their table 4 , for compressible analogues of jacobi ellipsoids , a configuration with @xmath96 and @xmath97 has @xmath98 ; the closest match among our @xmath57 models ( see table [ n05 ] ) with @xmath99 and @xmath100 has @xmath101 . other quantities also match well , after a proper normalization is applied as prescribed by their equation ( 3.27 ) . we omit more comparisons here and leave the rest to interested readers . other models that are of interest are the @xmath102 irrotational sequence @xcite . it turns out that the @xmath45 parameter is very sensitive to the axis ratios , so we could not build models that are very close to the irrotational sequence due to the discrete nature of our computational grids . our model that comes closest has @xmath103 ; it is a direct configuration with @xmath57 and its other parameters are listed in table [ n05 ] . we have presented a new method to construct 3d models for riemann s - type ellipsoids ( @xmath47 ) , and the method has been extended to construct compressible counterparts that share the same velocity field as that of riemann s - type ellipsoids . with this method , we have been able to build 3d models that cover almost the entire parameter space of riemann s - type ellipsoids . our results are in good agreement with previous studies . we expect that this method can also be straightforwardly extended to the construction of roche - riemann ellipsoids . by design , our compressible models satisfy the steady - state euler s equation . however , it is not guaranteed that they simultaneously satisfy the steady - state continuity equation exactly , because this additional constraint was not built into our method . because the adopted velocity flow - field @xmath11 in our models is divergence - free , the violations of the steady - state continuity equation will only arise to the extent that the iso - density contours deviate from concentric ellipses . as we can see from figures [ direct05 ] - [ adjoint ] , the outer low - density shell in compressible models becomes somewhat pointed along the major axis . however , the deviation from concentric ellipses is small and almost negligible in the inner , high - density regions . furthermore , a comparison between fig . [ direct05 ] and fig . [ direct ] shows that this deviation becomes smaller in less compressible models . in order to quantitatively measure the violation of the steady - state continuity equation , we calculated @xmath104 for the models shown in figs [ direct05 ] , [ adjoint05 ] , and [ direct ] . as long as @xmath105 for a given configuration , we can consider that the model has a reasonablly good quasi - equilibrium structure . in figure [ divrhov ] , we have plotted @xmath106 in the equatorial plane as a function of cylindrical radius @xmath50 along different azimuthal angles , @xmath107 and @xmath108 . in fact , this quantity remains small ( @xmath109 for @xmath57 and @xmath110 for @xmath58 ) throughout the majority of the configuration , but not surprisingly near the surface it asymptotically approaches infinity because @xmath111 goes to zero . as the figure illustrates , @xmath106 is smaller for the stiffer equation of state . we also note that @xmath106 is nearly zero along the x and y axes because the velocity vectors in the rotating frame along the x ( y ) axis only have a nonzero y ( x ) component and are therefore perpendicular to the density gradient there . thus , we expect our compressible models to represent fairly good quasi - equilibrium configurations , especially in nearly incompressible cases . on the other hand , this point was brought to our attention by an anonymous referee : the relatively large violation of the steady - state continuity equation on the very surface layer can potentially change the shape of the surface in hydrodynamical evolutions . to understand how the shape of the surface might be changed by the flow field , it is important to know the sign of @xmath112 along the surface . since we are adopting a divergence free velocity field , the sign of @xmath112 is determined by the sign of @xmath113 , which also measures to what degree the velocity vectors are orthogonal to density gradient . if the velocity vector points slightly outward ( inward ) from the surface , @xmath113 will be negative ( positive ) , because the density gradient is negative outward . [ divrhov2 ] and [ divrhov3 ] show contour maps of @xmath112 in the equatorial planes of the direct and adjoint models shown in figs . [ direct05 ] and [ adjoint05 ] , respectively . ( we note that the contour pattern for direct configurations will switch from fig . [ divrhov2 ] to fig . [ divrhov3 ] if @xmath10 flips sign . ) a @xmath114 symmetry is observed for both patterns . we also note that , in each quadrant , @xmath112 flips sign , which implies that the direction of velocity vectors will change from inward ( outward ) to outward ( inward ) from the surface within one single quadrant . accordingly , fig . [ shape ] shows a combined schematic drawing of how the velocity vectors in the rotating frames are aligned with the surface for these two patterns ( the panel on the left is for the adjoint figure shown in fig . [ adjoint05 ] , the panel on the right is for the direct figure shown in fig . [ direct05 ] ) . in this figure , the black ellipse denotes the actual surface ( which should not be a perfect ellipse ) , red arrows denote velocity vectors on the surface , purple arrows denote density gradient vectors , which are drawn pointing inward because of their negative sign . for direct configurations , in which the figure rotation is larger than the internal motion , these mismatches ( nonvanishing @xmath113 ) on top of a large rotation are likely to generate spiral arms ; whereas , for adjoint configurations , the internal motion is larger than figure rotation , so these mismatches tend to generate waves along the surface . using the 3d hydrodynamics code developed by the lsu group @xcite , we have carried out nonlinear 3d hydrodynamical evolutions for a subset of the nearly incompressible models discussed here in table [ n05 ] . despite the fact that the models do not fully satisfy the steady - state continuity equation , they are in very good force - balance initially and , as predicted by the above discussion , the models are fairly long - lived , holding their initial ellipsoidal shapes over the time that is followed by our simulations ( @xmath115 15 - 20 dynamical times ) . we also observed the formation of narrow spiral arms ( low amplitude surface waves ) containing low density materials in direct ( adjoint ) configurations . the results of these hydrodynamical evolutions will be presented in a separate paper in the context of our ongoing study of the stability of riemann s - type ellipsoids . since our ellipsoidal models are fairly long - lived and have exactly the same elliptical flow field as riemann s - type ellipsoids , they should provide good initial states for a nonlinear , hydrodynamical study of dynamical instabilities in riemann s - type ellipsoids . we especially have in mind the elliptical strain instability that is associated with elliptical stream lines , which should set in on a dynamical time scale . through such a study , we hope to provide insight into the stability of ellipsoidal configurations that will complement the work of @xcite . this should also help answer whether the evolutionary path of a grr - driven star toward the dedekind sequence is viable . i would like to thank joel tohline for valuable suggestions , discussions and carefully reading through this paper . i also thank juhan frank and patrick motl for helpful comments . it is a pleasure to thank norman lebovitz for drawing our attention to the particular analytical expression of the velocity field of riemann s - type ellipsoids used in this study . i would also thank howard cohl for developing the poisson solver that was an important ingredient in this work . i thank the anonymous referee for insightful suggestions to analysis our models . this work was partially supported by nsf grants ast-0407070 and phy-0326311 , and by the center for computation & technology at lsu . binney , j. , & tremaine , s. , 1987 , galactic dynamics , new york : academic press cazes , j. e. , & tohline , j. e. 2000 , , 532 , 1051 chandrasekhar , s. 1969 , equilibrium figures of equilibrium , new haven , ct : yale univ . press cohl , h. s. , & tohline , j. e. 1999 , , 527 , 86 friedman , j. , & schutz , b. f. 1978 , , 222 , 281 hachisu , i. 1986 , , 61 , 479 hachisu , i. 1986 , , 62 , 461 hachisu , i. , eriguchi , y. , 1982 , prog . theor.phys . , 68 , 206 hachisu , i. , eriguchi , y. , 1984 , , 36 , 239 james , r. a. , 1964 , , 140 , 552 lebovitz , n. r. , & lifschitz , a. 1996 , , 458 , 699 lebovitz , n. r. , & saldanha , k. i. 1999 , physics of fluids , 11 , 3374 lifschitz , a. , & lebovitz , n. 1993 , , 408 , 603 motl , p. m. , tohline , j. e. , & frank , j. 2002 , , 138 , 121 0.90 & 0.923 & 1.030 & 0.7550 & 6.6e-3 & 0.03183 & 0.8232 & -0.6989 & 0.4989 & -1.5 & & + 0.90 & 0.795 & 1.144 & 0.4310 & 0.046 & 0.07018 & 0.7086 & -0.5429 & 0.4992 & -0.76 & 1.14704 & 0.997 & 0.43181 & 0.998 + 0.90 & 0.641 & 1.119 & 0.1394 & 0.101 & 0.07764 & 0.5720 & -0.3759 & 0.4988 & -0.25 & 1.13137 & 0.989&0.15077 & 0.925 + 0.90 & 0.590 & 1.099 & 0.0569 & 0.122 & 0.07589 & 0.5261 & -0.3249 & 0.4987 & -0.10 & 1.10661 & 0.993 & 0.06406 & 0.888 + 0.90 & 0.564 & 1.090 & 0.0210 & 0.132 & 0.07452 & 0.5033 & -0.3007 & 0.4989 & -0.04 & 1.09034 & 0.999&0.02033&1.033 + 0.90 & 0.538 & 1.069 & -0.0250 & 0.144 & 0.07268 & 0.4801 & -0.2768 & 0.4991 & 0.05 & 1.07148 & 0.998&-0.02324 & 1.075 + 0.90 & 0.487 & 1.015 & -0.1205 & 0.167 & 0.06814 & 0.4341 & -0.2316 & 0.4992 & 0.24 & 1.02639&0.989 & -0.10880 & 1.108 + 0.90 & 0.333 & 0.797 & -0.3885 & 0.246 & 0.04860 & 0.2970 & -0.1167 & 0.4983 & 0.98 & 0.79257 & 1.006&-0.39224&0.990 + 0.74 & 0.769 & 1.099 & 0.5604 & 0.030 & 0.03807 & 0.5682 & -0.3744 & 0.4982 & -1.1 & & + 0.74 & 0.692 & 1.133 & 0.3840 & 0.058 & 0.04678 & 0.5116 & -0.3134 & 0.4986 & -0.71 & 1.13215 & 1.001&0.38599&0.995 + 0.74 & 0.538 & 1.094 & 0.0774 & 0.121 & 0.04836 & 0.3982 & -0.2034 & 0.4990 & -0.15 & 1.09508 & 0.999&0.08149 & 0.950 + 0.74 & 0.513 & 1.078 & 0.0303 & 0.133 & 0.04734 & 0.3791 & -0.1857 & 0.4991 & -0.06 & 1.07883 & 0.999&0.03474&0.872 + 0.74 & 0.487 & 1.054 & -0.0208 & 0.145 & 0.04594 & 0.3560 & -0.1700 & 0.4988 & 0.04 & 1.05883 & 0.995 & -0.01363 & 1.526 + 0.74 & 0.436 & 1.003 & -0.1184 & 0.171 & 0.04284 & 0.3223 & -0.1400 & 0.4987 & 0.25 & 1.00963 & 0.993&-0.10852 & 1.091 + 0.74 & 0.333 & 0.847 & -0.3230 & 0.228 & 0.03389 & 0.2463 & -0.0863 & 0.4983 & 0.80 & 0.85686 & 0.988&-0.31006&1.042 + 0.74 & 0.282 & 0.715 & -0.4523 & 0.259 & 0.02842 & 0.2081 & -0.0634 & 0.4968 & 1.3 & 0.72756 & 0.983&-0.43781 & 1.033 + 0.59 & 0.718 & 0.948 & 0.7677 & 0.014 & 0.01197 & 0.4208 & -0.2243 & 0.4924 & -1.9 & & + 0.59 & 0.538 & 1.087 & 0.2338 & 0.093 & 0.02876 & 0.3157 & -0.1374 & 0.4982 & -0.49 & 1.08732 & 0.999 & 0.23529 & 0.994 + 0.59 & 0.436 & 1.029 & 0.0108 & 0.145 & 0.02691 & 0.2554 & -0.0949 & 0.4985 & -0.02 & 1.03198&0.997 & 0.01472&0.734 + 0.59 & 0.410 & 1.008 & -0.0417 & 0.159 & 0.02592 & 0.2404 & -0.0853 & 0.4991 & 0.09 & 1.00774&1.000&-0.04012&1.039 + 0.59 & 0.359 & 0.943 & -0.1534 & 0.188 & 0.02342 & 0.2104 & -0.0673 & 0.4989 & 0.37 & 0.94638 & 0.996 & -0.14830&1.034 + 0.59 & 0.231 & 0.584 & -0.5426 & 0.267 & 0.01475 & 0.1351 & -0.0300 & 0.4910 & 2.1 & 0.60698 & 0.962&-0.51563&1.013 + 0.41 & 0.385 & 0.971 & 0.1369 & 0.132 & 0.01220 & 0.1563 & -0.0407 & 0.4978 & -0.40 & 0.97108 & 1.000&0.14159&0.967 + 0.41 & 0.333 & 0.930 & 0.0016 & 0.165 & 0.01112 & 0.1347 & -0.0313 & 0.4998 & -0.005 & 0.92963 & 1.000 & 0.00331&0.483 + 0.41 & 0.308 & 0.900 & -0.0685 & 0.183 & 0.01061 & 0.1254 & -0.0275 & 0.4991 & 0.22 & 0.89982&1.000 & -0.06224&1.101 + 0.41 & 0.256 & 0.808 & -0.2074 & 0.217 & 0.00896 & 0.1046 & -0.0199 & 0.4937 & 0.73 & 0.81275&0.994 & -0.20138&1.030 + 0.28 & 0.256 & 0.809 & 0.0246 & 0.183 & 0.00472 & 0.0717 & -0.0103 & 0.4982&-0.12 & 0.80944&0.999 & 0.03668&0.671 + 0.28 & 0.231 & 0.780 & -0.0572 & 0.204 & 0.00431 & 0.0646 & -0.0086 & 0.4988&0.28 & 0.77651&1.004&-0.04714&1.213 + 0.28 & 0.205 & 0.726 & -0.1505 & 0.224 & 0.00382 & 0.0575 & -0.0069 & 0.4967&0.80 & 0.72853&0.996&-0.13511&1.114 + 0.90 & 0.795 & 0.970 & 0.3660 & 0.045 & 0.0258 & 0.3812 & 0.5398 & -0.1751 & 0.5000 & -0.76 + 0.90 & 0.744 & 0.975 & 0.2831 & 0.061 & 0.0273 & 0.3541 & 0.5371 & -0.1544 & 0.5000 & -0.58 + 0.90 & 0.692 & 0.970 & 0.2050 & 0.079 & 0.0276 & 0.3266 & 0.5344 & -0.1344 & 0.5001&-0.43 + 0.90 & 0.641 & 0.958 & 0.1312 & 0.097 & 0.0227 & 0.2987 & 0.5309 & -0.1150 & 0.5001 & -0.28 + 0.90 & 0.590 & 0.940 & 0.0617 & 0.117 & 0.0256 & 0.2695 & 0.5263 & -0.0964 & 0.5001 & -0.13 + 0.90 & 0.564 & 0.929 & 0.0287 & 0.127 & 0.0246 & 0.2546 & 0.5240 & -0.0873 & 0.5001 & -0.06 + * 0.90 & 0.538 & 0.920 & 0.0001 & 0.137 & 0.0233 & 0.2391 & 0.5198 & -0.0783 & 0.5001 & -0.0002 + 0.90 & 0.513 & 0.908 & -0.0281 & 0.147 & 0.0218 & 0.2234 & 0.5169 & -0.0695 & 0.5001 & 0.06 + 0.90 & 0.487 & 0.898 & -0.0513 & 0.157 & 0.0201 & 0.2066 & 0.5115 & -0.0607 & 0.5001 & 0.12 + 0.74 & 0.692 & 0.958 & 0.3232 & 0.055 & 0.0166 & 0.2719 & 0.5380 & -0.0991 & 0.4997 & -0.71 + 0.74 & 0.641 & 0.959 & 0.2343 & 0.074 & 0.0171 & 0.2489 & 0.5345 & -0.0852 & 0.4999 & -0.51 + 0.74 & 0.590 & 0.949 & 0.1514 & 0.094 & 0.0167 & 0.2252 & 0.5314 & -0.0717 & 0.5001 & -0.33 + 0.74 & 0.538 & 0.932 & 0.0750 & 0.114 & 0.0156 & 0.2005 & 0.5263 & -0.0587 & 0.5001&-0.17 + * 0.74 & 0.487 & 0.911 & 0.0081 & 0.135 & 0.0137 & 0.1737 & 0.5192 & -0.0458 & 0.5002 & -0.02 + 0.74 & 0.462 & 0.904 & -0.0171 & 0.144 & 0.0123 & 0.1588 & 0.5156 & -0.0393 & 0.5002 & 0.04 + 0.59 & 0.718 & 0.792 & 0.6631 & 0.012 & 0.0037 & 0.2241 & 0.5418 & -0.0713 & 0.4943 & -1.9 + 0.59 & 0.538 & 0.922 & 0.2021 & 0.085 & 0.0090 & 0.1587 & 0.5304 & -0.0397 & 0.4997 & -0.50 + 0.59 & 0.487 & 0.911 & 0.1137 & 0.105 & 0.0081 & 0.1373 & 0.5256 & -0.0313 & 0.5001 & -0.29 + 0.59 & 0.462 & 0.905 & 0.0754 & 0.114 & 0.0073 & 0.1250 & 0.5230 & -0.0265 & 0.5002 & -0.19 + 0.59 & 0.436 & 0.871 & -0.0337 & 0.153 & 0.0105 & 0.1382 & 0.5453 & -0.0031 & 0.5088 & 0.09 + 0.90 & 0.795 & 0.803 & 0.3029 & 0.039 & 0.00893 & 0.2053 & 0.2921 & -0.0577 & 0.5001 & -0.76 + 0.90 & 0.744 & 0.804 & 0.2329 & 0.053 & 0.00910 & 0.1870 & 0.2867 & -0.0492 & 0.5001 & -0.58 + 0.90 & 0.692 & 0.796 & 0.1684 & 0.067 & 0.00872 & 0.1679 & 0.2799 & -0.0409 & 0.5002 & -0.43 + 0.90 & 0.641 & 0.782 & 0.1090 & 0.081 & 0.00788 & 0.1476 & 0.2711 & -0.0328 & 0.5002 & -0.28 + 0.90 & 0.590 & 0.760 & 0.0555 & 0.093 & 0.00657 & 0.1255 & 0.2611 & -0.0249 & 0.5002 & -0.14 + 0.90 & 0.564 & 0.747 & 0.0314 & 0.097 & 0.00572 & 0.1134 & 0.2563 & -0.0210 & 0.5002 & -0.08 + 0.74 & 0.641 & 0.784 & 0.1917 & 0.060 & 0.00513 & 0.1262 & 0.2805 & -0.0254 & 0.5003 & -0.51 + 0.74 & 0.590 & 0.769 & 0.1241 & 0.074 & 0.00450 & 0.1087 & 0.2731 & -0.0197 & 0.5003 & -0.34 + 0.74 & 0.538 & 0.743 & 0.0626 & 0.084 & 0.00349 & 0.0893 & 0.2680 & -0.0143 & 0.5003 & -0.18 + 0.74 & 0.512 & 0.722 & 0.0322 & 0.086 & 0.00289 & 0.0786 & 0.2676 & -0.0115 & 0.5003 & -0.09 + 0.59 & 0.590 & 0.748 & 0.244 & 0.046 & 0.00237 & 0.0870 & 0.2836 & -0.0136 & 0.5007 & -0.75 + 0.59 & 0.564 & 0.743 & 0.2011 & 0.052 & 0.00220 & 0.0797 & 0.2817 & -0.0118 & 0.5009 & -0.62 + 0.90 & 0.795 & 0.631 & 0.2351 & 0.030 & 0.00273 & 0.1068 & 0.1531 & -0.0181 & 0.5002&-0.75 + 0.90 & 0.744 & 0.622 & 0.1752 & 0.040 & 0.00260 & 0.0945 & 0.1467 & -0.0147 & 0.5003&-0.57 + 0.90 & 0.692 & 0.604 & 0.1187 & 0.048 & 0.00229 & 0.0817 & 0.1397 & -0.0115 & 0.5003&-0.40 + 0.90 & 0.641 & 0.574 & 0.0640 & 0.055 & 0.00183 & 0.0683 & 0.1324 & -0.0085 & 0.5003&-0.22 + 0.90 & 0.615 & 0.554 & 0.0365 & 0.057 & 0.00157 & 0.0614 & 0.1292 & -0.00711 & 0.5003&-0.13 + 0.74 & 0.692 & 0.609 & 0.2024 & 0.033 & 0.00148 & 0.0717 & 0.1491 & -0.00931 & 0.5006&-0.70 + 0.74 & 0.641 & 0.592 & 0.1378 & 0.041 & 0.00130 & 0.0612 & 0.1430 & -0.00713 & 0.5005&-0.49 + 0.64 & 0.670 & 0.587 & 0.2503 & 0.023 & 0.00086 & 0.0586 & 0.1524 & -0.00666 & 0.5014&-0.94 + 0.64 & 0.640 & 0.584 & 0.2116 & 0.027 & 0.00084 & 0.0542 & 0.1499 & -0.00584 & 0.5013&-0.80 + 0.90 & 0.923 & -0.755 & -1.030 & 0.0064 & 0.0307 & 0.8232 & 0.99999 & -0.6989 & 0.4987 & -2.744 + 0.90 & 0.641 & -0.139 & -1.119 & 0.100 & 0.0770 & 0.572 & 0.99999 & -0.3758 & 0.4980 & -16.1 + 0.90 & 0.590 & -0.0569 & -1.099 & 0.121 & 0.0754 & 0.5261 & 0.99999 & -0.3249 & 0.4979 & -38.8 + 0.90 & 0.564 & -0.0211 & -1.090 & 0.132 & 0.0741 & 0.5033 & 0.99999 & -0.3007 & 0.4981 & -104.2 + 0.90 & 0.538 & 0.0250 & -1.069 & 0.143 & 0.0723 & 0.4801 & 0.99999 & -0.2768 & 0.4982 & 86.0 + 0.90 & 0.487 & 0.121 & -1.015 & 0.166 & 0.0678 & 0.4342 & 0.99999 & -0.2314 & 0.4983 & 16.9 + 0.90 & 0.436 & 0.219 & -0.945 & 0.191 & 0.0623 & 0.3886 & 0.99999 & -0.1900 & 0.4979 & 8.689 + 0.90 & 0.333 & 0.389 & -0.797 & 0.246 & 0.0485 & 0.2970 & 0.99999 & -0.1167 & 0.4978 & 4.128 + 0.74 & 0.744 & -0.495 & -1.117 & 0.035 & 0.0376 & 0.5496 & 0.99999 & -0.3540 & 0.4943 & -4.7 + 0.74 & 0.590 & -0.171 & -1.116 & 0.093 & 0.0465 & 0.4357 & 0.99999 & -0.2377 & 0.4933 & -13.6 + 0.74 & 0.513 & -0.030 & -1.078 & 0.127 & 0.0452 & 0.3791 & 0.99999 & -0.1866 & 0.4932 & -74.3 + 0.74 & 0.487 & 0.021 & -1.054 & 0.139 & 0.0440 & 0.3595 & 0.99999 & -0.1699 & 0.4928 & 106.8 + 0.74 & 0.436 & 0.118 & -1.002 & 0.165 & 0.0414 & 0.3223 & 0.99999 & -0.1400 & 0.4929 & 17.7 + 0.74 & 0.385 & 0.210 & -0.944 & 0.193 & 0.0376 & 0.2841 & 0.99999 & -0.1117 & 0.4934 & 9.4 + 0.74 & 0.333 & 0.323 & -0.847 & 0.223 & 0.0332 & 0.2463 & 0.99999 & -0.0863 & 0.4940 & 5.5 + 0.74 & 0.282 & 0.453 & -0.714 & 0.257 & 0.0281 & 0.2081 & 0.99999 & -0.0634 & 0.4944 & 3.4 + 0.59 & 0.615 & -0.417 & -1.085 & 0.045 & 0.0199 & 0.3608 & 0.99999 & -0.1729 & 0.4834 & -5.5 + 0.59 & 0.487 & -0.119 & -1.063 & 0.100 & 0.0239 & 0.2853 & 0.99999 & -0.1152 & 0.4807 & -20.4 + 0.59 & 0.436 & -0.011 & -1.029 & 0.126 & 0.0235 & 0.2553 & 0.99999 & -0.0949 & 0.4801 & -217.5 + 0.59 & 0.410 & 0.042 & -1.008 & 0.140 & 0.0229 & 0.2404 & 0.99999 & -0.0853 & 0.4807 & 55.2 + 0.59 & 0.359 & 0.153 & -0.943 & 0.170 & 0.0213 & 0.2104 & 0.99999 & -0.0672 & 0.4811 & 14.1 + 0.59 & 0.282 & 0.271 & -0.856 & 0.204 & 0.0190 & 0.1803 & 0.99999 & -0.0509 & 0.4822 & 7.2 + 0.59 & 0.231 & 0.402 & -0.729 & 0.242 & 0.0161 & 0.1502 & 0.99999 & -0.0365 & 0.4843 & 4.1 + 0.90 & 0.923 & -0.642 & -0.8766 & 0.0063 & 0.0116 & 0.4476 & 0.5442 & -0.2297 & 0.4997 & -2.7 + 0.90 & 0.590 & -0.062 & -0.9401 & 0.116 & 0.0255 & 0.2695 & 0.5263 & -0.0964 & 0.4997 & -30.7 + 0.90 & 0.538 & -1.47e-4&-0.9195 & 0.137 & 0.0232 & 0.2391 & 0.5197 & -0.0782 & 0.4999 & -12606.1 + 0.90 & 0.513 & 0.028 & -0.9080 & 0.147 & 0.0218 & 0.2234 & 0.5169 & -0.0695 & 0.5000 & 65.2 + 0.74 & 0.795 & -0.537 & -0.9101 & 0.018 & 0.0102 & 0.3167 & 0.5421 & -0.1283 & 0.4966 & -3.5 + 0.74 & 0.692 & -0.323 & -0.9584 & 0.052 & 0.0154 & 0.2719 & 0.5379 & -0.0991 & 0.4963 & -6.2 + 0.74 & 0.538 & -0.075 & -0.9317 & 0.112 & 0.0150 & 0.2005 & 0.5263 & -0.0587 & 0.4975 & -25.9 + 0.74 & 0.487 & -8.2e-3 & -0.9114 & 0.134 & 0.0133 & 0.1737 & 0.5192 & -0.0458 & 0.4992 & -233.1 + 0.74 & 0.462 & 0.0171 & -0.9044 & 0.145 & 0.0121 & 0.1587 & 0.5157 & -0.0393 & 0.5009 & 110.2 + 0.59 & 0.615 & -0.355 & -0.9180 & 0.044 & 6.93e-3 & 0.1877 & 0.5368 & -0.0529 & 0.4894 & -5.9 + 0.59 & 0.538 & -0.202 & -0.9222 & 0.076 & 7.66e-3 & 0.1587 & 0.5304 & -0.0397 & 0.4910 & -10.4 + 0.59 & 0.487 & -0.114 & -0.9111 & 0.100 & 7.21e-3 & 0.1373 & 0.5256 & -0.0310 & 0.4948 & -18.3 + 0.59 & 0.462 & -0.075 & -0.9050 & 0.113 & 6.65e-3 & 0.1250 & 0.5230 & -0.0265 & 0.4992 & -27.4 + 0.90 & 0.795 & -0.303 & -0.8030 & 0.039 & 0.00887 & 0.2053 & 0.2921 & -0.0577 & 0.5000 & -5.33 + 0.90 & 0.692 & -0.168 & -0.7963 & 0.067 & 0.00869 & 0.1679 & 0.2799 & -0.0409 & 0.5003 & -9.51 + 0.90 & 0.590 & -0.056 & -0.7605 & 0.094 & 0.00657 & 0.1255 & 0.2611 & -0.0249 & 0.5009 & -27.59 + 0.90 & 0.564 & -0.031 & -0.7469 & 0.099 & 0.00573 & 0.1134 & 0.2563 & -0.0210 & 0.5014 & -47.78 + 0.74 & 0.641 & -0.192 & -0.7842 & 0.061 & 0.00499 & 0.1262 & 0.2805 & -0.0254 & 0.5013 & -8.54 + 0.74 & 0.538 & -0.062 & -0.7429 & 0.091 & 0.00354 & 0.0893 & 0.2680 & -0.0142 & 0.5071 & -24.80 + 0.74 & 0.512 & -0.032 & -0.7220 & 0.096 & 0.00295 & 0.0786 & 0.2676 & -0.0115 & 0.5097 & -46.74 + 0.59 & 0.590 & -0.244 & -0.7480 & 0.050 & 0.00217 & 0.0870 & 0.2836 & -0.0136 & 0.5049 & -7.01 + 0.59 & 0.564 & -0.201 & -0.7433 & 0.059 & 0.00210 & 0.0797 & 0.2817 & -0.0118 & 0.5083 & -8.47 +

the field of ultracold atoms in optical lattices has been a promising new opportunity for studying many - body effects which are important for condensed - matter physics in controlled environments @xcite . in particular , fermionic atoms such as @xmath2k may provide a direct path towards a `` quantum simulation '' of the hubbard model which itself has a paradigmatic role in condensed matter physics and is a key in understanding phenomena such as high - temperature superconductivity and strongly correlated magnetism . in these experiments , some novel possibilities to study physical correlations between constituents of the model are being explored . in many cases , such experiments@xcite drive the systems substantially beyond thermal equilibrium , so that they are inaccessible to methods of conventional equilibrium or linear - response theory . usually , one also encounters spatially inhomogeneous situations , since the atoms in the optical lattice are being held in a trap potential which co - exists with the lattice potential . in one - dimensional systems , many opportunities to provide computational benchmarks for such experiments exist , such as via the density - matrix renormalization group @xcite . however , it is a challenging problem to describe two- and three - dimensional systems out of thermal equilibrium , especially when they are also inhomogeneous . in this paper we present a nonequilibrium strong - coupling approach to inhomogeneous systems of ultracold atoms in optical lattices . the paper is structured as follows . section [ sec : model ] discusses the hubbard model for an optical lattice in a trap . in section [ sec : formalism ] , we outline the strong - coupling approach which enables us to simulate inhomogeneous higher - dimensional hubbard systems out of equilibrium . in section [ sec : algo ] , we develop the massively parallel algorithm which is used to solve the resulting equations on a supercomputer . section [ sec : results ] presents results of the algorithm for the example of a modulated lattice depth and validates them by comparing to the previously introduced strong - coupling method for homogeneous systems@xcite within the local - density approximation ( lda ) . conclusions are given in section [ sec : conclusion ] . we consider a fermi hubbard model in the presence of a trap potential , i.e. @xmath3 with @xmath4 where the on - site single - particle energy levels @xmath5 are determined by the trap potential @xmath6 and a global chemical potential @xmath7 which characterizes the initial equilibrium state at time @xmath8 . the initial state is assumed to have a temperature @xmath9 . the time - dependent interaction @xmath10 and the time - dependent hopping @xmath11 are chosen to be results of a tight - binding calculation for maximally localized wannier functions computed from the translationally invariant case . in the future , we plan to include corrections to the tight - binding parameters which result from generalized wannier functions for the inhomogeneous problem . we assume the system to be in thermal equilibrium at time @xmath8 and to be driven out of equilibrium subsequently by the time - dependence of the model parameters . we employ a second - order self - consistent expression for the self - energy @xcite around the atomic limit which is described by @xmath12 . the self - consistency takes advantage of a resummation of diagrams which yields a better approximation . for the simulation time . between the dashed blue lines , the system is driven out of equilibrium by a time - dependent hamiltonian . in the numerical part of the paper , we refer to the two points in time as @xmath8 and @xmath13 for simplicity . the real times between the matsubara branch and the point at which the system is driven out of equilibrium can be used to check for numerical convergence , since expectation values of observables have to be constant here . the time discretization is shown in red in the implementation itself , the point @xmath8 is identical to the point to which the matsubara branch of the contour is attached . the total number of time slices is @xmath14 . ] it can be extended to an inhomogeneous system in the following way : @xmath15 here , @xmath16 is the contour - ordered on - site no - hopping green s function at site @xmath17 with spin @xmath18 ( in the paramagnetic phase , the green s function is independent of the spin @xmath18 ) . times @xmath19 and @xmath20 are located on the kadanoff - baym - keldysh contour @xmath21 depicted in fig . [ fig : keldysh ] . @xmath22 is the second - order cumulant for the on - site no - hopping two - particle propagator @xmath23 at site @xmath24 . in a typical numerical implementation , one can not store this tensor for a reasonable grid of time slices . however , it is easy to compute it on the fly in the particle - number basis @xmath25 by inserting the time evolutions @xmath26 for each possible @xmath27 time ordering of the creation and annihilation operators . note that it is crucial to the applicability of the approach that the expression in eq . does not involve any time - ordered products , because @xmath28=0 $ ] for any combination of @xmath29 . the on - the - fly evaluation of the action of the operators in the basis can be realized by a fast multiplication with and division by tabulated @xmath30 values . for example , when @xmath31 , @xmath32 with @xmath33 in the calculation , one requires the expressions for all possible time orderings . the relation in eq . is solved iteratively . at a given step of this procedure , the self - energy at site @xmath34 is given by @xmath35 where @xmath36 the local green s function at site @xmath37 , @xmath38 is given by the @xmath37-th block - diagonal element of the lattice green s function @xmath39^{-1}(\vec r_i , t;\vec r_{i'},t ' ) , \end{split } \label{eq : glatt}\ ] ] i.e. @xmath40 here , the hatted quantities @xmath41 , @xmath42 , and @xmath43 denote the non - interacting green s function , the hopping , and the self - energy as operators acting on both , time and space coordinates . we comment now on the measurement of some important physical observables within our method . spin-@xmath18 occupancy _ of site @xmath34 may be evaluated by @xmath44 from the local green s function . the _ kinetic energy contribution of spin @xmath18 at lattice site @xmath34 _ can be deduced from the lattice green s function @xmath45 it is thus required to measure the equal - time hopping green s functions from site @xmath34 to its nearest neighbors @xmath46 . a quantity of particular interest is the _ double occupancy @xmath47 _ as a function of time @xmath19 and lattice site @xmath34 . we can derive it from some elements of the lattice green s function and its equal - time derivative , using the following relation ( see appendix [ app : doccformula ] ) : @xmath48 in this section , we outline the massively parallel algorithm required to solve eqs . , , and iteratively on a supercomputer . and @xmath49 are introduced in sec . [ sec : algo ] and correspond to evaluating eqs . and ( both @xmath50 ) , and eq . ( @xmath49 ) . ] in order to represent the contour - ordered green s function on a computer , the kadanoff - baym - keldysh contour is discretized to @xmath51 time slices , as shown in fig . [ fig : keldysh ] . in the implementation used here , we chose @xmath14 , where @xmath52 . here , @xmath53 is the number of time steps on each real branch of the kadanoff - baym - keldysh contour and @xmath54 the number of time steps on the imaginary time axis . furthermore , the system consists of a finite number @xmath55 of lattice sites . assuming that spatial symmetries such as reflection and rotation symmetries are given for not only the hamiltonian but also the quantum - statistical states , we can reduce the actual number of lattice sites @xmath55 within the algorithm by symmetry maps to the number @xmath56 which is the number of sites in the irreducible wedge of the lattice . many sites can then be represented by the equivalence class with respect to the symmetry . note that the full number of lattice sites still plays a role in computational complexity when the propagation of excitations is considered . since dyson s equation performs an infinite resummation of such processes , it is relevant for the calculation of the green s function . appendix [ app : symmetries ] describes the exploitation of symmetries in more detail . computationally , the self - consistency condition in eq . is solved iteratively using an alternating sequence of self - energy and green s function evaluations . the self - energy evaluations are performed via eqs . and and will be described in detail in sec . [ subsec : selfenergyeval ] . the green s functions required for self - consistency and measurements of observables are evaluated via eq . which will be described in sec . [ subsec : greensfunctionevaluation ] . due to the structure of eq . , the computation of @xmath57 at lattice site @xmath34 requires only the green s functions of neighboring lattice sites . however , the second step , i.e. the dyson s equation evaluation in eq . of the lattice green s requires self - energy information from all lattice sites . from a computational perspective , these demands are quite different in terms of optimal memory arrangement and distribution of tasks over a large set of compute nodes . as a consequence , we define two different configurations of the simulation . the green s function evaluation in eq . is performed in what we call the @xmath49 configuration of the simulation . the self - energy computation is done in the @xmath50 configuration . [ fig : algorithm ] shows a schematic flowchart for the computation . the two different global machine states are sketched in fig . [ fig : processorgroups ] . all processors are thought to be arranged along the direction of the abscissa in fig . [ fig : processorgroups ] . each configuration defines groups of processors which share tasks , we call them self - energy units @xmath58 and green s function units @xmath59 . the self - energy unit @xmath58 evaluates self - energies for the @xmath34-th range of representative sites . the green s function units evaluate relevant information for the @xmath34-th range of symmetry - representative site indices for blocked rows of the lattice green s function . there is an optimal value for the size of the units which is usually different for @xmath58 and @xmath59 . the determining factors for the optimal size will be elaborated below . by a given self - energy unit @xmath58 , the self - energy @xmath60 is computed for a certain range of ( representatives of ) sites @xmath61 for all @xmath29 using eqs . and . the expressions involve an on - the - fly evaluation of the cumulant @xmath62 using the tabulated values of eq . , as well as local green s functions which do not require any permanent storage on @xmath58 . memory - wise , the unit is required to have access to the local green s functions of neighboring sites , as well as to the value of the self - energy from the previous iteration . the latter is necessary , because in our implementation , eq . is regularized as follows : @xmath63 where @xmath64 is a linear mixing parameter which stabilizes the convergence by averaging between the current and the previous iteration . convergence is reached once @xmath65 where @xmath66 is the frobenius norm of the self - energy matrix with @xmath67 entries on the block diagonal , and @xmath68 is element - wise maximum norm of the matrix . a reasonable value for the accuracy of the self - consistency condition is @xmath69 where @xmath70 is some typical value of the interaction strength ( which often serves as the energy unit ) . convergence is usually reached after approximately ten iterations . in all relevant cases we have encountered so far , the self - energy evaluation step is computationally more costly than the green s function evaluation , if the latter has been optimized appropriately . this is due to the fact , that the contraction of cumulant indices in eq . scales with the fourth power of @xmath51 . the evaluation of the cumulant contraction is however massively parallelizable , i.e. no significant communication even within the @xmath58 is required during the computation . thus , the size of the cumulant units can be chosen rather freely regarding the aspect of communication between cpus within the unit . a small size is preferable , due to the typically small memory consumption . for convenience , when switching between configurations @xmath50 and @xmath49 , the self - energies just remain within the compute nodes of the respective processors of the self - energy units . however , it is better to ensure that each self - energy unit deals with more than one lattice site , because we encounter the situation that the time to evaluate the self - energy varies substantially from lattice site to lattice site . if the site indices are assigned randomly , this effect averages out for a sufficient number of sites . as a rule of thumb , it is reasonable to assign self - energies for sixteen random lattice sites to each @xmath58 . let us have a closer look at the dyson s equation in eq . . it contains a block - diagonal part @xmath71 and a sparse off - diagonal part given by the hopping matrix @xmath42 . the latter is composed of a contour-@xmath72-function in time and a tight - binding structure in real space . in particular , @xmath42 is a small quantity , due to the very nature of the strong - coupling expansion . this fact is exploited algorithmically . it is insightful to change notation in eq . . we can write the lattice green s function in a dirac type notation , as a matrix element @xmath73 where @xmath74 is an orthonormal basis associated with space and time variables . each @xmath59 computes @xmath75 at the sites @xmath76 assigned to it for all @xmath77 and stores the local and hopping green s functions as required . determining the lattice green s function corresponds to solving @xmath78 linear equations in @xmath78 variables . that is , the @xmath79 matrix elements of the green s function are determined by solving such linear equations @xmath80 times . typically , in our application , @xmath51 is at least 512 or 1024 and @xmath55 is around 10000 . note that since the simulation discretizes the points on the kadanoff - baym - keldysh contour with @xmath51 time slices , there is a finite size of time slices @xmath81 . the effect of a finite @xmath81 usually affects the accuracy of the calculations , and thus we perform a quadratic extrapolation of the simulation results from three finite values of @xmath81 to @xmath82 . cross - checks with linear extrapolations show the quadratic extrapolation is superior in most instances . the matrix to be inverted in eq . , @xmath83 , is typically around @xmath84 dimensional . its block structure can be written as @xmath85 where each block represents an @xmath86 matrix and @xmath87 as given in eq . . each row and each column in eq . contains at most 4 hopping matrices @xmath88 to the nearest neighbors @xmath89 , @xmath90 ( in two dimensions ) . the hopping matrices @xmath88 are essentially diagonal matrices whose diagonal elements are the time - dependent hopping amplitudes . the matrix in eq . is extremely sparse , while its inverse @xmath91 is dense . however , due to the small numerical value of the hopping in the strong - coupling expansion , matrix elements of @xmath91 fall off as a function of spatial distance and eventually become irrelevant ; in other words , the lattice green s function is typically block diagonal dominant . this fact can be exploited in a numerically controlled way by utilizing an iterative procedure . we choose the generalized minimal residue ( gmres ) method as a solver @xcite . the gmres method considers an equation system @xmath92 as well as an `` almost '' correct solution @xmath93 where @xmath94 is the so - called _ preconditioner_. @xmath94 is an arbitrary operator whose action on @xmath95 is cheap to compute but is a good approximation to @xmath96 . in our case , due to the small numerical value of the hopping , a natural choice for the preconditioner is the inverse block diagonal matrix @xmath97 where @xmath98 as defined earlier in eqs . and . further details of the gmres method as applied to the dyson s equation in the inhomogeneous strong - coupling expansion are discussed in appendix [ app : gmres ] . we distribute the task of applying gmres to each unit vector @xmath99 of the vector space @xmath100 within the set of green s function units @xmath101 . each green s function unit must store all blocks of the block diagonal @xmath102 . memory allocated to each cpu within a unit thus defines a lower bound to the size of a green s function unit through this requirement in the following way : @xmath103 assuming a double precision representation of complex numbers . if one also decides to store the preconditioner , this value doubles . of the block - diagonal matrix @xmath102 within each unit @xmath59 ( darkly shaded area ) . if symmetries apply , only the blocks associated with representatives of lattice sites need to be stored , and @xmath55 is then taken as @xmath56 . ] we store the blocks of @xmath102 as depicted in fig . [ fig : storageb ] . practically , within each @xmath59 , an efficient application of @xmath102 and @xmath104 to a given vector @xmath105 has to be achieved . the performance of these operations crucially relies on the small size of @xmath59 and on a good load - balancing within @xmath59 . the former is due to an increased number of communication events through relatively slow connections between cpus and also due to its role in the latter . in order to avoid an unnecessarily large number of cpus in @xmath59 , a minimization of memory consumption plays a key role . we construct the representation of @xmath102 within a green s function unit by distributing respective self - energies to the cpus dealing with specific parts of @xmath102 . a practical example is shown in table [ tab : para ] . .table of typical parallelization parameters on a cray xe6 ( 2 gb memory per cpu ) for the application presented in sec . [ subsec : numresults ] . the number @xmath106 denotes the block size in the parallel gmres implementation . [ cols="^,^,^,^,^",options="header " , ] a further important optimization when applying gmres takes advantage of both cache optimizations in basic linear algebra subprograms ( blas)@xcite level 3 routines and a reduction of network latency effects rather than letting @xmath59 apply gmres individually to each unit vector @xmath99 of its concern serially , we choose a parallel implementation . it applies the algorithm to blocks rather than vectors . thus , highly optimized blas level 3 rather than multiple calls of blas level 2 are used . in addition , less communication processes between cpus occur . more specifically , we can write this blockwise procedure as solving the blocked equation system @xmath107 where @xmath108 with appropriately defined unit vectors @xmath99 which belong to a single lattice site in the notation introduced above with eq . . the blocks @xmath109 are the respective blocks of @xmath110 . typical block sizes are listed in table [ tab : para ] . in the gmres procedure , this definition of blocked equations which are solved one after another has the following advantage . the method operates iteratively , where the start value is the preconditioner solution , which is still localized at some lattice site : @xmath111 ( or @xmath112 ) . the computation of @xmath104 is done with lapack calls@xcite for the diagonal blocks by the cpus assigned to them according to fig . [ fig : storageb ] . by means of the spatial structure of this then iteratively refined approximation to the solution , only the application of @xmath113 introduces further lattice sites to the problem . as a consequence , if gmres converges quickly for a given numerical accuracy of the gmres method , non - zero elements only occur within the spatial vicinity associated to the considered @xmath114 block in the blocked representation of the converging gmres solution @xmath115 ( see app . [ app : gmres ] for details ) . indeed , the gmres procedure converges very rapidly , because the hopping @xmath42 between adjacent lattice sites is required to be small in the strong - coupling expansion . for the block representations @xmath116 of iteratively refined gmres solutions , we thus choose to only store non - zero subblocks and distribute them within the @xmath59 units using the storage scheme for @xmath102 ( see above and fig . [ fig : storageb ] ) . in the case that @xmath116 contains non - zero contributions for reducible lattice sites ( this happens when @xmath114 is located near the boundary of the range of site representatives , see appendix [ app : symmetries ] ) , the respective block is handled by the cpu associated to its equivalence class . this minimizes both communication and memory consumption . the former is because applying the preconditioner is only a local operation and applying @xmath117 requires only communications with units storing neighbors of the respective lattice sites . the size of the blocks in @xmath116 is to be chosen as large as possible . however , memory in @xmath59 is usually very limited , so a tradeoff in unit size and block length has to be made . let us also comment on reasonable values for the gmres convergence parameter . the gmres method for our matrix inversion runs until a certain accuracy for the result is achieved , i.e. @xmath118 where @xmath119 is the desired numerical precision and @xmath66 is the euclidian norm . for all practical purposes we encountered so far , a value @xmath120 has been sufficient . this surprisingly large value was verified by comparing to simulations with higher accuracy , that is @xmath121 , for the physical systems studied in this paper at several parameter values . the plots of numerical results of interest are identical to the eye . similar tests were done for completely homogeneous systems by comparing to a numerically exact implementation in momentum space . it may be that for different applications than the one presented here a smaller value of @xmath119 is required . in order to understand the meaning of @xmath119 better it may be useful to compare it to the dimension of the vector it constrains . in our case , the dimension of @xmath122 is @xmath123 . thus , if one chooses to normalize the convergence criterion in eqs . and by the dimension , the constraint reads @xmath124 . in this context it may also be worthwhile to consider the fact that the gmres procedure only involves transformations with @xmath104 , @xmath102 , and @xmath42 . that is , it applies only transformations which comply with the causal structure of the green s function and do not introduce artificial discontinuities with respect to the time variables in the end result for the green s function which are in principle part of the vector space which is being searched by the algorithm . in other words , the physical choice of the preconditioner already constrains the solution space so drastically that even a relatively large value of @xmath119 might be sufficient . with all these optimizations , the green s function evaluation typically consumes no more than five to ten percent of the time required for the self - energy evaluation on a cray xe6 machine in the application to lattice depth modulation spectroscopy described below . the optimizations are necessary to speed up the green s function evaluation appropriately , because we encountered increases in speed by a factor of at least 10 and up to 1000 , for each blockwise application of gmres , distributed storage of the gmres vector blocks , and random assignment of site indices . in other applications than lattice depth modulation , with a large value of the hopping applied for a longer period of time , the requirement of computer time for the green s function evaluation may still exceed the time to evaluate the self - energy . however , we find these requirements to be within reasonable bounds . the only time when global communication and synchronization across all processors is required is when either the self - energy or the green s function evaluations are finished . then , convergence has to be checked , and the global configurations @xmath50 and @xmath49 have to be replaced by each other . this requires point - to - point communications of individual processors across the machine and broadcasts within smaller groups of processors which all require the same data . + the occuring communication events are displayed in fig . [ fig : globalconfigswitches ] . during the switch @xmath125 ( fig . [ subfig : cgtocs ] ) , the green s function contents on the nearest neighbors of the sites assigned to a given @xmath58 have to be sent from the green s function unit which computed them . due to the possibly random assignment of site indices @xmath24 to spatial coordinates @xmath126 , the input to @xmath58 is collected from various @xmath127 . the storage scheme used for the results within @xmath127 determines the actual processors which send the data . when switching from @xmath50 to @xmath49 , spatial ranges tasked to specific parts of each green s function unit have to be sent from the self - energy units which contain this information ( fig . [ subfig : cstocg ] ) . the storage pattern for @xmath59 is already the one specified for the block diagonal matrix denominator of the dyson s equation ( fig . [ fig : storageb ] ) . to finish the change of configurations , an in - place substitution of the self - energies by the respective blocks @xmath128 is performed by computing the respective atomic green s functions and inserting @xmath129 , as defined in eq . . the process @xmath130 is typically more time - consuming than @xmath125 . however , it can be optimized by using broadcasts between groups of processors with the same data requirements : each @xmath46-th cpu in @xmath59 requires the same data set to operate . the switching processes cost no more than five percent of the total computation time and are thus negligible . however , the implementation involves a considerable amount of book - keeping . let us summarize the algorithm and also provide some implementation details on the way . as the method implements the self - consistent solution of eq . , the algorithm is split in two steps : the self - energy evaluation ( eqs . and ) and the green s function evaluation ( eq . ) . since the computation of the self - energy and the evaluation of the green s function have different requirements in terms of computational resources on the supercomputer , they use different data structures and collaboration patterns amongst the cpus . we refer to the data structures of the algorithm in the self - energy evaluation state as the configuration @xmath50 and to the data structures of the algorithm in the green s function evaluation state as the configuration @xmath49 . the respective configurations are subdivided into mutually independent units @xmath58 and @xmath59 spanning several cpus and the memory associated with them , respectively . this approach is tailored to a cluster , rather than a shared - memory architecture . if one chooses to employ the message passing interface ( mpi ) standard in order to implement the algorithm , it is advantageous to use the ` mpi_group ` feature to ensure an efficient communication within the units @xcite . it turns out to be useful to define internal communicators for the @xmath59 and @xmath58 , respectively , as well as for sets of processors with shared requirements , such as the @xmath61-th processor of each @xmath59 , since they all require the same self - energies . within these shared - interest communicators , data can be broadcasted efficiently . it may also be reasonable to use advanced mpi features to perform an optimization with respect to the network topology of the supercomputer , such that communication within the @xmath59 units is optimal and equally fast for all @xmath34 . in contrast , the communication within @xmath58 is not time - critical , because the main effort , computing the integrals in eq . , is done by each processor in @xmath58 independently . let us now comment on the implementation of eq . . each @xmath58 computes @xmath131 for a certain range of sites . here , the main effort is the contraction of the time indices . the atomic - limit cumulant green s function @xmath132 is computed on the fly using tabulated values of the exponentials in eq . . the computational effort of eq . scales with @xmath133 and is the computationally most costly operation . however , it may also be implemented on gpus , due to little memory and bandwidth requirements . after having evaluated eq . , the units @xmath58 compute the updated self - energy using eqs . and . then , as described in section [ subsec : confswitch ] , the resulting local self - energies are sent to the @xmath59 units which may or may not overlap with the respective @xmath58 . within each @xmath59 , the self - energy for all sites has to be available and is thus equally distributed over the cpus according to the storage pattern depicted in fig . [ fig : storageb ] . it is advised to keep the self - energy results in @xmath58 for the next update as described by eq . , even though the machine changes to configuration @xmath49 in the meantime . this is because the @xmath134 in eq . has to be available in the self - energy computation of the next iteration of the self - consistency loop . in order to minimize the memory consumption of the relevant range of @xmath134 , it can be distributed equally within each @xmath58 . in order to establish the configuration @xmath49 to compute eq . , the self - energies in @xmath59 are then replaced by the blocks of the matrix @xmath102 according to eq . . optionally , the elements of the preconditioner @xmath104 can also be computed and stored at this point in time , also using the previous storage pattern . however , this competes with the requirement to keep @xmath59 small , because the action of the preconditioner can also be computed on the fly from @xmath102 with a smaller memory requirement . having fully set up the @xmath49 configuration , eq . is written as the vectorized linear eq . as described in section [ subsec : greensfunctionevaluation ] . the key variable is the bundle of gmres vectors @xmath116 whose initial value @xmath135 is the preconditioner @xmath136 applied to a bundle of unit vectors , as in eq . . @xmath135 has only non - zero entries at a single site index . @xmath116 is also stored according to the scheme in fig . [ fig : storageb ] . a good optimization here is to store only nonzero components of @xmath116 emerging from @xmath135 due to the application of the hopping matrix . for this purpose , each processor in @xmath59 can keep track of the sites with non - zero elements in @xmath116 based on the site index of @xmath135 and the number of hopping events applied to @xmath116 during the gmres procedure elaborated in app . [ app : gmres ] . once a hopping occurs due to the application of @xmath137 , a given processor may have contributions to be stored and/or added to a value assigned to another processor within the storage scheme ( compare to the block structure of @xmath138 in eq . ) . such transmissions are the major communication events within @xmath59 . the required communication bandwidth within @xmath59 can only be minimized by assigning connected spatial domains with a minimal surface to single processors within @xmath59 . however , doing so is strongly disadvantageous in the case that the gmres is converging very rapidly , i.e. if the hopping is small and @xmath139 is small . in this case , all but one of the cpus in @xmath59 will remain idle , because the non - zero elements in @xmath116 do not leave the cpu storing the non - zero elements of @xmath135 . thus , for a rapidly converging gmres , a random assignment of the site leading to largely scattered domains is more appropriate . once each @xmath59 has computed the green s functions on the spatial range assigned to it , the different spatial components of @xmath140 are distributed to the @xmath58 units which require them for evaluating the green s function sum in eq . in order to start the next iteration . in this section , we present results of the algorithm for trapped atoms in a two - dimensional optical lattice which is subject to a periodic modulation of the lattice depth . we compare this method to a homogeneous version of the algorithm which was previously successfully applied to a lattice depth modulation experiment within the lda @xcite . the lda is generally expected to yield good results for systems without mass transport . this is a well - established observation in equilibrium @xcite . we show that also in a nonequilibrium scenario without mass transport , the high accuracy of the lda can be explicitly demonstrated . at the same time we validate our direct computational approach . in lattice - depth modulation spectroscopy , the atoms are subject to a time - dependent optical lattice potential @xmath141 the trap potential does not depend on time and has the parabolic shape @xmath142 the lattice potential satisfies @xmath143 which contains the time - dependent lattice depth @xmath144 } ( t ) \cdot \delta v \sin \omega t.\ ] ] we assume that the lattice is modulated over a finite time interval @xmath145 $ ] and that the system is in an initial thermal state at time @xmath8 . numerically , we start the simulation at an earlier point in time , in order to be able to check for convergence , as discussed in fig . [ fig : keldysh ] . the lattice constant @xmath146 is defined by the laser wavelength @xmath147 . the single - particle hamiltonian @xmath148 yields the recoil energy @xmath149 as a natural choice for an energy unit . in order to compute the coefficients of the many - body hamiltonian in eq . from the single - particle hamiltonian , we insert the constant hoppings @xmath150 and interactions @xmath151 of a translationally invariant lattice . these can be computed easily with maximally localized wannier functions @xcite . due to the time - dependence of the lattice depth @xmath152 , we also obtain a time - dependent interaction @xmath153 and hopping @xmath154 . we write the initial values of the interaction and the hopping as @xmath70 and @xmath155 , respectively . in these units , the trap potential can be written as @xmath156 hence , @xmath157 can be interpreted as the length scale on which the trap potential reaches the strength of the initial hopping amplitude . it is important to keep @xmath157 larger than a couple of lattice spacings , since otherwise the trap potential interferes drastically with the hopping between neighboring sites and the density changes too fast for the lda to be accurate . + + + as a test system , we set the lattice depth to @xmath158 and modulate it with an amplitude @xmath159 at the resonant frequency @xmath160 . the interaction strength is chosen to be @xmath161 , and we assume an initial temperature @xmath162 . we choose to study two cycles of the modulation , that is @xmath163 . for the simulations , we use up to 1024 time slices and a lattice with up to 1024 symmetry - irreducible lattice sites , that is up to 7844 actual lattice sites . the computational effort for a system with 512 symmetry - irreducible sites and a maximum of 1024 time slices is approximately @xmath164 cpu - hours on a cray xe6 . for instance , this involves 32768 cpus for approximately 12 hours by the main simulation and some further cpu time for the cheaper simulations at larger @xmath81 which are required for the extrapolation @xmath165 . figure [ fig : trapresults ] shows simulation results for distribution of the double occupancy in a trapped system with @xmath166 and the global chemical potential @xmath167 . each subfigure displays the distribution at a different point in time . due to the lattice depth modulation , the hopping in units of the interaction @xmath168 drives the system . the increases in the double occupancy occur as @xmath168 is decreasing . to provide a better picture of the time dependence , fig . [ fig : integrateddocc ] shows the fraction of atoms on doubly occupied sites , @xmath169 where @xmath170 as a function of time for several values of the trap curvature . in the cases @xmath166 and @xmath171 , the results lie on top of each other , whereas for @xmath172 a slight deviation occurs . this agrees with the results found in our previous publication ref . @xcite for homogeneous systems . in order to perform the comparison to the lda , we solve numerous mutually independent homogeneous versions of the problem at several chemical potentials @xmath173 and compare to local observables obtained from the full trap simulation at a position @xmath76 . we consider a set of test systems with three different trap curvatures , that is different values of the characteristic length @xmath157 of the trap potential , namely @xmath174sites , @xmath175sites , and @xmath176sites . in the simulations , the real part of the kadanoff - baym - keldysh contour extends over an interval @xmath177=[-2\hbar / u_0 , 14 \hbar / u_0]$ ] , whereas the modulation acts over the interval @xmath178 $ ] . figure [ fig : ldacompadocc ] shows a comparison of the double occupancy @xmath179 as a function of the initial chemical potential @xmath7 at times before ( @xmath180 ) and after ( @xmath181 ) it has been driven out of equilibrium by the lattice depth modulation . as we see for both , the equilibrium ( @xmath180 ) and nonequilibrium ( @xmath181 ) situations , the numerical results for the inhomogeneous system agree well with the lda , even for the rather steep trap potential with @xmath174sites . the slight deviation of the solution for @xmath182sites from the other nonequilibrium curves may still be due to numerical imperfections . the agreement with the lda indicates that the creation of doubly occupied sites in a mott insulator subject to a modulated lattice depth is caused by strongly local excitation processes . we presented a computational approach to an inhomogeneous mott - insulating system of ultracold atoms . a major challenge is to compute a large matrix inverse in the dyson equation . we show that a gmres - based inversion approach exploiting the small numerical value of the hopping as compared to the many - body interaction yields a feasible implementation on supercomputers . a comparison to the lda shows that both methods are well - suited for the problem of lattice - depth modulation spectroscopy . this hints towards mainly local processes being involved in the coherent excitations between lower and upper hubbard bands in this particular setting , as might have been anticipated . in the future , we will apply the inhomogeneous method to problems with mass transport where the lda is expected to fail . at present , the computational complexity of the algorithm is proportional to @xmath133 . it may be worthwhile to investigate the possibility to extend the time range by truncating certain parts of the self - energy at a given threshold for @xmath183 . this measure could increase the applicability of the algorithm greatly but requires further efforts . this work was supported by a muri grant from the air force office of scientific research numbered fa9559 - 09 - 1 - 0617 . supercomputing resources came from a challenge grant of the dod at the engineering research and development center and the air force research and development center . the collaboration was supported by the indo - us science and technology forum under the joint center numbered jc-18 - 2009 ( ultracold atoms ) . jkf also acknowledges the mcdevitt bequest at georgetown . hrk acknowledges support of the department of science and technology in india . point - group symmetry imposed by the circular trap potential on the 2d lattice . the circles denote lattice sites . solid circles are representatives of an equivalence class of lattice sites with respect to the symmetry . the symmetry partners of the red / gray solid circle are displayed in red / gray . ] we illustrate the exploitation of symmetries for the example of spherical symmetry on a two - dimensional square lattice with an @xmath184-orbital basis . in this case , a @xmath185 point group symmetry @xcite is imposed on the lattice . [ fig : symmetry ] displays the lattice sites of a lattice . for the @xmath185 symmetry , all sites can be represented by the sites in the irreducible wedge @xmath186 . these representatives of the equivalence classes are denoted by solid black circles . if the green s function and self - energy transform as the identity representation of the point group , the computations only need to be performed for those representatives . the representative of a given site can be retrieved by reflections with respect to the coordinate axes and the diagonals which are shown as dashed lines in fig . [ fig : symmetry ] . we provide a brief derivation of eq . . let us start by assuming an equidistant discretization @xmath187 ( @xmath188 ) of the forward part of the kadanoff - baym - keldysh contour . we obtain @xmath189 with @xmath190.\ ] ] the operator @xmath191 simplifies as follows : @xmath192 + \mathcal{o}(\delta t^2 ) \\ = & -{\ensuremath{\mathrm{i}}}\delta t \cdot \bigg ( \sum_j j_{lj } c^\dagger_{l\sigma}c_{j\sigma } - { \varepsilon}_{l\sigma}(t ) n_{l\sigma } - \\ & \qquad\quad - u_l(t ) n_{l\uparrow } n_{l\downarrow}\bigg ) + \mathcal{o}(\delta t^2 ) . \end{split}\ ] ] taking the limit @xmath82 results in eq . . numerically , this limit must be performed via linear and/or quadratic extrapolation of multiple simulations for different @xmath81 values . the generalized minimal residue method was introduced by saad and schultz @xcite to solve a linear equation @xmath193 a good introduction to the method can be found in ref . @xcite , and a useful c++ implementation is provided by the nist iml++ template library @xcite . in order to solve the equation , gmres operates in @xmath194-dimensional krylov subspaces @xmath195 which are successively built up , starting with @xmath196 . @xmath197 is the residue of the initial guess @xmath198 for the solution . the gmres method forms a refined approximate solution @xmath199 defined by the minimization requirement @xmath200 where the expression `` @xmath201 '' demands @xmath202 to be minimal . the krylov space dimension @xmath194 is increased until the convergence criterion is reached or @xmath194 approaches the threshold @xmath203 at which the minimization is considered too expensive . if @xmath204 , gmres is restarted with a krylov space size @xmath196 , where the initial guess @xmath198 is taken to be @xmath205 from the previous gmres iteration before restarting . with a preconditioner @xmath94 , one applies gmres to the system @xmath206 rather than eq . . if @xmath207 this system is better conditioned than eq . . in the case that @xmath117 is a sparse matrix with large entries on the diagonal and some randomly occuring off - diagonal entries , the diagonal matrix @xmath102 defined by the diagonal entries of @xmath117 yields a good preconditioner @xmath104 , because the condition number @xmath208 is not as close to @xmath209 as the regularized @xmath210 here , @xmath211 and @xmath212 represent the respective maximal and minimal singular values . also , the application of the diagonal matrix @xmath102 to a vector is a cheap operation . the same argument can be made in the case that @xmath117 has large entries on the block diagonal @xmath102 , as it is the case in the nonequilibrium inhomogeneous strong - coupling expansion , where @xmath213 . the block diagonal @xmath102 is defined in eq . , and the hopping matrix @xmath42 is a sparse matrix with small numerical values ( see also eq . ) . due to the latter , in fact @xmath214 , so the equation system is well - conditioned . 1 . for the initial guess @xmath215 compute the preconditioned residue @xmath216 , as well as @xmath217 . initialize the hessenberg matrix @xmath218 2 . for @xmath219 do * @xmath220 * for @xmath221 do * * * @xmath222 * * @xmath223 * @xmath224 * if @xmath225 then proceed with step 3 to compute the result . * otherwise use step 3 to check for convergence , eq . . continue if not converged . * @xmath226 3 . solve the @xmath194-dimensional linear minimization problem @xmath227 where @xmath228 is the upper left @xmath194-dimensional square of @xmath229 , to obtain the result @xmath230 . + then set @xmath231 , with @xmath232 . + practically , the gmres with preconditioner scans for solutions in the modified affine krylov spaces @xmath233 here , the basis vectors are @xmath234 in the case of the inhomogeneous nonequilibrium strong - coupling expansion , @xmath235 is a unit vector localized at a given lattice site . the @xmath61-th basis vector @xmath236 of the krylov space @xmath237 corresponds to @xmath61 hopping processes , because @xmath238 , and in @xmath236 , @xmath117 is applied @xmath61 times to @xmath235 . that is , the gmres method includes exactly @xmath194 iterated hopping processes in iteration @xmath194 until convergence is reached . if gmres is restarted , the @xmath194th iteration includes @xmath239 iterated hopping processes , where @xmath24 is the number of restarts . this is crucial for the computational optimizations used in the implementation . the procedure can be seen as a numerically controlled analogue to a partial summation of the dyson series @xmath240 99 i. bloch , j. dalibard , and w. zwerger , rev . phys . * 80 * , 885 ( 2008 ) t. esslinger , ann . rev . condens . mat . * 1 * , 129 ( 2010 ) th . stferle , h. moritz , c. schori , m. khl , and t. esslinger , phys . * 92 * , 130403 ( 2004 ) r. jrdens , n. strohmaier , k. gnter , h. moritz , and t. esslinger , nature * 455 * , 204 ( 2008 ) d. greif , l. tarruell , th . uehlinger , r. jrdens , and t. esslinger , phys . lett . * 106 * , 145302 ( 2011 ) s.r . clark and d. jaksch , phys . a * 70 * , 043612 ( 2004 ) c. kollath , u. schollwck , w. zwerger , phys . * 95 * , 176401 ( 2005 ) c. kollath , a. iucci , t. giamarchi , w. hofstetter , and u. schollwck , phys . lett . * 97 * , 050402 ( 2006 ) ; c. kollath , a. iucci , i.p . mcculloch , and t. giamarchi , phys . a * 74 * , 041604(r ) ( 2006 ) k. winkler , g. thalhammer , f. lang , r. grimm , j. hecker denschlag , a. j. daley , a. kantian , h. p.bchler , p. zoller , nature * 441 * , 853 ( 2006 ) a. j. daley , p. zoller , and b. trauzettel , phys . lett . * 100 * , 110404 ( 2008 ) a. j. daley , j. m. taylor , s. diehl , m. baranov , and p. zoller , phys . lett . * 102 * , 040402 ( 2009 ) k. mikelsons , j. freericks , and h. r. krishnamurthy , phys . * 109 * , 260402 ( 2012 ) a. dirks , k. mikelsons , h. r. krishnamurthy , and j. freericks , arxiv:1304.7802 w. kohn , phys . rev . * 115 * , 809 ( 1959 ) y. saad and m.h . schultz , siam j. sci . . comput . * 7 * , 856 ( 1986 ) e. anderson , z. bai , c. bischof , s. blackford , j. demmel , j. dongarra , j. du croz , a. greenbaum , s. hammarling , a. mckenney , and d. sorensen , _ lapack users guide _ ( society for industrial and applied mathematics , philadelphia 1999 ) e. assmann , s. chiesa , g. g. batrouni , h. g. evertz , and r. t. scalettar , phys . rev . b * 85 * , 014509 ( 2012 ) w. gropp , e. lusk , and a. skjellum , _ using mpi _ , mpi press , 1999 , isbn 9780262571326 f.a . cotton , _ chemical applications of group theory _ , wiley and sons ltd . , 1990 , isbn 997151267x https://lp.uni-goettingen.de/get/text/2024 and https://lp.uni-goettingen.de/get/text/2025 ( in german , opened on sept . 10 , 2013 ) ( sept . 10 , 2013 )

the electronic transport properties of quasi - onedimensional ( 1d ) systems , described by a localized orbitals basis set , can be calculated using the nonequilibrium green s function ( negf ) method.@xcite the system is usually divided into two semi - infinite left- and right - hand side leads , and a scattering region joining them . the effect of the leads onto the scattering region is taken into account by the so called self - energies ( se ) , which can be calculated from the surface green s function ( sgf ) of the semi - infinite leads . these can be obtained either with recursive methods @xcite or by using a semi - analytic formula.@xcite recursive methods are affected by poor convergence for some critical systems , typically when the hamiltonian for the leads is rather sparse . semi - analytical methods instead bypass those problems by construction , however major difficulties arise if the hopping matrices are singular or , more generally , ill conditioned . unfortunately the condition of the hamiltonian is set by the electronic structure of the leads and by the unit cell used , and thus it is largely not controllable . for this reason an algorithm that performs under the most generic conditions is highly desirable . here we present an improved semi - analytical method that overcomes these limitations and thus represents a robust algorithm for quantum transport based on _ ab initio _ electronic structure . in the first part of the paper the extended algorithm for the calculation of the se is presented . first the construction of the green s function of an infinite 1d system as derived in reference [ ] is recast into a more general form based on the notion of a complex group velocity . then we present an extension of such method to the calculation of the sgf and se . the new algorithm is defined also for the case of singular hopping matrices . this largely improves the numerical accuracy . however we find that even such an improved scheme sometimes fails if the hopping matrices are close to being singular . we overcome this problem by performing a transformation of the hopping matrix that reduces its condition number @xmath0 , defined as the ratio between its largest to its smallest singular value.@xcite this transformation limits the maximum absolute value of the imaginary part of the bloch wave - vectors , increasing both accuracy and stability . two approaches are presented , the first is based on a singular value decomposition ( svd ) , which is also used to significantly reduce the dimension of the hamiltonian , while the second consists in adding a random noise matrix . this extended scheme is implemented in the negf _ ab initio _ transport code _ smeagol_,@xcite based on the density functional theory ( dft ) code siesta.@xcite in the second part of this work we present three examples of calculations performed with our new implementation . we compare the results to the ones obtained by using the original method of reference [ ] , finding a considerable improvement . however , although the algorithm appears very robust , our detailed error analysis reveals that for a given system the accuracy is lost at some specific energies . this is caused by the divergence of one of the se eigenvalues . the physical origin of this behavior lies in the presence of surface states very weakly coupled to the semi - infinite leads . surface states appear whenever at a given energy the set of bloch functions ( with both real and imaginary wave - vectors ) for the infinite quasi-1d system is linearly dependent . in the simplest case this corresponds to two bloch functions being equal inside the unit cell . a small imaginary part is thus added to the energy in a small energy range around the surface state . it is shown that this has little effect on the transport properties in the high transmission regime , whereas for low transmission it has a substantial influence on the results . crucially only a very small imaginary part is used , and moreover this is added only around the energy of the surface state and thus the error can be carefully controlled . following the scheme introduced in reference [ ] the construction of the retarded green s function for an infinite quasi-1d system is now recalled . this is the starting point for the calculation of the sgf . it is assumed that the hamiltonian is written over a localized orbitals basis set and that the interaction has finite range . the size of the unit cell can be chosen to guarantee interaction only to the first nearest neighboring unit cells . the total hamiltonian of the system @xmath1 ( the integers @xmath2 and @xmath3 label the unit cells ) can then be written as @xmath4 where @xmath5 and @xmath6 are @xmath7 matrices , with @xmath8 being the number of orbitals comprised in the unit cell ( see figure [ fig : onedimh ] ) . and hopping @xmath9 . the overlap matrix has the same structure.,width=264 ] if time - reversal symmetry holds then @xmath10 , and @xmath11 , however the solutions presented here are valid also in the more general case when @xmath12 and/or @xmath13 . we further assume that the overlap matrix @xmath14 has the same structure and range of the hamiltonian @xmath15 where @xmath16 and @xmath17 are again @xmath18 matrices with the same meaning of their hamiltonian counterparts . the solutions of the hamiltonian equation for the associated infinite periodic system @xmath19 are bloch functions @xmath20 , where @xmath21 and @xmath22 are @xmath8-dimensional vectors and @xmath23 is the wave - vector , which in general is a complex number . for a given real or complex energy @xmath24 there are @xmath25 solutions with wave - vectors @xmath26 and corresponding wavefunctions @xmath27 . each of them satisfies @xmath28 if we define @xmath29 , ( @xmath30 ) , the equation above can be rewritten as @xmath31 where the additional index r denotes explicitly that the solution is a right eigenvector . the corresponding left eigenvector @xmath32 satisfies @xmath33 time - reversal symmetry gives @xmath34 , so that in the case of real @xmath26 ( propagating states ) left and right eigenvectors are equal . for complex @xmath26 left and right eigenvectors are different , describing left- and right - decaying states . the sets @xmath35 , @xmath36 and @xmath37 that satisfy eqs . ( [ eq : scheq ] ) and ( [ eq : scheql ] ) at a given energy can be found by solving a quadratic eigenvalue problem@xcite of the form @xmath38 where @xmath39 here @xmath40 and @xmath41 are respectively the @xmath7 unit and zero matrices . the normalization constant is the square root of the complex group velocity @xmath42 ( @xmath43 ) equal to @xmath44 in the following we assume that the eigenvectors @xmath45 and @xmath32 are always normalized to give @xmath46 . if time - reversal symmetry holds then @xmath47 , so that if the imaginary part of @xmath26 is zero the group velocity is real . note that , at variance with reference [ ] , eqs . ( [ eq : qep1 ] ) and ( [ eq : qep2 ] ) avoid the inversion of @xmath48 , so that they eliminate a possible source of singularities in the calculation of @xmath26 , @xmath45 and @xmath32 . the full sets of left @xmath49 and right eigenvectors @xmath50 form a complete and orthogonal basis . the orthogonality relation is @xmath51 where @xmath52 is a constant . this leads to @xmath53 for @xmath54 this equation is only satisfied if @xmath55 , in which case it corresponds to the definition of @xmath56 . with the chosen normalization the basis is therefore orthonormal . the corresponding completeness relation then reads @xmath57 and provides the three following useful relations @xmath58 note that in eqs . ( [ eq : ortho]-[eq : rel3 ] ) the sums run over all @xmath59 solutions . if @xmath60 and @xmath61 eqs . ( [ eq : rel2 ] ) and ( [ eq : rel3 ] ) are equivalent . the retarded green s function @xmath62 of the system is defined by means of the green s equation @xmath63=\delta_{zz '' } , \label{eq : gf}\ ] ] with @xmath64 real . in what follows we present and expand , by using left and right bloch functions , the solution to eq . ( [ eq : gf ] ) given in reference [ ] only in terms of the right eigenvectors @xmath65 . first we divide the @xmath59 @xmath45 vectors into @xmath8 right - going states with either im@xmath66 ( right decaying ) or im@xmath67 and @xmath68 ( right propagating ) , and @xmath8 left - going states with either im@xmath69 ( left decaying ) or im@xmath67 and @xmath70 ( left propagating ) . as a matter of notation in order to distinguish left- from right - going states , in what follows we indicate the right - going states with @xmath23 , @xmath22 and @xmath71 , and the left - going states with a bar over these quantities , i.e. @xmath72 , @xmath73 and @xmath74 . as in reference [ ] we introduce the duals @xmath75 of the right - going states @xmath45 defined by @xmath76 , and the duals @xmath77 of the left - going states @xmath78 defined by @xmath79 . if we define the matrices @xmath80 and @xmath81 as @xmath82 then the duals can be obtained by simple inversion @xmath83 the inversions in eqs . ( [ eq : adjbasisl ] ) are usually well defined , unless @xmath80 and @xmath81 do not have full rank . we will return on this aspect in section [ sec : sstates ] , for the moment we assume that the duals can always be constructed . the green s function calculated in reference [ ] is then @xmath84 with the matrix @xmath85 given by @xmath86 we now introduce the right transfer matrices@xcite @xmath87 and @xmath88 @xmath89 note that both @xmath87 and @xmath88 have eigenvalues with complex modulus @xmath90 . for an integer @xmath2 the following relations hold @xmath91 which allow us to write the green s function of equation ( [ eq : gfstefano ] ) as @xmath92 in the same way @xmath93 is rewritten as @xmath94 note that although the matrices @xmath87 and @xmath88 are in general well defined , the inverse of these matrices is not . in fact if @xmath48 and @xmath95 are singular there are some @xmath26 with im@xmath96 , so that @xmath97 ( see section [ sec : svdreduce ] ) . in this case @xmath87 does not have full rank and is therefore singular . the same argument holds for @xmath88 . equation ( [ eq : g00a ] ) can therefore be used only if the matrices @xmath48 and @xmath95 are not singular . a possible way for overcoming such limitation is by using an equivalent form for the green s function based on the left and right eigenvectors . the starting point is the relation ( [ eq : rel1 ] ) that will allow us to find the connection between the duals and the left eigenvectors . ( [ eq : rel1 ] ) contains a sum over both left- and right - going states . by moving the contribution of the left - going states to the right side of the equation we obtain @xmath98 , where we have introduced the auxiliary matrix @xmath99 . by multiplying @xmath99 from the left with either @xmath100 or @xmath101 we obtain respectively @xmath102 and @xmath103 . the matrix @xmath99 is determined by inserting these relations into eq . ( [ eq : v ] ) and by using eq . ( [ eq : rel3 ] ) , the result is @xmath104 . the relation between the dual basis and the left eigenvectors is therefore @xmath105 this result allows us to rewrite the green s function of eq . ( [ eq : gfstefano ] ) in a shorter form @xmath106 this result represents a generalization to complex energies and to systems breaking time - reversal symmetry of the solution given in references [ ] for hermitian hamiltonians , real energy and an orthogonal tight binding model . this derivation shows that the green s function can be equivalently expressed by using the right eigenvectors and their duals ( eq . ( [ eq : gfstefano ] ) ) , or both the right and left eigenvectors ( eq . ( [ eq : gfansatz ] ) ) . it is thus possible to move from one representation to the other through eq . ( [ eq : phitildeofphil ] ) that relates the duals to the left eigenvectors . one can then decide which representation to use , depending on the specific problem investigated . we note that eq . ( [ eq : gfansatz ] ) has the benefit that @xmath107 can be calculated also in the case where the two matrices @xmath48 and @xmath95 are singular . for those @xmath26 where im@xmath96 the group velocity becomes @xmath108 and is therefore well defined ( @xmath109 for im@xmath110 ) . as a matter of completeness we show that a representation entirely based on the left bloch functions and their duals @xmath111 and @xmath112 is also possible . by multiplying eq . ( [ eq : gfansatz ] ) respectively by @xmath111 and @xmath112 from the right we obtain the two relations @xmath113 again the left transfer matrices @xmath114 and @xmath115 are defined as @xmath116 and the green s function of eq . ( [ eq : gfansatz ] ) can be rewritten as @xmath117 the structure of eq . ( [ eq : gleft ] ) is the same as that of eq . ( [ eq : gfa ] ) , with the difference that now @xmath107 is multiplied to the left of the transfer matrix . finally we extend eq . ( [ eq : g00a ] ) and present four equivalent relations for the inverse of @xmath107 @xmath118 the second of these relations can be shown by multiplying eq . ( [ eq : rel2 ] ) by @xmath119 from the right and then by using eq . ( [ eq : phitildeofphil ] ) . in the same way the third and fourth equations can be obtained by multiplying equations ( [ eq : rel2 ] ) and ( [ eq : rel3 ] ) by @xmath119 from the left . in the following we will use mostly the quantities expressed in terms of the right eigenvectors only , however the same conclusions can be derived using the left eigenvectors . as an example of the use of the green s function in the form of eq . ( [ eq : gfansatz ] ) we determine the spectral function @xmath120 and the density of states ( dos ) of the infinite quasi-1d system for the special case where the hamiltonian and the overlap matrices are hermitian . the spectral function is defined as@xcite @xmath121_{zz'}=i\left[g_{zz'}-\left(g_{z'z}\right)^\dagger\right ] . \label{eq : azzg}\ ] ] the dos @xmath122 projected on the unit cell @xmath2 then is @xmath123.\ ] ] by using eq . ( [ eq : overlap ] ) this becomes @xmath124 . \label{eq : dosg}\ ] ] in general the main contribution originates from the first term in the sum , which can be interpreted as the onsite dos @xmath125 @xmath126 . \label{eq : onsitedos}\ ] ] we now calculate @xmath120 and @xmath127 for the special case where @xmath128 and @xmath61 . in this case for im@xmath67 we have @xmath129 , whereas if im@xmath130 then @xmath131 . in the same way for im@xmath132 we have @xmath133 , whereas if im@xmath134 then @xmath135 . therefore for each right decaying state with im@xmath66 there is a left decaying state with @xmath136 and @xmath137 . using these relations when inserting the green s function of eq . ( [ eq : gfansatz ] ) in the definition of @xmath138 , the contribution from all the decaying states cancels out . the only remaining contributions come from the propagating states , also denoted as open channels . for these @xmath139 , @xmath140 and @xmath141 . with these constraints , and by using eq . ( [ eq : gfansatz ] ) , the spectral function becomes @xmath142 where @xmath143 is the number of open channels ( number of bloch functions at a given energy with real positive @xmath23 vector ) . if there are no open channels @xmath144 and the green s function is hermitian . finally , by using eqs . ( [ eq : azz ] ) and ( [ eq : dosg ] ) , and the fact that the eigenvectors are normalized so to give @xmath46 ( see eq . ( [ eq : ln ] ) ) , the dos at the site @xmath145 is simply @xmath146 this is the well known result for the dos of infinite periodic 1d systems.@xcite the retarded green s function @xmath147 for a quasi - periodic system , where the left and right sides are separated at the position @xmath145 ( the left - hand side part extends from @xmath148 to @xmath149 and the right - hand side part from @xmath150 to @xmath151 , with no coupling between the cells at @xmath149 and @xmath150 ) , can be constructed from the green s function @xmath152 for the infinite chain as demonstrated in reference [ ] @xmath153 the left - hand side sgf is then defined as @xmath154 , and the right sgf as @xmath155 . the sgf can be obtained by using eq . ( [ eq : gfa ] ) @xmath156 this corresponds to the form derived in reference [ ] . this result can be simplified by using the relations ( [ eq : g00 ] ) for @xmath107 to @xmath157 these equations unfortunately are only defined if @xmath48 and @xmath95 are not singular . the same problem however does not affect the left and right se , @xmath158 and @xmath159,@xcite since they simply are @xmath160 in complete analogy the same expressions obtained by using the left transfer matrices are @xmath161 and @xmath162 . this result is equivalent to those obtained in references [ ] and derived with different approaches , demonstrating the equivalence of those to our semi - analytical formula . since negf - based transport codes simply require @xmath163 and @xmath164 , our scheme allows the calculations of system with arbitrarily complicated electronic structure . a schematic tree diagram describing the steps involved in obtaining the se is shown in figure [ fig : diagram ] ( basic algorithm ) . ( [ eq : senel ] ) and ( [ eq : sener ] ) demonstrate that the se can be calculated directly without explicitly calculating @xmath165 and @xmath166 . in situations where also the sgf are needed , these can be obtained by using the relation @xmath167^{-1 } , \label{eq : rgfl}\\ g_\mathrm{r } & = & - \left[k_0 + \sigma_\mathrm{r } \right]^{-1}. \label{eq : rgfr}\end{aligned}\ ] ] this can be derived by adding one layer to the left and one to the right surfaces respectively.@xcite in appendix a we show that the se calculated with eqs . ( [ eq : senel ] ) and ( [ eq : sener ] ) indeed fulfill the above equation . moreover with the use of eqs . ( [ eq : gfs ] ) and ( [ eq : rgfl ] ) we can now regularize equation ( [ eq : g00a ] ) also for the case where @xmath87 is singular by writing it as @xmath168 we have therefore a scheme where the se are identified as the principal quantities , whereas the sgf and @xmath107 are derived from these . when we compare the method of reference [ ] with the equations derived above , we notice that now it is not necessary to calculate the matrix @xmath107 and its inverse using eq . ( [ eq : g00a ] ) in order to obtain the se . this is not defined in the case of singular @xmath48 and @xmath95 , and therefore we expect the new method to be more stable and accurate . also the problems caused close to band edges by the van hove singularities in @xmath107 are avoided . moreover the method in reference [ ] relies on the calculation of the sgf in order to obtain the se , whereas here the sgf is not needed . as we will show in section [ sec : sstates ] close to surface states the error in the sgf is much larger than the one for the se , so that we also expect a large improvement in the accuracy for those particular states . the accuracy with which the se are calculated depends on the accuracy involved in solving eq . ( [ eq : qep1 ] ) , a quadratic eigenvalue problem extensively studied in the past.@xcite however most solution methods have problems if @xmath48 or @xmath95 are close to being singular , or more generally if their condition number @xmath0 is large . in this case some of the complex eigenvalues tend to infinity and others to zero at the same time , and this results in a loss of accuracy in numerical computations . when calculating @xmath87 ( @xmath88 ) however the contributions from the states with im@xmath169 ( im@xmath170 ) are vanishingly small . it is therefore useful to limit the range of the eigenvalues @xmath171 in such a way that the important eigenstates with small @xmath172 and @xmath173 can be calculated accurately , while losing precision for the less important eigenstates with large @xmath172 and @xmath173 . in this section we show how this can be achieved by decreasing @xmath174 and @xmath175 . here we assume that @xmath60 , so that @xmath176 . minor modifications are needed for the general case ( see appendix b ) . in order to obtain @xmath174 first a svd of the matrix is performed @xmath177 @xmath178 and @xmath93 are unitary matrices , and @xmath179 is a diagonal matrix , whose diagonal elements @xmath180 are the singular values . these are real and positive , and ordered so that @xmath181 . if @xmath182 is the largest singular value , and @xmath183 the smallest one , then the condition number is defined as @xmath184 , with @xmath48 singular if @xmath183 is zero . we now replace @xmath179 with an approximate @xmath185 , whose diagonal elements @xmath186 are @xmath187 and accordingly @xmath48 with @xmath188 . the tolerance parameter @xmath189 is a real positive number that determines the condition number of @xmath190 . we now present two possible choices for @xmath191 . the first is to set @xmath192 , resulting in @xmath190 being singular . we can then perform a unitary transformation in order to eliminate the degrees of freedom associated to @xmath193 , and obtain an effective @xmath48 matrix ( @xmath194 ) with reduced size for which @xmath195 . the second possibility is to set @xmath196 , so that by definition we have @xmath197 . the accuracy obtained with both strategies is similar , the advantage of using the first however is that the size of the matrices is reduced , so that for big systems the computation is much faster . in our implementation we use both methods together , first we reduce the size of the system by setting @xmath192 , and then , if necessary , we further reduce the condition number for the effective system by limiting the smallest singular value . here we set all the @xmath198 singular values @xmath180 smaller than @xmath199 to zero , so that there are @xmath200 singular values @xmath180 with @xmath201 . the transformations needed in order to obtain the right se are now presented ( the procedure for the left se is analogous ) . we apply the unitary transformation @xmath202 , @xmath203 , and we define @xmath204 , @xmath205 , @xmath206 . since @xmath198 singular values of @xmath190 are zero the transformed matrices have the structure @xmath207 where the dimensions of the new matrices are : @xmath208 for @xmath209 , @xmath210 and @xmath120 , @xmath211 for @xmath212 and @xmath99 , @xmath213 for @xmath214 and @xmath215 , and @xmath216 for @xmath217 . finally @xmath218 is a column vector of dimension @xmath219 , and @xmath220 is of dimension @xmath198 . the transformed form of eq . ( [ eq : scheq ] ) is @xmath221 due to the structure of @xmath222 there are @xmath198 solutions to this equation with @xmath97 and @xmath223 . we therefore split up the right - going states into those with finite @xmath224 and those with @xmath97 . for the first set , from eq . ( [ eq : scheqv ] ) , we obtain @xmath225 with @xmath226 the @xmath218 are then solutions of an effective system with reduced size @xmath227 where the effective matrices are @xmath228 we can now solve the quadratic eigenvalue problem ( eq . ( [ eq : qep1 ] ) ) for this effective system to get the set of @xmath219 eigenvectors @xmath229 and eigenvalues @xmath171 for the right - going states . the @xmath198 eigenvectors of the second set of solutions with @xmath97 are given by @xmath223 with a general @xmath220 . the set of eigenvectors of the full @xmath230 matrix therefore is @xmath231 with @xmath232 , and @xmath233 is a general matrix of solution vectors for the states with @xmath97 . from this we obtain the set of duals @xmath234 using these results we can now calculate the transfer matrix @xmath235 of the transformed system @xmath236 where we have also used the fact that @xmath97 for the second set of solutions . we note that setting the @xmath198 smallest singular values @xmath180 to zero causes the last @xmath198 columns of @xmath235 to be zero too . moreover the explicit calculation of @xmath233 is not needed in order to obtain @xmath235 . from this and eq . ( [ eq : sener ] ) we obtain the right se @xmath237 where @xmath238 is the se of the effective system . of the diagonal elements of the transformed right se for different values of @xmath239.,width=245 ] the structure of @xmath240 shows that by applying this unitary transformation we have ordered the elements of the se by absolute size , moving those columns ( rows ) with the smallest values to the right ( bottom ) . by setting the smallest singular values of @xmath48 to zero those columns and rows of the se with small values have also been set to zero . this is illustrated in figure [ fig : sigmadiag ] , where the absolute value of the diagonal elements of the transformed selfenergy @xmath241 is shown for a ( 8,0 ) zigzag carbon nanotube at the fermi energy @xmath242 ( see section [ sec : erranalysis ] for a detailed description of the system ) . the @xmath241 are basically identical for different @xmath189 up to @xmath243 , and indeed an increasing value of @xmath189 results in more diagonal elements of @xmath240 set to zero . we note that @xmath219 is of similar size as @xmath8 in figure [ fig : sigmadiag ] , since the system is rather short along @xmath2 and a small basis set is used ( i.e. @xmath8 is small ) . for large systems and rich basis sets the ratio @xmath244 will decrease . the physical interpretation of the zero columns and rows in the se is that the @xmath198 states with @xmath245 decay infinitely fast , so that the interaction of those states is limited to the site they are localized at . finally the se of the original system can be obtained by applying the inverse unitary transformation @xmath246 and in contrast to @xmath247 the matrix @xmath164 is a dense @xmath7 matrix . note that in order to obtain the left se we perform the unitary transformation @xmath248 , @xmath249 , and then follow an analogous procedure . in this case however instead of the right - going states the left - going ones are used . we can limit the lower bound of the singular values @xmath180 by setting @xmath250 in eq . ( [ eq : ssvd ] ) . in this case the approximated @xmath251 matrix is obtained by replacing @xmath48 with @xmath190 . the error introduced is now of the order of @xmath252 . ideally @xmath252 should be of the order of the machine numerical precision , so that the error is minimal . however sometimes increasing @xmath252 beyond that value improves the results , therefore @xmath189 is left as a parameter to adjust depending on the material system investigated . this will be discussed extensively in the next section . a simpler but equally effective possibility for limiting the smallest singular value of a matrix is that of adding a small random perturbation.@xcite thus another strategy for reducing the condition number of @xmath48 is that of replacing @xmath253 with @xmath254 , where @xmath255 is a matrix whose elements are random complex numbers with an average absolute value @xmath256 . in particular we choose the @xmath257 in such a way that both @xmath258 and @xmath259 are random numbers in the range @xmath260 $ ] . we find that if @xmath261 the addition of noise usually gives results as accurate as those obtained with the svd procedure , but the calculation is faster since instead of performing a svd we just perform a sum of the matrices . in figure [ fig : diagram ] we present our final extended algorithm as it has been implemented in _ smeagol_. this now includes the following regularization procedure of @xmath48 . first the size of @xmath48 and hence of the whole problem is reduced by using the scheme described in section [ sec : svdreduce ] , with a tolerance parameter @xmath262 . this generates an effective matrix @xmath263 whose condition number @xmath264 is reduced by adding a small noise matrix @xmath265 . such a step is extremely fast and enhances considerably the numerical stability of the calculation . in most cases the se for the effective system can then be calculated and no further regularization steps are needed . however , in some cases the calculation of the se still fails . this , for example , happens when the solution of eq . ( [ eq : qep1 ] ) for the effective system fails , or else when the calculated number of left - going states erroneously differs from the number of right - going states . in these critical situations we further decrease @xmath264 by limiting the smallest singular value of @xmath263 as described in section [ sec : svdlimit ] with a tolerance parameter @xmath266 . the code automatically adjusts @xmath267 , @xmath268 and @xmath269 within a given range until the se is calculated . in our test calculations for a number of different systems we found no situation where such a scheme has failed . in contrast when the standard algorithm of reference [ ] is employed the number of failures was considerable . note that our extended algorithm can also be used in conjunction with recursive methods for evaluating the se.@xcite also in this case it will decrease the computing time for large systems due to the reduced size of the effective @xmath251 matrix . when recursive algorithms are used the accuracy of the se is automatically known as it coincides with the convergence criterion . poor convergence is found when the error can not be reduced below a given tolerance . direct methods , as the one presented here , are in principle error free in the sense that when the solution is found , this is in principle exact . for this reason the numerical errors arising from direct schemes usually are not estimated . in this section we perform this estimate and present a detailed error analysis for three different material systems . in order to estimate the numerical accuracy we use the recursive relations of eqs . ( [ eq : rgfl ] ) and ( [ eq : rgfr ] ) , written as @xmath270^{-1 } k_{1}\\ \sigma_\mathrm{r}^{\mathrm{out } } & = - k_1 \left[k_0 + \sigma_\mathrm{r}^\mathrm{in } \right]^{-1 } k_{-1 } , \end{split } \label{eq : sigmaout}\ ] ] where @xmath271 are calculated with our extended algorithm , and @xmath272 are obtained by evaluating the right - hand side term of the above equations . when the solution is exact then @xmath273 and @xmath274 . therefore we can define a measure of the error @xmath275 as @xmath276 where @xmath277 stands for the max norm,@xcite the corresponding relative error is @xmath278 . the accuracy criterion used in the extended algorithm is the following . we first set @xmath267 , @xmath269 and eventually @xmath268 and compute @xmath279 . this should be lower than a target accuracy @xmath280 . if this is not the case then the se will be recalculated with a different set of tolerance parameters , until @xmath279 reaches the desired accuracy . if this condition is never achieved the final se is the one with to the smallest @xmath279 . we now calculate the se for different variations of the method , chosen in order to highlight the problems arising from @xmath48 and @xmath95 and to show the difference between the basic method of reference [ ] and the extensions presented here . there are two main differences between the two methods . the first is that here we solve eq . ( [ eq : qep1 ] ) without inverting @xmath48 , whereas in reference [ ] @xmath281 is used to solve the inverse band - structure relation @xmath282 . clearly this second choice is less accurate if @xmath48 is close to singular . however it is much faster computationally , so that it might be of advantage for big systems . the second difference is that here it is not necessary to calculate @xmath107 via eq . ( [ eq : g00a ] ) , so that one does not need to invert @xmath87 and @xmath88 . in order to investigate the effect of these two aspects independently , we have calculated the se using the following four methods . in method 1 we use the algorithm presented in this work . in particular we use eq . ( [ eq : qep1 ] ) to solve the quadratic eigenvalue problem and eqs . ( [ eq : senel]-[eq : sener ] ) to obtain the se ( for the right se we actually use a different form of eq . ( [ eq : qep1 ] ) , see appendix c ) . method 2 is essentially the same , with the only difference that instead of solving eq . ( [ eq : qep1 ] ) we use the eigenvalue method of reference [ ] . in method 3 we solve eq . ( [ eq : qep1 ] ) , but we use eq . ( [ eq : sgfstefano ] ) to calculate the sgf , with @xmath107 obtained from eq . ( [ eq : g00a ] ) . finally method 4 is the algorithm of reference [ ] . in order to obtain a statistically significant average of the errors , we plot a histogram of the calculated errors for both @xmath163 and @xmath164 for a large energy range . here we use the absolute error , since it can readily be compared to the energy scale of the problem . note that although the relative error might be small , the absolute error can be very large if @xmath283 ry . furthermore in order to keep the analysis simple in all the calculations of this section we do not reduce the system size nor do we add noise ( @xmath284 ) . we regularize @xmath48 and @xmath95 by using @xmath196 in eq . ( [ eq : ssvd ] ) . since the error depends on the chosen @xmath189 , here we calculate @xmath285 for a set of @xmath189 in the range @xmath286 $ ] . we then present the smallest @xmath275 found for @xmath189 taken in that range . this is the smallest possible error achievable with a given method and allows us to extract informations on the range of optimal svd values for a given method . , i.e. the direction of the transport.,width=264 ] for three different systems . ( a ) ( 8,0 ) zigzag carbon nanotube , ( b ) bcc fe , ( c ) fcc au . @xmath8 is the number of times a given error @xmath275 occurs ( not normalized).,width=302 ] giving the smallest error in the self - energy . ( a ) ( 8,0 ) zigzag carbon nanotube , ( b ) bcc fe , ( c ) fcc au . @xmath8 is the number of times a given @xmath189 generates the smallest error ( not normalized).,width=302 ] as first example a ( 8,0 ) zigzag carbon nanotube@xcite is presented ( the unit cell is shown in figure [ fig : ucellc](a ) ) . the length of the periodic unit cell is 4.26 along the nanotube , with 32 carbon atoms in the unit cell . the lda approximation ( no spin - polarization ) is used for the exchange correlation potential . we consider @xmath287 and @xmath288 orbitals for carbon with double-@xmath289 and a cutoff radius @xmath290 for the first @xmath289 of @xmath291 bohr . higher @xmath289 are constructed with the split - norm scheme with a split - norm of 15%.@xcite the real space mesh cutoff is 200 ry . the matrices @xmath292 , @xmath9 , @xmath293 and @xmath294 are extracted from a ground state dft calculation for an infinite periodic nanotube . we calculate the se for the semi - infinite nanotube at 1024 energy points in a range of @xmath295 ev around the fermi energy . figure [ fig : deltaall](a ) shows the histogram of the errors in the se , where @xmath8 is the number of times a given error @xmath275 appears . in general the figure shows that for this system the average error increases when going from method 1 to method 2 and method 3 , and finally to method 4 . the error obtained with method 1 is on average about 6 orders of magnitude smaller than the one obtained with method 4 . the main reason behind this dramatically improved accuracy is that method 1 does not involve any steps where a singular @xmath48 leads to divergencies . method 4 on the other hand is strongly dependent on the condition number of @xmath48 , since it necessitates to invert @xmath48 and @xmath87 ( or @xmath88 ) . methods 2 and 3 are on average about one order of magnitude more precise than method 4 . since they both still involve one of the two inversions the difference is however not large . figure [ fig : fsall](a ) shows the histogram of the optimum @xmath296 used for the calculations of the se . here we plot the number of times @xmath8 a particular @xmath189 has given the smallest error in the set of calculations . a larger optimal value for @xmath296 indicates a stronger dependence of the computational scheme on @xmath174 . for method 1 the range of used @xmath296 is smaller than @xmath297 . if we force @xmath296 to be zero we get almost the same level of accuracy as shown in figure [ fig : deltaall](a ) , which confirms that the accuracy of for method 1 depends little on @xmath174 for this system . however also for this method there is a set of energies ( a few percent of the total number ) where the solution of eq . ( [ eq : qep1 ] ) fails if @xmath296 is too small . the optimal @xmath296 for the other methods is orders of magnitude larger than that of method 1 , and it is never smaller than @xmath298 . the absolute error induced by replacing @xmath48 by @xmath190 is of the order of @xmath299 . usually @xmath182 is of the order of 1 ry , so that the error is of the order of @xmath189 ry . therefore since in methods 2 to 4 a large value of @xmath296 is needed in order to improve @xmath300 , also the resulting error is large . the second example is bcc fe ( figure [ fig : ucellc](b ) ) , oriented along the ( 100 ) direction . the lattice parameters are the same as in reference [ ] . there are 4 fe atoms in the unit cell . we apply periodic boundary conditions in the direction perpendicular to the stacking , so that these correspond to 4 fe planes . the length of the cell along the stacking direction is 5.732 . a double-@xmath289 @xmath301 ( @xmath290=5.6 bohr ) , single-@xmath289 @xmath302 ( @xmath290=5.6 bohr ) and single-@xmath289 @xmath303 ( @xmath290=5.2 bohr ) basis is used . the real space mesh cutoff is 600 ry , and the dft calculation is converged for 7x7 @xmath23-points in the brillouin zone orthogonal to the stacking . the se have been calculated for the converged dft calculation at 32 different energies in a range of @xmath3041 ev around the fermi energy , and for 10,000 @xmath23-points in the 2d brillouin zone perpendicular to the stacking direction . for each @xmath23-point there is a different set of matrices @xmath305 , @xmath48 and @xmath95 , so that for each @xmath23-point there is a different se . the histogram for the error of the calculated self - energy @xmath275 is shown in figure [ fig : deltaall](b ) , and the histogram for the optimal @xmath296 in figure [ fig : fsall](b ) . the general behavior is similar to the one found for the carbon nanotube . we note that , although for the vast majority of the calculations the error in the se is small , there is a long tail in the histograms of figure [ fig : deltaall](b ) indicating the presence of a small number of large errors . this is present for all the methods , with a maximum error of @xmath306 ry for method 1 , and @xmath307 ry for method 4 . closer inspection shows that the reason for the increase of the error for certain energies and @xmath23-points is caused by a divergence in @xmath308 . this will be illustrated in more detail in the next section . finally we consider fcc au ( figure [ fig : ucellc](c ) ) , with the stacking along the ( 111 ) direction . the unit cell consists of three planes of nine gold atoms each . these are the typical leads used for the calculations of the transmission properties of molecules attached to gold.@xcite we use double-@xmath289 @xmath301 ( @xmath290=6.0 bohr ) and single-@xmath289 @xmath303 ( @xmath290=5.5 bohr ) and four @xmath23-points in the brillouin zone perpendicular to the stacking . the mesh cutoff is 400 ry . the se have been calculated for 418 energy points , from about 15 ev below to about 10 ev above the fermi energy . the general behavior ( figures [ fig : deltaall](c ) and [ fig : fsall](c ) ) is again similar to that of the previous examples . also here the error for method 1 is about 6 orders of magnitude smaller than that of method 4 , with method 2 and 3 giving some marginal improvement . our results show that the new scheme in general allows the calculation of the se with high accuracy . the main advantage of method 1 is rooted in the possibility of using a much smaller @xmath296 . for big systems sometimes one might prefer to use method 2 , since it is considerably faster than method 1 and gives the second best accuracy . in this case we first calculate the se with method 2 and check the error . only for those energy points where the error is above some maximum value ( of the order of @xmath309 ry for example ) the calculation is repeated with method 1 to improve the accuracy . finally the results show that for all methods the svd transformation of @xmath48 is necessary , although for method 1 it is needed only a few percent of the times . for big systems , in particular if the unit cell is elongated along the stacking direction , or if a rich basis set is used , @xmath174 will generally increase as there will be some singular values of @xmath48 going to zero . in these cases also method 1 will require a svd transformation for most energies . the range of @xmath296 should however be similar to the one shown in figure [ fig : fsall ] , so that also the error in the se should be of the same order of magnitude . we also note that in order to keep the analysis simpler here we have not used the reduction of system size described in section [ sec : svdreduce ] , for such large systems it is however crucial in order to decrease the computational effort and regularize @xmath48 at the same time . the center of the error distribution for method 1 ( figure [ fig : deltaall ] ) is located at small @xmath310 , usually smaller than 10@xmath311 ry . however the histogram has also a tail reaching up to very large errors . these are found only at some critical energies as demonstrated in figure [ fig : ev_gf](a ) , where we show @xmath275 for the carbon nanotube calculated over 1024 energy points in a range of 2 ev around the fermi energy . the average error is of the order of 10@xmath312 ry , but at energies around -0.8 ev and -0.34 ev the error drastically increases . indeed a finer energy mesh at these points suggests a divergence . the origin of the large errors at particular energies can be investigated by looking at the eigenvalues @xmath313 of the sgf @xmath314 . in figure [ fig : ev_gf](b ) the largest and the smallest absolute value for the eigenvalues , respectively @xmath315 and @xmath316 , are plotted as function of energy ( @xmath317 ) . it can be seen that @xmath315 diverges close to the energies where the error increases , i.e. we can associate large errors in @xmath165 with a divergence in its spectrum . since @xmath318 is calculated from eq . ( [ eq : senel ] ) the only possible origin for the divergence is in the norm of some of the @xmath75 . as these are obtained by inverting the matrix @xmath319 ( eq . ( [ eq : q ] ) ) , one deduces that the set of vectors @xmath320 is not linearly independent . for these energies we therefore can simply check the magnitude of @xmath322 to determine whether there is a divergence of the se close to a particular energy . physically the divergence of the se translates into the presence of a surface state at that particular energy.@xcite consider the spectral representation of @xmath165 @xmath323 where @xmath324 are the eigenvalues and @xmath325 are the right eigenvectors of the effective surface hamiltonian matrix @xmath326 with overlap @xmath293 , and @xmath327 are the left eigenvectors of the same hamiltonian . a localized surface state is found when there is a real eigenvalue @xmath328 at @xmath329 ( or more generally if im(@xmath328 ) is very small ) . from the recursive relation ( [ eq : rgfl ] ) one can deduce that for an infinite eigenvalue there is also a corresponding vanishing eigenvalue . therefore in figure [ fig : ev_gf](b ) for energies where @xmath330 we have also @xmath331 . close to the singularity we can therefore expand the two eigenvalues as @xmath332 and @xmath333 . for @xmath334 the largest eigenvalue in eq . ( [ eq : gldiagonal ] ) is then equal to @xmath335 , and the smallest is equal to @xmath336 . to avoid divergence therefore the magnitude of the @xmath165 eigenvalues can be bounded to a finite value @xmath335 by introducing a small imaginary part to the energy for energies in the vicinity of a surface state . another possibility for limiting the size of @xmath315 is to bound the singular values of @xmath80 from below in the same way as it is done for @xmath48 ( section [ sec : svdlimit ] ) . this essentially imposes the @xmath45 to be linearly independent from each other . however , with this scheme it is not possible to conserve the green s function causality , so that the sgf might have eigenvalues lying on the positive imaginary axis . moreover we loose control over the accuracy of the computed sgf and se . both these problems are avoided when using a finite @xmath336 . error analysis for the carbon nanotube of figure [ fig : ucellc ] : ( a ) absolute error @xmath275 of the self - energy as function of the energy @xmath24 , ( b ) maximum ( @xmath315 ) and minimum ( @xmath316 ) eigenvalues of @xmath165.,width=264 ] density of states and transmission coefficient for the carbon nanotube of figure [ fig : ucellc ] . ( a ) density of states of the surface layer @xmath337 as function of energy @xmath24 , calculated for different broadenings @xmath336 . the inset is a zoom at energies around -0.34 ev . ( b ) transmission coefficient @xmath338 for different values of @xmath336.,width=264 ] we now investigate the dos and transport properties of a system when the finite imaginary part @xmath336 ( broadening ) is added to the energy . we consider as an example the carbon nanotube of figure [ fig : ucellc ] . in figure [ fig : trho0](a ) the onsite surface dos @xmath337 as defined in eq . ( [ eq : onsitedos ] ) is shown for @xmath339 ry , @xmath340 ry , @xmath341 ry and @xmath342 ry . for @xmath339 the surface dos vanishes for energies between -0.37 ev to + 0.45 ev , indicating the presence of a gap around the fermi energy . note that there are no van hove singularities in @xmath337 , since we never divide by the group velocity when calculating the sgf . for finite @xmath336 and energies away from the band gap , the dos is essentially identical to that calculated for @xmath339 , however inside the gap @xmath337 does not vanish but saturates to a small value proportional to @xmath335 . moreover whereas the surface states are not visible for @xmath339 , they appear in the dos for finite @xmath336 , with a full width at half maximum ( fwhm ) equal to @xmath343 . we then move to the transport by calculating the transmission coefficient@xcite @xmath344 for a carbon nanotube attached to semi - infinite leads made from an identical carbon nanotube . since this is a periodic system @xmath344 must equal the number of open channels , so that it can only have integer values . this is indeed the case for @xmath339 ( figure [ fig : trho0](b ) ) . for finite @xmath336s the transmission coefficient is only approximately an integer , especially inside the energy gap region where the finite surface dos introduced by @xmath336 leads to a non zero transmission . the transmission in the gap is proportional to @xmath345 ( note that the scale is logarithmic ) , since on both sides of the scattering region the artificial surface dos is proportional to @xmath336 . in this region of small transmission therefore the results might change by orders of magnitude depending on the value of @xmath336 . for all values of @xmath336 however we find no contributions to the transmission coming from the surface state , indicating that these do not carry current . these results show that adding a finite value @xmath336 to the energy has little effect on the actual transmission if this is large . however when the transmission is small , as in the case of tunnel junctions , the finite @xmath336 introduces an additional contribution to the conduction that might arbitrarily affect the results . it is thus imperative for those systems to identify surface states and use the imaginary @xmath336 only in a narrow energy interval around them . finally we can give an estimate of the relative accuracy @xmath346 at the energy corresponding to the surface state . as discussed before the origin of the error is the inversion of @xmath80 needed to calculate the duals . the relative error introduced by the inversion of @xmath80 is proportional to @xmath322.@xcite close to a surface state the smallest singular value is of the order of @xmath336 , so that @xmath347 . as this is the dominant source of error in the calculation of the se close to a surface state , we can approximate the relative error as @xmath348 where @xmath349 is a constant that depends on the machine precision and on the details of the algorithm . the label `` in '' explicitly indicates that this is the error in the se calculated with the extended algorithm ( @xmath350 in eq . ( [ eq : sigmaout ] ) ) . the absolute error @xmath351 is equal to the relative error times @xmath352 , which is itself proportional to @xmath335 , so that we get @xmath353 . when using eq . ( [ eq : error ] ) to estimate the error in the se we introduce an additional error due to the inversion involved in obtaining @xmath165 . the largest singular value of @xmath165 is proportional to @xmath335 , and the smallest one is proportional to @xmath336 , so that the relative error introduced by the inversion is proportional to @xmath354 . for small @xmath336 we can therefore write for the error in @xmath355 @xmath356 where @xmath357 is again a constant . since the errors are random the total estimated error can be approximated by adding the contributions from the two inversions @xmath358 @xmath279 is therefore a good estimate for the true error @xmath359 if @xmath360 is small . close to surface states however @xmath361 , so that @xmath279 largely overestimates the true error . ( a ) relative error of the self - energy ( @xmath362 represents the true error ) , ( b ) condition numbers of @xmath80 and @xmath165 , as a function of the broadening @xmath336 for the carbon nanotube of figure [ fig : ucellc ] calculated at the surface state energy.,width=264 ] to verify these estimates numerically we present a scheme for calculating @xmath359 and @xmath360 independently . for each se we perform a second calculation where we add a small amount of noise to the input matrices @xmath363 , and @xmath95 , so that we obtain the self - energy @xmath364 for a slightly perturbed system . the noise is added as a random relative perturbation of each element of the matrices . as we decrease the magnitude of the noise the difference between @xmath163 and @xmath364 is reduced until it becomes constant for noise smaller than a critical value . in this range of minimum noise even if the difference in the input matrices decreases , the difference in the output matrices is constant , it therefore corresponds to the error in the calculation . as one might expect we find that this critical value of noise is of the same order of magnitude as the numerical accuracy used ( approximately @xmath365 in our calculations ) . we can therefore obtain @xmath366 and @xmath367 , with the magnitude of the noise equal to the critical value . we have calculated the maximum error for a set of 128 energy points located within @xmath368 ry around the energy of the surface state at -0.34 ev for different values of @xmath336 . the result is shown in figure [ fig : errss](a ) . indeed for small @xmath336 @xmath362 follows eq . ( [ eq : dsin ] ) with @xmath369 ry , @xmath370 follows eq . ( [ eq : dsout ] ) with @xmath371 ry@xmath372 , and @xmath373 . in figure [ fig : errss](b ) the condition numbers @xmath322 and @xmath374 are shown , confirming @xmath347 and @xmath375 . this demonstrates that close to surface states @xmath279 is mainly caused by the calculation of @xmath165 . thus @xmath279 largely overestimates the real error @xmath362 , which even for @xmath376 ry has an acceptable size of @xmath377 . since @xmath349 and @xmath357 are generally system dependent , in practical calculations we use a value of @xmath336 ranging between @xmath378 ry and @xmath379 ry for energies in the vicinity of surface states , mainly in order to limit the absolute error . moreover @xmath336 is added in an energy range corresponding approximately to the fwhm of the imaginary part of @xmath380 , which is equal to @xmath343 . although this range is only of the order of @xmath381 ry , in practical calculations where both energy and @xmath23-point sampling is fine the number of times when this prescription is applied can be rather large ( see figure [ fig : deltaall ] ) . the above analysis confirms that close to surface states also direct methods have the same accuracy problems of recursive methods . this fact is usually ignored in the literature,@xcite where it is assumed that the accuracy is constant for a given algorithm . here we show that the accuracy of a method is solely determined by the value of @xmath349 , which , as indicated in section [ sec : erranalysis ] , can vary over many orders of magnitude . our analysis also shows that methods requiring the explicit calculation of @xmath165 from its inverse are much less accurate close to surface states than those calculating @xmath163 directly . by extending the scheme proposed in reference [ ] we have presented a different but equivalent form for calculating the green s functions of an infinite quasi-1d system , as well as the sgf and se for the semi - infinite system . we have then constructed an extended algorithm containing also the necessary steps to regularize the ill conditioned hopping matrices . this is found to be crucial in order to obtain a numerically stable algorithm . by applying a unitary transformation based on a svd we remove the rapidly decaying states and calculate the se for an effective system with reduced size . we further decrease the condition number of the hopping matrices by adding a small random perturbation and by limiting the smallest singular value . we have performed a detailed error analysis on the numerical calculation of the se , showing that if the algorithm does not involve an inversion of the hopping matrices @xmath48 ( or @xmath95 ) high accuracy is obtained . we also find that the error is not constant as function of energy . it is shown that an increase of accuracy is needed especially close to energies where the se and sgf diverge , which corresponds to the presence of surface states in the semi - infinite system . at these energies we improved the accuracy by adding a small imaginary part to the energy . we have shown that this procedure affects the transport properties little in the high transmission limit . however , for low transmission this adds some spurious surface density of states contributing significantly to the total transmission . the transport can therefore be strongly affected , so that the imaginary part should be added only in a small energy range around the poles and it should be as small as possible . our final algorithm is therefore highly numerically stable and extremely accurate . most importantly errors and accuracy can be closely monitored . we believe that this is an ideal algorithm to be used with _ ab initio _ transport schemes , where the condition of the hamiltonian and its sparsity is controlled by the convergence of the electronic structure and therefore can not be fixed _ a priori_. here we demonstrate that @xmath163 calculated using eq . ( [ eq : senel ] ) indeed fulfills the recursive relation for @xmath165 of eq . ( [ eq : rgfl ] ) . insert eqs . ( [ eq : gfs ] ) and ( [ eq : senel ] ) into eq . ( [ eq : rgfl ] ) and take the inverse to obtain @xmath382 using the definition of the matrix @xmath88 ( eq . ( [ eq : bara ] ) ) we write @xmath383 this equation corresponds to the defining equation for the @xmath78 and is therefore fulfilled by definition . the same is therefore true for eq . ( [ eq : rgfl ] ) . ( [ eq : rgfr ] ) for @xmath166 can be demonstrated similarly . in section [ sec : svdreduce ] we assume that @xmath60 in order to write the transformed matrices @xmath385 and @xmath222 in form of eq . ( [ eq : transfmat ] ) . if @xmath384 the same can be done by performing a generalized svd of the hamiltonian and overlap matrices as described in reference [ ] . here we present a different approach , based on two standard svd transformations , one for @xmath48 and one for @xmath386 @xmath387 here @xmath388 and @xmath389 are unitary matrices , @xmath390 and @xmath391 are diagonal matrices with the singular values on the diagonal . in general there are @xmath392 singular values of @xmath48 smaller than @xmath393 , and @xmath394 singular values of @xmath95 smaller than @xmath395 , with @xmath396 and @xmath397 being respectively the largest singular value of @xmath48 and @xmath95 . if @xmath398 , we obtain @xmath190 by setting the smallest @xmath198 singular values of @xmath48 to zero . in the same way we obtain @xmath399 by setting the smallest @xmath198 singular values of @xmath95 to zero . a transformation @xmath400 brings both @xmath385 and @xmath222 to the form of eq . ( [ eq : transfmat ] ) . all the results of section [ sec : svdreduce ] are then valid also for @xmath384 . if the hamiltonian and overlap matrices are real and hermitian , but the energy is complex , then @xmath401 . by using eq . ( [ eq : k1m1svd ] ) , and the fact that @xmath390 and @xmath391 are real , we obtain @xmath402 , so that @xmath403 . if the hamiltonian and overlap matrices are hermitian but not real , then in general @xmath404 . however in all the calculations performed the difference between @xmath390 and @xmath405 was very small , so that in practice we always had @xmath406 . in section [ sec : svdlimit ] we limit the singular values of @xmath48 from below without reducing the size of the system . if @xmath384 we simply apply the transformations described in section [ sec : svdlimit ] to both @xmath48 and @xmath95 independently . we find that in the solution of eq . ( [ eq : qep1 ] ) the numerical accuracy for those eigenvalues with @xmath407 ( im@xmath408 ) is better than for those with @xmath409 ( im@xmath410 ) , especially when @xmath411 . for @xmath163 we only need the left - going states , for which eq . ( [ eq : qep1 ] ) gives the better accuracy . for @xmath164 the right - going states are needed . in this case , in order to increase the accuracy for the right decaying states ( im@xmath410 ) , instead of eq . ( [ eq : qep1 ] ) we solve the equivalent equation @xmath412 with @xmath413 the eigenvalues of the states with im@xmath410 now have an absolute value larger than one and therefore a higher accuracy .

the quantum mechanical properties of the vacuum state in quantum field theories are presently attracting the attention of researchers from very different communities , from astrophysics and gravitation , to quantum optics and condensed matter physics . a central idea of all these studies is the possibility of converting zero - point quantum fluctuations into observable quantum vacuum radiation by some spatial and/or temporal dependence of the background over which the quantum field is propagating @xcite . a number of fundamental physical effects belong to this category , from the dynamical casimir effect when the boundary conditions of the quantum field are rapidly varied in time @xcite , to the hawking radiation when it propagates on a curved space - time showing a black hole horizon @xcite . in the wake of the pioneering work by unruh @xcite , researchers have started investigating condensed - matter systems where the propagation of some low - energy excitation field follows an effectively curved space - time geometry . in suitable configurations showing a black hole horizon for this low - energy excitation , a quantum vacuum emission is expected to appear via a mechanism analogous to hawking radiation . among the many systems that have been investigated in this perspective @xcite , low - temperature superfluids and nonlinear optical systems are nowadays considered the most promising ones . dilute superfluids such as bose - einstein condensates of ultracold atoms join a very simple spectrum of elementary excitations with sonic dispersion , with the possibility of pushing the sensitivity of measurements down to the quantum limit . their potential as analog models was first proposed in refs . and later confirmed by _ ab initio _ numerical simulations of the condensate dynamics @xcite . even if acoustic black hole configurations have been experimentally realized @xcite , no evidence of hawking radiation has been reported yet . analog models based on nonlinear optical systems were pioneered in ref . : a strong pulse of light propagating in a kerr nonlinear medium can be used to generate a moving spatial interface separating regions of different light velocity . for suitably chosen parameters , the interface then behaves as a horizon that can be crossed by light in one direction only , which should result in hawking radiation being emitted . some experimental claim in this direction was recently reported @xcite and has raised a number of interesting questions concerning the interpretation of the observed radiation in terms of hawking processes @xcite . a completely new perspective to analog models based on nonlinear optical systems was opened by marino @xcite , who first proposed the use of quantum fluids of light to generate acoustic black hole configurations and , then , to look for the hawking radiation of bogoliubov phonons on top of the photon fluid . in the following years , this idea has been pushed forward by a number of authors , who have considered different geometrical configurations @xcite and material systems , and very recently exciton - polariton fluids in semiconductor microcavities @xcite . in the present work , we report a comprehensive study of analog hawking radiation effects in quantum fluid of light , specifically exploiting the quantum fluid properties of an exciton - polaritons condensate . our theoretical model fully includes their intrinsically non - equilibrium nature @xcite and describes quantum fluctuations of the polariton field within the so - called truncated - wigner formalism of degenerate quantum gases @xcite . as a result , our numerical calculations are able to provide quantitative predictions for the observable quantities , and to point out clear and accessible signatures of analog hawking radiation in the emitted light from the cavity . even if our discussion is mostly focused on the specific case of semiconductor microcavity devices @xcite where polariton superfluidity has been first demonstrated @xcite , all our conclusions straightforwardly extend to generic planar cavity devices filled with or made of a kerr nonlinear medium . the article is organized as follows . in sec . [ theory ] we introduce the physical system under consideration and we summarize the theoretical tools that are used to describe it . in sec . [ horizon ] we describe the original laser beam configuration that we propose to generate an analog black hole horizon with a large surface gravity . the observable consequences of hawking mode - conversion processes are illustrated in the following sections . in sec . [ scattering ] we discuss the scattering of a coherent phonon wavepacket on the horizon : the signature of classical hawking processes is visible as an additional wavepacket emerging from the horizon on a negative - norm branch . in sec . [ hawking ] , we present the numerical evidence of the hawking radiation due to the conversion of zero - point quantum fluctuations into observable radiation , and we compare it with the theoretical expectations . conclusions are finally drawn in sec . [ concl ] . we consider a polariton wire device where cavity photons propagating along one dimension are resonantly coupled to the fundamental exciton transition in one or more quantum wells embedded in the cavity layer ( in - plane polarized heavy - hole to conduction band transition ) . starting from a planar microcavity where light is confined along the growth axis @xmath0 by a pair of distributed bragg reflectors , a suitable etching procedure is used to laterally pattern the device in the @xmath1 direction , and create the ridge structure that is schematically shown in fig . [ fig1](a ) . provided the losses are weak enough , the bosonic excitations that result from the strong coupling of the cavity photon with the quantum well exciton have the typical mixed light - matter nature of one - dimensional exciton - polaritons , free to propagate along the wire axis according to the dispersion law that is plotted in fig . [ fig1](b ) . a high quality experimental realization of this polariton wire concept was reported in ref . , from which we shall take the state - of - the art system parameters to be used in the numerical simulations . along the wire axis . the thin dashed and dash - dotted lines indicate the bare cavity photon and exciton dispersions , respectively . the thick dashed line indicates the parabolic approximation of the lower polariton dispersion . system parameters are inspired from ref . : @xmath2 mev , @xmath3 mev , @xmath4 ( with @xmath5 the free electron mass ) , @xmath6 . , title="fig:",scaledwidth=45.0% ] + along the wire axis . the thin dashed and dash - dotted lines indicate the bare cavity photon and exciton dispersions , respectively . the thick dashed line indicates the parabolic approximation of the lower polariton dispersion . system parameters are inspired from ref . : @xmath2 mev , @xmath3 mev , @xmath4 ( with @xmath5 the free electron mass ) , @xmath6 . , title="fig:",scaledwidth=48.0% ] + a theoretical description of the dynamics of this system can be developed in terms of the standard hamiltonian for the two coupled bosonic fields describing the quantum well exciton and the cavity photon @xcite @xmath7 where @xmath8 is the longitudinal coordinate along the wire axis and the two - component operator vector @xmath9 summarizes the quantum fields describing the exciton @xmath10 and photon @xmath11 fields . each of them satisfies one - dimensional bosonic commutation rules , e.g. @xmath12=\delta_{i , j}\delta(x - x')$ ] , with @xmath13 . throughout the whole paper , we restrict ourselves to the case where the system is pumped on a single spin state , so that the spin degrees of freedom can be neglected in the theoretical model . the single - particle hamiltonian @xmath14 describing the evolution of the non - interacting exciton and cavity photon fields has the simple representation @xmath15 in terms of the bare cavity photon and exciton dispersion law @xmath16 in a spatially homogeneous one - dimensional system . the rest energy @xmath17 and the effective mass @xmath18 of the cavity photon are determined by the spatial confinement along the @xmath0 and @xmath1 axis by the dbr mirrors and by the total internal reflection at the etching interfaces , respectively . in practical calculations the exciton dispersion can be safely neglected , @xmath19 , as the exciton mass @xmath20 is orders of magnitude larger than the photon mass . the strength of the light - matter coupling is quantified by the rabi frequency @xmath21 , proportional to the square root of the oscillator strength per unit area of the excitonic transition @xcite . in spatially homogeneous systems , for which the external potentials acting on the cavity photon and the quantum well exciton vanish ( @xmath22 ) , diagonalization of the single particle hamiltonian @xmath14 as a function of wave vector @xmath23 leads to the well - known upper and lower polariton branches of dispersions @xmath24 and @xmath25 . as it is shown in fig . [ fig1](b ) the bottom part of the lower polariton dispersion is accurately approximated by a parabolic form @xmath26 where the @xmath27 and @xmath28 parameters have the clear physical meaning of a rest energy and an effective mass for the ( lower ) polariton . in the resonant case @xmath29 , one has the simple relation @xmath30 . several methods to generate external potentials @xmath31 with almost arbitrary spatial shapes along the @xmath32 direction have been experimentally demonstrated in the last years , from the in - plane patterning of the microcavity layer @xcite , to the all - optical potential created by a strong pump laser with opposite polarization @xcite , to the application of a mechanical stress to the device @xcite . exciton - exciton interaction , due to the coulomb interactions between the electron and holes forming the excitons , are the main source of optical nonlinearity in this problem . experimental measurements @xcite report a value on the order of @xmath33 , which translates into a reduced one - dimensional interaction parameter to be used in the hamiltonian ( [ eq : model ] ) @xmath34 , where the integral is over the transverse wire width ( @xmath35 ) , and the forth power of the exciton wave function comes from the kerr - type nature of the nonlinearity . assuming the transverse confinement of the exciton envelope function to be described by @xmath36 , a good estimate can be given to be @xmath37 . taking into account the exciton weight in the polariton , this translates in the resonant @xmath38 case into a one - dimensional polariton - polariton interaction constant @xmath39 . polaritons are coherently injected into the system by an external laser drive coupled to the cavity photon via the non - perfect reflectivity of the dbr mirrors . in the theoretical model , such processes are described by the terms in the last row of eq . ( [ eq : model ] ) , where @xmath40 is the spatio - temporal profile of the coherent pump beam including the coupling coefficient proportional to the dbr transmission amplitude . correspondingly , photons ( excitons ) are subjected to radiative losses at a rate @xmath41 ( @xmath42 ) that have to be described at the level of the master equation for the density matrix , @xmath43 , for the coupled quantum fields @xmath44+\mathcal{l}[\rho].\ ] ] performing the standard born - markov approximation @xcite , the dissipation superoperator can be cast into the lindblad form @xmath45 \ , .\end{gathered}\ ] ] the resulting decay rate of polaritons is a weighted average of the exciton and photon ones @xmath46 : in the vicinity of the bottom of the lower polariton branch of fig . [ fig1](b ) , it has an almost momentum - independent value @xmath47 . in a regime of weak exciton - exciton interactions , the dynamics of the system can be accurately captured by the mean - field approximation : the quantum fields @xmath48 are replaced by the classical fields corresponding to their expectation values @xmath49 and evolving according to the pair of nonlinear partial differential equations @xmath50 \left ( \begin{array}{c } \phi_{\mathrm{x } } ( x , t ) \\ \phi_{\mathrm{c } } ( x , t ) \end{array } \right ) \ , , \end{gathered}\ ] ] which generalize to the non - equilibrium context of polaritons the well - known gross - pitaevskii equation ( gpe ) of dilute bose condensed gases @xcite . [ sec : homog ] in the simplest case of a spatially homogeneous system under a coherent pump in a plane wave form @xmath51\ ] ] of frequency @xmath52 , wavevector @xmath53 and amplitude @xmath54 , the analog of the equation of state can be obtained by simply injecting the same plane wave ansatz @xmath55 $ ] into the generalized gpe ( [ eq : mf_gpe ] ) . @xmath56m@xmath57 , and frequency @xmath58 mev tuned slightly above the lower polariton branch at @xmath59 mev . equal loss rate are assumed for the exciton and the photon fields , @xmath60 mev . the value @xmath61 of the nonlinearity corresponds to a @xmath62 m wide polariton wire ( see text ) . lower panel : spatial intensity profile of the coherent pump along the wire axis , with excitation parameters : @xmath63 , @xmath64 , @xmath65 , @xmath66 . , scaledwidth=50.0% ] this leads to the simple form @xmath67 for the steady - state excitonic density @xmath68 as a function of the pump parameters , and to an analogous expression for the cavity - photon density @xcite . the pump detuning from the cavity photon and exciton modes at wavevector @xmath69 are indicated here as @xmath70 and @xmath71 . in the quantum fluid language , this equation plays the role of the equation of state for the polariton fluid under a coherent pump in a plane wave form : while the polariton flow speed is controlled by the incident wavevector @xmath69 , the density has a more complex dependence on the pump intensity @xmath72 and frequency @xmath52 . a most remarkable example is illustrated in fig . [ fig2]a : the choice of a pump frequency slightly above the lower polariton branch leads to an optical bistable behavior . for reasons that will soon become clear , we shall concentrate our attention onto the case where the pump intensity is tuned in the close vicinity of the end - point of the upper branch of the hysteresis loop ( indicated as @xmath73 in fig . [ fig2]a ) . in order to reach this working point in a given spatial region , a pump with a non - trivial spatial envelope , @xmath74 , multiplying the plane wave can be used , as shown in the lower panel of fig . [ fig2](b ) . upstream of the region of interest ( i.e. for @xmath75 the pump amplitude is tuned to a value @xmath76 higher than the switch - on point of the hysteresis loop ( i.e. the end point of the lower branch ) , so to switch the system to the upper branch of the hysteresis loop @xcite . in the long region of interest ( @xmath77 ) , the pump amplitude is maintained at a spatially constant value @xmath73 , so to keep the system sits at the desired working point with the desired in - plane flow wavevector @xmath69 . the behavior of the density @xmath78 and the wavevector @xmath79 in the downstream region ( @xmath80 ) is determined by a complex interplay of propagation and losses . first numerical and analytical studies of this regime have appeared in refs . and : the spatial drop of the polariton density away from the pump results in an increase of the polariton speed under the effect of the interactions . the lower the losses , the slower this acceleration effect . the dispersion of collective excitations on top of the steady state of the coherently pumped condensate is obtained as usual by linearizing the gpe around the steady state solution @xcite . a complete discussion of the many different behaviors that are possible depending on the pump frequency and intensity can be found in refs . . here , we shall restrict our attention to the two cases of present interest , namely ( i ) a coherent pump whose intensity is tuned right at the turning point of the upper branch of the hysteresis loop , and ( ii ) the ballistic flow regime in the downstream region . if the polariton population is restricted to the lower polariton branch and we perform a parabolic approximation of the branch bottom with effective mass @xmath28 , the dispersion of the collective excitations on top of a condensate of polariton density @xmath81 in uniform flow at speed @xmath82 has , in both cases , the usual bogoliubov form @xmath83 in the first ( i ) case , the local wavevector @xmath84 is fixed to the in - plane component @xmath53 of the pump wavevector . in the second ( ii ) case , hydrodynamic effects @xcite make the local wavevector @xmath85 to slowly vary in space . in both cases , the local in - plane flow speed @xmath82 is related to the local wavevector by @xmath86 , and the polariton speed is defined as @xmath87 in terms of the polariton current @xmath88 \simeq j_{\rm c}=\frac{1}{2i m_{\rm c}}[\hat{\psi}^*_{\rm c}\,\nabla { \psi}_{\rm c}- \psi_{\rm c}\,\nabla{\psi}^*_{\rm c}]$ ] . given its very heavy mass , @xmath6 , the exciton fraction does not significantly contribute to the polariton current . ] . for small wave vectors , @xmath89 , the bogoliubov dispersion tends to a ( lossy ) sonic dispersion @xmath90 with a sound speed @xmath91 of fluid excitations such that @xmath92 , and a doppler shift term due to the background fluid flow at @xmath82 . for large wavevectors , it recovers a ( lossy ) parabolic single - particle dispersion @xmath93 with a hartree energy shift @xmath94 due to the interactions with the condensate . plots of the bogoliubov dispersion at different spatial positions are shown in the small panels of figs . [ fig : nobh_scheme ] and [ fig : bh_scheme ] . in the following of the paper , the will be used to interpret the numerical results for the evolution of excitation wavepackets and for the density - density correlations . before proceeding , it is important to emphasize the conditions underlying the dispersion ( [ eq : sonic ] ) of the collective excitations . in the coherent pump case ( i ) , this form only holds at the end point of the upper branch of the hysteresis loop , while gaps and/or band sticking regions appear in all other cases because of the coherent pump locking the condensate phase @xcite . on the other hand , the presence of a soft excitation branch of sonic nature is guaranteed in the ballistic case ( ii ) by the fact that the condensate phase is completely free to evolve both in space and time @xcite . however , as the density is spatially varying , the very concept of dispersion relation holds only locally and is limited to excitations whose wavelength is small as compared to the characteristic length scale of the density profile . a standard technique to go beyond the mean - field approximation and include the fluctuations of the polariton field around its mean value is based on the wigner representation of quantum fields . within the _ truncated wigner approximation _ , the dynamics of the quantum field problem can be described by a stochastic partial differential equation that can be numerically solved . this technique was first introduced in the context of quantum fluids in ref . and its application to atomic condensates was fully developed and characterized in ref . . its extension to the polariton case was developed in ref . : remarkably , the presence of loss and pump terms in the stochastic equations suppresses some of the difficulties of the truncated wigner method and guarantees its accuracy as long as interactions between individual polaritons are weak enough . if a spatial grid of real - space spacing @xmath95 is used to numerically solve the stochastic partial differential equation , the weak interaction condition can be formulated as @xmath96 . while this condition rules out the possibility of using the truncated wigner method to study strongly correlated polariton states , e.g. in the polariton blockade regime @xcite , it does not hinder the study of quantum hydrodynamics effects such as the analog hawking radiation , which originate from the collective dynamics of a large number of polaritons . the stochastic partial differential equations of the truncated wigner method for the coupled exciton and polariton fields has the explicit form @xcite @xmath97 \left ( \begin{array}{c } \phi_{\mathrm{x } } \\ \phi_{\mathrm{c } } \end{array } \right)\right\}\,dt + \frac{\hbar}{\sqrt{4\,\delta x } } \left ( \begin{array}{c } \sqrt{\gamma_{\mathrm{x}}}\,dw_{\mathrm{x } } \\ \sqrt{\gamma_{\mathrm{c}}}\,dw_{\mathrm{c}}\end{array } \right ) \ , , \ ] ] where @xmath95 is the spacing of the real - space grid and @xmath98 and @xmath99 are complex valued , zero - mean , independent gaussian noise terms , with white noise correlation in both space and time @xmath100 with @xmath13 . in the numerical simulations , we reconstruct the equilibrium wigner distribution by first letting the system evolve to its steady state under the monochromatic pump , and then taking a large number of independent configurations by sampling the stochastic evolution at different times spaced by @xmath101 . in order to ensure statistical independence of the different realizations , a large enough @xmath101 has to be taken such that @xmath102 . in practice , a number of realization on the order of @xmath103 is used , taken at time intervals on the order of @xmath104 . as usual in wigner approaches @xcite , the stochastic averages over the configurations of different functions of the fields provide the expectation value of the corresponding symmetrically ordered operator . inspired from previous numerical studies of acoustic black holes in atomic condensates @xcite , we shall look for the signature of analog hawking radiation in the normalized , zero - delay correlation of the cavity field intensity defined as @xmath105 within the wigner formalism , the different contributions to ( [ eq : g2def ] ) have the form @xcite @xmath106 and @xmath107 where the @xmath108 averages indicate the classical averages over the different stochastic configurations of the field sampled at interval times @xmath101 . after having reviewed the technical tools to study the dynamics of the polariton quantum fluid , we can now proceed to discuss realistic configurations that can be used to generate an analog acoustic black hole with a large surface gravity . all simulations are performed by solving the generalized gross - pitaevskii equation ( [ eq : mf_gpe ] ) using experimental parameters taken from the recent work in ref . . the most straightforward configuration to generate an acoustic horizon was proposed by solnyshkov et al . @xcite and is illustrated in fig . [ fig : nobh_scheme](b ) : the presence of losses is responsible for a spatially decreasing density profile of a ballistically flowing polariton condensate and a correspondingly increasing velocity profile . eventually , the polariton density tends to zero and the flow becomes necessarily super - sonic . the polariton density and speed in the pumped region is fixed by the pump beam parameters : if these are suitably chosen to have an initially sub - sonic flow , a horizon must necessarily appear at some point . in spite of the simplicity of this configuration , some care has to be paid to the value of hawking temperature that can be expected . for smooth horizons within the hydrodynamic limit @xcite , the hawking temperature @xmath109 is determined by the surface gravity @xmath110 \right|_{x_{\rm hor } } { \label{eq : surf_grav}}\ ] ] according to @xmath111 , with @xmath112 the boltzmann constant . according to our numerical simulations and analytical approximations @xcite , the characteristic length scale of the variation of the flow parameters ( density and flow speed ) is inversely proportional to the loss rate @xmath113 . direct combination of these two general facts shows that the achievable values of the hawking temperature is limited from above by the loss rate : for the case in the figure , one indeed has @xmath114 mev . and the sound @xmath115 speeds at steady state . the parameters and the spatial shape of the pump laser are the same as in fig . ( c ) dispersion of the collective excitations at two spatial positions : point @xmath116 ( left panel ) is chosen to be in the sub - sonic flat - top region ; point @xmath117 ( right panel ) is located downstream of the horizon in the super - sonic region . red solid ( blue dashed ) lines refer to positive ( negative ) norm bogoliubov modes . ( d ) spatial profile of the exciton density at a time @xmath118 ps after the arrival of the probe pulse . the probe laser parameters are chosen so to resonantly excite the bogoliubov branch indicated as @xmath119 in panel ( c ) : frequency @xmath120 mev , wavevector @xmath121 . the probe spot has a gaussian shape centered at @xmath122 m with a waist @xmath123 m and a peak amplitude @xmath124 mev@xmath125m@xmath126 . its temporal shape is also gaussian of duration @xmath127 ps . all numerical calculations are performed in a 800-@xmath56 m simulation box with 2048 lattice points and absorbing boundary conditions at the edges of the box . , scaledwidth=46.0% ] a possible solution to overcome this difficulty , is to insert a narrow repulsive potential in the ballistic flow region , close to the edge of the pump spot . the efficiency of this technique is illustrated by the numerical calculations shown in fig . [ fig : bh_scheme](b ) . flow across the defect mostly occurs via tunneling processes through the potential barrier : this produces a sudden drop of the condensate density and a corresponding sudden increase of the flow speed . as a result , a horizon appears in the vicinity of the defect center , separating a sub - sonic upstream region from a super - sonic downstream one , with a surface gravity as high as @xmath128 mev , i.e. about a factor @xmath129 larger than the purely ballistic flow case of fig . [ fig : nobh_scheme ] . . the parameters and the spatial shape of the pump laser are again the same as in fig . the defect consists of a square potential barrier of thickness @xmath130 m and is located at a position @xmath131 m close to the flat - top edge . the height of the defect potential is @xmath132 mev and acts on both exciton and photon components of the polariton field . the horizon appears in the close vicinity of the defect with a large surface gravity . the probe laser parameters are chosen so to resonantly excite the bogoliubov branch indicated as @xmath119 in panel ( c ) : frequency @xmath133 mev , wavevector @xmath134 . the shape and the amplitude of the pump spot are the same as in fig . [ fig : nobh_scheme](d ) . , scaledwidth=47.0% ] another difficulty of the purely ballistic flow configuration of fig . [ fig : nobh_scheme](b ) was the significant spatial variation of the polariton density on both sides of the horizon , which may strongly distort the geometrical structure hawking signal . the defect configuration of fig . [ fig : bh_scheme](b ) appears to be favorable also in this respect : as the flow in the upstream region is sub - sonic , no bogoliubov - cerenkov emission can take place from the defect in this direction @xcite and the polariton density remains almost flat across the whole flat - top region . the slow spatial variation of the density in the downstream super - sonic region is hardly avoided unless more complex laser configurations are used , but does not seem to prevent identification of the hawking effect . on the other hand , the use of a coherent pump raises the crucial issue of avoiding all those branch - sticking and gap effects typical of non - equilibrium condensates @xcite that would spoil the analogy with gravitational systems . as we have reviewed in sec . [ sec : bogoliubov ] , choosing the value of the laser amplitude in the flat - top region in the vicinity of the end - point of the upper branch of the hysteresis loop is enough to have a linear , sonic dispersion of the low-@xmath135 dispersion of the form ( [ eq : sonic ] ) at all spatial positions around the horizon , and hence has no influence on the final results , which further supports the robustness of our scheme . ] . as it was illustrated in fig . [ fig2 ] , keeping the system stably on the upper branch of the hysteresis loop requires including a region of higher pump intensity upstream of the flat - top . exception made for its role in switching on the optical bistability , the high pump intensity region play no role in the hawking physics . as we have discussed in sec . [ sec : bogoliubov ] , the bogoliubov dispersion in the downstream region is guaranteed to be sonic at low wavevectors by the ballistic nature of the flow in this region . the flow speed is guaranteed to be super - sonic by the suppressed value of the density beyond the defect . ps , ( b ) @xmath136 ps , and ( c ) @xmath137 ps after the arrival of the probe pulse . each panel displays the excitonic density modulation with respect to the time - independent steady state . the pump and probe parameters are the same as for the purely ballistic flow shown in fig . [ fig : nobh_scheme](d ) . the labels on the wavepackets refer to the bogoliubov dispersions of fig . [ fig : nobh_scheme](c ) , while the arrows indicate the propagation direction of each wave packet . , scaledwidth=50.0% ] ps , ( b ) @xmath136 ps , and ( c ) @xmath137 ps after the arrival of the probe pulse . each panel displays the excitonic density modulation with respect to the time - independent steady state . the pump and probe parameters correspond to the defect configuration shown in fig . [ fig : bh_scheme](d ) . the labels on the wavepackets refer to the bogoliubov dispersions of fig . [ fig : bh_scheme](c ) , while the arrows indicate the propagation direction of each wave packet . , scaledwidth=50.0% ] the numerical simulations that we have presented in the previous section demonstrate the possibility to create configurations of a polariton flow through a horizon separating regions of sub- and super - sonic regimes with a sizable surface gravity . this configuration will be a powerful workbench for studies of hawking phenomena . inspired by related theoretical work on atomic condensates @xcite and experiments in classical hydrodynamics @xcite , in this section we shall first consider the scattering of a phonon wavepacket off the horizon , a process where the classical counterpart of the hawking effect manifests itself as an additional reflected wavepacket . this discussion will be the starting point for the next section , where we shall investigate the very hawking effect , namely the conversion of zero - point quantum fluctuations into observable radiation by the horizon . the scattering dynamics of bogoliubov phonons hitting the horizon can be studied in terms of the generalized gpe ( [ eq : mf_gpe ] ) : far from the horizon , the propagation of weak wavepacket perturbations follows the bogoliubov dispersion discussed in sec . [ sec : bogoliubov ] albeit with spatial dependent flow and sound speeds . figs . [ fig : nobh_scheme ] to [ fig : bh_snap ] illustrate the different aspects of this physics for the two previously mentioned cases , namely a smooth flow with a low surface gravity a defect configuration with a sizable surface gravity . for each configuration , the structure of the flow is illustrated in figs . [ fig : nobh_scheme](b ) and [ fig : bh_scheme](b ) , while the smaller ( c , d ) panels show the bogoliubov dispersion at different spatial positions . once the system is in its steady state , a low - frequency wavepacket perturbation is generated in the downstream super - sonic region by means of an extra coherent laser pulse with a spatio - temporal shape of the form @xmath138 the probe wave vector ( @xmath139 ) and frequency ( @xmath140 ) are chosen to be resonant with the bogoliubov branch labeled as @xmath119 in the ( c ) panels . the probe beam is centered at @xmath117 ; its waist @xmath141 determines the spatial size of the generated wavepacket ( and its inverse width in @xmath135-space ) . the temporal duration @xmath142 has to be long enough not to excite the upper polariton branch nor the negative - norm bogoliubov branch . its amplitude @xmath143 is chosen weak enough to remain within the validity domain of a linearized description of excitations . the generated wavepacket shown in figs . [ fig : nobh_scheme](d ) and [ fig : bh_scheme](d ) then propagates in the leftward direction against the horizon . a series of snapshots of this evolution are presented in figs . [ fig : nobh_snap ] and [ fig : bh_snap ] . the difference between the behaviors of the weak and of a large surface gravity cases is apparent : in the former case , only two wavepackets visibly emerge from the horizon at long times , located on the @xmath144 and @xmath145 bogoliubov branches . on the other hand , in the latter case , an extra wavepacket appears on the @xmath146 branch . remarkably , the bogoliubov norm of this wavepacket is negative , which signals the occurrence of hawking conversion processes . the identification of the different wavepackets in terms of their bogoliubov branch is confirmed by the numerical measurement of their group velocity and of their wavevector . of course , the physical reason why we did not observe a @xmath146 wavepacket in fig . [ fig : nobh_snap ] is that the low value of the surface gravity strongly suppresses the amplitude of the hawking process . as it was expected on general grounds @xcite and then explicitly verified @xcite , the hawking scattering amplitude for a bogoliubov wavepacket of given carrier frequency @xmath147 scales down exponentially with the surface gravity according to the boltzmann factor @xmath148 . on the other hand , excitation of the @xmath146 branch would be totally forbidden if the flow was everywhere sub - sonic and no horizon was present : in this case , in fact , energy could not be conserved in the hawking scattering process . this quantity directly reflects in the intensity correlations of a near - field image of the emitted light from the microcavity . the upper ( a ) panel refers to the case of a purely ballistic flow considered in figs . [ fig : nobh_scheme ] and [ fig : nobh_snap ] . the lower panel ( b ) refers to the case with a defect in the ballistic flow considered in figs . [ fig : bh_scheme ] and [ fig : bh_snap ] . the colored dashed lines in ( b ) indicate the different hawking features as discussed in the text.,scaledwidth=50.0% ] the process studied in the previous section provides a solid evidence of the classical counterpart of the hawking effect , namely the possibility of inter - converting positive and negative norm waves at the horizon . now we shall proceed with the study of the hawking effect _ stricto sensu _ , i.e. the emission of radiation by the horizon via the inter - conversion of zero - point quantum fluctuations into observable radiation . to this purpose , one has to fully include in the model the quantum fluctuations of the fields around the mean - field value predicted by the gross - pitaevskii equation . in this work , we adopt the truncated wigner approach described in sec . [ sec : montecarlo ] , where expectation values of quantum operators are calculated as classical averages over a suitably chosen stochastic partial differential equation . inspired from previous theoretical @xcite and numerical @xcite work , we shall consider the spatial correlation function of the quantum fluctuations of the fluid density : the smoking gun of hawking radiation consist of a negative long - range correlation signal between points located on opposite sides of the horizon . numerical plots of the normalized density - density correlation function @xmath149 are shown in the two panels of fig . [ fig : g2 ] for the two cases of a weak and strong surface gravity @xmath150 , respectively . while in the upper panel , fig . [ fig : g2](a ) ( weak surface gravity , @xmath151 mev ) there is no clear feature emerging over noise , the lower panel , fig . [ fig : g2](b ) ( strong surface gravity , @xmath152 mev ) shows several features that can be interpreted as a direct consequence of the analog hawking radiation @xcite : the two bogoliubov excitations forming the hawking pair are simultaneously emitted from the horizon . at later times , the temporal correlation between the emission time of the hawking partners reflects into a long - distance correlation in the density fluctuations due to each member of the pair . while in the atomic case this correlation extends at long times to indefinitely large distances from the horizon , the finite decay rate of bogoliubov excitations in polariton fluids predicted in ( [ eq : sonic ] ) restricts the correlation signal to a finite region of size @xmath153 around the horizon , @xmath154 being the group velocity of the bogoliubov branch under consideration . away from this region around the horizon , the hawking phonons have decayed and the corresponding density fluctuations have disappeared . the dashed lines in fig . [ fig : g2](b ) indicate the location at which one would expect the maximal density correlations due to the quantum vacuum emission process into the @xmath155 ( white ) , the @xmath156 ( purple ) , the @xmath157 ( yellow ) pair of modes . the first process is the traditional hawking emission process where the two quanta are emitted into opposite directions from the horizon . the main properties of the density correlation pattern due to this process is well captured by the non - dispersive quantum field theory on curved space - time as proposed in ref . . for @xmath158 and @xmath159 , the correlation is peaked along the half - line @xmath160 here , @xmath161 and @xmath162 are the flow speed and speed of sound in the upstream and downstream regions , so that @xmath163 and @xmath164 are the group velocity of the @xmath146 and @xmath144 modes in the low-@xmath135 , sonic region . while in the atomic bec case of ref . their value did not depend on space in the asymptotic regions away from the horizon , the driven - dissipative nature of polaritons is responsible for the spatial dependence of the flow @xmath82 and sound @xmath165 speeds that is visible in fig . [ fig : bh_scheme](b ) , mostly in the downstream region . however , as the space dependence remains moderate within the regions of thickness @xmath166 and @xmath167 on either side of the horizon where correlations are significant , no appreciable curvature of the hawking tongues is visible in fig . [ fig : g2](b ) . the orientation of the purple and yellow dashed lines in fig . [ fig : g2](b ) is analogously obtained from the group velocity of the corresponding modes , namely @xmath168 and @xmath169 the excellent agreement of the geometrical location of the numerically observed features with the analytical predictions ( [ eq : ud2][eq : d1d2 ] ) confirms our interpretation . remarkably , the peak value of the @xmath155 hawking correlation signal is on the order of @xmath170 , which is not too far from the analytical prediction of ref . . as the surface gravity is here comparable to the interaction energy @xmath171 , the quantitative discrepancy can be easily traced back to the breakdown of the hydrodynamical approximation underlying the analytical model @xcite . thanks to the continuous - wave nature of the proposed experimental setting , the quantitatively low value signal to be observed can be overcome by a sufficiently long integration time . while agreement with the atomic case is good also for the @xmath156 feature , the @xmath157 one has a different sign . a possible explanation of this behavior can be traced back to the non - universality of the @xmath172 back - scattering effect that is responsible for the conversion at the horizon of the standard @xmath155 hawking correlation into the @xmath157 one . before concluding , it is important to note that throughout the whole discussion , we have implicitly assumed that the @xmath173 correlation function is assumed to be measured at the same time : a rough estimate of the required temporal resolution of the detector is given by @xmath174 , where @xmath175 is the speed of sound in the polariton gas ( on the order of @xmath176m / ps from figs . [ fig : bh_scheme ] ) and @xmath177 is the spatial width of the correlation signal . using the value @xmath178 m taken from fig . [ fig : g2 ] , one can estimate the needed temporal resolution to be on the order of @xmath179ps , which is within the state - of - the - art of optical technology . in this article we have presented a comprehensive study of classical and quantum hydrodynamic properties of acoustic black holes in superfluids of exciton - polaritons in semiconductor microcavities . inspired from on - going experimental research , we have identified and characterized one - dimensional polariton wire devices as model systems , where the polaritons injected by a single monochromatic laser beam with a suitable spot profile generate a flow configuration showing an acoustic black hole horizon . inserting a repulsive potential defect provides a dramatic enhancement of the analog surface gravity at the horizon . even if our calculations have been performed for the specific case of a semiconductor microcavity in the strong - coupling regime , all conclusions straightforwardly extend to a generic planar cavity filled by a nonlinear optical medium , as originally suggested in ref . , or photonic crystal polariton states with engineered mode dispersion @xcite . in analogy with previous works on acoustic black holes in atomic condensates and classical hydrodynamics of surface waves , two experiments have been proposed and numerically simulated to assess the efficiency of hawking wave - conversion processes at the horizon . when a coherent wavepacket of bogoliubov phonons is incident on the horizon , the hawking effect is visible as the production of an extra emerging wavepacket on the negative - norm branch . this is the classical evidence for the hawking process taking place at the horizon . on the other hand , the usual hawking radiation is a purely quantum effect consisting of the conversion of zero - point fluctuations into a stream of observable phonons at the horizon : the correlation between the emission times of a phonon and its hawking partner is visible as a long - distance correlation between the density fluctuations on either side of the horizon . the density correlations provide the clearest experimental signature of hawking radiation in superfluids : in the present polariton case , in - cavity density correlations directly transfer into the secondary emission from the cavity photon , and can be observed as analogous correlations in the intensity fluctuations of the near - field emission . as soon as a sufficient signal over noise ratio is achieved , an experiment along the proposed lines will hopefully provide a clear evidence of a fundamental effect of quantum field theory that so far has eluded all experimental observation , overcoming the barrier to direct observation imposed by the negligibly small hawking temperature of astrophysical black - holes . we are grateful to a. amo , r. balbinot , j. bloch , a. bramati , s. finazzi , s. robertson , d. sanvitto for continuous stimulating discussions , and d. sarchi for a fruitful collaboration at an early stage of the work . dg warmly acknowledges the friendly hospitality of the ino - cnr bec center in trento , where most of this work has been conceived and carried out , and l.c . andreani for supporting this research . r. schtzhold and w. g. unruh , phys . lett . * 107 * , 149401 ( 2011 ) ; f. belgiorno , s. l. cacciatori , m. clerici , v. gorini , g. ortenzi , l. rizzi , e. rubino , v. g. sala , and d. faccio , phys . rev . lett . * 107 * , 149402 ( 2011 ) . a. amo , d. sanvitto , f.p . laussy , d. ballarini , e. del valle , m. d. martin , a. lematre , and j. bloch , d. n. krizhanovskii , m. s. skolnick , c. tejedor , and l. via , nature * 457 * , 291 ( 2009 ) ; a. amo , j. lefrere , s. pigeon , c. adrados , c. ciuti , i. carusotto , r. houdr , e. giacobino , and a. bramati , nat . phys . * 5 * , 805 ( 2009 ) ; a. amo , s. pigeon , d. sanvitto , v. g. sala , r. hivet , i. carusotto , f. pisanello , g. lemenager , r. houdr , e. giacobino , c. ciuti , and a. bramati , science * 332 * , 1167 ( 2011 ) . o. el daf , a. baas , t. guillet , j .- p . brantut , r. idrissi kaitouni , j. l. staehli , f. morier - genoud , and b. deveaud , appl . . lett . * 88 * , 061105 ( 2006 ) ; d. lu , j. ahn , s. freisem , d. gazula , and d. g. deppe , appl . . lett . * 87 * , 163105 ( 2005 ) . d. f. walls and g. j. milburn , _ quantum optics _ ( springer verlag , berlin , 2006 ) ; c. cohen - tannoudji , j. dupont - roc , and g. grynberg , _ atom - photon interactions . basic processes and applications . _ ( wiley science paperback series , new york , 1998 ) .

fu orionis is the prototype of a small , but quite remarkable class of low - mass young stellar objects ( ysos ) normally referred to as fu ori objects ( fuors ) . for the first members of this class an outburst in optical light of up to 4 - 6 magnitudes over short time scales , followed by a decrease in luminosity over several years or decades , was observed @xcite . other objects were included in the class as they shared common specific spectroscopic features , e.g. , double - peaked line profiles , a spectral type varying with wavelength and often co bandhead absorption features . as for one object ( v1057 cyg ) the pre - outburst spectrum is known and as it resembles that of a classical ttauri star @xcite , it is commonly assumed that fuors should be low - mass ysos . most observational data can be explained by the presence of an accretion disk surrounding the young stars ( * ? ? ? * however , found that some spectral properties are well explained in the context of rapidly rotating late - type stars with strong stellar winds ) . a dramatic temporal increase in the accretion rate , where the disk outshines the star by several orders of magnitude , can account for the observed outbursts in luminosity . several scenarios , possibly triggering such an increased accretion rate , have thus far been proposed . they include ( a ) interactions of binary or multiple systems where tidal forces disturb the circumstellar disk @xcite , ( b ) planet - disk interactions , where thermal instabilities in the disk are caused by the presence of a massive planet @xcite , or ( c ) thermal instabilities in the disk alone @xcite . for a detailed overview concerning the fu ori phenomenon we refer to @xcite . apart from revealing the mechanism leading to the observed outbursts , it is also important to investigate whether all ttauri stars undergo such epochs of enhanced accretion or whether fuors are a special class of ysos . most observations of classical ttauri stars show that the derived accretion rates of @xmath1 ( e.g. , * ? ? ? * ) are not sufficient to build up a low - mass star over time scales of a few myr . even if part of the matter is supposed to be accreted in the very early phases of a yso , fuor - phases might provide an elegant solution to this problem as they would speed up the accretion process . as so far only a dozen or so well established fuors are known and as the fuor - phase has significant impact on the young star - disk system it is crucial to combine as much observational information as possible for these objects to eventually derive a coherent theoretical picture . based on near - infrared ( nir ) and/or mid - infrared ( mir ) interferometry as a technique to study the inner few au of the accretion disks new insights to some of the best studied fuors were provided recently . @xcite found that accretion disks alone can not reproduce the sed and observed low k - band visibilities for v1057 cyg , v1515 cyg and z cma - se simultaneously . they concluded that additional uncorrelated flux may arise due to scattering by large dusty envelopes . @xcite presented the first vlti/ midi observations of v1647 ori whose eruptive behavior suggests that it is either an fuor or an ex lupi ( exor ) type object . in this case it was possible to fit both , the sed and the observed mir visibility with a simple disk model with moderate disk flaring . for fu ori itself it was @xcite who analyzed a wealth of nir interferometric data . they showed that the nir visibilities and the sed could be fitted with two models : one consisting simply of an optically thick and geometrically thin accretion disk and a second one consisting of an accretion disk and an embedded `` hot spot '' . from their error statistics these authors concluded that the latter model was more likely . in summary it becomes clear that up to now no coherent picture can be derived from the interferometric observations of fuors and the group of objects seems to be rather inhomogeneous . in this paper we present the first mir interferometric measurements of fu orionis . the data are thus complementary to the nir observations of malbet et al . ( 2005 ) . in section 2 we briefly describe the observations and the data reduction process . in section 3 we discuss the findings derived from the n - band acquisition images . the 8 - 13@xmath0 m spectrum of fu ori is analyzed in section 4 . in sections 5 and 6 we discuss the results from the interferometric measurements , i.e. the visibilities and the correlated flux spectra , respectively . a simple analytical disk model and its fit to the sed and the visibilities is presented in section 7 . finally , we summarize our conclusions and mention some future prospects in section 8 . the observations were carried out between october 31@xmath2 and november 4@xmath3 2004 with the mid - infrared interferometric instrument ( midi ) at eso s very large telescope interferometer ( vlti ) on paranal / chile . together with the keck interferometer nuller that recently started to produce first scientific results @xcite , midi is currently worldwide the only instrument able to conduct spectrally resolved interferometric observations in the mid - infrared @xcite . for the observations midi was used in high - sens mode using the nacl prism as dispersive element yielding a spectral resolution of 30 . the maximum projected baseline was 86.25 m and the minimum projected baseline 44.56 m leading to an angular resolution at 10@xmath0 m of 0.@xmath4029 and 0.@xmath4056 , respectively . for an assumed distance of 450 pc these values correspond to 13.1 au and 25.2 au . a journal of the observations including projected baselines , position angles and calibrator stars is given in table [ journal ] . for completeness we also mention the observations from december 2004 although they are disregarded in the following sections . these observations had almost exactly the same baseline and position angle as the october observations but showed in general a lower level of total and correlated flux due to rather poor observing conditions in terms of seeing and transparency . the calibrator stars for the data reduction were chosen by analyzing all calibrator stars observed over the whole night . we selected those showing a good agreement in their transfer functions , i.e. their instrumental visibilities after the assumed sizes were taken into account . the data reduction was carried out with the software package mia+ews-1.3 . this software consists of two independent reduction programs ( mia and ews ) which were both applied for comparison . the software , further information and manuals can be downloaded from the internetnevec / midi / index.html ] . a general description of the basic data reduction steps is also given in @xcite and we refer the reader to this paper . as the results derived with mia and ews agreed quite well ( overall differences in the correlated flux @xmath5 ) , we decided to show only plots resulting from the mia package . fu ori was found to be a binary system by @xcite . @xcite confirmed the detected companion and concluded that it was a young star of spectral type k showing clear nir excess . in the same paper it was also stated that it was very unlikely that the binary component ( fu ori s ) triggered the outburst of fu ori observed in the late 1930s . knowing about the existence of the fainter companion the integration time of some midi acquisition images was increased in order to derive n - band photometry for both components . in three images fu ori s was clearly visible ( figure [ image ] ) and aperture photometry could be applied to the observations . the results are summarized in table [ fluxes ] . interestingly , fu ori s shows relatively a higher n - band excess than fu ori itself . this can be explained in terms of differences in the geometry of an assumed circumstellar disks ( e.g. , larger flaring angle or smaller disk inclination ) . we furthermore derive a separation between the components of 0.@xmath4484@xmath60.@xmath401 and a position angle of 162.5@xmath74.1@xmath8 ( measured from north eastwards ) which is in good agreement with the results from @xcite and @xcite . our errors are standard deviations based on gaussian fitting of 3 independent images . for an assumed distance of 450 pc the separation corresponds to 217.8@xmath64.5 au . figure [ spectrum_short ] shows the mir spectrum of fu ori . the spectrum obtained with midi is the average over the first three observing nights using both unit telescopes ( uts ) . for comparison _ spitzer irs _ data are plotted additionally . those data were available from the _ spitzer _ data archive . for the data reduction we used the _ droopres _ intermediate data product processed through the ssc pipeline s12.0.2 . the smart reduction package developed by the irs instrument team at cornell @xcite was used to extract the spectrum . within the error bars the spectra agree quite well . a broad and weak silicate emission feature is present between 8 - 13@xmath0 m . however , the averaged midi spectrum appears to be a little flatter than the more roundish _ spitzer _ spectrum . in contrast to other ysos the observed silicate emission feature of fu ori is rather weak ( see also * ? ? ? * ) . such a broad and flat mir dust spectrum can be explained by grains that already underwent some coagulation process @xcite . to test this hypothesis we derived a possible dust composition by fitting a dust model to the normalized and continuum subtracted _ spitzer _ spectrum from 6 - 13@xmath0 m . this analysis method is similar to approaches that were successfully applied in previous studies for determining the dust composition of observed dust emission features of other ysos ( e.g. , * ? ? ? in the model we assume that the observed emission can be computed from the sum of the emission of individual dust species . the species we used are summarized in table [ dust_species ] together with references for their optical properties . we furthermore took into account different grain sizes as for each of the dust species the opacities were calculated for 0.1 , 1.5 , and 6.0@xmath0 m sized particles . to estimate the possible contribution of polycyclic aromatic hydrocarbons ( pahs ) to the emission we used a template spectrum based on profiles that were derived from observations @xcite . finally , in order to reproduce the observed spectrum we fitted the following emission model to the _ spitzer _ data using a least square minimization @xmath9 here , @xmath10 is the observed flux , @xmath11 denotes the planck function at the temperature of the underlying continuum , @xmath12 the planck function at the temperature of the silicate grains , @xmath13 is the mass absorption coefficient for the silicate species @xmath14 with the grain size @xmath15 , @xmath16 is the template emission spectrum for the pahs , and @xmath17 , @xmath18 , and @xmath19 are the weighting factors for the continuum , the silicate emission and the pah contribution , respectively . a more detailed description of the dust model and plots showing the different dust opacities is given in @xcite . the model results are shown in figure [ dust_spectra ] and table [ dust_compo ] summarizes the corresponding dust composition . as for the underlying continuum we could fit a temperature of 880 k , we believe that this optically thick emission most likely arises from the inner parts of the disk close to the star . the outer and thus cooler regions of the disk would then be responsible for the weak silicate emission for which we fitted a temperature of 230 k to the optically thin emission layer . the results strongly support the idea that most particles are amorphous and much larger in size than those in the ism . pahs , silica particles and crystalline silicates do not seem to be present at all . although the derived fractions of different dust species do depend on the model applied to the data ( e.g. , in terms of the precise grain sizes considered and also in terms of particle structures ) the main conclusions , that we do not find evidence for crystalline silicates and that the grains are considerably larger than those found in the ism , remain unaltered . for a more detailed discussion about the influence of different model parameters we refer to @xcite and @xcite . as the disk is much brighter in the optical and nir than the central star ( section 7 ) , stellar radiation is probably not able to produce the observed weak emission feature in an optically thin disk surface layer . however , it was shown that the silicate feature of fu ori can be reproduced taking into account self - irradiation of the disk @xcite . in this case the hot inner parts of the accretion disk serve as flux source and illuminate the disk surface layer under a certain angle . in figure [ visibilities ] and table [ visibilities_table ] the visibilities measured for fu ori at three different baselines are shown . the errors are computed as standard deviations resulting from the use of different calibrators for each night ( see table [ journal ] ) . in most of the following plots the position of the atmospheric ozone band is indicated as this part of the spectrum suffered sometimes from imperfect corrections during the data reduction process . here , small `` dips '' and `` bumps '' in otherwise flat spectra are remnants from the data reduction . the calibrated visibility of the shortest baseline ( ut2-ut3 ) remains almost constant at @xmath200.95 over the whole wavelength range and the source is only marginally resolved . as expected , for the longest baseline ( ut2-ut4 ) the lowest visibility is observed ranging from @xmath200.73 at 8.3@xmath0 m to @xmath200.6 at 13@xmath0 m . for the intermediate baseline ( ut3-ut4 ) the visibility shows a slight increase from @xmath200.73 at 8.3@xmath0 m to @xmath200.85 at 13@xmath0 m . the fact that the object is clearly resolved with two baselines in the mir supports the argument of @xcite that we are observing an extended circumstellar structure , i.e. a disk , and not only the stellar photosphere @xcite . for the two baselines ut2-ut3 and ut2-ut4 the observations are consistent with expectations from thermal disk emission as despite the decreasing resolution for longer wavelengths the visibilities indicate larger sizes for the emitting regions ( see also section 5.3 . ) . the observed increase in visibility for the ut3-ut4 baseline implies that for this baseline , however , the object appears smaller at 13@xmath0 m than at 8@xmath0 m . this seems difficult to imagine in the context of a circumstellar disk as emitting source and certainly a second observation for this baseline configuration seems eligible . theoretically it is possible that the photometric measurements carried out directly after the interferometric observations are corrupted due to technical problems or different weather conditions . this in turn might then lead to a change in the visibility function . however , the calibrators observed at this night did not show any sign of poor photometric measurements and their transfer function was very stable over the whole night . in addition , also other ysos with circumstellar disks showed an increasing visibility for certain baselines when observed with midi . thus , since we do not find any evidence for excluding this dataset due to bad quality we decided to keep it in our analyses . as thus far no midi visibilities for ttauri stars have been published we are limited to a comparison between fu ori and circumstellar disks around herbig ae / be stars ( haebes ) . it shows that these stars as well as models applied to them normally show a prominent drop in the visibility between 8@xmath0 m and 10@xmath0 m from where the curve remains almost constant @xcite . qualitatively , this drop results from the intensity distribution of the passive disks around the haebes : at the short wavelength end the hot inner rim of the disks provides an overproportional contribution to the flux and is at the same time confined to a small spatial region leading to a high visibility . most of the rest of the mir emission originates from a large area of the hot , illuminated surface layer of the flared circumstellar disk . the visibilities for fu ori show hardly any wavelength dependence and are very flat from 8 - 13@xmath0 m regardless of baseline . thus , its seems as if the flux distribution of fu ori is smoother and the visibilities show no sign of a significant contribution from a hot inner rim . however , as fu ori is surrounded by a heavily active accretion disk where the disk alone produces the majority of the observed flux at almost all wavelengths ( see section 7 ) differences in the intensity distribution and thus in the visibility can be expected . apart from differences in the shape of the visibility curves most haebes are much better resolved , i.e. show lower visibilities than fu ori @xcite . this , however , can at least partly be explained by the distance to these objects which is in general less than 200 pc . fu ori on the other hand has an assumed distance of 450 pc , and if it was closer to the earth we would observe lower visibilities also for this object . to derive a simple model for the geometry of the emitting regions we assume a simple gaussian brightness distribution for each baseline . this is a reasonable first approximation for objects showing high visibilities . the fwhm , and hence the physical size , of this gaussian in arcsec can be computed by @xmath21 where @xmath22 is the measured visibility for a certain spatial frequency @xmath23 ( in arcsec@xmath24 ) derived from the projected baseline @xmath25 and the wavelength @xmath26 . equation [ visi_eq ] results from a simple fourier transformation of the assumed brightness distribution . we computed the fwhm for all three baselines at three different wavelengths ( 9.0 , 11.0 , and 12.5@xmath0 m ) . for this we averaged for each baseline 5 visibility points from table [ visibilities_table ] centered on the specified wavelengths . table [ table_sizes ] gives the resulting sizes of the emitting regions in au for an assumed distance to fu ori of 450 pc . the results are also visualized in figure [ figure_sizes ] where the fwhm are shown in their orientation on the sky . as expected for thermal disk emission the fwhm increases with wavelength for a given baseline . the only exception is the 12.5@xmath0 m size for the ut3-ut4 baseline which is surprisingly a little smaller than that seen at 11.0@xmath0 m . in addition to size estimations for each individual baseline our measurements based on different position angles allow to constrain the geometry of the disk . we fitted the derived fwhm with an ellipse for each of the considered wavelengths in order to derive a simple model for the spatial orientation of an assumed disk - like structure . the resulting best fit ellipses are overplotted in figure [ figure_sizes ] and their parameters are summarized in the lower half of table [ table_sizes ] . as one would expect from a disk - like structure showing a decrease in temperature with radius the semimajor axes of the ellipses increase with wavelength . however , due to the increasing visibility of the ut3-ut4 baseline the semiminor axis and also the total area of the ellipse are slightly smaller at 12.5@xmath0 m than at 11.0@xmath0 m . an inclination angle for an assumed circular disk can be derived by computing the @xmath27 of the ratio of the semiminor and semimajor axis . the resulting angles agree very well ( 55.4@xmath72.4@xmath8 ) and are also in good agreement to what was found for fu ori disk models based on nir observations @xcite . by combining the derived inclination angle with rotational velocity measurements in the optical spectral region it is possible to estimate the central mass . based on @xcite we derive for fu ori @xmath280.36 which is reasonably consistent with typical values for low - mass pre - main sequence stars . in addition to the inclination the fitted ellipses provide also information on the position angle of the disk . the position angles ( measured from north eastwards ) we find for the semimajor axes at different wavelengths show a larger scatter ( 109.1@xmath711.6@xmath8 ) than the inclinations and differ significantly from the nir findings of @xcite ( see also section 7 ) . it seems as if the position angle becomes smaller for longer wavelengths , although one has to keep in mind that , again , at least for the ut3-ut4 baseline the increasing visibility is the main reason for the observed rotation . at this point we leave it to future investigations to derive a 3-d disk model based on a more complex dust distribution possibly able to confirm the apparent changes in the position angle for different wavelength regimes . the correlated flux is directly linked to the total flux ( i.e. the flux from a single ut telescope ) via the visibility : @xmath29 figure [ correlflux ] depicts the results for the correlated flux for the three different baselines . the total spectrum is plotted for comparison . taking into account the spatial resolution of the three baselines it is possible to estimate from where the correlated 8 - 13@xmath0 m flux originates . our ut2-ut3 baseline has a spatial resolution of @xmath2025 au at the distance of fu ori , and if we assume an average visibility of 0.95 we hence know that 95% of the 8@xmath0m-13@xmath0 m flux must come from within this 25 au . similarly , we find that for the ut2-ut4 baseline 65% of the flux is emitted from the inner 13 au . for the ut3-ut4 baseline 75% of the 8 - 13@xmath0 m flux comes from the inner 20 au . the spatially resolved mir spectra of the midi instrument make it possible to study the radial dependence of the dust composition in the protoplanetary disk @xcite . in section 4 we showed that the spatially unresolved ( total ) spectrum of fu ori can be fitted with mainly large amorphous dust grains . the shape of the spatially resolved ( correlated ) flux spectra contains information about the dust composition of the inner parts of the disk . to compare the shape of the correlated spectra to the total spectrum we first subtracted the continuum for which we fitted a straight line between the flux at 8.25 and 12.95@xmath0 m . then we normalized all continuum subtracted correlated flux spectra and the total _ spectrum by @xmath30 where @xmath31 and @xmath32 denote the continuum subtracted spectra and their mean value , respectively . by applying this formula all correlated spectra conserve their shape and can be compared to the total spectrum ( figure [ shape_correl ] ) . to identify those parts where the spectra differ significantly we furthermore computed the deviation from the total _ spectrum for each wavelength in units of the data error : @xmath33 @xmath34 and @xmath35 are the normalized correlated spectra and the normalized total spectrum and @xmath36 denotes the standard deviation of the correlated spectra ( midi errors ) at a given wavelength . we applied the same normalization factors to the errors as we did to the spectra . figure [ shape_correl ] shows that no clear deviations from the total unresolved spectrum are observed for any of the correlated spectra . this means that with the given spatial resolution and sensitivity no significant chemical difference in the dust composition is observed and the spectra of the inner parts of the disk look very similar to the total unresolved disk spectrum . this finding is somewhat surprising . as seen in section 4 the silicate dust grains apparently already underwent noticeable coagulation . thus , the disk does not consist of purely ism dust anymore and must already have a significant age . in addition it is known that apart from grain growth also thermal annealing takes place within protoplanetary disks and transform the amorphous silicates into crystalline silicates ( see , * ? ? ? * ; * ? ? ? however , neither the total spectrum nor the correlated spectra indicate the presence of crystalline silicates . this is even more surprising considering the high temperatures in the disk interior due to viscous heating of this actively accreting star ( section 7 ) . annealing processes transforming ism amorphous dust into crystalline dust set in in disk regions where @xmath37 ( e.g. , * ? ? ? * ) in addition to that , high accretion rates directly relate to a faster vertical and horizontal mixing within the protoplanetary disk @xcite . this effect also eases the detection of crystalline silicates within the accretion disk surrounding fu ori . but why do nt we find any evidence for their existence ? a possible explanation might be that they simply can not be detected as the related feature is rather weak and simply dominated by the underlying continuum . this hypothesis can be checked by analyzing the longer wavelength regime of the _ spitzer _ spectrum as shown in figure [ spectrum_long ] . as at longer wavelengths the temperature of the continuum is decreasing the contrast between the continuum and possible emission features due to the presence of crystalline silicate particles increases . hence , spectral signatures of these particles should be more easily detectable . figure [ spectrum_long ] , however , shows that also between 14 and 34@xmath0 m the spectrum of fu ori does not show any clear signs of crystalline silicates . normally crystalline forsterite particles have emission bands at 16.4 , 23.9 , 27.7 and 33.8@xmath0 m and enstatite grains show features at 18.8 , 21.5 , and 24.5@xmath0 m @xcite . if one argued that possibly around 27.7 - 27.8@xmath0 m a weak forsterite feature was present one should keep in mind that the features at 23.9 and 33.8@xmath0 m are normally stronger @xcite but are not seen at all . thus , also the longer wavelength mir spectrum of fu ori shows that crystalline silicates are not present . this shows that apparently the non - detection is at least not only due to contrast problems between the emission features and the underlying continuum . what else can prevent us from detecting crystalline silicates ? in case fu ori underwent its first ( of possible multiple ) outburst(s ) and phase(s ) of enhanced accretion , it is possible that not enough time has yet elapsed for the crystalline silicate particles to reach the disk surface from where they can be detected by means of mir spectroscopy . this hypothesis is , however , questionable as the large sizes of the amorphous silicates indicate that enough time , at least for dust coagulation , has elapsed . recently , @xcite found that assuming the crystallization by thermal annealing happens at the very early phases of disk evolution , the level of crystallinity is linked to the properties of the molecular cloud core from which the disk formed . it was found that rapidly rotating cores produce rather massive disks with a high accretion rate but a low level of crystallinity . as we observe these properties for fu ori the proposed model , although yet very simple and with limited predictive power , might provide an interesting way to explain at least partly the lack of crystalline silicates in the fu ori accretion disk . finally , another possibility might be that the vertical and radial mixing within the disk does not work as efficient as expected . in consequence not enough crystalline dust grains reach the disk surface at radii where also the contrast between the emission feature and the dust continuum allows their detection . the results from the correlated flux allow a comparison to disks around other ysos that have also been observed with vlti / midi . not only seem crystalline silicates be commonly present in protoplanetary disks , but for three intermediate - mass haebes stars @xcite found that the degree of crystallization within the disk depends on the distance from the central star . the inner parts of the disks are more crystalline than the outer disk regions . this can naturally be explained by assuming that in the inner disk regions where the disk has higher temperatures dust processing and annealing occurs and that over time crystalline particles are transported outwards . not only for intermediate mass stars but also for the ttauri star ry tau the same radial dependence of the crystallinity was observed @xcite . thus , crystalline silicates should in general be more easily detectable in the correlated spectra of protoplanetary disks . for reasons mentioned in the previous section it seems thus surprising that the disk around fu orionis does not seem to contain any detectable amount of crystalline dust . as described in the introduction fuors are believed to be low - mass or ttauri - like stars in a state of enhanced accretion . from sub - millimeter and millimeter observations @xcite concluded that fuors are possibly in an evolutionary youger age than classical ttauri stars . this younger age is also supported by theoretical models where the protostellar disk is continuously replenished by a dusty envelope . this material infall on the disk can lead to thermal instabilities in the disk and thus provides a possible mechanism for explaining the observed outbursts @xcite . the puzzling thing is now that apparently in the disk surrounding fu ori the amorphous grains have already grown considerably . hence , the disk can not be in its native dust composition and might already have a significant age . furthermore , from the _ spitzer _ feps ( _ formation and evolution of planetary systems _ ) legacy program it is known that apparently the amount of crystalline silicates in ttauri disks is not correlated with the average size of amorphous dust particles @xcite . hence , crystalline dust particles are created in the _ early _ phases of disk evolution where the accretion rate and the disk temperature are thought to be higher . if thus the fuor - phase is indeed common to most / all ttauri stars then the non - detection of crystalline particles can not be explained in the current picture of protoplanetary disk evolution . the present paper is more intended to show the observational results rather than to present a sophisticated disk model possibly able to match the observed sed and visibilities with highest accuracy . thus , instead of applying a complex radiative transfer model we decided to use in a first step a simple analytical disk model . in this model the effective temperature and also the surface density are prescribed by the following broken power law distributions @xmath38 @xmath39 for the temperature law the temperature at 1 au and the exponent @xmath40 are free parameters . for the surface density the total disk mass @xmath41 and the exponent @xmath42 are free parameters . as the integrated surface density must equal the total mass of the disk the radial dependence of the surface density @xmath43 can be computed . the shape of the total sed ( and finally also the observed visibilities ) require a double power - law to fit the data . at a radius of 3 au the temperature and surface density distribution change . however , we made sure that for both parameters the transition was smooth and continuous . figure [ sed ] shows the sed and the resulting fit from the simple disk model . the model parameters are summarized in table [ model_parameter ] . based on the value for the visual extinction @xmath44 we interpolated the extinction for the other wavelengths following the findings of @xcite . the disk inclination @xmath15 was derived from the ellipses fitted to the fwhm of assumed gaussian brightness distributions ( see section 5.3 . ) . for the position angle @xmath45 we used two values : one was also derived in section 5.3 . , and for the second one we fitted the disk model to the observed visibilities with @xmath45 being the only free parameter ( see also section 7.2 . ) . it is clear that this simple approach is not able to reproduce all features of the sed ( e.g. , the weak silicate emission feature ) . for the moment , however , we restrict ourselves to this simple approach and , as mentioned above , a detailed modeling is left for further investigations . it is striking that rather high disk temperatures are required to fit the sed . in the inner 0.35 au no dust can withstand the high temperatures @xmath461500 k even on the disk surface and only the gaseous disk component can survive . it is interesting to note that the power law index of @xmath47 of the inner disk region equals exactly that what was derived in analytical disk models for geometrically flat , optically thick _ passive _ disks ( e.g. , * ? ? ? * ) as well as for flat , steady state , optically thick _ accretion _ disks ( e.g. , * ? ? ? thus , from the temperature distribution alone one can not distinguish between the two different cases ( see also * ? ? ? . however , in the context of fu ori the passive disk model where the disk is merely re - processing radiation from the central star is unlikely and only an accretion disk model matches most observations @xcite . a nice example is the recently detected magnetic field in the innermost regions of the disk surrounding fu ori which can only be explained in the picture of a heavily accreting disk @xcite . as the disk temperature is directly related to the mass accretion rate @xmath48 in classical accretion disk models it is possible to derive the @xmath48 required to produce the observed high temperatures . @xcite analyzed the effective disk temperature and also the midplane temperature of an accretion disk as a function of the mass accretion rate . figure 1 of their paper shows that one needs @xmath49 to get close to the effective temperature we find in our model . such high values for @xmath48 were also found by @xcite and @xcite . due to the high accretion rate the midplane temperature of the disk in the inner @xmath2010 au is even much higher than the effective disk temperature @xcite . concerning the temperature law farther out than 3 au the value we adopt here for @xmath50 is in good agreement to what can be found in the literature for isothermal flared disks @xcite . in their model the index value diverges from the flat solution where @xmath51 at a certain point in the inner disk and approaches @xmath52 for radii much larger than the stellar radius . at intermediate radii the temperature index might be approximated by values in between these most extreme cases depending on the actual flaring of the disk . in summary , the temperature distribution indicates that in the innermost disk regions the disk luminosity is accretion dominated whereas further out it is irradiation dominated . this is expected for an active accretion disk since the luminosity released by accretion is a much steeper function of distance to the star than that by absorption of radiation from the central star and the hot innermost disk regions ( @xmath53 , whereas @xmath54 , roughly ) . for this simple disk model it is straightforward to compute the visibilities as a function of projected baseline and wavelength . assuming that each part of the disk surface is emitting as a blackbody with an effective temperature @xmath55 the corresponding total flux and visibility for an axisymmetric face - on disk is given by ( see also equations ( 2 ) , ( 3 ) , and following in * ? ? ? * ) : @xmath56 @xmath57dr\ ] ] with @xmath58 being the flux , @xmath59 the planck function , @xmath60 the zeroth - order bessel function of the first kind , @xmath61 the projected baseline and @xmath62 the distance to the object . taking into account the inclination of the disk and the position angle as given in table [ model_parameter ] we computed the corresponding visibilities for the three different baselines . the results for two different position angles are shown in figure [ model_visi ] . in the upper panel the disk inclination and the position angle are derived from the ellipses fitting in section 5.3 . in the lower panel the position angle was fitted to the observed visibilities with a least squares fit . while the ellipse fitting yielded a position angle of 109.1@xmath6311.6@xmath64 a least square analysis leads to a value of 93.4@xmath636.8@xmath64 . from the plots in figure [ model_visi ] one can see that changing the position angle mostly affects the visibility of the shortest baseline while the curves for the other baselines change only marginally . fitting the ut3-ut4 baseline proofs quite a challenge as the observed visibility is increasing over the 8 - 13@xmath0 m wavelength interval while all other modeled visibilities are decreasing . especially for this baseline better fits are expected from more sophisticated disk models . however , given the simplicity of the current models the fits seems in reasonable agreement with the observations . it clearly shows that interferometric measurements can put additional constraints on the structure and geometry of circumstellar disks that can not be derived from sed fitting alone . interestingly , @xcite were able to fit their nir visibilities of fu ori with a similar disk model and also with a disk and an embedded `` hot spot '' . they concluded that the latter model was more likely based on statistical arguments . the major differences between their models and ours is that ( 1 ) they fit the sed with single power laws for the temperature neglecting the flux longwards of 20@xmath0 m and ( 2 ) their best fits yield disk position angles of 47@xmath65 and 8@xmath66 . the inclination angles ( 55@xmath67 and 48@xmath68 , respectively ) , are in good agreement with our value . in figure [ malbetmodels ] we compare both of their models to our observations . we find that both models fit the midi visibilities not as good as the model presented here . even more we find that our mir interferometric measurements rule out their first model of a simple accretion disk . this model predicts a higher visibility for the ut3-ut4 baseline than for the ut2-ut3 baseline which is not observed . the model with an embedded `` hot spot '' in the accretion disk shows the right trend for the visibilities , but the values for the ut3-ut4 and ut2-ut4 baseline are higher than for our models which already give slightly too high values compared to the observations . the main reason for the higher visibilities resulting from the model applied by @xcite is the single power law approach which provides less mir flux compared to our models . we also compared our models to the measured nir visibilities from @xcite . we restricted ourselves to those observations where the object was clearly resolved . the results are given in table [ table_nirvisi ] . except for the observations done with north - west baseline of the palomar testbed interferometer ( pti / nw ) both of our models agree with the nir visibilities within the error bars . concerning the predicted `` hot spot '' in the accretion disk @xcite our observations do not yield new insights into its nature as they do not have the required spatial resolution to confirm its existence . also the calibrated phase of our observations do not contain any information on which basis we could speculate about this possible second companion . we presented the first multi - baseline mir interferometric observations of fu orionis with midi / vlti . the findings can be summarized as follows : 1 . fu orionis was clearly resolved in the mir with two vlti baselines and marginally resolved with a third baseline indicating the presence of warm dusty material surrounding fu ori out to several tens of au . the inclination and the position angle of the accretion disk can be inferred from multi - baseline measurements with midi / vlti . thus , the instrument provides means for deriving disk parameters that are otherwise only poorly constrained from models and sed fits . the correlated flux indicates that 95% of the 8 - 13@xmath0 m flux comes from within the inner 25au of the disk for the shortest of our baselines while 65% of this flux arise from the inner 13au for our longest baseline . the shapes and strengths of the total 8 - 13@xmath0 m spectrum and the ( spatially resolved ) correlated spectra indicate that most dust particles within the accretion disk are amorphous and already significantly larger than typical particles in the ism . no spectra , neither the total nor the spatially resolved correlated spectra bear significant traces of crystalline silicates . given the high accretion rate and disk temperature of the system this is unexpected and requires further investigations . the sed and the observed mir visibilities can be fitted reasonably well with a simple analytical disk model prescribing a broken power law for the effective temperature distribution of the protoplanetary disk . within the innermost 3au the temperature decreases with @xmath69 , farther out the temperature goes with @xmath70 . the derived power law index for the inner regions is equal to what is derived theoretically for flat , steady state , optically thick accretion disks @xcite . for the outer disk region the derived value is in good agreement to what can be found for isothermal flared disks @xcite . 7 . based on points 4 . and 5 . and recent spectroscopic observations of protoplanetary disks around ttauri stars and haebes it seems questionable that fu ori is a very young classical ttauri star as the size of amorphous dust grains in combination with the non - detection of crystalline dust contradicts this assumption . 8 . from the acquisition image n - band aperture photometry could be derived for the companion fu ori s. it shows that this young k - type star has more relative mir excess than fu ori itself hinting at a circumstellar disk with larger flaring angle or smaller disk inclination . 1 . most of the published nir visibilities of fu ori @xcite do agree with our newly derived disk models . 2 . our interferometric observations rule out one out of two disk models for fu ori presented by @xcite . due to the lack of sufficient spatial resolution the presence of the proposed second very close companion can neither be confirmed nor disproved with the current set of midi data . 3 . in contrast to nir interferometric observations of other fuors @xcite our data do not seem to require the introduction of an extended dusty structure providing additional uncorrelated flux . 4 . for the inner 3au we require a significantly different disk model ( e.g. , @xmath71 , @xmath69 ) to fit our visibilities and sed in comparison to what was found for the outbursting star v1647 ori ( @xmath72 , @xmath70 ) @xcite . although the present results provide important new insights into the fu ori system new questions arose as well . future investigations will include a more thorough and detailed modeling of the observations with numerical radiative transfer disk models . these models will not only need to consider the high accretion rates but also such effects as self - irradiation of the disk . it will be interesting to see whether these models are able to confirm the apparent rotation of the disk s position angle with wavelength and whether they can reproduce the increasing visibility curve for the ut3-ut4 baseline . also , these disk models will have to reproduce the observed nir and mir visibilities simultaneously with higher accuracy than the current simple models are able to do . from the observational side re - measuring the ut3-ut4 baseline might be eligible and any other new baseline configuration will certainly help to constrain the disk geometry ( especially the position angle ) even better . concerning the dust structure the lack of crystalline silicates needs to be analyzed in greater detail . is it still just a contrast effect that we do not see any crystalline features in the spectra or are they simply not there ? eventually we will have to face the problem that currently all fuors that have been observed with high spatial resolution techniques draw a rather inhomogeneous picture of the group . apparently , some are still surrounded by nearby dense envelopes while for others the circumstellar disks suffice to explain the observations . however , even those disk - systems ( i.e. , fu ori and v1647 ori ) seem to be different in terms of physical properties as different temperature and surface density profiles are derived . without any doubt interferometric observations of circumstellar disks in the nir and mir provide an unprecedented means for deriving geometrical disk properties and put new constraints on the physical ( and also chemical ) processes taking place within protoplanetary disks . and as more and more interferometric data of circumstellar disks are published our view on the cradle of planetary systems will be once more refined . s. p. quanz kindly acknowledges financial support by the german _ friedrich - ebert - stiftung_. th . henning and j. bouwman acknowledge financial support by the ec - rtn on `` the formation and evolution of young stellar clusters '' ( hprn - ct-2000 - 00155 ) . we are most grateful to r. khler and f. lahuis for their support during the data reduction and to r. lachaume and f. malbet for providing mir data derived from their disk models ( section 7.2 . ) . we thank our anonymous referee for a detailed and thorough review that improved the manuscript . this research has made use of the simbad database , operated at cds , strasbourg , france , and was partly based on observations made with the _ spitzer space telescope _ , which is operated by the jet propulsion laboratory , california institute of technology , under nasa contract 1407 . lllll 31.10.2004 & fu ori & 44.56 m ( ut2-ut3 ) & 46.54@xmath8 & - + & hd37160 & & & calibrator + 02.11.2004 & fu ori & 86.25 m ( ut2-ut4 ) & 84.24@xmath8 & - + & hd 31421 & & & calibrator + & hd 37160 & & & calibrator + & hd 50778 & & & calibrator + 04.11.2004 & fu ori & 56.74 m ( ut3-ut4 ) & 106.64@xmath8 & - + & hd 25604 & & & calibrator + & hd 20644 & & & calibrator + & hd 37160 & & & calibrator + & hd 50778 & & & calibrator + 29.12.2004 & fu ori & 44.80 m ( ut2-ut3 ) & 46.61@xmath8 & - + & hd 37160 & & & calibrator + & hd 94510 & & & calibrator + llr j [ mag ] & 6.30@xmath73 & 10.75@xmath74 + h [ mag ] & 5.64@xmath75 & 9.92@xmath76 + k [ mag ] & 5.25@xmath77 & 9.15@xmath78 + l [ mag ] & 4.18@xmath79 & 8.09@xmath80 + n [ mag ] & 2.75@xmath81 & 5.28@xmath82 + llll amorphous olivine & mgfesio@xmath83 & homogeneous & ( 1 ) + amorphous pyroxene & mgfesi@xmath84o@xmath85 & homogeneous & ( 1 ) + crystalline forsterite & mg@xmath84sio@xmath83 & inhomogeneous & ( 2 ) + crystalline enstatite & mgsio@xmath86 & inhomogeneous & ( 3 ) + amorphous silica & sio@xmath84 & inhomogeneous & ( 4 ) + lllll amorphous olivine & 0.17 & @xmath87 0.01 & 0.18 & 0.35 + amorphous pyroxene & @xmath87 0.01 & 0.07 & 0.57 & 0.64 + crystalline silicates & @xmath87 0.01 & @xmath87 0.01 & @xmath87 0.01 & @xmath87 0.01 + cccc 12.95 & 0.98@xmath60.02 & 0.60@xmath60.02&0.84@xmath60.02 + 12.83 & 0.93@xmath60.02 & 0.61@xmath60.03&0.83@xmath60.02 + 12.70 & 0.94@xmath60.03 & 0.63@xmath60.03&0.84@xmath60.02 + 12.57 & 0.96@xmath60.02 & 0.62@xmath60.02&0.84@xmath60.02 + 12.44 & 0.96@xmath60.02 & 0.62@xmath60.02&0.81@xmath60.02 + 12.31 & 0.94@xmath60.02 & 0.60@xmath60.01&0.81@xmath60.02 + 12.17 & 0.92@xmath60.01 & 0.61@xmath60.02&0.79@xmath60.02 + 12.03 & 0.94@xmath60.02 & 0.61@xmath60.02&0.81@xmath60.02 + 11.89 & 0.93@xmath60.03 & 0.62@xmath60.02&0.79@xmath60.02 + 11.75 & 0.94@xmath60.03 & 0.62@xmath60.01&0.78@xmath60.02 + 11.60 & 0.94@xmath60.03 & 0.63@xmath60.01&0.77@xmath60.02 + 11.45 & 0.91@xmath60.02 & 0.63@xmath60.01&0.79@xmath60.02 + 11.30 & 0.94@xmath60.03 & 0.62@xmath60.01&0.78@xmath60.02 + 11.14 & 0.92@xmath60.03 & 0.62@xmath60.01&0.76@xmath60.02 + 10.98 & 0.93@xmath60.02 & 0.62@xmath60.01&0.77@xmath60.02 + 10.82 & 0.92@xmath60.02 & 0.62@xmath60.01&0.76@xmath60.01 + 10.66 & 0.92@xmath60.02 & 0.62@xmath60.01&0.75@xmath60.01 + 10.49 & 0.91@xmath60.01 & 0.62@xmath60.01&0.75@xmath60.02 + 10.32 & 0.92@xmath60.02 & 0.63@xmath60.01&0.75@xmath60.02 + 10.15 & 0.92@xmath60.01 & 0.63@xmath60.01&0.75@xmath60.02 + 9.97 & 0.94@xmath60.02 & 0.65@xmath60.01&0.73@xmath60.01 + 9.80 & 0.97@xmath60.02 & 0.64@xmath60.01&0.73@xmath60.01 + 9.62 & 0.97@xmath60.03 & 0.64@xmath60.01&0.74@xmath60.01 + 9.43 & 0.93@xmath60.02 & 0.61@xmath60.01&0.76@xmath60.02 + 9.25 & 0.93@xmath60.01 & 0.67@xmath60.01&0.76@xmath60.02 + 9.06 & 0.93@xmath60.01 & 0.69@xmath60.01&0.74@xmath60.02 + 8.86 & 0.94@xmath60.01 & 0.69@xmath60.01&0.74@xmath60.02 + 8.67 & 0.95@xmath60.01 & 0.70@xmath60.01&0.73@xmath60.02 + 8.47 & 0.95@xmath60.01 & 0.70@xmath60.01&0.74@xmath60.02 + 8.27 & 0.94@xmath60.01 & 0.73@xmath60.01&0.73@xmath60.01 + llll ut2-ut3 ( 44.56 m ) & 2.47@xmath88 & 3.27@xmath89 & 3.43@xmath90 + ut2-ut4 ( 86.25 m ) & 3.25@xmath91 & 4.34@xmath92 & 4.93@xmath93 + ut3-ut4 ( 56.74 m ) & 4.19@xmath94 & 5.00@xmath95 & 4.83@xmath96 + semimajor axis @xmath97 [ au ] & 2.28 & 2.51 & 2.54 + semiminor axis @xmath98 [ au ] & 1.21 & 1.54 & 1.42 + ellipse area [ au@xmath99 & 8.67 & 12.14 & 11.33 + position angle @xmath45 & 122.2@xmath64 & 111.1@xmath64 & 93.9@xmath64 + inclination @xmath15 & 58.0@xmath64 & 52.2@xmath64 & 56.0@xmath64 + lll inner disk radius & @xmath100 & @xmath101 + outer disk radius & @xmath102 & 100 au + disk inclination & @xmath15 & @xmath103 + disk position angle & @xmath45 & @xmath104 / @xmath105 + extinction & a@xmath106 & @xmath107 + temperature at 1 au & @xmath108 & @xmath109 + power law index for temperature & @xmath110 & @xmath111 + power law index for surface density & @xmath112 & @xmath113 + disk mass & @xmath114 & @xmath115 + lcllcc pti / ns & h & 103.7 & 61.8 & 0.83@xmath60.04 & 0.80 / 0.82 + pti / nw & h & 85.6 & 14.1 & 0.79@xmath60.05 & 0.92 / 0.94 + pti / ns & k & 102.7 & 63.6 & 0.72@xmath60.08 & 0.78 / 0.80 + pti / nw & k & 84.2 & 14.0 & 0.79@xmath60.05 & 0.91 / 0.93 + pti / sw & k & 82.5 & -55.6 & 0.82@xmath60.08 & 0.85 / 0.87 + vlti / ut1-ut3 & k & 89.6 & 54.2 & 0.87@xmath60.05 & 0.83 / 0.85 +

one of the most important diagnostics in understanding the evolution of galaxies is the star - formation rate ( sfr ) . the evolution of the sfr of galaxies can give clues as to how galaxies were enriched with heavy elements , how they build up their stellar mass through cosmic time , and helps us to understand the bolometric output of galaxies . at redshift @xmath7 , when the universe was just @xmath8 3gyr old , star - formation activity in the universe was at its peak and galaxies were in the process of assembling most of their stellar mass ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? studying this critical epoch is essential to gaining a better understanding of the evolution of the progenitors of the local galaxy population . the ultra - violet ( uv ) continuum ( 1500 to 2800 ) intensity of a galaxy is one of the most commonly used diagnostics for the sfr as it is observable over a wide range of redshifts and intrinsic luminosities . it is sensitive to massive stars ( @xmath9 ) , making it a direct tracer of current sfr . by extrapolating the formation rate of massive stars to lower masses , for an assumed form of the initial mass function ( imf ) , one can estimate the total sfr @xcite . another widely used diagnostic for measuring the sfr is nebular emission , with h@xmath1 being the most common because of its higher intensity compared to the other hydrogen recombination lines such as h@xmath10 , pa@xmath1 , pa@xmath10 , etc . , and it is easier to interpret than the ly@xmath1 line . h@xmath1 is an `` instantaneous '' tracer of sfr because it is sensitive only to the most massive stars ( @xmath11 ) . however , it becomes more challenging to observe h@xmath1 from the ground at @xmath12 because the line is redshifted to the near - ir where the terrestrial background is much higher than at optical wavelengths . the main disadvantage of using uv / optical luminosities as tracers of the sfr is their sensitivity to dust attenuation . the dust absorption cross - section is larger for shorter wavelengths and choosing the appropriate attenuation curve to correct the observed luminosities plays an important role in determining intrinsic physical quantities . aside from the assumed attenuation curve , the geometry of dust with respect to the stars can lead to different color excesses , @xmath13 , between the ionized gas and the stellar continuum . @xmath13 is the color excess measured between the _ b _ and _ v _ bands , @xmath14 , where @xmath15 is the total extinction at wavelength @xmath16 in magnitudes . in particular , the nebular recombination lines arise from the hii regions around the most massive o and early - type b stars ( with masses of m@xmath17 and main sequence lifetimes of @xmath18myr ) . on the other hand , for a salpeter imf , solar metallicity , and a constant or rising star - formation history , the uv continuum in starburst galaxies originates from stars over a broader range of mass that includes later - type b stars with lifetimes @xmath19myr @xcite . these older non - ionizing stars have more time to migrate to regions of lower dust density in the galaxy , while h - ionizing stars with shorter lifetimes do not have enough time to escape from their dusty birthplace or let the parent molecular clouds to dissipate . as a result , the nebular lines can be subject to a higher degree of reddening than the uv continuum . @xcite found that the nebular emission is more attenuated than the stellar continuum at the same wavelengths for a sample of local uv - bright galaxies . subsequently , @xcite studied a similar sample of local galaxies and argued that the _ color - excess _ is 2.27 times larger for the nebular emission lines than for the stellar continuum . this relationship was derived under the assumption that the calzetti curve is applied to the stellar continuum and a galactic extinction curve is applied to the nebular emission lines . in a separate study , @xcite found that for an sdss sample of @xmath20 galaxies , @xmath21 , both assuming a @xcite extinction curve . there have been several studies of relative attenuation of the stellar continuum and nebular regions in high - redshift galaxies . in most of the high - redshift studies , the calzetti curve is assumed for both the nebular and stellar regions . frequently , the @xcite relation between the stellar and nebular @xmath13 is used to correct the nebular lines , while this relation was originally derived based on using a galactic extinction curve for the nebular lines . at present it is unclear what type of attenuation curve should be applied to the nebular regions . these issues are discussed in detail in @xcite . at @xmath7 , @xcite selected a sample from various imaging surveys in the optical , near - ir , mid - ir , and sub - millimeter regime , with sed - derived sfrs of @xmath8 10 - 1000 @xmath22 and stellar masses @xmath23 , assuming a chabrier imf . in their study , the predicted h@xmath1 luminosities from the best - fit sed models for 62 galaxies were compared with the observed line luminosities measured from integral field spectroscopy . the latter were then corrected for reddening assuming the @xmath24 returned from stellar population modeling of their galaxies . @xcite found that assuming a factor of two redder color excess towards the hii regions relative to the stellar continuum yielded the best agreement between the predicted and observed h@xmath1 luminosities . similarly , @xcite investigated 25 star forming galaxies at @xmath7 and compared the h@xmath1 sfrs , dust - corrected by the calzetti attenuation curve , with the sed modeled sfrs . they showed that h@xmath1 sfrs are consistent with the sed sfrs , provided that @xmath25 when assuming the calzetti curve for both . @xcite presented two galaxies at redshifts @xmath26 and 2.024 , and added ten other galaxies at similar redshifts from @xcite , @xcite , and @xcite to compare a@xmath27 derived from the balmer decrement ( @xmath28 ) with a@xmath29 derived from the sed models , both assuming the calzetti curve . the only two galaxies with detected h@xmath10 line measurements showed evidence for higher nebular attenuation , but about half of the non - detections were consistent with equal attenuation for the nebular and stellar regions . in @xcite , h@xmath1 attenuation was investigated through several comparisons for a sample of optically - selected star - forming galaxies at @xmath7 . aided with near - infrared spectroscopic observations , @xcite used the balmer decrement along with the calzetti curve to correct the h@xmath1 sfr . the dust - corrected sfr was then compared with the h@xmath1 sfr corrected by the sed color - excess and also with the sfrs inferred from uv , ir , radio , and x - ray . the comparisons yield contradictory results . their analysis favored a higher nebular color - excess for galaxies with larger intrinsic sfrs . in separate studies , @xcite and @xcite used a large sample of uv - selected galaxies at @xmath7 and corrected both the h@xmath1 and uv sfrs with the color - excess derived from the best - fit sed models and applied the calzetti attenuation curve . they showed that using different color excesses for the continuum and lines generally resulted in h@xmath1-inferred sfrs that over - predicted those calculated from the dust - corrected uv emission and direct ir measurements . the apparently conflicting results mentioned above may be reconciled if the relation between the nebular and stellar color - excesses depends on the physical properties of galaxies , such as their sfrs @xcite or specific sfrs @xcite . for example , @xcite showed that the ratio of the line optical depth to the continuum optical depth decreases with increasing ssfr and at ssfr@xmath30 the line - to - continuum optical depth ratio reaches the @xcite ratio assuming the @xcite attenuation curve . @xcite investigated the attenuation of the nebular regions at @xmath31 using the balmer decrement from stacked _ hst _ grism spectra assuming the calzetti curve , and found no strong trend of @xmath32 with sfr or ssfr . in @xcite study , at ssfr@xmath33 the ratio is consistent with 1 . the primary goal of this paper is to understand the relationship between the uv and h@xmath1 emission in high - redshift star - forming galaxies , with a large sample that is immune to uncertainties in slit loss corrections that affects the h@xmath1 flux estimation ( c.f . * ? ? ? * ; * ? ? ? * ) , the small sample sizes inherent in previous spectroscopic studies of h@xmath1 @xcite , and not subject to the bias of selecting high equivalent width objects from narrow - band selected samples @xcite . we consider in our analysis only spectroscopically - confirmed galaxies , enabling us to estimate the h@xmath1 flux from broad - band photometric excess techniques without the additional uncertainties that plague photometric redshifts @xcite . the impact of nebular lines on the broad - band photometry was known and studied for many years . using the photometric excesses to determine the line strengths was pioneered by @xcite , where they showed that the excess in _ 3.6@xmath3 m relative to the sed model continuum is due to the redshifted h@xmath1 emission line for their sample of galaxies at @xmath34 . @xcite also investigated a sample of galaxies at the same redshift range of @xcite and inferred the h@xmath1 emission line strengths by comparing the observed flux in _ 3.6@xmath3 m band and the continuum flux as expected from the sed model . following that , @xcite verified the photometric excess technique by applying it to a small sample of 9 galaxies at @xmath35 ; for 8 galaxies the [ o iii ] line fluxes inferred by the same technique as @xcite were within a factor of 2.5 of the spectroscopically measured [ o iii ] fluxes . an additional advantage of our study is that we include ir data to independently assess the dust - obscured sfr . comparing h@xmath1 , uv , and ir - inferred sfrs in a statistical sense allows us to understand how to correct extinction - sensitive measures of sfr for the effects of dust . the outline of this paper is as follows . in [ sec : sample ] we discuss the properties of our sample , the assumptions that have been made to model the stellar populations using the rest - frame uv to near - ir photometry , the _ spitzer_/mips photometry and stacking method . a detailed description of how we estimated the h@xmath1 line flux is provided in [ sec : sec3 ] . the analysis of the mips 24@xmath3 m data and ir luminosities is described in [ sec : irlum ] . in [ sec : sec5 ] we compare the two different tracers of sfr - h@xmath1 and uv - and discuss the dust correction recipe most consistent with our measured values . [ sec : sec6 ] focuses on combining sfr diagnostics ( e.g. , h@xmath1 and uv with ir ) to deduce bolometric sfrs . the results are summarized in [ sec : sum ] . throughout this paper , a @xcite imf is assumed and a cosmology with h@xmath36 = 70 km s@xmath37 mpc@xmath37 , @xmath38 = 0.7 , and @xmath39 = 0.3 is adopted . all magnitudes are given in the ab system @xcite . the galaxies used in this study are drawn from a parent sample that is part of an imaging and spectroscopic survey of uv - selected galaxies at z @xmath8 2 - 3 @xcite . the galaxies were selected based on the bx , bm , and lyman break galaxy rest - uv color criteria @xcite , where @xmath40 optical data were obtained with the palomar large format camera ( lfc ) or keck low resolution imaging spectrograph ( lris ; * ? ? ? * ; * ? ? ? * ) . rest - uv spectroscopic follow - up with keck / lris was conducted for galaxies brighter than @xmath41 @xcite . near - ir @xmath42 and @xmath43 imaging was obtained using the palomar / wirc and magellan / panic instruments @xcite . _ h_-band ( f160w ) data were obtained with the _ hubble space telescope _ wfc3 camera @xcite . all galaxies in the sample have coverage in at least one of the _ spitzer_/irac four channels ( 3.6 , 4.5 , 5.8 , and 8.0 @xmath3 m ; @xcite ) . the objects are located in the goods - north field and 11 additional fields that were primarily selected to have one or more relatively bright background qsos for studying the interface between the intergalactic medium ( igm ) and galaxies at z @xmath8 2 - 3 @xcite . out of the final sample of 2283 objects with spectroscopically confirmed redshifts , a subset of galaxies is selected based on the following criteria : ( 1 ) the galaxy must be covered by the _ k_-band imaging , ( 2 ) it must have a redshift in the range 2.08 @xmath44 z @xmath44 2.51 so that the h@xmath1 line falls into the _ @xmath2 _ band , and ( 3 ) it must be detected in at least two of the irac channels or one of the irac channels and either j or f160w bands . the third condition ensures a more robust estimate of the stellar mass and the continuum level at 6564 . furthermore , agns ( making up @xmath45 of the parent sample ) were identified by either strong uv emission lines ( e.g. , ly@xmath1 , civ ) or by a power law sed through the irac bands . these agns are removed from our sample . eventually , 262 galaxies remain that satisfy the aforementioned criteria . for each galaxy in our sample , the best - fit stellar population model is found by using the rest - frame uv through near - ir broad - band photometry . as mentioned above , all the galaxies in our sample have spectroscopically - confirmed redshifts , thus removing a key degeneracy in the modeling of the stellar populations . in addition , for a better estimation of stellar mass and age , all galaxies in our sample have at least two detections long - ward of the 4000 break - excluding _ @xmath2_-band . c. charlot & g. bruzual ( 2007 ) models with a @xcite imf and solar metallicities are used for the fitting . for each individual galaxy , different star - formation histories are assumed , including constant , exponentially declining , and exponentially rising with characteristic timescales of @xmath46 = 10 , 20 , 50 , 100 , 200 , 500 , 1000 , 2000 , 5000myr for exponentially declining and @xmath46 = 100 , 200 , 500 , 1000 , 2000 , 5000myr for exponentially rising histories . ages are allowed to vary between 50myr and the age of the universe at the redshift of each galaxy . the @xmath47 50myr limit corresponds to the typical dynamical timescale of star - forming galaxies at @xmath7 as inferred from velocity dispersion and size measurements of these galaxies @xcite . for interstellar dust obscuration , the @xcite attenuation curve is used , allowing @xmath13 to vary between 0.0 and 0.6 . the @xmath48 values have been determined for each set of observed broad - band and model magnitudes . the best - fit model is determined through @xmath48 minimization . there is generally no significant difference between the best - fit @xmath48 values of the six different population models ( constant , exponentially rising , and exponentially declining star - formation histories , for each considering all ages and ages greater than 50myr ) . as previous studies have shown , the assumption of declining star - formation histories at these redshifts results in systematically lower sed - inferred sfrs compared to the observed ir+uv sfrs @xcite . furthermore , @xcite showed that assuming a constant star - formation history for @xmath7 galaxies predicts specific sfrs ( sfr / m@xmath49 ) at higher redshifts ( @xmath50 ) that are substantially larger than the observed values . given these , we adopt the models that assume exponentially rising star - formation histories with ages greater than 50myr . the sed models used to fit the observed magnitudes do not include nebular emission lines . for example , the h@xmath1 line can significantly affect the photometry and we use its contribution to the _ k_-band to estimate the h@xmath1 line flux . the [ o iii ] emission line is the other strong line which falls into the f160w filter given the redshift range of our galaxies . only 20% of the galaxies have f160w observations , for which we did not correct the broadband photometry for the contamination . the sed - inferred sfr of these galaxies is consistent with the sfr(uv ) estimates within the uncertainties . at this redshift , j band is contaminated by the [ o ii ] emission line , but this line is generally weaker than the h@xmath1 line and its effect on the sed inferred sfrs is negligible compared to the uncertainties . , the typical [ o ii ] line flux is @xmath51 erg s@xmath37 @xmath52 @xcite . the ratio of the [ o ii ] flux to the typical j - band flux errors in our observations is only @xmath53 . ] to further investigate the bolometric properties of our sample , we use _ spitzer_/mips 24@xmath3 m wherever available . out of 12 fields , goods - north @xcite , and four other fields ( q1549 , q1623 , q1700 , and q2343 ; reddy & steidel 2009 ) have mips 24@xmath3 m coverage to a typical 3@xmath54 depth of 10 - 15@xmath3jy . photometry on 24@xmath55 images is performed by using point - spread function ( psf ) fitting with priors determined by the locations of the objects in the higher resolution irac images ( irac data exist in all fields ) . a 40@xmath5640 pixel region centered on each target is extracted with pixel size of @xmath57 . psfs are then fitted simultaneously to all known sources in the sub - image and one random background position . this procedure is repeated many times to obtain sufficient statistics for proper background estimation based on the random background flux measurements . the other source of uncertainty is poisson noise , which for objects in our sample is negligible compared to the background dispersion . we remove objects whose photometry may be compromised due to blending with nearby sources , using the criteria specified in @xcite . this results in 115 galaxies with secure psf fits . out of 115 objects with mips data , 47 are detected with s / n @xmath58 . undetected objects , those with s / n @xmath59 , are considered using either survival analyses or through stacking of the 24@xmath55 data . throughout the paper , we employ a stacking method to determine the median 24@xmath3 m fluxes of objects that are individually detected and undetected , following the procedures described in @xcite . we performed aperture photometry on the stacked images , and applied an aperture correction based on the @xmath60@xmath3 m psf . the average background level and noise were determined by placing many apertures of the same size used for the stacked signal at random positions in the stacked image and measuring the average flux level and dispersion in flux of these `` background '' apertures . furthermore , we used bootstrap resampling simulations to estimate the dispersion in the fluxes of objects contributing to each stack . this was accomplished by creating 100 samples of random images in each bin and measuring the standard deviation of the median stacked fluxes . the intrinsic dispersion in the stacked flux is larger than the background error by a factor of @xmath61 . using the procedure discussed in [ sec : sedfit ] , we fit all the photometry , excluding _ @xmath2_-band , in order to determine the continuum level at 6564 . the continuum magnitude is calculated by multiplying the best - fit sed model by the _ @xmath2 _ filter transmission curve . the difference between the observed _ @xmath2_-band magnitude and the continuum magnitude is used to extract the h@xmath1 line flux as follows . the observed _ @xmath2 _ magnitude is considered as the sum of the continuum and the h@xmath1 fluxes , while the sed - inferred _ @xmath2 _ magnitude represents only the continuum flux . assuming a gaussian form for the redshifted h@xmath1 line , the change between these fluxes will yield the h@xmath1 flux ( figure [ fig : sed ] ) : @xmath62 here , @xmath63 is the flux density of the h@xmath1 line in units of ergs@xmath37@xmath52hz@xmath37 , @xmath64 is the _ @xmath2 _ filter transmission curve , and @xmath65 and @xmath66 are the observed and the continuum magnitudes , respectively . the h@xmath1 flux is corrected for contamination from the [ nii ] line based on the mass - dependent [ nii]-to - h@xmath1 flux ratios of @xcite . the stellar masses are determined from the sed models and the corresponding [ nii ] line contamination , as listed in table [ tab : nii ] , is used to correct the h@xmath1 flux . the sfrs are then calculated using the @xcite relation to convert the h@xmath1 line luminosity to an sfr . there are three main sources of uncertainty in the derived h@xmath1 fluxes . the largest uncertainty is the photometric error . the typical error in the observed _ @xmath2_-band magnitude is @xmath67 . the second source of uncertainty is the [ nii ] correction . the uncertainty on the @xcite [ n ii]-to - h@xmath1 line ratios ( see table [ tab : nii ] ) is added in quadrature to the photometric error . the third source of uncertainty is the error associated with the continuum flux . in order to account for this uncertainty , we estimated the continuum flux at 6564 from the best - fit model assuming six different star - formation histories : exponentially rising , exponentially declining , and constant , for each considering all ages and ages greater than 50myr . for 94% of the galaxies the difference in the mean of the continuum fluxes assuming different star - formation histories to those assuming a rising star - formation history with ages @xmath68 50myr is less than 0.1 magnitude . the average error in the estimated continuum magnitude is @xmath69mag , which is negligible compared to the observed _ @xmath2_-band magnitude errors . combining the three sources of uncertainty discussed above yields a typical relative error in h@xmath1 flux of @xmath70 . cc @xmath71 & @xmath72 + 0.88 2.00 & @xmath73 + 2.00 3.69 & @xmath74 + 3.69 6.03 & @xmath75 + 6.03 8.82 & @xmath76 + @xmath77 & @xmath78 [ tab : nii ] the equivalent width of the h@xmath1 line is estimated by dividing the h@xmath1 flux derived from equation [ equ : ha ] by the continuum flux density at the wavelength of the h@xmath1 line : @xmath79 the @xmath80 is the continuum flux density in units of erg s@xmath37 @xmath52 @xmath37 that is estimated through the best - fit sed model , @xmath81 is the h@xmath1 line flux in erg s@xmath37 @xmath52 , @xmath82 is the redshift of the galaxy , and @xmath83 is the rest - frame equivalent width in . we define whether a galaxy has a `` detected '' h@xmath1 line according to the following . galaxies whose _ @xmath2_-band photometry exceeds the continuum level by more than the _ @xmath2_-band magnitude error are referred to as `` h@xmath1 detections '' . galaxies where the _ @xmath2_-band photometry is consistent with the continuum level to within @xmath84@xmath54 are referred to as `` h@xmath1 non - detections '' , and an upper limit of @xmath85 above the measured k photometry is used for these objects . out of 262 galaxies in our sample , 149 are detections and 94 are non - detections . there are 19 objects with _ @xmath2 _ magnitudes fainter than the continuum by more than @xmath85 . we removed these galaxies from our discussion due to their _ @xmath2 _ photometry being inconsistent with the photometry from adjacent bands . rest - frame equivalent widths as a function of the error of the observed _ @xmath2_-band magnitude . the circles are detected galaxies and are color coded based on their observed h@xmath1 fluxes ( uncorrected for extinction ) . the red triangles show @xmath85 upper limits . detecting the h@xmath1 emission requires that the observed _ @xmath2 _ magnitude lies more that @xmath85 above the underlying continuum , where @xmath85 is defined by the _ @xmath2 _ magnitude error . the _ @xmath2 _ magnitude includes the light both from the h@xmath1 emission line and the brightness of the continuum at 6564 . there is a tight correlation between the continuum brightness and the _ @xmath2 _ magnitude error ; the fainter continuum results in larger uncertainty in the observed magnitude . as a result , there are more undetected objects with large _ @xmath2_-band errors . on the other hand , the brightest h@xmath1 objects have bright continuum and small _ @xmath2_-band uncertainties . the deficiency of objects in the upper - left corner is mainly because very bright h@xmath1 objects that also have bright continuum are rare.,scaledwidth=50.0% ] as discussed in [ sec : intro ] , our method of computing h@xmath1 fluxes and ews has the advantage of being immune to slit - loss corrections . our sample covers a wide range of h@xmath1 fluxes and ews . figure [ fig : ew - omagerr ] shows the distribution of h@xmath1 ews and fluxes as a function of the _ @xmath2 _ magnitude @xmath85 uncertainty . the detection of h@xmath1 with the method adopted here depends on both the h@xmath1 line flux and the brightness of the continuum at 6564 . as we have defined detections to be those objects where the _ @xmath2 _ magnitude exceeds the continuum level by more than @xmath85 , the number of undetected objects increases for objects that are fainter in the continuum ( figure [ fig : ew - omagerr ] ) . fluxes of the objects in common between our sample and that of @xcite . the diamonds are the detected objects and downward arrows show @xmath86 upper limits , as discussed in the text . the fluxes are measured based on the difference between the observed and the continuum _ @xmath87 _ magnitudes predicted from the best - fit model . the error bars are dominated by the uncertainty in the observed _ @xmath87 _ photometry . the @xcite spectroscopic fluxes are multiplied by a factor of 2 to account for slit losses . the dashed line denotes the one - to - one relationship.,scaledwidth=50.0% ] the h@xmath1 observed fluxes of 44 objects in common with the @xcite spectroscopic sample are compared in figure [ fig : fluxcompare ] . the @xcite spectroscopic fluxes are multiplied by a factor of two to account for slit losses and the aperture used to extract the spectra @xcite . the slit loss correction factor applied to the spectroscopic fluxes depends on various factors such as the accuracy of the astrometry , the size of the object convolved with the seeing at the time of observation in comparison with the size of the slit , and the accuracy of the mask alignment during the observation . as noted in @xcite , the factor of two slit - loss correction is an average estimate and the slit losses will of course vary from object to object , likely accounting for some of the scatter in figure [ fig : fluxcompare ] . on the other hand , while broadband photometry has its advantages , our _ @xmath2_-band measurements suffer from a larger measurement ( random ) uncertainty relative to the other methods given the depth of our ground - based images . although most of the detected h@xmath1 fluxes in this study ( the black symbols in figure [ fig : fluxcompare ] ) are higher that the spectroscopic fluxes , about half of the sample is estimated as upper limits ( the red arrows ) . considering both the detections and non - detections , the broad - band estimated fluxes are generally consistent with the spectroscopic fluxes . to quantify the degree of the correlation we conduct a generalized kendall s tau ( @xmath46 ) statistic . the generalized kendall s @xmath46 rank correlation coefficient is a non - parametric test that permits non - detections in both variables ( here , our estimated h@xmath1 fluxes ) . kendall s @xmath46 correlation coefficient for our flux estimation with the @xcite measurements , excluding the one outlier object , is @xmath46 = 0.17 , with significance of p - value = 0.11 . uv luminosities are calculated using the fluxes of the best - fit sed models at 1700 . the @xcite relation to convert the uv luminosity ( over the wavelength range 1500 - 2800 ) to sfr applies only to galaxies where star formation proceeds for @xmath88myr . however , there are galaxies in our sample with inferred ages between 50 and 100myr . for these younger galaxies , we used an age- and star - formation - history - dependent relation in the conversion of the uv luminosity to sfr , as discussed in @xcite . the age - dependent conversion resulted in uv sfrs that are at most @xmath89 larger than the @xcite sfrs for young galaxies . throughout this analysis , we also use sfrs inferred from the best - fit sed models . as expected , within the uncertainties , the sed sfrs are highly correlated with dust corrected uv sfrs . to estimate the bolometric luminosities of galaxies in our sample , we infer the infrared luminosity ( integrated between 8 - 1000 @xmath3 m ) as discussed below , and add this to the unobscured uv luminosity ( i.e. , observed uv luminosity ; [ sec : uvlum ] ) , as in @xcite . infrared luminosities , @xmath90 , are estimated by using the _ spitzer_/mips 24@xmath3 m observations . the @xmath60@xmath3 m band is sensitive to the rest - frame 7.7 @xmath3 m pah emission , which correlates with @xmath90 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? to convert observed 24@xmath3 m magnitudes to total ir luminosity ( l@xmath91 ) , we used several dust sed templates , including those of @xcite , @xcite , and @xcite . luminosities determined from the rest - frame 8 @xmath3 m flux density alone , regardless of the dust template used , tend to over predict l(ir ) , particularly for lirgs ( e.g. , * ? ? ? * ) and ulirgs ( e.g. , * ? ? ? * ; * ? ? ? * ) . in order to account for the luminosity - dependent overestimation of the derived l(ir ) , we use a correction equation described in @xcite . the correction is calculated by comparing the l(ir ) computed from 24 , 100 , 160@xmath3 m , and 1.4ghz fluxes , and the l(ir ) determined solely from 24@xmath3 m data for a similarly - selected sample of galaxies : @xmath92 all luminosities are in l@xmath93 . equation[equ : it - tot ] is based on the @xcite models . there is a factor of @xmath61 variation in l(ir ) derived from 24@xmath3 m data using @xcite and @xcite dust templates ( e.g. , see * ? ? ? however , once corrected using the appropriate equations similar to equation[equ : it - tot ] , all the templates result in l(ir ) estimates that are consistent with each other and with the l(ir ) computed from 24 , 100 , 160@xmath3 m , and 1.4ghz fluxes , within the uncertainties of the measured ir luminosity as explained in [ sec : sample - b ] ( see also * ? ? ? in the subsequent analysis , we use infrared luminosities inferred from the @xcite templates . the ir luminosity is converted to ir ( dust - obscured ) sfrs using the @xcite relation . the sum of ir- and uv - inferred sfrs are then used to estimate the bolometric sfrs . sfrs inferred from the uv luminosity and the h@xmath1 fluxes are shown in the left panel of figure [ fig : sfrhauv ] . without dust corrections , the uv and h@xmath1 sfrs have a large discrepancy due to the smaller dust absorption cross - section at 6564 relative to that at 1700 ( see the inset panel in figure [ fig : sfrhauv ] ) . the main left panel of figure [ fig : sfrhauv ] shows h@xmath1 and uv sfrs , both corrected for dust attenuation based on the calzetti attenuation curve and assuming that the same color - excess , @xmath13 , as derived from the best - fit sed model , applies to the stellar continuum and the nebular regions . to better quantify the mean trend , measurements have also been performed in bins of uv dust - corrected sfr ( filled circles in figure [ fig : sfrhauv ] ) . the bins are @xmath94log(sfr@xmath95 ) = 0.3dex wide . for all the galaxies in each bin , regardless of their redshift , we calculate the median flux densities in the observed filters . with the new set of median fluxes , the best - fit stellar population model is determined through @xmath48 minimization for the mean redshift of the galaxies in each bin . the h@xmath1 and uv luminosities are then derived from the stacked sed in the same way as for individual galaxies ( see [ sec : sec3 ] ) . reported errors are calculated based on the @xmath86 standard error of the mean of the _ @xmath2 _ magnitudes contributing to each bin . ccccccc @xmath96 & 26 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & @xmath101 + @xmath102 & 63 & @xmath103 & @xmath104 & @xmath105 & @xmath106 & @xmath107 + @xmath108 & 83 & @xmath109 & @xmath110 & @xmath111 & @xmath112 & @xmath113 + @xmath114 & 56 & @xmath115 & @xmath116 & @xmath117 & @xmath118 & @xmath119 + @xmath120 & 15 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & @xmath125 [ tab : sfr ] the appropriate dust corrections to apply to the nebular emission lines is still a subject of debate , as discussed in section [ sec : intro ] . the attenuation of the nebular lines and the uv stellar continuum are not completely decoupled as both arise from dust around massive stars . on the other hand , whether the color - excess is the same for the nebular and stellar regions ( e.g. , * ? ? ? * ; * ? ? ? * ) or not ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) in high - redshift galaxies has yet to be fully investigated . @xcite found that @xmath126 for a sample of local star - forming galaxies . in the absence of direct measurements of the nebular color - excess ( e.g. , via the balmer decrement ) , this relationship can in principle be used to estimate the nebular color excess and apply a dust correction to the h@xmath1 line ; e.g. , @xmath127 where @xmath128 is the @xcite galactic extinction curve , assuming that the nebular regions in the high - redshift galaxies abide by such a dust curve . @xmath129 is the absolute extinction of the h@xmath1 emission line , and @xmath130 is the sed - inferred stellar color excess . we investigate several of the more commonly used dust - correction recipes , as described below . the h@xmath1 sfrs are corrected using five different methods : 1 . the calzetti attenuation curve for both gas and stars , but 2.27@xmath56 larger color - excess for the nebular lines than the stellar continuum ( as we call it `` different @xmath13 '' ) , 2 . the calzetti attenuation curve for the stellar continuum , and the @xcite galactic curve for the nebular lines assuming 2.27@xmath56 larger color - excess , 3 . the calzetti attenuation curve for the stellar continuum , and an smc extinction curve @xcite for the nebular lines assuming 2.27@xmath56 larger color - excess , 4 the calzetti attenuation curve for both gas and stars , and the same color - excess ( as we call it `` equal @xmath13 '' ) , and 5 . the calzetti attenuation curve for the stellar continuum , an smc extinction curve @xcite for the nebular lines assuming the same color - excess . we choose to compare these different dust - correction scenarios for the stacked values instead of the individual galaxies as the former represent an average over many points including both detected and undetected quantities , each of which may be relatively uncertain . the results are plotted in the right panel of figure [ fig : sfrhauv ] and are reported in table [ tab : sfr ] . to summarize , these are the values that are used to correct the observed h@xmath1 sfrs in each model according to equation [ diffatt ] : @xmath131 in these equations , @xmath13 is the sed - inferred color - excess observed for the stellar continuum and a@xmath132 is the total nebular extinction at 6564 . in model ( a ) with a larger color excess for the nebular regions with the calzetti curve , h@xmath1-inferred sfrs are significantly larger than uv sfrs . taking the average of the sfr(h@xmath1 ) ( model a ) to the corrected sfr(uv ) in the five bins of sfr(uv ) indicated in tables [ tab : sfr ] yields a value of @xmath133 which shows a @xmath134 discrepancy from unity . using either model ( c ) or ( d ) , results in sfr@xmath135 that are in good agreement with sfr@xmath136 . model ( b ) reproduce systematically higher sfrs but is still consistent with the range of sfr@xmath136 in higher bins , and model ( e ) estimations are lower than expected . based on the data provided in table [ tab : sfr ] and the right panel of figure [ fig : sfrhauv ] , model ( d ) provides the best agreement between sfrs among the five models . the average of the sfr(h@xmath1)-to - sfr(uv ) ratios in the five sfr(uv ) bins of this model shows less than @xmath85 discrepancy from unity . this suggests that _ on average _ the best recipe to correct the h@xmath1-inferred quantities of star - forming galaxies at @xmath7 is to use @xmath137 with the calzetti reddening curve . line luminosity as a function of bolometric sfr . the stars indicate the average values in bins of sfr(sed ) . galaxies with sfr(sed ) @xmath138 32 @xmath139 yr@xmath37 are in the first bin ; those with sfr(sed ) between 32 @xmath139 yr@xmath37 and 100 @xmath139 yr@xmath37 are in the middle bin ; and last bin consists of galaxies with sfr(sed ) @xmath47 100 @xmath139 yr@xmath37 . the error in sfr(uv)+sfr(ir ) is determined by the dispersion error in ir luminosity of the objects contributing to each stack . l(ir ) dispersion error is estimated through bootstrap resampling simulations . the solid blue line is the best linear fit . the slope of this fit gives us the conversion factor for l(h@xmath1)@xmath140-to - sfr@xmath141 relation . the shaded area is the 68% confidence interval of our estimated slope . the orange and dark - red dashed lines show the conversion factors of @xcite and @xcite , respectively . ] as discussed above , a primary disadvantage of using h@xmath1 and uv sfr diagnostics is that we must account for dust extinction , though this is less of a problem for h@xmath1 as it is for the uv . the attenuation curves ( e.g. , the @xcite curve ) that are used to correct the luminosities encode information regarding the dust grain size distribution and the geometrical distribution of dust with respect to stars ( e.g. , whether the dust is clumpy or uniform , a homogeneous mixture or a foreground screen , etc . ) . on the other hand , the ir luminosity is a direct tracer of dust and thus provides an independent and more robust diagnostic of dust attenuation . adding the ir luminosity , which accounts for the obscured star formation , to the unobscured ( uv and h@xmath1 ) tracers of sfr can give a more reliable estimate of the bolometric sfr . to convert the observed h@xmath1 luminosity to total sfr , the most commonly used calibration is that of @xcite , which is computed using an evolutionary synthesis model , assuming case b recombination with @xmath142=10,000k . in the absence of balmer decrement measurements , the observed h@xmath1 sfr must be corrected for dust with some assumption of the attenuation curve and color - excess of the ionized gas . in this section , we examine the l@xmath132-to - sfr conversion by relying only on the observed data , without making any assumptions about the dust correction or electron temperature of the ionized gas . with this conversion , one can use the h@xmath1 observed luminosity as a proxy for total sfr without assuming a dust attenuation curve and its associated uncertainties . the conversion applies to stellar populations with constant star formation at least over 100myr . for bursty star formation histories over timescales shorter than 100myr , the h@xmath1 and uv+ir luminosities diverge as the h@xmath1 luminosity traces more instantaneous star formation over timescales of @xmath143myr , while ir+uv is not sensitive to the change of star formation on timescales @xmath144myr . here , we are deriving the conversion factor based on the stacked data that represents the `` average '' quantities and thus should be used with caution for galaxies that may be undergoing bursty star formation . bolometric sfr ( sfr(ir)+sfr(uv ) ) is plotted as a function of the h@xmath1 luminosity in figure [ fig : lha ] . as both the h@xmath1 and ir measurements have non - detections , we decided to bin the data with respect to the sed - inferred sfr . we chose the bins so that in each bin the sfr(sed ) is consistent with the stacked sfr(ir)+sfr(uv ) within the uncertainties , because ultimately we are using sfr(ir)+sfr(uv ) to calculate the h@xmath1 luminosity - to - sfr conversion factor . mips data are stacked in each bin and the total ir luminosity is extracted in the same way as for individual galaxies . the uv and h@xmath1 luminosities are derived from the stacked seds . a weighted least squares regression method is used to find the slope of the relation : @xmath145 this relation can be used to convert the observed h@xmath1 luminosity in units of ergs@xmath37 to the bolometric sfr . the factor given in @xcite and @xcite , to convert observed / intrinsic luminosity to observed / intrinsic sfr , is 7.9 @xmath146 @xmath139 yr@xmath37 erg@xmath37 s. @xcite reported a conversion factor of 6.3 @xmath146 @xmath139 yr@xmath37 erg@xmath37 s. these factors are computed using evolutionary synthesis models and are subject to those models uncertainties as well as the initial assumptions that went into these models , such as imf , star - formation history , and the stellar evolution and atmosphere models . for the same galaxy type and assumed imf , @xcite reports @xmath147 variation among different calibrations , which mainly reflects the sensitivity to the sed modeling stellar evolution input data . in our analysis , converting l(ir ) and l(uv ) to sfr still rely on the model assumptions , but previous studies have shown that in the absence of the balmer decrement measurements , the sum of sfr(ir ) and sfr(uv ) is the most reliable estimate of total sfr @xcite . our derived slope for a salpeter imf is consistent with @xcite and @xcite within the errors . excluding galaxies with ages @xmath138 100myr ( see [ sec : sfrsed ] ) results in a best - fit slope of 7.8 @xmath148 yr@xmath37 erg@xmath37 s that is consistent with the slope found above . m ; open blue circles include all objects and filled red circles include only those galaxies older than 100myr . diamonds are individual detected objects and downward arrows represent @xmath149 upper limits in sfr(ir)+sfr(uv ) for the mips undetected galaxies . young galaxies ( ages @xmath138 100myr ) are colored blue and galaxies with ages @xmath47 100myr are orange . sfr(ir ) error bars are estimated through bootstrap simulations and show the dispersion in the sfrs contributing to each stack . the last bin consists of galaxies with sfr(sed ) @xmath47 100 @xmath139 yr@xmath37 ; by removing young galaxies ( ages @xmath138 100myr ) about half of the sample in this bin is dismissed and we see the change between the two measurements ( including and excluding the young ones ) . the dotted line indicates a one - to - one correspondence . ] figure [ fig : sed_bol ] shows the bolometric sfr vs. median sfr inferred from the best - fit sed model , in bins of sfr(sed ) . both quantities are representative of the total sfr of the galaxies , the only difference being the method by which dust is accounted for . the bolometric sfr , as mentioned before , is the sum of sfr(ir ) and sfr(uv ) , and is independent of any dust correction . on the other hand , for sfr(sed ) , the effect of dust is considered by using the calzetti attenuation curve in the sed fitting procedure . we would like to determine whether for all types of galaxies at @xmath7 the bolometric sfrs agree with sed sfrs that are corrected by the calzetti attenuation curve . we stack the mips data of individual detected and undetected galaxies in three bins of sfr(sed ) as described in [ sec : sample - b ] . the three bins in log(sfr(sed ) ) are @xmath138 1.5 , 1.5 to 2 , and @xmath47 2 , with respectively 44 , 56 , and 15 galaxies in each bin . sfr(sed ) and sfr(uv ) are the median sfrs of the individual galaxies in each bin . the results of stacking is shown with open blue circles in figure [ fig : sed_bol ] . in the last two bins , the sed sfr is overestimated compared to the bolometric sfr . stacking the mips images for only objects that are inferred to be older than 100myr produces different results . the number of objects in bins with galaxies older than 100myr reduces to 43 , 50 , and 8 , respectively from the lowest sfr(sed ) bin to the highest . once we remove galaxies with ages @xmath138 100myr , sfr(sed ) agrees well with sfr(uv)+sfr(ir ) . an age - dependent l(uv)-to - sfr calibration ( as we use here ) does not decrease the difference between sfr(sed ) and the bolometric sfr for stacks of all galaxies . it seems younger galaxies are at the same time less dusty ( smaller bolometric sfr ) but redder ( larger sfr(sed ) ) than their older counterparts . these results are consistent with those of @xcite , @xcite , and @xcite , who find that galaxies with young stellar population ages are less dusty for a given uv slope than older galaxies . an smc - like curve may be more appropriate for these galaxies . one possibility is that young galaxies have larger dust covering fractions than older galaxies @xcite . a large dust covering fraction makes the uv slope redder for a given amount of dust attenuation . this study confirms that at @xmath7 the calzetti attenuation curve is applicable to star - forming galaxies older than 100myr , but a steeper attenuation curve may be necessary for the younger galaxies . we have studied the multi - wavelength properties of a sample of @xmath8 200 galaxies with spectroscopically confirmed redshifts at @xmath0 , to study the validity of commonly used dust correction recipes and to compare sfrs inferred from the h@xmath1 line luminosity , the uv continuum luminosity , and _ spitzer_/mips 24@xmath3 m measurements that are converted to the total ir luminosity . in this study , we benefit from using the broad - band photometry excess techniques to determine the h@xmath1 fluxes and , hence , conduct a large sample of @xmath7 h@xmath1 measurements that is immune to uncertainties in the spectroscopic slit - loss corrections . the galaxies properties are determined from stellar population model fitting to the rest - frame uv through near - infrared data . the main conclusions are as follows . * by investigating several recipes for dust correcting the nebular emission lines , we find that assuming the same color excess of the ionized gas and the stellar continuum ( i.e. , @xmath150 ) , and assuming that the calzetti attenuation curve applies to both , results in the best agreement between sfr(h@xmath1 ) and sfr(uv ) . if we assume the calzetti curve to both the stellar and nebular regions but use @xmath151=2.27 @xmath152 , the corrected sfr(h@xmath1 ) measurements are inconsistent with the corrected sfr(uv)s at @xmath134 level , averaged on the five bins of sfr(uv ) ( see the right panel of figure [ fig : sfrhauv ] ) . * using the available _ spitzer_/mips data for @xmath8 100 galaxies in our sample , we derive an observed l@xmath132-to - sfr@xmath141 conversion factor of @xmath153 @xmath139 yr@xmath37 erg@xmath37 s. this calibration is independent of any assumptions on the dust correction and can be used to convert the observed ( extincted ) h@xmath1 luminosity to a bolometric sfr when no dust attenuation measurements for the h@xmath1 luminosity is available . * by comparing the stacks of sfr(uv)+sfr(ir ) with the sed - inferred sfrs that are corrected for dust by the locally - derived calzetti curve we conclude that a steeper attenuation curve ( such as an smc curve ) may be necessary for galaxies younger than 100myr , as previous studies @xcite have found . we find that applying the calzetti curve to the stacks of all galaxies , including the young ones , results in sfr(sed)s that are inconsistent with the sfr(uv)+sfr(ir ) at @xmath154 significance ( the sfr(sed ) in three mass bins of figure [ fig : sed_bol ] are inconsistent with sfr(uv)+sfr(ir ) by 1 , 2 , and 3@xmath54 , from the lowest to highest mass bins respectively ) . the young galaxies have redder uv slope and at the same time lower bolometric sfr compared to their older counterparts . as a result , applying the calzetti curve to the young galaxies overestimates the sed - inferred sfr when compared with their bolometric sfr . the calzetti attenuation curve shows a good overall agreement for galaxies older than 100myr . a more detailed investigation will include direct tracers of nebular dust extinction ( i.e. , the balmer decrement ) . in the future studies , we plan to investigate this aspect with the mosfire deep evolution field ( mosdef ) survey , which uses the near - ir multi - object spectrograph mosfire on the keck i telescope to obtain spectroscopic measurements of the nebular emission lines for a sample of @xmath155 galaxies @xcite . is thanks valentino gonzlez for useful discussions . support for is is provided through the national science foundation graduate research fellowship under grant no . nar is supported by an alfred p. sloan research fellowship .

the discovery of a new class of faint dwarf satellites around the milky way ( mw ; see e.g. , ( * ? ? ? * ; * ? ? ? * and references therein ) ) and the andromeda galaxies ( m31 ; see e.g. , ( * ? ? ? * and references therein ) ) has opened a new window for the study of the formation history of large spirals . the new systems show a number of remarkable differences with respect to the `` classical '' dwarf spheroidals ( dsphs ) surrounding the mw and m31 : i ) they have much lower surface brightnesses ( @xmath11 mag ) , for which they were named `` ultra - faint '' dwarfs ( ufds ) ; ii ) they are very metal poor , with large dispersions and [ fe / h ] values as low as @xmath12 dex ( see * ? ? ? * and references therein ) . such extreme abundances are not observed among the classical dsphs where only a few stars with [ fe / h]@xmath13 have been detected @xcite compared to the large number found in the mw halo ; iii ) they generally contain rr lyrae stars that conform to the subdivision into oosterhoff types i ( oo i ) and ii ( ooii ; * ? ? ? * ) d , whereas type ii ( oo ii ) clusters have @xmath14 d ] observed for field and cluster mw variables . so far , the only exception among the ufds is canes venatici i ( cvn i , * ? ? ? * ) , the brightest of the mw ufds , that , like the classical mw dsphs , has instead oosterhoff - intermediate ( oo - int ) properties ( * ? ? ? * ; * ? ? ? * and references therein ) ; and , finally : iv ) the ufds discovered so far outnumber by almost a factor of two the classical dsphs , thus partially reducing the so - called `` missing satellites problem '' @xcite affecting the @xmath15-cold - dark - matter ( @xmath15cdm ) scenario of galaxy formation . with their properties the ufds are potentially much better analogues than the classical dsphs of the `` building blocks '' that contributed to the formation of the two large spirals in the local group . they have absolute luminosities similar to the globular clusters ( gcs ; @xmath16 mag , on average ) but they are much more spatially extended than gcs . with typical half - light radii of @xmath17 100 pc , in fact they equal in size the `` classical '' dsphs . the ufds are found in groups on the sky ( see e.g. fig . 1 of * ? ? ? * ) , have small velocity dispersions , and high mass - to - light ratios . all ufds host an ancient population around 10 gyr old . they have gc - like color - magnitude diagrams ( cmds ) resembling the cmds of metal - poor galactic gcs ( ggcs ) such as m92 ( ngc 6341 ) , m15 ( ngc 7078 ) and m68 ( ngc 4590 ) . some of the mw ufds have a distorted shape due to the tidal interaction with the mw . we are carrying out an extensive observational campaign of the new mw and m31 ufds to study structural parameters and stellar population properties , as well as the variable stars of these systems . we have already published results for bootes i ( boo i , * ? ? ? * ) , cvn i @xcite , canes venatici ii ( cvnii , * ? ? ? * ) , coma @xcite , leo iv @xcite , and ursa major ii ( umaii * ? ? ? * ) among the mw ufds . with the only exception of cvn i , all these dwarfs contain rr lyrae stars with pulsation periods suggesting an oo ii classification . however , only boo i and cvn i contain sufficiently large numbers of variables to be safely classified into oosterhoff types . the classification of the other ufds is less certain given the small numbers of variables they contain . nevertheless , their few variables clearly tend to have oosterhoff type ii properties . thus , in terms of stellar metallicity and pulsation properties of the variable stars , systems similar to the ufds , as they were at earlier times , could resemble the building blocks of the galactic halo . in this paper we extend our analysis to the hercules ufd galaxy ( r.a.@xmath18 , decl.@xmath19 , j2000.0 ) . the galaxy was discovered by @xcite from the analysis of sdss data . the cmd of hercules ( based on follow - up isaac newton telescope - int- data ) shows besides the red giant branch ( rgb ) , also blue and red horizontal branches ( hbs ) . @xcite interpreted this evidence as a possible signature of multiple stellar populations present in the galaxy . a number of studies have been devoted to this galaxy after the discovery paper . cmds of hercules reaching well below the galaxy s main sequence turn - off were published by ( * ? ? ? * in the @xmath1 and @xmath0 bands ) and ( * ? ? ? * in the @xmath20 and @xmath1 bands ) , based on very wide - field observations obtained with the red and blue channels of the large binocular camera ( lbc ; @xcite ) of the large binocular telescope ( lbt ) . according to the @xcite recovery of the galaxy s star formation history ( sfh ) , hercules is old ( @xmath21 12 gyr , with negligible star formation in the last 12 gyr ) and metal - poor ( [ fe / h]@xmath22 dex ) , with an intrinsic spread in metallicity and both [ fe / h ] = @xmath23 dex and @xmath24 dex populations contributing to the sfh . in the literature there are several spectroscopic and photometric determinations of the metallicity of hercules . these studies confirm that hercules shows a spread in metal abundance , with values of the mean metallicity @xmath25 } \rangle$ ] ranging from about @xmath262.0 to @xmath262.7 dex @xcite . the line of sight towards the hercules dsph galaxy is heavily contaminated by galactic foreground stars , making it hard to determine membership from the cmd alone . even when radial velocities are added the selection remains uncertain , because the mean velocity of the hercules dsph galaxy coincides with the velocity of the thick disk ( @xcite ) . indeed , in the @xcite cmd , based on strmgren photometry obtained with the wide field camera of the int , the rgb of hercules is not easily identified due to the halo foreground contamination . these authors used the c1 index in the strmgren system to disentangle the galaxy s rgb and hb stars from the foreground contamination . hercules appears to be highly elongated . the galaxy s structural parameters were obtained by a number of different authors @xcite who , fitting different stellar profiles to their independent photometric datasets , found similar values for the galaxy s central position , position angle ( @xmath27 ) , and ellipticity ( @xmath28 ) , but rather different half - light radii . in a recent analysis , @xcite suggest that the density profiles of relaxed , tidally stripped dwarf spheroidals like hercules are better approximated by a plummer law . moreover , @xcite clearly point out the need for deep photometry in order to properly constrain the structural parameters of the new faint mw satellites . in the following , we shall adopt the structural parameters obtained by @xcite assuming a plummer stellar distribution , namely @xmath29 , @xmath30 and @xmath31 . hercules has no evidence of internal rotation , and a very low velocity dispersion of @xmath32 km s@xmath33 for @xcite or 3.72 km s@xmath33 for @xcite . the explanation of such a large ellipticity in absence of a rotational support might imply that hercules is not in dynamical equilibrium due to strong tidal distortions ( see discussion in * ? ? ? * ; * ? ? ? the latest studies estimate distances in the range of @xmath34 kpc to @xmath35 kpc ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and a total absolute magnitude ranging from @xmath36 mag to @xmath37 mag ( see e.g. * ? ? ? * ; * ? ? ? the study presented in this paper is based on @xmath38 photometric time - series imaging covering a field of view ( fov ) of @xmath39 ( see table [ t : obs ] ) , extending well beyond hercules half - light radius . these data have allowed us to obtain a complete inventory of the variable stars belonging to the galaxy and to trace the corresponding parent stellar populations . this paper is organized as follows : in section 2 we present the observations and the data reduction procedures ; section 3 is devoted to the variable stars , whereas the cmds and the implications for the structure of the hercules ufd are discussed in section 4 . a new estimate of the distance to hercules based on the galaxy s rr lyrae stars is presented in section 5 . finally , the summary and conclusions in section 6 close the paper . time - series observations in @xmath1 and @xmath0 of the hercules ufd were obtained over the period 2007 april to 2009 june , using a number of different telescopes . the collected data and related telescopes / instrumental set - ups are summarized in table [ t : obs ] , where @xmath40 and @xmath41 are the number of frames in the @xmath1 and @xmath0 bands , respectively . pre - reduction of the images was performed by following standard procedures ( bias subtraction and flat - field correction ) with iraf , except for the wfi data , for which the pre - reduction was carried out using the vst - tube pipeline @xcite . we then performed psf photometry using the daophotiv / allstar / allframe packages ( @xcite ) . after an accurate evaluation of the psf of each individual frame , a reference image was built by averaging all the available frames and a source catalogue was extracted from the stacked image . the source list was then passed to allframe , in order to obtain a homogeneous photometry of all images simultaneously , thus providing @xmath42 and @xmath43 instrumental magnitude catalogues for each telescope . typical internal errors of the single - frame photometry for stars at the hb magnitude level ( @xmath44 mag ) are of about 0.01 mag in both bands . the absolute photometric calibration was derived using observations of standard stars in the landolt fields sa 101 , sa 107 , sa 110 and pg1323 @xcite , as extended by @xcite , which were obtained at the int during the night of 2007 april 22 . errors of the absolute photometric calibration are @xmath45 and @xmath46 mag . as the various telescopes we have used define slightly different photometric systems , each individual dataset was tied to the int standard calibration independently . the calibrated catalogues were then combined using daomaster @xcite to create a final list containing positions , robust intensity - weighted mean @xmath1 and @xmath0 magnitudes @xcite , and weighted average values of the @xmath47 and @xmath48 parameters parameter of each star is a robust estimate of the observed pixel - to - pixel scatter of the fitting residuals to the expected scatter , whereas the @xmath48 parameter is related to the intrinsic angular size of the astronomical objects , and for stellar objects should have a value close to zero @xcite . ] for all objects . we have adopted @xmath49 and @xmath50 to select the stellar - like objects imaged in our total fov of about @xmath10 . this selection is reliable for magnitudes brighter than @xmath9 24 mag , while the uncertainty increases at fainter magnitudes . variable stars were identified from the @xmath1 and @xmath0 time - series data , separately . first , we calculated the fourier transforms ( in the @xcite formulation ) of the stars having at least 12 measurements in each photometric band , then we averaged these transforms to estimate the noise and calculated the signal - to - noise ratios ( s / ns ) . results from the @xmath1 and @xmath0 photometries were cross - correlated , obtaining a list of 860 stars with @xmath51 in both photometric bands ( see e.g. @xcite ) . a visual inspection of the light curves allowed us to confirm the variability of 10 of these candidates . we also analyzed in detail all stars falling on the hb , in the region above the hb , where the anomalous cepheids are generally located , and in the blue straggler region , where one may expect to find variables of the sx phoenicis type . no further variable stars were identified , thus confirming the reliability of our procedure to identify variable stars . to study the light curves we used the software graphical analyzer of time series ( gratis ; see e.g. , @xcite ) that , beyond confirming the variability for all the 10 candidates , also provided periods accurate to 4 - 6 decimal places for all of them . the hercules variables include 1 anomalous cepheid ( ac ) and 9 rr lyrae stars , of which 6 are fundamental - mode ( rrab ) and 3 are first - overtone ( rrc ) pulsators . classification and properties of the confirmed variable stars are summarized in table [ t : rr ] , whereas their light curves are shown in figs . [ f : fig1 ] and [ f : fig2 ] . we have assigned to the variables increasing numbers starting from the galaxy center that was set at r.a.@xmath52 , decl.@xmath53 , j2000.0 @xcite . time - series data for the variable stars are provided in table [ t : pho ] . the light curves are very well sampled , and show very little scatter , except for star v10 . the star does not appear to be blended with other stars on the images . we suspect that v10 might be affected by the blazhko effect @xcite . we note that , according to the period , the ac ( star v2 ) could as well be a fundamental - mode rr lyrae star . however , the star s average magnitude is about 0.4 - 0.5 mag brighter than the hb level . this occurrence , together with a too small amplitude for the star s short period , suggest that the variable either suffers from blending by a contaminant star or is , perhaps , an ac . the photometric parameters @xmath47 ( goodness of the psf fitting ) and @xmath48 ( estimate of the object shape ) provided for v2 by the allframe package seem to rule out the presence of detectable companions around the star , thus favoring the ac hypothesis . another possibility is that v2 could be an overluminous rrc as , for instance , v70 in the m3 ggc . however , the period of v2 ( 0.53777 d ) is definitely longer than that of m3-v70 ( @xmath54 d , @xcite , @xcite ) . on the other hand , the mean magnitude and the period of v2 are in very good agreement with the ac period - luminosity relations by ( * ? ? ? * and references therein ) , @xcite and ( * ? ? ? * and references therein ) , thus favoring an ac classification for v2 . finally , as described in section [ s : cmd ] , the ac classification is further supported by the comparison with evolutionary tracks . the ac , all the rrc variables and three of the rrab stars lie inside the half - light radius of the hercules galaxy . one rrab is just on the border of this region and the remaining two rrab stars lie well outside the half - light radius , with the farthest from the center being confirmed as a member by @xcite on the basis of both radial velocity and strmgren photometry ( see section [ s : cmd ] and figs . [ f : cmd ] and [ f : ad ] for details ) . all the rr lyrae stars fall on the hb of the hercules ufd , and their mean magnitude is consistent , within the errors , with the average luminosity of the hb inferred by fitting the galaxy cmd with the ridge line of the ggc m68 ( see section [ s : cmd ] for details ) . the mean period of the rrab stars is @xmath55 d ( @xmath4=0.03 d ) , thus classifying hercules as an oosterhoff type - ii system . [ f : bailey ] shows the position of the hercules rr lyrae stars on the @xmath0-band period - amplitude ( bailey ) diagram . we have also reported the rr lyrae stars identified in the other mw ufds we have studied so far , for comparison . rrab stars ( star symbols ) lie closer to the loci of the oo ii systems ( from * ? ? ? * solid line ) and the rrc stars fall on the long - period tail of the bell - shaped distribution defined by rrc stars in oo ii systems ( see , e.g. , fig . 9 of @xcite , and fig . 4 of @xcite ) . as shown by @xcite , it is possible to use the shape of an rr lyrae @xmath0-band light curve to obtain an estimate of the star s metallicity . we have performed a fourier decomposition of the @xmath0-band light curves of the variables in hercules and calculated amplitude ratios @xmath56 and phase differences @xmath57 . we list in table [ t : fou ] these quantities up to the term @xmath58 . the fourier parameters were then used along with the formulae provided by @xcite and @xcite for ab- and c - type rr lyrae stars , respectively , to obtain individual metallicities for the hercules rr lyrae stars , except for v10 whose light curve is too noisy for a reliable application of the jurcsik & kovacs method . the metallicity estimates obtained with this technique are reported in column 2 of table [ t : met ] ; they are on the @xcite scale . these individual metallicities were transformed to the @xcite metallicity scale ( using the @xcite transformations for the ab - type variables , and the @xcite recalibration for the c - type rr lyrae stars , see column 3 of table [ t : met ] ) . we then averaged these values ( weighted average ) to obtain our best estimate for the metallicity of hercules old population . our resultant metallicity is @xmath25_{rr , c09 } } \rangle= -2.30 \pm 0.15 $ ] dex . literature values for the hercules ufd s metal abundance are summarized in table [ t : lit_met ] . they were obtained using a variety of different methods and span a rather large range . in particular , we note that the values from @xcite and @xcite are photometric estimates based on fitting the galaxy s cmd with evolutionary tracks , while all other entries in the table are spectroscopic determinations . specifically , the @xcite value is based on abundance analysis of medium - high resolution ( @xmath59 ) spectra of two stars in hercules . the @xcite analysis is based on medium - resolution spectra of 30 red giants in hercules , and the measurement of the equivalent widths of the ca triplet absorption lines . the @xcite study is a re - analysis of the @xcite spectroscopic data , based on an automated spectral synthesis technique . this latter method likely provides systematically lower metallicities ( by @xmath60 dex ) , as discussed by the authors themselves ( see * ? ? ? * ; * ? ? ? finally , in a recent study , @xcite confirm that a large metallicity spread ( @xmath61}\leq -2.0 $ ] dex , for an average value of [ fe / h]=@xmath62 dex ) exists in hercules , from the analysis of medium - high resolution ( @xmath63 ) spectra of 11 rgb stars . however , taking into account only the red giants with metallicity measurements based on a significant number of iron lines ( @xmath64 ; four stars according to table 6 in @xcite ) , we obtain @xmath65}\rangle=-2.3 \pm 0.2 $ ] dex ( and a range of ( @xmath66}\leq -2.0 $ ] dex ) . this value is in excellent agreement with our estimate from the rr lyrae stars . on this basis , we will adopt the metallicity estimate @xmath67_{rr , c09 } \rangle$ ] obtained from the rr lyrae stars in the following analysis . hercules reddening can be estimated from the galaxy s rr lyrae stars . using the relation for rrab stars by @xcite , that connects the color excess to the star s b - band amplitude , the logarithm of the period and the metallicity , we obtain a mean color excess @xmath68 mag , where we have adopted the individual metallicities on the @xcite scale listed in table [ t : met ] . we also use the method by @xcite , which is based on the rr lyrae s @xmath69 color at minimum light , and obtain @xmath70 mag , by applying the metallicity - dependent relation by @xcite . this estimate is larger than the previous value but , as already pointed out by @xcite and @xcite , sturch s method overestimates the color excess by 0.01 - 0.03 mag . we therefore adopt @xmath71 mag for the reddening . this is in very good agreement with the value of @xmath72 mag derived from the @xcite maps . [ f : cmd ] shows the @xmath0 , @xmath69 cmd of hercules obtained in the present study . in the left panel we plot stellar - like objects within the galaxy s half - light radius ; in the right panel , instead , we show objects outside this region over our fov of @xmath10 . all the rr lyrae stars have been plotted in the left panel of the figure including those falling outside the galaxy half - light radius . the cmd reaches @xmath73 25 mag and appears to be heavily contaminated at each magnitude level by field objects belonging to the mw halo and disk , as well as by background galaxies . our photometry reaches a few magnitudes deeper than the sdss photometry ( see fig . 2 of @xcite ) ; while it is roughly 1 mag shallower than the photometry obtained using the lbt ( fig . 1 in @xcite , and fig . 8 in @xcite ) however , within the errors , it is consistent both in @xmath1 and @xmath0 , with the photometry reported in table 2 of @xcite . in fig . [ f : cmd ] , the main branches of the hercules cmd are barely distinguishable due to the overwhelming contamination by the mw field . to identify stars belonging to hercules we used the method that we have already successfully applied in our previous papers ( see e.g. , @xcite ) . specifically , we used the mean ridgelines of the ggc m68 ( dashed black lines in figure [ f : cmd ] ) , obtained from @xcite @xmath74 photometry to fit the hb and rgb of the hercules ufd , by letting the m68 ridgelines vary within the values of the cluster reddening and distance modulus available in the literature . the `` by eye '' best fit was obtained for a shift of @xmath75 mag in magnitude and @xmath76 mag in color . we adopted m68 for identifying members of the hercules ufd because , like hercules , it is very metal - poor . its metallicity , @xmath77_{m68 } } = -2.27 \pm 0.04 $ ] dex @xcite , is in very good agreement with the mean metallicity obtained in section [ s : met ] from the hercules rr lyrae stars . furthermore , the cluster has a well - defined and tight rgb , as well as an extended hb including stars both redder and bluer than the rr lyrae instability strip ( @xcite ) , thus resulting better suited to identify the galaxy s hb than , for instance , m92 , which instead was used as a fiducial by @xcite . with this procedure we selected as most probable members of the hercules galaxy the sources lying within @xmath78 mag in @xmath69 from the ridgelines of m68 ( black dots in the left panel of figure [ f : cmd ] ) . to account for the larger photometric errors , below @xmath79 mag we extended this range within @xmath80 mag of the ridgeline of m68 ( blue dots ) adopting for m68 a reddening value of @xmath81 mag @xcite the color shifts needed to match the hb and rgb of hercules imply a reddening of @xmath82 mag for the galaxy , in excellent agreement with the value obtained from the rrab stars ( see section [ s : met ] ) . totally similar results are obtained using the ridgelines of the ggc m15 that is slightly more metal - poor than m68 and matches equally well the main branches of the hercules cmd , whereas the ridgeline of the metal - intermediate gc m3 ( ngc 5272 ; * ? ? ? * ; * ? ? ? * ) is too red and would require a negative reddening to match the galaxy rgb . with the help of m68 ridgelines , it is possible to determine the average luminosity of the hercules hb in the region of the so - called rr lyrae gap ( @xmath83 mag ) , and to locate the galaxy s main - sequence turnoff at @xmath84 mag . our identification of hercules members is supported by the spectroscopic study of ( * ? ? ? * stars ) and the spectrophotometric analysis of ( * ? ? ? * stars ) , with membership to the hercules ufd confirmed by radial velocity measurements and , for @xcite s sample , also by strmgren photometry ( violet open circles in fig . [ f : cmd ] ) . the excellent agreement between these studies and our results supports the reliability of the procedure we have used to select hercules members and to identify the hb . @xcite compared the galaxy s cmd with the ridge - line of the ggc m92 ( @xmath5 \sim -2.24 $ ] dex ) and m13 ( @xmath5 \sim -1.65 $ ] dex ) and concluded that the morphological features of hercules cmd are generally well described by the ridge line of the old , metal - poor globular cluster m92 . using suitable metal - poor evolutionary tracks and their deep lbt photometry to recover the galaxy s sfh , @xcite and @xcite find that hercules hosts a single old stellar population ( @xmath85 gyr ) , and shows no evidence of significant subsequent star formation events . this is in contrast with the @xcite claim that hercules extended hb showing both a blue and a red component might be the signature of possible multiple stellar populations in the galaxy . however , @xcite s argument can not be considered conclusive , as the morphology of the hb is driven by a complex interplay of different effects , traditionally known in the literature related to ggcs as the `` second parameter problem '' ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . on the other hand , in the region of the cmd above the hb ( @xmath86 and @xmath87 mag ) , there seems to be an overabundance of stars that might represent the signature of an intermediate - age population in hercules . unfortunately , the contamination by field stars in this region of the cmd is very high ( see right panel of fig . [ f : cmd ] ) , thus we can not reach any firm conclusion with our data . an analysis of the existent literature data shows that the hb and the brighter portions of hercules cmd are very poorly populated in the lbt photometry of @xcite and @xcite , likely due to saturation effects , and can not be used to check the possible existence of such an excess above the hb . however , an overabundance of stars brighter than the hb was observed in hercules also by @xcite , who investigated whether they might be variables , based on the magnitude variation in their few ( no more than 3 ) epoch data for the stars . they provide a list of possible variable stars in their table 7 . though we note that these candidate variables are not included in the @xcite final list of hercules members , we checked them in our time series photometry , and do not confirm their variability . however , we have identified an ac variable in this region above the hb , star v2 ( asterisk in fig . [ f : cmd ] , see section [ s : variables ] ) . in fig . [ f : cmd_tracce ] we show an enlargement of the hb region of the cmd in fig . [ f : cmd ] ( with the same color coding ) , where we have plotted the helium burning evolutionary tracks of the stellar model database basti @xcite , for stellar masses in the range of 0.7 to 1.4 @xmath88 ( with a step of 0.1 @xmath88 ) and metal abundance @xmath89 . in particular , the thick ( black and green ) lines are the hb evolutionary tracks for 0.7 and 0.8 @xmath88 , respectively , whereas the thin ( grey ) lines are for @xmath90 including the range of masses corresponding to the hb turnover @xcite . the magenta dot - dashed line represents the zero age horizontal branch ( zahb ) for the same chemical composition . the location of the rr lyrae stars is consistent with the model predictions and suggests an evolutionary effect for some of them . on the other hand , the ac , star v2 , is , as expected , consistent with the evolution of a @xmath91 from the turnover region of the zahb . the origin of acs is still debated ( see e.g. * ? ? ? * and references therein ) and the most widely accepted interpretations are : ( 1 ) they are young ( @xmath92 gyr ) single stars due to recent star formation ; ( 2 ) they formed from mass transfer in binary systems as old as the other stars in the same stellar system . our analysis does not allow us to discriminate between these two scenarios , however , we note that , in the first hypothesis , the comparison of v2 with the evolutionary tracks would suggest the existence in hercules of an intermediate - age population of stars as old as @xmath93 gyrs . fig . [ f : ad ] shows the position of the stars we consider to be members of the hercules galaxy in our fov . symbols and color - coding are the same as in fig . [ f : cmd ] , and the symbol size is inversely proportional to the object s magnitude . the black ellipse corresponds to the half - light radius , the angle position , and the ellipticity obtained for hercules by @xcite . this map confirms that the galaxy is elongated and has an irregular and extended shape . likely , hercules was disrupted due to tidal interaction and is now embedded in the mw halo . our discovery of two rrab variables outside the half - light radius , together with the spectrophotometric identification in the same external region of some stars with confirmed membership by @xcite , support the tidal interaction scenario . [ f : adzoom ] shows an enlargement of the map in fig . [ f : ad ] with the galaxy half - light radius region and all the variable stars identified in the present work . the rr lyrae stars we have detected in hercules give us the opportunity to estimate the distance to the galaxy using these variables as standard candles . the position of the hercules rr lyrae stars in the cmd is in satisfactory agreement with the fiducial - line hb of m68 , although with a significant spread . the most deviating ( brightest ) of the rr lyrae variables , star v9 , is located well outside the galaxy s half - light radius ( see figs . [ f : ad ] and [ f : adzoom ] ) . thus , its high luminosity could be caused by a projection effect . the average magnitude of the rr lyrae stars lying inside the galaxy s half - light radius is @xmath94 mag ( average on 7 stars ) , where the error is the standard deviation of the mean . to obtain the absolute visual magnitude of hercules hb stars , we adopt for the slope of the rr lyrae magnitude - metallicity calibration the value @xmath95 } = 0.214 ( \pm 0.047)$ ] mag dex@xmath33 @xcite , and consider two different calibrations of the zero point . at [ fe / h]=@xmath96 dex , we adopt : @xmath97 mag , that is based on the large magellanic cloud distance modulus of 18.52 @xmath98 mag of @xcite , and , alternatively , @xmath99 mag , from @xcite . adopting for metallicity and reddening the values derived in section [ s : met ] ( @xmath100_{rr , c09}\rangle}= -2.30 \pm 0.15 $ ] dex , and @xmath101 mag , respectively ) we obtain true distance moduli of @xmath102}=20.6\pm 0.1 $ ] mag ( @xmath103}=132 \pm 6 $ ] kpc ) , and @xmath104}=20.7 \pm 0.1 $ ] mag ( @xmath105}=138 \pm 6 $ ] kpc ) , for the @xcite and @xcite calibrations , respectively . the errors on the distance values include the contribution of the uncertainties on the metallicity , the reddening , the adopted slope of the @xmath106}$ ] relation , and the average apparent visual magnitude of the hercules rr lyrae stars . an additional estimate of the distance , based on the rr lyrae stars , can be obtained using the theoretical wesenheit relation in the @xmath38 bands , as defined by @xcite , with the assumption of a suitable evolutionary mass for the metallicity of hercules rr lyrae stars . the resulting distance modulus of @xmath107 mag is in perfect agreement with @xmath102}$ ] . moreover , a distance of @xmath108 kpc is in excellent agreement with the estimates by @xcite , @xcite , @xcite , and in statistical agreement with the distance by @xcite . on the other hand , the distance @xmath105}$ ] appears to be only in statistical agreement with all previous determinations . to conclude this discussion , we note that the very long estimate of @xmath109 kpc , obtained by @xcite is likely due to their hb mean magnitude being based on a few stars , of which some are variables observed at random phase , as we have confirmed in our study . finally , the apparent distance modulus adopted in fig . [ f : cmd_tracce ] , @xmath110 mag , corresponds to a true distance modulus of 20.7 mag . however , we should take into account the theoretical uncertainties on the luminosity of the zahb and , in particular , the remark by @xcite that , adopting updated conductive opacities , might cause an increase in the @xmath111 of about 0.06 mag at the metallicity of hercules , and a corresponding decrease in the inferred distance modulus . on this basis , and considering also the results of the theoretical wesenheit relation , we thus conclude that our most reliable estimate for the distance modulus of hercules is : @xmath112 mag . in this paper we have presented the first time - series analysis of the hercules ufd . using a variety of telescope / instrument facilities we secured @xmath113 80 and 50 epochs in @xmath1 and @xmath0 . these data allowed us to detect and characterize 9 rr lyrae stars ( 6 ab- and 3 c - type , respectively ) and one anomalous cepheid . the same observations allowed us to build a deep cmd extending well beyond the galaxy s half - light radius . the main results of this study are listed below : * the average period of the ab - type rr lyrae stars , @xmath2p@xmath1140.68 d , qualifies hercules as an oosterhoff ii system , in good agreement with the vast majority of the ufds investigated so far . this occurrence favors the hypothesis that the ufds could be the `` building blocks '' of the galactic halo , since the pulsation characteristics of their rr lyrae stars are in agreement with the properties of the mw halo variables . * hercules cmd is dominated by a stellar population at least as old and metal - poor as the ggc m68 . this result is in agreement with previous findings . the hb shows some spread . we also detected an overabundance of stars above the hb , thus confirming the previous finding by @xcite . this , along with the detection of an anomalous cepheid very likely belonging to hercules , hints at the possible presence of an intermediate - age population about @xmath93 gyrs old in hercules . * the spatial distribution of hercules stars confirms the elongated shape of this galaxy . the signature that hercules is undergoing tidal disruption is provided by the absence of a clearcut difference between galaxy and field star properties , and by the presence of two rr lyrae stars lying well beyond the galaxy s half - light radius . * the rr lyrae variables were used to obtain independent estimates of the metallicity , reddening and distance to hercules , for which we find [ fe / h]=@xmath115 dex , @xmath6 mag , and @xmath116 mag respectively , in very good agreement with literature values . we thank an anonymous referee for carefully reading the paper and for providing comments that helped to improve the clarity of the manuscript . financial support for this research was provided by cofis asi - inaf i/016/07/0 , by the agreement asi - inaf i/009/10/0 , and by prin inaf 2010 , ( p.i . : g. clementini ) . support for m.c . is provided by proyecto fondecyt regular # 1110326 ; basal center for astrophysics and associated technologies ( pfb-06 ) ; fondap center for astrophysics ( 15010003 ) ; the chilean ministry for the economy , development , and tourism s programa iniciativa cientfica milenio through grant p07 - 021-f , awarded to the milky way millennium nucleus ; and proyecto anillo act-86 . has thanks the u.s . nsf for support under grants ast0607249 and ast0707756 . adn , d. , feltzing , s. , koch , a. , et al . 2009 , , 506 , 1147 adn , d. , eriksson , k. , feltzing , s. , et al . 2011 , , 525 , a153 baldacci , l. , clementini , g. , held , e. v. , marconi , m. , & rizzi , l. 2004 , communications in asteroseismology , 145 , 32 belokurov , v. , zucker , d. b. , evans , n. w. , et al . 2006 , , 647 , l111 belokurov , v. , zucker , d. b. , evans , n. w. , et al . 2007 , , 654 , 897 belokurov , v. , walker , m. g. , evans , n. w. , et al . 2010 , , 712 , l103 benedict , g. f. , mcarthur , b. e. , feast , m. w. , et al . 2011 , , 142 , 187 bersier , d. , & wood , p. r. 2002 , , 123 , 840 blazhko , s. , 1907 , astron . nachr . 175 , 325 buonanno , r. , corsi , c. , bellazzini , m. , ferraro , f. r. , fusi pecci , 1997 , aj , 113 , 706 cacciari , c. , corwin , t. m. , & carney , b. w. 2005 , , 129 , 267 caputo , f. , & deglinnocenti , s. 1995 , , 298 , 833 carretta , e. , cacciari , c. , ferraro , f. r. , fusi pecci , f. , & tessicini , g. 1998 , , 298 , 1005 carretta , e. , bragaglia , a. , gratton , r. , dorazi , v. , lucatello , s. , 2009 , a&a , 508 , 695 cassisi , s. , potekhin , a. y. , pietrinferni , a. , catelan , m. , & salaris , m. 2007 , , 661 , 1094 catelan , m. , 2009 , ap&ss 320 , 261 clement & rowe 2000 , aj , 120 , 2579 clementini , g. 2010 , in variable stars , the galactic halo and galaxy formation , eds . n. samus , c. sterken , l. szabados , sternberg astronomical institute publications , p. 111 , ( arxiv:1002.1575 ) clementini , g. , et al . , 2000 , aj , 120 , 2054 clementini , g. , gratton , r. g. , bragaglia , a. , carretta , e. , di fabrizio , l. , & maio , m. 2003 , aj , 125 , 1309 coleman , m. g. , et al . , 2007 , aj , 134 , 1938 dallora , m. , et al . 2006 , apj 653 , l109 dallora , m. , et al . 2012 , apj , accepted , arxiv:1203.5321 di criscienzo , m. , greco , c. , ripepi , v. , clementini , g. , dallora , m. , marconi , m. , musella , i. , federici , l. , di fabrizio , l. 2011 , , 141 , 81 di criscienzo , m. , marconi , m. , caputo , f. 2004 , apj 612 , 1092 durrell , p. r. , & harris , w. e. 1993 , , 105 , 1420 ferraro , f. r. , carretta , e. , corsi , c. e. , fusi pecci , f. , cacciari , c. , buonanno , r. , paltrinieri , b. , & hamilton , d. 1997 , a&a , 320 , 757 frebel , a. , kirby , e. n. , & simon , j. d. 2010 , , 464 , 72 giallongo , e. , ragazzoni , r. , grazian , a. , et al . 2008 , , 482 , 349 grado , a. , capaccioli , m. , limatola , l. , & getman , f. 2012 , memorie della societa astronomica italiana supplementi , 19 , 362 gratton , r. g. , carretta , e. , bragaglia , a. , lucatello , s. , dorazi , v. 2010 , a&a , 517 , 81 greco , c. , et al . 2008 , apj 695 , l83 johnson , j. a. , & bolte , m. 1998 , apj , 115 , 693 jurcsik , j. & kovcs , g. 1996 , a&a , 312 , 111 kaluzny , j. , hilditch , r. w. , clement , c. , & rucinski , s. m. 1998 , , 296 , 347 kapakos , e. , hatzidimitriou , d. , & soszyski , i. 2011 , , 415 , 1366 kirby , e. n. , simon , j. d. , geha , m. , guhathakurta , p. , & frebel , a. 2008 , apj , 685 , l43 kirby , e. n. , guhathakurta , p. , bolte , m. , sneden , c. , & geha , m. c. 2009 , , 705 , 328 kirby , e. n. , guhathakurta , p. , simon , j. d. , et al . 2010 , , 191 , 352 klypin , a. , kravtsov , a. v. , valenzuela , o. , & prada , f. 1999 , , 522 , 82 koch , a. , mcwilliam , a. , grebel , e. k. , zucker , d. b. , & belokurov , v. 2008 , , 688 , l13 kuehn , c. , et al . 2008 , apj 674 , l1381 landolt , a. u. 1992 , , 104 , 340 landolt , a. u. 2009 , , 137 , 4186 lee , y .- w . , demarque , p. , zinn , r 1994 , apj , 423 , 248 marconi , m. , fiorentino , g. , & caputo , f. 2004 , , 417 , 1101 martin , n. f. , de jong , j. t. a. , rix , h .- w . 2008 , apj 684 , 1075 moore , b. , ghigna , s. , governato , f. , et al . 1999 , , 524 , l19 moretti , m.i . , et al . 2009 , apj 699 , l125 morgan , s. m. , wahl , j. n. , & wieckhorst , r. m. 2007 , , 374 , 1421 musella , i. , et al . 2009 , apj 675 , l73 oosterhoff , p. t. 1939 , the observatory , 62 , 104 pearrubia , j. , navarro , j. f. , mcconnachie , a. w. , & martin , n. f. 2009 , , 698 , 222 piersimoni , a. m. , bono , g. , & ripepi , v. 2002 , , 124 , 1528 pietrinferni , a. , cassisi , s. , salaris , m. , & castelli , f. 2004 , , 612 , 168 pietrinferni , a. , cassisi , s. , salaris , m. , & castelli , f. 2006 , , 642 , 797 richardson , j.c . 2011 , , 732 , 76 sand , d. j. , olszewski , e. w. , willman , b. , zaritsky , d. , seth , a. , harris , j. , piatek , s. , saha , a. , 2009 , apj , 704 , 898 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , apj , 500 , 525 schwarzenberg - czerny , a. 1996 , , 460 , l107 simon , j. d. , & geha , m. 2007 , apj , 670 , 313 stetson , p. b. 1987 , , 99 , 191 stetson , p. b. 1994 , , 106 , 250 stetson , p. b. 2000 , , 112 , 925 sturch , c. 1966 , , 143 , 774 tolstoy , e. , hill , v. , tosi , m. 2009 , , 47 , 371 walker , a. r. 1994 , aj , 108 , 555 walker , a. r. 1998 , , 116 , 220 zinn , r. & west , m. j. 1984 , apjs , 55 , 45

since the advent of infrared arrays in the early 1990s , there have been many low resolution spectroscopic atlases published covering the @xmath3 , @xmath4 , or @xmath5 bands . previous efforts that included the full o m spectral classes and i v luminosity classes made use of moderate resolution fourier transform spectroscopy techniques ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? an additional window of atmospheric transparency between these traditional near - infrared bands and the optical window , dubbed the @xmath6 band , was highlighted by @xcite . with existing observational capabilities on instruments at the irtf , keck , las campanas , and ukirt telescopes , and with upcoming wide - field survey facilities such as pan - starrs and vista , the @xmath6 band is becoming an increasingly appreciated and scientifically important near - infrared bandpass . several low resolution ( @xmath7 ) spectral atlases for m , l , and t dwarfs have been published including @xmath6 band data , e.g. @xcite , @xcite , and @xcite . however , a high - resolution spectroscopic atlas covering earlier spectral types and the full range of luminosity classes has not yet appeared . to complement existing moderate resolution catalogs in the @xmath3 , @xmath4 , and @xmath5 bands , we present here a high - resolution infrared spectral atlas for the @xmath6 band . our data cover the @xmath8 m region sampled by nirspec @xcite at the w. m. keck observatory . we obtained spectra of 20 mk - classified stars ranging in spectral type from f through m and in luminosity class from i v for use as spectroscopic standards in future work . the data were processed as briefly described in section [ sec : data ] . we present a representative grid of spectroscopic standards , a table of identified absorption features , and the resulting atlas of spectral lines in section [ sec : atlas ] . in section [ sec : analysis ] , we present relevant atomic data and illustrate as well as discuss line sensitivity to stellar effective temperature and surface gravity . near - infrared echelle spectra of 20 stars with well - established mk classifications were acquired between 2001 and 2009 with nirspec @xcite on keck ii . the targets , listed in table [ tab : stds ] , were drawn largely from the lists of standards utilized in the @xcite and @xcite studies of stellar photospheres in the @xmath4 and @xmath3 bands , while the remainder were selected by simbad searches on spectral type and luminosity class , brightness , and accessibility . we used the n1 filter and have wavelength coverage from @xmath9 to @xmath10 @xmath11 m at r = 25,000 ( @xmath12v = 12 km s@xmath13 ) . the detector is a 1024 @xmath14 1024 pixel insb chip , and the projected spectrograph slit we used was @xmath15 in angular size . the @xmath6 band is covered in 12 full and 2 partial orders , without gaps . the wavelength coverage of each order is given in table [ tab : coverage ] . spectra were generally taken in nodded pairs using an abba pattern , with the astronomical target offset by @xmath16 in each ab pair . exposure times ranged from 1 to720 s , integrated in the case of each spectral standard to acquire signal - to - noise ratios @xmath17 in the reduced one - dimensional spectra . data reduction , including wavelength calibration and spatial rectification , extraction of one - dimensional spectra from the images , and removal of telluric emission and absorption features , is discussed in @xcite and @xcite . we used a modified version of an iraf script reduceit " provided by marianne takamiya to reduce the 2002 and 2005 data , and we used the idl package redspec by s. kim , l. prato , and i. mclean to reduce data acquired in 2006 and 2009 . from our 20 spectroscopic standards we identified a grid representing each of the four spectral classes ( f , g , k , m ) and three luminosity classes ( i - i - i , iii , iv v ; hereafter supergiants , giants , dwarfs ) . the star used to typify each class is indicated in table [ tab : stds ] and was chosen based on having a good signal - to - noise ratio and relatively low rotational velocity ( @xmath18 ) . figures [ fig : grid_80][fig : grid_69 ] show the grid of representative spectroscopic standards displayed from shorter to longer wavelengths for orders 80 through 69 . each spectrum is normalized to the local continuum and shifted to line rest velocity . orders 81 and 68 at the edges of our wavelength coverage are excluded due to the low signal - to - noise ratio and the relatively high telluric contamination towards the edges of the @xmath6 band window . as can be seen , the 1 @xmath11 m spectral region is rich in spectral features . the absorption lines originate from both atomic and molecular species . in order to generate a complete list of the atomic and molecular lines seen in the grid of spectroscopic standards , we began with a candidate list of transitions from the infrared atlas of the arcturus spectrum @xcite . arcturus is a k2 iii star and so its associated line list should be relevant for the majority of the cool stars in our sample , with some supplementing needed as described below especially for the hotter stars . of the 786 lines in our final candidate line list , 209 were added beyond those present in the arcturus atlas list . because the arcturus atlas does not contain the complete details of each atomic transition ( only line frequency and species ) , the candidate line list was cross - referenced with the atomic line database @xcite in order to obtain excitation potentials and log @xmath19 values , as well as the atomic line list @xcite , to obtain the transition terms . the kurucz line list was developed from theoretical computations and it is far more extensive in line transitions ( over 16,000 within our wavelength range ) than necessary given the moderate spectral resolution and signal - to - noise ratio typical of our ( and most ) empirical astrophysical data . thus great care was taken to properly match the atomic data to the lines identified in the arcturus atlas . each line in the arcturus atlas was matched by species , ionization , and wavelength within @xmath20 to lines in the kurucz atomic line database ( in vacuum wavelengths ) . in some cases this resulted in multiple potential matches between the lists among lines of the same species but having different ionization states . taking ionization into account generally resolved the situation but ambiguities often remained for lines such as the numerous cr i , fe i , and ti i transitions . in such cases we chose to associate at least initially the arcturus atlas line with the kurucz line possessing the largest log @xmath19 among those with a realistic ( @xmath2110ev ) lower excitation potential . after the excitation potential , log @xmath19 , and transition term details of the atomic lines were added from the various sources to the candidate line list , the spectra were visually inspected . we observed that some lines identified in the arcturus atlas did not correspond to any observed spectral features . lines that were not present at @xmath22 of the continuum level were flagged appropriately ( as having strength 0 on the scale of 03 described below ) , as were any multiply matched to those in the kurucz list but decided not to be the correct identification . these lines are ignored in our further analysis but remain in the candidate line list for reference . visual inspection of the spectra relative to the line list also identified strong absorption features seen in the spectra but not present in the assembled atomic line list . for later type stars ( and especially prominent in the supergiants ) such features were mostly identifiable as molecular lines based again on comparison to the arcturus atlas . we do not list these individually herein , but we found that nearly all unidentified strong lines in the later type stars appear to be molecular . for the earlier type f stars , additional atomic lines were identified using the kurucz line list , now allowing for higher excitation potentials . in the case that there were multiple possible lines within the spectral resolution of the data ( @xmath23 ) the line with lowest ground state energy level was chosen . if this line was a species that had not been previously identified in the spectra , then other line candidates near the observed line center were considered . after the initial matching using the arcturus atlas ( hinkle ) and atomic line database ( kurucz ) , all lines that were part of the same multiplet were identified using the on - line atomic line list ( van hoof ) . any lines in multiplets of those already identified based on the above exercises and within the spectral range of the data were added to the line list if not already present , and are flagged as so in the final catalog . as above , we considered only the expected range of log @xmath19 and excitation potential values when initially assigning new candidate lines based on multiplets . we then also considered the relative line strengths among the multiplets . specifically , for optically thin ( weak ) lines in local thermodynamic equilibrium , the expected line intensity ratio @xmath24 is @xmath25 or @xmath26 where the symbols have their usual meanings . for stronger lines the line broadening factors should be taken into account but we have ignored this aspect for ease of analysis . in the case of multiplets ( lines arising from the same lower level ) the above expression simplifies because of the very small energy differences , to @xmath27 which is what we use as a check on the line identifications . if during the multiplet analysis the case arose that an already - identified line could also be part of a multiplet from a different line species , we chose the multiplet identification over the alternative species unless both could be justified via multiplets or previous identification , in which case an additional flag for line blending was added . we note that the line wavelengths as well as their other data as given among the arcturus atlas ( hinkle ) , the cfa atomic line database ( kurucz ) and the on - line atomic line list ( van hoof ) were not always identical within quoted errors . further , there are lines appearing in some of the above sources that are not available in the others . we believe sensible choices for line identifications have been made , and we have indicated remaining uncertainties in the tabulation . we also note that in some cases we have been unable to discern the transition term from the existing sources of atomic data . the 1 @xmath11 m region is a relatively unscrutinized wavelength regime in stellar astrophysics and we are offering some of the first data that can help refine the available atomic line lists and parameters . finally , all lines in the catalog were assigned a numerical grade based on their relative strength in the spectroscopic standard grid . the grades range from zero to three , where the zero corresponds to lines that are not seen in any of the spectroscopic standards and are left in the catalog only for completeness , while the three is assigned to the deepest 210 lines per order ( often deeper than 70% of the continuum level across multiple spectral types and luminosity classes ) . throughout this paper , we consider a one to be a weak line ( shallower than 90% of the continuum ) , while a two ( deeper than 90% of continuum ) or a three ( up to 50% of continuum ) is a strong line . lines that were blended ( at greater than the 50% level ) , or lines that were only seen in the f standards were also noted in the catalog . we note that for late - type and especially low - gravity stars , some expected atomic lines can be easily lost amidst strong molecular absorption features , generally cn lines which can exceed 80% depth in some orders . the number of lines in each strength classification , and the fractions that are molecular , blended , or only seen in the f standards are indicated in table [ tab : strengths ] . the strongest ( deepest ) atomic and ionic lines , those given grades two and three above , plus their related atomic data are presented in table [ tab : lines ] . the log @xmath19 values among this strong line set range from @xmath28 and up for fe i and tii lines , to -2.5 and up for cr i lines , to @xmath29 and up for si i lines , etc . we show in figure [ fig : lowerep ] the lower excitation potential as a function of wavelength and of log @xmath19 for these strong lines . the general trends in figure [ fig : lowerep ] are as expected from atomic physics considerations . in figure [ fig : ephisto ] , we show a histogram of the number of atomic lines in 1ev bins of the lower excitation levels , for both the entire line catalog and for just the strong lines ( those presented in table [ tab : lines ] and figure [ fig : lowerep ] ) . this figure is discussed in more detail below . the table and figures include only the 0.951.11@xmath11 m spectral region , where our final spectra are most uniform in signal - to - noise ratio across the standards grid . the strong atomic lines identified above as well as the molecular lines are illustrated in figures [ fig : fmgiant_pg1][fig : fmdwarf_pg2 ] using the extremes of our spectral type grid ( f and m , giants and dwarfs ) for comparison . orders 79 and 71 appear to cover particularly useful spectral ranges for discriminating between hotter and cooler stars ( see section [ sec : diagnostic ] ) based on the trade off in strong features as a function of spectral type . summarizing , notably present in the 1@xmath11 m spectral region are numerous low and moderate excitation lines of neutral ca i , cr i , fe i , k i , mg i , na i , ni i , si i , and ti i along with singly ionized sr ii . lines of higher excitation potential like h i ( and he i , though not within the effective temperature range sampled here ) , c i , n i , p i , and s i and singly ionized lines of al ii , ca ii , fe ii , and mg ii are also pervasive in the spectroscopic standards . based on visual examination of the spectral grid , many of these lines are temperature sensitive and a few , such as sr ii ( @xmath30 ) , are gravity sensitive . such trends among the atomic lines are discussed more quantitatively below . the spectra also feature many molecular lines , including ( for all g k m types ) a very prominent cn 00 transition molecular series with band heads near 1.093 @xmath11 m ( @xmath31 ) and 1.097 @xmath11 m ( @xmath32 ) according to @xcite ; these occur in the red part of order 70 and the series covers much of order 69 . at the later spectral types ( m stars ) , the feh molecule has a prominent 00 band head ( the wing ford " band which is apparent longward of @xmath33 ) and @xmath34-branch ( at @xmath35 ) . at even later spectral types , for which we do not have data , the @xmath6 band contains tio bands which appear at 0.942 and 0.950 @xmath11 m , feh appearing at 1.025 and 1.064 @xmath11 m @xcite , and vo appearing at 0.971 and 1.064 @xmath11 m through @xmath36 @xcite . there is also h@xmath37o toward the short- and long - wavelength edges of the @xmath6 band atmospheric window . late m and l dwarfs also feature very prominent atomic k i , na i , and ca i features in the @xmath6 band that are only weakly ( grade one on our system ) or not present ( grade zero ) yet in the latest types among our spectra . we examine in figure [ fig : ephisto ] the distribution of energy levels for the suite of atomic spectral features identified in the @xmath6 band ( complete candidate line list as well as the strong lines of table [ tab : lines ] ) . the number of lines as a function of lower excitation potential is peaked at approximately 5 ev . the distribution declines more rapidly on the high excitation potential side of the peak than the lower , largely due to the presence of many strong moderate excitation si i and fe i transitions throughout the f , g , k , m spectral range . heavier metals and higher ionization states generally contribute to the high excitation potential tail , while a large number of ti i and some cr i lines contribute to the lower excitation potential tail . notable exceptions to this general trend include several c i lines at @xmath38ev , and a few h i lines in the upper excitation potential range . as expected , many of these high excitation potential lines are only seen ( with strength values of 3 or 2 on our system ) among the earlier type f stars within our spectroscopic grid . in figure [ fig : aveprofs ] , we have plotted for the spectroscopic standards grid the average line profile for four different bins in excitation potential . the atomic lines were divided into bins such that roughly the same number of @xmath6 band lines are contained in each bin , but with an attempt to minimize the splitting of lines of the same species into multiple bins . this resulted in divisions at @xmath39 , @xmath40 , and @xmath41 . only strong , unblended lines were included . however , hydrogen lines as well as any lines within @xmath42 of it were excluded since these particularly strong / broad lines dominate in the averaged profiles over nearby line features . there are 35 lines in the 0 - 2.4 ev bin , 41 lines in the 2.44.9 ev bin , 29 lines in the 4.96.8 ev bin , and 26 lines in the @xmath43 ev bin . as expected , the hotter stars tend to have stronger lines with higher excitation potentials while cooler stars are dominated by low excitation potential lines . supergiants also tend to have stronger lines overall relative to the dwarfs . specifically , the f stars are dominated by the 6.8evand higher lines , the g stars are dominated by the 4.96.8ev lines , and the m stars are dominated by the 02.4ev lines . one would expect that the k stars would lie in sequence and be dominated by the 2.44.9ev lines . instead , we find that the line strengths are roughly constant from 0 to 6.8ev and comparable to the 4.96.8ev bin in the g standards . although the k standards do fit within the overall excitation - potential - temperature trend , they do not match the observed pattern in detail in the expected way . among the strong lines there are some obvious temperature dependencies apparent in figures [ fig : fmgiant_pg1]-[fig : fmdwarf_pg2 ] . for example , in order 79 , there are several strong c i lines that appear in the early spectral types and are not seen in the later types . in the same order there are also tii lines that complement this trend , appearing very strongly in the late spectral types , but only weakly , if at all , in the early spectral types . a similar trend can be seen in order 73 for si i relative to ti i , and in order 71 for both c i and si i relative to ti i. we investigated through equivalent width analysis the behaviors with both effective temperature and surface gravity of the various strong lines in the 1 @xmath11 m region . several strong lines stand out over the others in table [ tab : lines ] as potentially useful for analytic or diagnostic purposes , the strongest of which have line depths greater than 50% of the continuum level . figure [ fig : equivwidths ] plots the equivalent widths of several such lines versus the standard star effective temperature . in constructing this plot we were particularly sensitive to including those lines with temperature dependencies discernable for dwarf stars . as discussed above , lines of note include c i and s i multiplets , which show a strong temperature dependence , and several tii multiplets , which show a strong inverse temperature dependence . this behavior was also highlighted by for the a5f z5go multiplet of ti i in the 1.04 - 1.07@xmath11 m region , though we find many of these long wavelength lines of the multiplet blended with other strong lines . ca i , cr i , and fe i have generally weaker ( inverse ) temperature dependence . several neutral h i lines and higher excitation potential neutral metals such as c i , p i , and s i and the singly ionized metals ca ii , fe ii , and mg ii appear only in the f standards and are therefore also good temperature indicators . the si i lines peak in strength in the g k spectral range and become weaker toward both f and m types , with some gravity sensitivity . in addition to their temperature dependence , the ti i lines are also gravity sensitive at log @xmath44 3.7 as are the c i and s i lines at log @xmath45 . srii shows gravity dependence at all temperatures but little trend with temperature . we have identified a few lines having line depth ratio that are useful indicators of spectral and luminosity class . these ratios versus standard star effective temperature are illustrated in figure [ fig : diagnostic ] . the ti i ( @xmath46 ) and sr ii ( @xmath47 ) lines , are the temperature and gravity sensitive lines , respectively , with the strongest trends apparent in figure [ fig : equivwidths ] . in this case , division of the temperature - sensitive line by the gravity - sensitive line spreads the spectroscopic standards such that dwarfs and then giants have increased line strength relative to supergiants at similar effective temperatures . the ti i ( @xmath46 ) and c i ( @xmath48 ) lines are both temperature sensitive , but the temperature dependence of c i is opposite that of ti i. this creates an amplified trend in line strength versus temperature relative to that of the single lines ( as in figure [ fig : equivwidths ] ) . we also investigated lines pairs that were closer together in wavelength and therefore suitable for temperature and gravity designation from single - order moderate to high dispersion spectra . thus the cai ( @xmath49 ) @xmath50 sr ii ( @xmath51 ) ratio is also illustrated as sensitive to gravity and the ti i ( @xmath52 ) @xmath50 ci ( @xmath53 ) ratio as sensitive to temperature . either of these line pair sets could be used in comparing the relative spectral typing of a set of previously unclassified stellar spectra , depending on the spectral dispersion although not present in our effective temperature range , there is a high excitation ( 21ev ) he i line at @xmath54 in addition to the high excitation potential ( 12 ev ) h i lines already noted here . these he i and h i lines are useful spectral type diagnostics for o , b , and a stars . both lines are also particularly sensitive probes at all spectral types of energetic circumstellar processes such as chromospheric activity and wind processes @xcite . we have presented a spectral atlas for the @xmath6 band region of the near infrared portion of the electromagnetic spectrum . our data set consists of a grid of mk - classified stars for use as spectroscopic standards spanning the f through m spectral classes and i v luminosity classes observed at r@xmath5525,000 . the associated catalog of identified lines contains numerous spectral absorption features , both atomic and molecular in origin , that span a wide range of excitation potentials . many of the atomic features show strong line depth and equivalent width trends across the spectral atlas . this atlas and the identified atomic spectral lines are potentially useful for two - dimensional spectral classification , especially line depth ratios involving ti i and c i or s i , which are temperature sensitive in opposite directions , and sr ii , which is gravity sensitive . ascii versions of all spectra are available to download with the online version of the journal ( an example of the form and content of these spectra can be found in table [ tab : asciiex ] ) . we thank the anonymous referee for his / her careful inspection of the manuscript . we acknowledge with gratitude the builders of nirspec and the hard work and assistance from the staff at w.m . keck observatory . the observatory is operated as a scientific partnership among the california institute of technology , the university of california and the national aeronautics and space administration and was made possible by the generous financial support of the w.m . keck foundation . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . some of the spectra presented herein were obtained by dan stark and richard ellis or colette salyk and geoff blake ; we are grateful for their contributions . we consulted the atomic line list ( http://www.pa.uky.edu/@xmath56peter/atomic/ ) and the cfa kurucz line list ( http://cfa-www.harvard.edu/amp/ampdata/kurucz23/sekur.html ) while conducting this research . we are grateful to andrew baker for his tolerance during the final analysis and writing stages of this paper . llrclrc hd 21770 & hr 1069 & 2006 nov 30 & f4 iii & 7170 & @xcite & x + hd 193370 & hr 7770 & 2006 nov 30 & f5 ib & 6640 & @xcite & + hd 38232 & bd+29 983 & 2005 dec 13 & f5 ii & 6640 & @xcite & x + hd 55052 & hr 2706 & 2002 nov 26 & f5 iii & 6470 & @xcite & + hd 210027 & hr 8430 & 2006 nov 30 & f5 v & 6530 & @xcite & x + hd21019 & hr 1024 & 2006 nov 29 & g2 v & 5830 & @xcite & x + hd 87 & hr 4 & 2009 jul 11 & g5 iii & 5072 & @xcite & x + hd 58367 & hr 2828 & 2005 dec 13 & g6 iib & 4720 & @xcite & x + hd 75935 & bd+27 1682 & 2006 nov 30 & g8 v & 5430 & @xcite & + hd 166 & hr 8 & 2006 nov 30 & k0 v & 5240 & @xcite & + hd 39400 & hr 2037 & 2005 dec 13 & k2 iib & 4330 & @xcite & x + bd + 59 2667 & & 2006 nov 30 & k2 iv & 5010 & @xcite & + hd 61772 & hr 2959 & 2005 dec 13 & k3 iii & 4320 & @xcite & x + hd 219134 & hr 8832 & 2006 nov 30 & k3 v & 4780 & @xcite & x + hd 201091 & 61 cyga & 2006 dec 01 & k5 v & 4340 & @xcite & + hd 201092 & 61 cygb & 2006 dec 01 & k7 v & 4040 & @xcite & + hd 232979 & bd+52 857 & 2006 nov 30 & m0.5 v & 5240 & @xcite & + hd 39225 & hr 2028 & 2005 dec 13 & m1 + iii & 3550 & @xcite & x + hd 37536 & hr 1939 & 2005 dec 13 & m2 iab & 3450 & @xcite & x + hd 1326 & gl 15a & 2006 nov 30 & m2 v & 3530 & @xcite & x + lrcl 81 & 0.94118 - 0.94813 + 80 & 0.94646 - 0.95985 + 79 & 0.95835 - 0.97192 + 78 & 0.97054 - 0.98543 + 77 & 0.98304 - 0.99696 + 76 & 0.99590 - 1.00998 + 75 & 1.00906 - 1.02332 + 74 & 1.02255 - 1.03709 + 73 & 1.03628 - 1.05122 + 72 & 1.05081 - 1.06578 + 71 & 1.06553 - 1.08065 + 70 & 1.08065 - 1.09604 + 69 & 1.09584 - 1.11184 + 68 & 1.11241 - 1.12224 + lcrrrllc 0.9504665 & mn i & -1.200 & 5.69633 & 7.00088 & ? ? & 2b & uncertain but in the arcturus list + 0.950867 & ti i & 0.078 & 3.58290 & 4.88689&y5fo - f5f & 2b & + 0.951329 & ti i & -0.316 & 3.55869 & 4.86196&y5fo - f5f & 2b & + 0.951587 & fe i & -1.491 & 5.03341 & 6.33641 & x5fo - f5f & 2 & + 0.952265 & n ii & -0.960 & 4.16750 & 5.46958&y3do - e3d & 2 & + 0.954859 & h i & -0.540 & 12.08751 & 13.38597 & 3 - 8 & 3f & + 0.954865 & ti i & -1.598 & 0.83605 & 2.13459&a5f - z5fo & 3 & * + 0.957253 & fe i & -0.542 & 4.99161 & 6.28690&x5fo - e5 g & 2 & + 0.957438 & cr i & -1.784 & 2.54444 & 3.83949&a5g - z5fo & 2b & + 0.957692 & cr i & -1.088 & 2.54479 & 3.83949&a5g - z5fo & 3b & + 0.960218 & ti i & -1.440 & 0.82591 & 2.11721&a5f - z5fo & 3 & * + 0.960567 & c i & -0.940 & 7.48090 & 8.77172&3po-3s & 3f & + 0.962342 & c i & -0.516 & 7.48278 & 8.77114&3po-3s & 3f & + 0.962903 & cr i & -3.451 & 2.91531 & 4.20301&a3p - z3po & 2b & + 0.962913 & fe i & -0.610 & 5.03340 & 6.32108&x5fo - e5 g & 2 & + 0.963682 & fe i & -0.607 & 5.06412 & 6.35078&x5fo - e5 g & 2 & + 0.964090 & ti i & -0.638 & 0.84848 & 2.13459&a5f - z5fo & 3b & only weak blend , * + 0.965002 & ti i & -1.462 & 0.81820 & 2.10309&a5f - z5fo & 3 & * + 0.965222 & s i & 0.350 & 8.41201 & 9.69661&3do-3d & 2f & + 0.965576 & fe i & -0.778 & 4.73346 & 6.01759&y3do - e3f & 2 & + 0.966108 & c i & -0.264 & 7.48830 & 8.77172 & 3po-3s & 3f & + 0.967314 & cr i & -1.230 & 2.54400 & 3.82626&a5g - z5fo & 2 & + 0.967494 & s i & -0.090 & 8.40872 & 9.69031&3do-3d & 2f & + 0.967815 & ti i & -0.831 & 0.83605 & 2.11721&a5f - z5fo & 3 & * + 0.968346 & s i & 0.100 & 8.40982 & 9.69028&3do-3d & 2f & + 0.969133 & ca i & -0.422 & 4.73889 & 6.01830&3do-3p & 2b & + 0.969146 & ti i & -1.638 & 0.81300 & 2.09240&a5f - z5fo & 3 & * + 0.969651 & fe i ] & -3.115 & 5.08593 & 6.36467&x5fo - e7f & 2b & + 0.970242 & fe i ] & -2.215 & 5.06412 & 6.34208&x5fo - e7f & 2 & + 0.970459 & ca i & -0.169 & 4.74385 & 6.02151&3do-3p & 2 & + 0.970834 & ti i & -1.037 & 0.82591 & 2.10309&a5f - z5fo & 3 & * + 0.971964 & ti i & -1.590 & 2.30552 & 3.58121&z5do - a5d & 2 & + 0.972167 & ti i & -1.272 & 1.50259 & 2.77801&a1g - z1fo & 2 & + 0.973100 & ti i & -1.234 & 0.81820 & 2.09240&a5f - z5fo & 3 & * + 0.973719 & cr i & -1.377 & 2.54400 & 3.81740&a5g - z5fo & 2 & + 0.974037 & ti i & -1.882 & 2.29697 & 3.56995&z5do - a5d & 2b & + 0.974124 & fe i & 0.037 & 4.99161 & 6.26447&x5fo - e5 g & 2b & + 0.974625 & ti i & -1.334 & 0.81300 & 2.08521&a5f - z5fo & 3 & * + 0.974956 & ti i & -1.258 & 2.31807 & 3.58984&z5do - a5d & 2b & + 0.976610 & fe i & -0.468 & 5.09989 & 6.36951&x5fo - e5 g & 2b & + 0.976658 & fe i ] & -1.123 & 5.03341 & 6.30297&x5fo - e7f & 2b & + 0.977293 & ti i & -1.604 & 0.84848 & 2.11721&a5f - z5fo & 3 & + 0.978602 & ti i & -1.452 & 0.83605 & 2.10309&a5f - z5fo & 3 & * as blend + 0.978623 & ti i & -1.643 & 0.81820 & 2.08521&a5f - z5fo & 3 & * as blend + 0.979030 & ti i & -1.469 & 0.82591 & 2.09240&a5f - z5fo & 3 & * blend + 0.980352 & fe i ] & -2.223 & 5.09989 & 6.36467&x5fo - e7f & 2 & + 0.985746 & ca ii & -0.228 & 7.505644 & 8.763499&2po-2s & 2f & + 0.986444 & fe i & -0.299 & 5.06412 & 6.32109&x5fo - e5 g & 2f & + 0.989175 & fe i & -0.125 & 5.03341 & 6.28690&x5fo - e5 g & 2 & + 0.989334 & ca ii & -0.601 & 8.438551 & 9.69184&2fo-2 g & 3f & + 0.992711 & fe i ] & -3.095 & 3.54676 & 4.79579&a1d - y3do & 2 & + 0.993006 & ti i & -1.580 & 1.87933 & 3.12799&a3g - y3fo & 2 & + 0.993410 & ca ii & 0.072 & 7.515348 & 8.763499 & 2po-2s & 2f & + 0.994412 & ti i & -1.924 & 2.16043 & 3.40732&a3d - x3do & 2 & + 0.995181 & ti i & -1.831 & 2.15370 & 3.39963&a3d - x3do & 2b & + 0.996401 & na i & -0.820 & 3.61721 & 4.86162&2d-2fo & 2b & doublet , @xmath61 + 1.000032 & fe ii & -1.826 & 5.48450 & 6.72439&z4fo - b4 g & 2f & + 1.000070 & ti i & -1.840 & 1.87329 & 3.11313&a3g - y3fo & 2b & + 1.000580 & ti i & -1.135 & 2.16043 & 3.39963&a3d - x3do & 2 & + 1.001460 & ti i & -1.258 & 2.15370 & 3.39182&a3d - x3do & 2 & + 1.003720 & ti i & -2.002 & 1.46024 & 2.69556&b3f - z3go & 3 & + 1.003940 & sr ii & -1.312 & 1.80482 & 3.03988 & 2d-2po & 3 & + 1.005160 & ti i & -2.131 & 1.44325 & 2.67681&b3f - z3go & 3 & + 1.005210 & h i & -0.303 & 12.08830 & 13.32180 & 3 - 7 & 3f & + 1.006050 & ti i & -0.900 & 2.17485 & 3.40732&a3d - x3do & 2 & + 1.006270 & ti i & -2.278 & 1.42985 & 2.66206&b3f - z3go & 2 & + 1.006780 & fe i & -2.098 & 4.83518 & 6.06676&y3do - e3f & 2 & + 1.009498 & mg ii & 1.020 & 11.63047 & 12.85873&2fo-2 g & 2f & + 1.0115253 & n i & 0.595 & 11.75833 & 12.98412 & 4do-4f & 2f & + 1.011680 & fe i & -3.771 & 2.75876 & 3.98438&a3g - z3fo & 2 & + 1.0117413 & n i & 0.755 & 11.76464 & 12.990176&4do-4f & 2f & + 1.012370 & ti i & -1.672 & 2.17485 & 3.39963&a3d - x3do & 2b & + 1.012660 & c i & -0.032 & 8.53767 & 9.76209 & 1p-1po & 3f & + 1.014830 & fe i & -0.298 & 4.79579 & 6.01759&y3do - e3f & 2 & + 1.015790 & fe i & -4.229 & 2.17609 & 3.39674&a5p - z5fo & 2 & + 1.017030 & fe i & -4.061 & 2.19801 & 3.41718&a5p - z5fo & 2 & + 1.017326 & ti i & -3.292 & 1.44325 & 2.66206&b3f - z3go & 2 & + 1.019192 & ti i & -3.264 & 1.46024 & 2.67681&b3f - z3go & 2 & + 1.019600 & n ii & -0.790 & 4.08956 & 5.30564&1po-5/2[5/2 ] & 2 & + 1.019790 & fe i & -3.699 & 2.72773 & 3.94359&a3g - z3fo & 2 & + 1.021910 & fe i & -0.215 & 4.73346 & 5.94680&y3do - e3f & 2 & * + 1.022120 & fe i & -2.760 & 3.07155 & 4.28464&c3p - z3po & 2 & + 1.026800 & fe i & -4.467 & 2.22286 & 3.43042&a5p - z5fo & 2b & + 1.029180 & si i & -1.360 & 4.92042 & 6.12519 & 3po-3s & 2b & + 1.033010 & sr ii & -0.353 & 1.83958 & 3.03988 & 2d-2po & 3 & * + 1.033310 & n ii & -1.228 & 4.10569 & 5.30564 & 3go-3f & 2 & + 1.034370 & fe i & -3.493 & 2.19801 & 3.39674&a5p - z5fo & 3 & * as blend + 1.034670 & ca i & -0.408 & 2.93271 & 4.13109 & 1po-1s & 3 & * as blend , @xmath61 + 1.037411 & sii & -0.447 & 4.92965 & 6.12519 & 3po-3s & 2 & * + 1.038150 & n ii & -0.965 & 4.08846 & 5.28282 & ? & 2b & + 1.038184 & fe i & -4.065 & 2.22286 & 3.41718&a5p - z5fo & 2b & + 1.039860 & fe i & -3.330 & 2.17609 & 3.36848&a5p - z5fo & 3b & * as blend + 1.039961 & ti i & -1.537 & 0.84848 & 2.04076&a5f - z5go & 3b & * as blend + 1.042590 & fe i & -3.776 & 2.69259 & 3.88187&a3g - z3fo & 2b & + 1.042660 & fe i & -2.916 & 3.07155 & 4.26074&c3p - z3po & 2b & * + 1.045544 & [ fe i ] & ? & 0.99 & 2.17 & a3f - a5p & 2 & in arcturus list + 1.045830 & s i & 0.260 & 6.86061 & 8.04620 & 3so-3p & 3f & + 1.045960 & s i & -0.430 & 6.86061 & 8.04605 & 3so-3p & 2f & + 1.046230 & s i & 0.040 & 6.86061 & 8.04575 & 3so-3p & 3f & + 1.047250 & fe i & -1.229 & 3.88377 & 5.06775&z3do-3p & 2 & * + 1.048910 & cr i & -0.972 & 3.01083 & 4.19294&b5dz5do & 2 & + 1.049900 & ti i & -1.651 & 0.83605 & 2.01705&a5f - z5go & 3 & + 1.050438 & fe ii & -1.997 & 5.54914 & 6.72953&z4fo - b4 g & 2f & + 1.051290 & cr i & -1.558 & 3.01351 & 4.19294&b5dz5do & 2 & + 1.05324 & p i & 0.200 & 6.95482 & 8.13207 & 4p-4do & 2f & + 1.053340 & n ii & -1.189 & 4.10569 & 5.28282 & ? ? & 2 & + 1.053510 & fe i & -1.532 & 3.92887 & 5.10582&z3do-3p & 2 & + 1.054410 & c i & -1.290 & 8.53767 & 9.71361 & 1p-1po & 2f & + 1.058000 & fe i & -3.268 & 3.30163 & 4.47357&b3h - z3go & 2 & + 1.058450 & p i & 0.480 & 6.98570 & 8.15716&4p-4do & 2f & + 1.058750 & ti i & -1.777 & 0.82591 & 1.99703&a5f - z5go & 3b & * as blend + 1.058800 & si i & -0.020 & 4.95413 & 6.12519 & 3po-3s & 3b & * as blend , @xmath61 + 1.060630 & si i & -0.380 & 4.92998 & 6.09902 & 3po-3p & 3 & * as blend , @xmath61 + 1.061060 & ti i & -2.676 & 0.84848 & 2.01705&a5f - z5go & 3 & * as blend + 1.06146 & fe i & -0.336 & 6.16964 & 7.33777&n7fo-9/2[13/2 ] & 2 & + 1.06196 & fe i & -3.257 & 3.26733 & 4.43491&b3h - z3go & 2 & + 1.063060 & si i ] & 0.000 & 5.86287 & 7.02925 & 1p-3po & 2 & + 1.063883 & al ii & -0.610 & 15.30358 & 16.46905 & 3fo-3d & 2f & + 1.065050 & cr i & -1.613 & 3.01083 & 4.17502&b5dz5do & 2 & + 1.066390 & si i & -0.350 & 4.92042 & 6.08315 & 3po-3p & 3b & * as blend , @xmath61 + 1.066460 & ti i & -1.916 & 0.81820 & 1.98085&a5f - z5go & 3b & * as blend + 1.067047 & cr i & -1.489 & 3.01301 & 4.17502&b5dz5do & 2 & + 1.067510 & cr i & -1.374 & 3.01351 & 4.17502&b5dz5do & 2 & + 1.068000 & ti i & -2.507 & 0.83605 & 1.99703&a5f - z5go & 3 & + 1.068600 & c i & 0.076 & 7.48328 & 8.64360 & 3po-3d & 3f & + 1.068830 & c i & -0.276 & 7.48090 & 8.64098 & 3po-3d & 3f & + 1.069260 & si i & -0.190 & 5.95413 & 7.11374 & 3d-3fo & 2 & * + 1.069420 & c i & 0.348 & 7.48830 & 8.64774 & 3po-3d & 3f & + 1.069720 & si i & -0.060 & 5.96433 & 7.12345 & 3d-3fo & 2 & + 1.071030 & c i & -0.401 & 7.48328 & 8.64098 & 3po-3d & 3f & + 1.072920 & ti i & -2.064 & 0.81300 & 1.98652&a5f - z5go & 3b & + 1.073030 & si i & 0.060 & 5.98441 & 7.13995 & 3d-3fo & 3b & + 1.073250 & c i & -0.401 & 7.48830 & 8.64360 & 3po-3d & 3f & + 1.073590 & ti i & -2.504 & 0.82591 & 1.98085&a5f - z5go & 3 & + 1.075230 & si i & -0.300 & 4.92998 & 6.08315 & 3po-3p & 3 & + 1.075595 & fe i & -1.902 & 3.95999 & 5.11277 & z3do-3p & 2 & + 1.075690 & c i & -1.590 & 7.48830 & 8.64098 & 3po-3d & 2f & + 1.077770 & ti i & -2.658 & 0.81814 & 1.96852&a5f - z5go & 3 & + 1.07836 & fe i & -3.414 & 3.23694 & 4.38676 & b3h - z3go & 2 & + 1.078600 & fe i & -2.597 & 3.11117 & 4.26074&c3p - z3po & 2 & + 1.078751 & si i & -0.910 & 5.96433 & 7.11374 & 3d-3fo & 2 & + 1.078980 & si i & -0.390 & 4.92998 & 6.07914 & 3po-3p & 3 & + 1.08043 & cr i & -1.715 & 3.01135 & 4.15897 & b5d - z5do & 2 & + 1.081410 & mg i & -0.320 & 5.94632 & 7.09290 & 3d-3fo & 2b & @xmath61 + 1.082130 & fe i & -1.998 & 3.95999 & 5.10582&z3do-3p & 2b & + 1.082470 & cr i & -1.678 & 3.01351 & 4.15897&b5d - z5do & 2 & + 1.083010 & si i & 0.220 & 4.95413 & 6.09902 & 3po-3p & 3 & + 1.083091 & ti i & -3.878 & 0.83605 & 1.98085&a5f - z5go & 2b & + 1.08378 & na i & -0.500 & 3.61721 & 4.76129&2d-2fo & 2 & @xmath61 + 1.084680 & si i & -0.310 & 5.86287 & 7.00599 & 1p-1do & 2 & + 1.086562 & fe ii & -2.121 & 5.58957 & 6.73071&z4fo - b4 g & 2f & + 1.086650 & fe i ] & -0.803 & 4.73346 & 5.87452&y3do - e5f & 2 & + 1.087180 & si i & 0.040 & 6.19142 & 7.33192&3fo-3/2[9/2 ] & 2b & + 1.087250 & si i & 0.320 & 5.08269 & 6.22311 & 1po-1d & 3b & @xmath61 + 1.08847 & fe i ] & -3.713 & 2.84523 & 3.98438 & b3p - z3fo & 2 & + 1.088578 & si i & -0.870 & 5.98441 & 7.12345 & 3d-3fo & 2 & + 1.08883 & si i & 0.040 & 6.18113 & 7.31990&3fo-3/2[7/2 ] & 2 & + 1.089930 & fe i & -2.729 & 3.07155 & 4.20917&c3p - z3po & 2 & + 1.090886 & cr i & -0.647 & 3.43820 & 4.57480&y7po - e7s & 2 & + 1.091723 & mg ii & 0.020 & 8.86425 & 10.00002 & 2d-2po & 2fb & + 1.091790 & sr ii & -0.638 & 1.80482 & 2.94051 & 2d-2po & 3 & + 1.091827 & mg ii & -0.930 & 8.86436 & 10.00002 & 2d-2po & 2fb & + 1.093287 & cr i & -2.330 & 3.01351 & 4.15897 & b5d - z5do & 2 & amidst cn bands + 1.094110 & h i & 0.002 & 12.08830 & 13.22160 & 3 - 6 & 3f & + 1.095477 & mg ii & -0.230 & 8.86436 & 9.99622 & 2d-2po & 2f & + 1.095630 & mg i & -0.852 & 5.93194 & 7.06364 & 3po-3d & 2b & @xmath61 + 1.096023 & cr i & -2.067 & 3.01351 & 4.15897 & b5d - z5do & 2b & + 1.096030 & mg i & -0.690 & 5.93235 & 7.06364 & 3po-3d & 2b & @xmath61 + 1.096845 & mg i & -0.250 & 5.93319 & 7.06363 & 3po-3d & 2b & cn band head confused , @xmath61 + 1.097038 & p i & 0.120 & 8.07889 & 9.20914 & 2d-2do & 2f & + 1.098231 & sii & -0.590 & 4.95413 & 6.08315 & 3po-3p & 3 & + 1.098284 & n ii & -0.929 & 4.15386 & 5.28282&y3do - e3d & 3f & + 1.09851 & si i & -0.050 & 6.19142 & 7.32015&3fo-3/2[7/2 ] & 2 & + 1.09875 & si i & -0.500 & 6.19142 & 7.31990&3fo-3/2[7/2 ] & 2 & + 1.10163 & fe i ] & -1.334 & 4.79579 & 5.92133&y3do - e5f & 2b & + 1.101870 & cr i & -0.515 & 3.43820 & 4.57480&y7po - e7s & 2 & + 1.102100 & si i & 0.610 & 6.20642 & 7.33148&3fo-3/2[9/2 ] & 3 & + 1.102287 & ki & -0.010 & 2.67014 & 3.79501 & 2d-2fo & 2 & doublet , wavelength off , @xmath61 + 1.104770 & cr i & -2.508 & 3.01135 & 4.13369&b5d - z5do & 2 & + cccl 1.0963141 & 1.07008 + 1.0963299 & 1.07359 + 1.0963458 & 1.10119 + 1.0963615 & 1.08921 + 1.0963775 & 1.11744 + 1.0963933 & 1.14190 + 1.0964092 & 1.08628 + 1.0964249 & 1.06581 + 1.0964408 & 1.08024 + 1.0964566 & 1.03010 + of the continuum level and appearing in table [ tab : lines ] ( bottom panel ) . the lines are divided into @xmath62 bins according to the energy of the lower transition level . most lines are in the @xmath63 range , but there is a tail extending to higher excitation potential , up to @xmath64 . while the strong lines are more uniformly distributed in the peak excitation potential range , they are still representative of the whole of the line list . [ fig : ephisto ] ] ( vertical dotted lines with element names ) identified relative to the supergiant spectroscopic standards hd 38232 , an f5 ii ( black line ) and hd 37536 , an m2 iab ( green line ) . weaker atomic lines from our completed catalog are marked at the top ( short black hashes ) as are molecular lines ( short red hashes ) ( a color version of this figure is available in the online journal . ) . [ fig : fmgiant_pg1],height=672 ] ( vertical dotted lines with element names ) identified relative to the dwarf spectroscopic standards hd 55052 , an f5 v ( black line ) and hd 1326 , an m2 v ( green line ) . weaker atomic lines from our completed catalog are marked at the top ( short black hashes ) as are molecular lines ( short red hashes ) ( a color version of this figure is available in the online journal . ) . [ fig : fmdwarf_pg1],height=672 ] ( black ; containing 26 lines ) . the three h lines are excluded from the calculation of the solid black profile but are shown separately for the f stars as the dotted black line ( in the cooler standards the h profile is too heavily blended with neighboring features for the absorption profile to be informative ) . only lines that are unblended and strong enough to be tabulated in table [ tab : lines ] are included . a clear trend can be seen with hotter stars having stronger high - excitation lines compared to cooler stars having stronger low - excitation lines . additionally , supergiants tend to have stronger lines then the dwarfs in the same spectral class . [ fig : aveprofs ] ] as those of strength three with significant temperature and/or gravity sensitivity within our spectral grid . generally , only one line from among those in a multiplet set is shown ; its behavior is typical of all lines in the multiplet . the three columns depict lines with higher excitation potential ( left ) , lines with lower excitation potential ( right ) , and lines sensitive to surface gravity ( center ) . within each column the panels are ordered from top to bottom alphabetically by line species and then by increasing wavelength . [ fig : equivwidths ] ] ) @xmath50 sr ii ( @xmath51 ) , the lines with the strongest temperature and gravity trends in our spectra , while the right side shows ca i ( @xmath49 ) @xmath50 sr ii ( @xmath51 ) which are chosen due to their proximity in wavelength . bottom panel : the ratio of line strengths for two temperature sensitive lines with opposite temperature dependence , ti i ( @xmath65 ) and c i ( @xmath66 ) on the left side and ti i ( @xmath52 ) and c i ( @xmath53 ) on the right side , as a function of stellar effective temperature . squares represent supergiants , triangles the giants , and crosses the dwarf standards . although strong dependences are evident , the illustrated lines connect the existing data only and should not be misinterpreted as defining the detailed trends with temperature . [ fig : diagnostic ] ]

challenging questions appear often at edges of traditional fields . as an example , the new branch of mathematical physics , the `` statistical topology '' emerged recently by absorbing ideas from the statistical physics , theory of integrable systems , and algebraic topology ( see , @xcite for review ) . the scope of the statistical topology includes , on the one hand , mathematical problems involved in the construction of topological invariants of knots and links based on solvable models and , on the other hand , the physical and statistical problems related to summation over knot ensembles . in this work , we dwell predominantly to problems of the latter kind , demonstrating the emergence of a critical behavior in ensemble of @xmath0 torus knots . this critical behavior is formulated in terms of knot invariants . torus knots @xmath3 seem to be among the simplest objects in the knot theory . it is difficult to overestimate their role in different branches of mathematical physics . the topology of a torus knot is uniquely determined by the pair @xmath4 , which fixes windings along two torus periods . in the fig.[fig:01 ] few particular examples of torus knots , @xmath5 from the series @xmath0 are depicted . the closed curves wrap around the torus which is not shown . : @xmath5.,width=302 ] various explicit expressions for knot invariants are known , ranging from classical jones kauffman polynomials @xcite , to recent superpolynomials of torus knots @xcite . new approaches to the construction of torus knot invariants for particular knots , based on the application of topological string theories and deformed matrix models , have been formulated relatively recently in @xcite . much less is known about properties of _ knot ensembles _ , where the particular topology of a knot diagram is considered as a topologically `` quenched '' variable similar to the quenched disorder in statistical physics . the weighted summation over different torus knot types ( i.e. different pairs @xmath4 ) means the consideration of the grand canonical ensemble ( i.e. of the generating function ) of torus knots . in this work we uncover the relation between three physical problems : i ) two - dimensional directed ( i.e. ( 1 + 1)-dimensional ) lattice random walks with fixed area under the curve , ii ) ensemble of @xmath0 torus knots , and iii ) a five - dimensional sqed with the chern simons term at the level one . the reason for random walks to appear in ii ) and iii ) can be intuitively explained as follows . the main tool for the evaluation of the torus knot superpolynomials @xcite and of nekrasov partition function in the susy gauge theory @xcite , is the `` equivariant localization '' approach , which reduces the integral over the particular moduli space to the summation over the young tableau . the last problem can be reformulated as a weighted sums over directed paths on a square lattice . the schematic relations between the problems considered in the paper is shown in the flowchart in the fig.[fig:02 ] . our main goal is the investigation of the critical behavior in these topological ensembles . from the `` random walk side '' the critical behavior occurs in the space of fugacities and we shall focus at the `` area+length+corners '' statistics of paths . we derive an explicit expression for the free energy in the random walks problem for the `` area+length+corners '' statistics , which provides the very nontrivial phase structure in the three - dimensional space of fugacities . in the generic case we find a cascade of phase transitions and in the degenerate cases we reproduce the previously known results . the relation between the random walks and the torus knot superpolynomials can be traced from the mathematical literature , however the relation with the particular observable in 5d sqed instanton ensemble is new . collecting different points of view , and the results of @xcite , we interpret the generating function of ensemble of @xmath0 knots as the weighted sum of instanton contributions to the particular observable in 5d sqed and analyze the structure of corresponding generating function . the interpretation of the random walk fugacities in terms of the generating parameters in the torus knot ensemble is quite straightforward , moreover , these fugacities are identified in the 5d sqed as well : the corresponding parameters turn out to be the gauge coupling , the mass of the hypermultiplet , and the parameters of the @xmath1-deformation . given an explicit expression for the free energy in the random walks problem for the `` area+length+corners '' statistics , we use it to analyze the ensembles of knots and instantons . we show that at the `` gauge theory side '' the particular third derivative of the instanton partition function with respect to the masses exhibits an unexpected critical behavior at some critical line in the space of parameters . we provide a physical interpretation of critical behaviors in all three theories considered here . having an exact expression for the @xmath2-narayana numbers ( the generating function of the area- and corner - weighted ( 1 + 1)d brownian excursion ) , we describe explicitly the phase portrait of different ensembles . we show that there is a 3rd order phase transition in the instanton ensemble , corresponding to the 1st order phase transition in the ensembles of random paths and torus knots . the paper is organized as follows . in the section 2 we describe the sum over the paths with different statistics and focus at the `` area+length+corners '' ensembles . in the section 3 we discuss the representation of the partition function of ( 1 + 1)d brownian excursions ( dyck paths ) for the superpolynomials of @xmath0 torus knots . the section 4 is devoted to the identification of the sum over the paths as the instanton contribution to the particular observable in the 5d sqed . in the section 5 we consider the physical interpretation of the critical behavior in the space of fugacities in all models discussed at length of the paper . the relation with the toda - like integrable system is mentioned in section 6 . our findings and the open questions are summarized in the conclusion . in the appendix a , using the combinatorial description , we derive the new explicit expression for the generating function of @xmath6-narayana polynomials ( the generating function of area- and corner - weighted dyck paths ) . in the appendix b we remind for completeness the relation of the critical behavior with the `` hydrodynamic description '' of the edge singularities in gue matrix ensembles . by definition , the dyck path of length @xmath7 on the square lattice starts at the origin @xmath8 , ends at point @xmath9 and consists of the union of sequential elementary `` @xmath10 '' and `` @xmath11 '' steps , such that the path always stays above the diagonal of the square see the fig.[fig:03 ] . the number of all dyck paths of length @xmath7 is given by the catalan number , @xmath12 . we also denote dyck paths as `` brownian excursions '' ( be ) , having in mind an image of a charged particle on a square lattice in an external transversal magnetic field ( after applied wick rotation ) , where the motion of a particle is subject to two restrictions : it moves only up and right and never intersects the diagonal . calculating the action for such a particle , we see that @xmath13 is the fugacity of the area , @xmath14 , under the dyck path , and the @xmath15 is the fugacity of the path length , @xmath16 . the information about statistics of area - weighted dyck paths can be easily extracted from the generating function , which is the sum over all path lengths . this model can be referred to as the `` chiral hofstadter system '' , considered in @xcite . -narayana polynomials.,width=226 ] to proceed , turn the lattice by @xmath17 and write the recursion relation for the partition function @xmath18 on a half - line , @xmath19 , where @xmath20 is the height of the path at the step @xmath21 . the area , @xmath14 , below the path is counted as a sum of filled plaquettes ( i.e. `` heights '' ) between the path and the @xmath22-axis , as shown in the fig.[fig:04]a . each plaquette has the weight @xmath2 . -step dyck path ( the same figure as in the fig.[fig:03 ] ) turned by @xmath23 , the heights are measured along the dotted lines and the @xmath24-corners have fugacity @xmath25.,width=302 ] the partition function of area - weighted dyck paths , @xmath18 , satisfies the relation \ { l z_k+1(x;q ) = q^x-1 z_k(x+1;q ) + z_k(x-1;q ) + z_k(0;q)=0 + z_k=0(x;q)=_x,1 . [ eq:01 ] equivalently , ( [ eq:01 ] ) can be written in a matrix form . define @xmath26 . then _ n(q ) = t^n(q ) * z*_0 where t(q)= ( ccccc 0 & 1 & 0 & 0 & + 1 & 0 & q & 0 & + 0 & 1 & 0 & q^2 & + 0 & 0 & 1 & 0 & + & & & & ) ; _ 0= ( c 1 + 0 + 0 + ) [ eq:02 ] we are interested in the value @xmath27 since at the very last step the trajectory returns to the initial point . evaluating powers of the matrix @xmath28 , we can straightforwardly check that the values of @xmath29 are given by carlitz riordan @xmath2-catalan numbers , namely , z_n(1;q ) = \ { ll c_n/2(q ) & + 0 & . [ eq:03 ] recall that @xmath30 satisfy the recursion c_n(q ) = _ k=0^n-1q^k c_k(q ) c_n - k-1(q ) [ eq:04 ] which is the @xmath2-extension of the standard recursion for catalan numbers . the generating function @xmath31 obeys the functional relation f(s , q ) = 1 + s f(s , q ) f(sq , q ) [ eq:05 ] it is known that the solution of ( [ eq:05 ] ) can be written as a continued fraction expansion , f(s , q ) = = [ eq:06 ] where @xmath32 is the @xmath2-airy function , a_q(s)=_k=0^ ; ( t;q)_k=_k=0^k-1 ( 1-t q^k ) [ eq:07 ] let us describe how the critical behavior emerges at the brownian excursion side . in the works @xcite it has been shown that in the double scaling limit @xmath33 and @xmath34 the function @xmath35 has the following asymptotic form c(z ) ~c_reg+(1-q)^1/3 ( 4z ) ; z= , [ eq : asymp1 ] where @xmath36 is the regular part at @xmath37 and @xmath38 is the airy function . the function @xmath39 is the generating function for the undeformed catalan numbers : c(s , q=1)= the generating function @xmath39 is defined for @xmath40 , and at the point @xmath41 the first derivative of @xmath39 experiences a singularity which is interpreted as the critical behavior . the limit @xmath42 , @xmath43 can be read also from the asymptotic expression for @xmath35 , eq.([eq : asymp1 ] ) : c(s , q)|_q1 ^ -~c_reg-2 . [ eq : limq1 ] where @xmath36 is the regular part of @xmath35 at @xmath44 and @xmath45 and @xmath46 at @xmath41 . note that the first non - singular term in ( [ eq : limq1 ] ) does not contain @xmath2 , so it is no matter in which order the limit in ( [ eq : asymp1 ] ) is taken . however to define the double scaling behavior and derive the airy - type asymptotic , the simultaneous scaling in @xmath47 and @xmath2 is required . the generating function , @xmath48 , for narayana numbers , which count dyck paths with fixed fugacity of corners , @xmath25 , demonstrates the behavior similar to ( [ eq : limq1 ] ) , namely the square - root singularity , f(a , s)= [ eq : nar ] this behavior at @xmath49 coincides with the one of catalans ( [ eq : limq1 ] ) . in appendix a we have derived the explicit expression for the generating function @xmath50 of @xmath2-narayanas with the airy - type asymptotic ( [ eq : asymp1 ] ) in terms of @xmath2-orthogonal polynomials related to rogers ramanujan continued fractions , f(q , a , s)= [ eq : crit ] where @xmath51 is the extension of the @xmath2-airy function @xmath32 defined in ( [ eq:07 ] ) . the function @xmath51 reads a_q(s;s(1-a ) ) = _ k=0^ [ eq : crit2 ] one can immediately see that @xmath52 , where @xmath53 is given by ( [ eq:07 ] ) . completing this section it is worth reminding that appearance of the singularity of type ( [ eq : asymp1 ] ) is the manifestation of the third - order phase transition . in the seminal paper @xcite it has been shown that the largest eigenvalue , @xmath54 , of the gaussian @xmath55 random matrix ensemble , converges at @xmath56 to @xmath57 , where the random variable @xmath58 has a limiting @xmath16-independent distribution , @xmath59 , being the so - called tracy - widom distribution for gue ensemble @xcite . so , the normalized value @xmath60 at large ( but finite ) @xmath16 has an uncertainty ( i.e. the width of the distribution ) of order of @xmath61 , typical for the 3rd order phase transitions . above and below the critical value @xmath62 , the tails of the distribution @xmath63 have different asymptotics , signifying existence of strong ( for @xmath64 ) and weak ( for @xmath65 ) couplings . in this section we briefly explain the relation between the invariants of knots and the random walks with three fugacities . we restrict ourselves by the torus knots , @xmath66 , and consider the superpolynomial introduced in @xcite . they depend on three generating parameters , which can be related to the fugacities of 2d directed random walk . remind that the superpolynomial is the poincar polynomial of the triply graded khovanov homologies , @xmath67 , which in the vector space can be attributed to the knot . the knot superpolynomials are the generalizations of the homfly knot polynomials and depend on three variables , corresponding to gradings @xcite p_n , m(a , q , t)=_ijka^i q^j t^k h_ijk at @xmath68 the superpolynomial @xmath69 reduces to the standard homfly polynomial . alternatively , it can be interpreted as the generating function for the multiplicities in the particular sector of bps states in the susy gauge theories @xcite . we will be interested in the critical behavior of the `` area+length+corner '' type statistics in the ensemble of torus knots and focus on the particular series of @xmath0 uncolored knots parameterized by one integer , @xmath16 . the main object , as before , is the generating function for superpolynomials , @xmath70 , in the ensemble of @xmath0 knots , where the fugacity , @xmath47 , is conjugated to the index @xmath16 , which defines the winding around the cycle on the solid torus ( see the fig.[fig:01 ] ) , z(s , a , q , t)=_n=0^ p_n , n+1(a , q , t ) s^n hopefully there is an explicit expression for the superpolynomial of the @xmath0 torus knots obtained in two different ways . the first one deals with the combinatorics of the young diagrams and can be related to the brownian excurion approach @xcite , while the second approach has been developed in @xcite via the localization on the fixed points of the torus action in the moduli space of @xmath16 points in @xmath71 . the second approach will be used in the next section to compare the generating function for superpolyniomials of @xmath0 uncolored torus knots with the instanton contribution to the particular observable in 5d sqed with the chern simons term . we focus on the be representation and provide a dictionary identifying the fugacities at be side with the ones at the knot side . this dictionary establishes the translation of the be language to the language of torus knot superpolynomials . the construction of torus knot invariants involves two more statistics for dyck paths , the statistics of `` up - corners '' , where the path changes direction from `` up '' ( @xmath10 ) to `` right '' ( @xmath11 ) , and the statistics of `` dinv '' , a definition for which can be found in @xcite . the superpolynomial for @xmath0 knots expressed in terms of dyck paths can be presented by the following partition function : z_n(q , t , a ) = _ _ ndyck paths q^a t^#dinv a^#corners , [ eq:1 ] where @xmath2 , @xmath25 and @xmath72 are the fugacities of area , corners and dinv correspondingly . the main object of our study is the generating function , z(q , t , a , s ) = _ n=0^ z_n(q , t , a ) s^n [ eq:1a ] where s is fugacity for the length of the path conjugated to the integer @xmath16 , which weights knot type ( at a knot side ) and dyck path length ( at a be side ) . as before , the area , dinv , and corner statistics in the ensemble of lattice paths are represented by the corresponding fugacities in the grand ensemble . to use the known critical behavior for the generating function of superpolynomials in the ensemble of torus knots , we have to consider the reduction of paths statistics to the `` length+area+corners '' one , switching off the `` dinv '' fugacity ( i.e. setting @xmath73 in ( [ eq:1 ] ) ) . due to @xmath74 duality , one can equivalently consider the `` length+area+corners+bounce '' statistics , switching off the `` bounce '' fugacity . this duality can be seen , for instance , in the lowest row in the expansion in the @xmath25 variable of the superpolynomial : p_n , n+1(a , q , t)= _ k a^k p^k_n , n+1(q , t ) . it turns out that the corresponding term at @xmath75 in the expansion coincides with the @xmath76-catalan number , namely , @xmath77 , when the duality is well - known . in the next section we explain the relation between the generating function for the torus knot superpolynomials and the some observables in the 5d sqed upon the particular identification of parameters . in the section 6 we shall use the generating function for @xmath2-narayanas to analyze the critical behavior in the torus knots ensembles classified by the homfly knot invariants . consider now the nekrasov - like partition function @xcite in the abelian 5d susy gauge theory with the massless hypermultiplet in fundamental representation and massive multiplet in the antifundamental representation in the @xmath1background . the coefficient in front of the chern simons term is fixed , @xmath78 , the coupling constant in 5d theory is dimensionful , and the fifth coordinate is compact . the nekrasov partition function is trivial in this theory , however we shall be interested in the vacuum matrix element @xmath79 of the particular operator @xmath80 . its evaluation involves the weighted sum of the integrals over the instanton moduli space , where the parameters of the @xmath1background , @xmath81 , serve as the equivariant parameters of two torus actions for the integration over the moduli space , @xmath82 , of @xmath16 point - like instantons , all located at the origin . the instanton number is weighted with the counting parameter @xmath83 , where @xmath84 and @xmath85 is the radius of the compact fifth dimension . the desired operator @xmath80 can be identified as follows . first , we have to recognize the deformed catalan numbers , @xmath86 , with @xmath79 in the @xmath16instanton sector . to this aim we use the following important result @xcite ^t(hilb^n(c^2,0 ) , v ^n v ) = c_n(q , t ) , which interprets the @xmath76-deformed catalans as equivariant integrals over the moduli space of @xmath16abelian instantons valued in the @xmath16th power of the @xmath16dimensional tautological bundle , @xmath87 . looking at the representation of the @xmath76catalans in terms of the paths on the young tableau , the desired operator , up to the normalization , can be written in the @xmath16instanton sector as o _ n ( tr e^)_n , where @xmath88 is the hypermultiplet , @xmath89 substitutes the integral over the @xmath90 in the @xmath1background , and @xmath91 is the `` long scalar '' in the @xmath1background @xcite . hence , the composite operator under consideration , is the product of local and nonlocal 4-observables . it is more convenient to use the equivalent , however a bit more symmetric formulation of the desired observable in 5d sqed @xcite . consider the 5d qed with two flavors in the fundamental representation with masses @xmath92 , @xmath93 , and the flavor in the antifundamental representation with a mass @xmath94 . the following relation between the second derivative of the nekrasov partition function and the generating function for the @xmath0 knot superpolynomials , holds : @xmath95 this is equivalent to the evaluation of the correlator of local and nonlocal operators defined above due to the relation . |_m = ( tr e^ ) , we keep the parameter @xmath25 in the superpolynomial arbitrary , hence the complete list of the identifications reads as follows a=- e^-m_a , t = e^-_1 , q = e^-_2 thus , as above , we have the generating function depending on four fugacities . the relation between the derivative of the nekrasov partition function and torus knots superpolynomials is important per se , however in this paper we are focusing at the critical behavior in the topological ensembles . therefore , what we need is i ) the identification of the parameters of the gauge theory as the fugacities of the random walk , and ii ) expression of the generating function for the @xmath2-narayana numbers . to make use the known expression for @xmath2-narayana s , we have to switch off one equivariant parameter , @xmath96 . the critical behavior is now the @xmath97phase space , and in this formulation it is clear that we are dealing with the 3rd order phase transition . however , the new point is that the 3rd derivative of the nekrasov partition function is taken with respect to three distinct variables @xmath98 ( or , equivalently , to @xmath99 ) . the physical interpretation of the critical behavior is given in the next section . in this section we discuss the physical interpretation of the critical behavior found above . as we have already argued , the critical behavior is exact due to the explicit expression of the generating function for @xmath2-narayana s , which depends on three fugacities ( chemical potentials ) controlling length of the path , area under the path and corners . the interpretations of the fugacities and of the critical behavior are different at brownian excursion side , knot side , and 5d sqed side , so we consider them separately . however , these seemingly different critical behaviors reflect one and the same generic pattern of the phase transition . in terms of the ( 1 + 1)d random walks the critical behavior in the @xmath100 space has the following interpretation . in the limit @xmath101 the fugacity of the area is switched off and we deal with the generating function @xmath48 ( see ( [ eq : nar ] ) ) defined in the @xmath102plane . at large finite lengths , @xmath16 , the average number of corners , @xmath103 diverges with the length , @xmath16 , as @xmath104 . the corresponding contour plot of the generating function @xmath105 is shown in the fig.[fig:05 ] in the @xmath42panel . the critical lines divide the @xmath102plane in three domains . in the first domain ( `` phase 1 '' ) since @xmath47 is small , one has short and ( in average ) sufficiently wrinkled paths , with varying number of corners controlled by the fugacity @xmath25 . in the second ( intermediate ) domain the trajectories are long since they exceed the critical value for the fugacity , again with varying number of corners , however since the generating function @xmath106 diverges , currently we have not any physical interpretation of this phase . in the third domain ( `` phase 2 '' ) the trajectories are long and essentially wrinkled , since for any @xmath47 the value of @xmath25 is bounded from below : @xmath107 . for @xmath108 the behavior of the system becomes much more rich and the ensemble of area- and corner - weighted dyck paths exhibits the cascade of airy - type ( 3rd order ) phase transitions as it is seen from the contour plots for the function @xmath109 drawn at few fixed values of @xmath2 ( @xmath110 ) see the corresponding panels in the fig.[fig:05 ] . above the @xmath102plane for different fixed values of the area fugacity , @xmath2 . for @xmath42 the phase 1 corresponds to short paths , and the phase 2 to long and rather wrinkled paths . for @xmath108 one sees in the @xmath102plane the cascade of transitions with the airy - type asymptotics.,width=302 ] the asymptotic behavior of the function @xmath50 as a function of @xmath47 for fixed values @xmath2 and @xmath25 near the singularities ( transition points ) is better seen in the figure fig.[fig:06 ] . one clearly distinguishes the finite cascade of phase transitions at different @xmath111 and @xmath25 . for few fixed values of @xmath2 and @xmath25.,width=302 ] for completeness , we provide in the fig.[fig:07 ] two contour plots of the function @xmath112 at two fixed values of corner fugacity , @xmath113 and @xmath114 , which correspond to weakly and highly wrinkled paths . above the @xmath115 plane for fixed values @xmath116 . the cascade of transitions in clearly seen.,width=302 ] the nature of existence of the cascade of transitions lies in the lattice origin of the problem and deals with the incommensurability of `` optimal '' trajectories with respect to the lattice spacing for particular values of @xmath25 and @xmath2 and @xmath47 . actually , it is very easy to imagine that there are conflicts between wrinkled ( with large number of corners ) and `` inflated '' ( with extended area ) trajectories at some specific length . on the lattice this competition leads to a finite number of phase transitions , manifested in the divergence of the generating function @xmath50 . in the ensemble of the @xmath0 uncolored torus knots , the situation when we can use the exact answer for the @xmath2narayana s corresponds to the unrefined case with 3d space of parameters @xmath100 . hence , we consider generating function for homfly polynomials , @xmath117 , for @xmath118 torus knots . what could be the interpretation of the phase transitions in the ensembles of the torus knots ? naively , different behavior of the generating function at different ridges of the transition point , means different ability to recognize a particular knot in the ensemble . speaking more formally , we could try to distinguish phases via the derivatives of the generating function with respect to the particular generating parameters , that is the average values of the corresponding `` topological numbers '' . the derivative with respect to the variables yields the average number of winding , or in other terms , the type of the torus knot . however the interpretation of other topological numbers is less evident . on the other hand , these quantum numbers are quite transparent if we remember the that knot invariants are related to the enumeration of bps states . along this approach , other generating parameters count the number of the corresponding branes . let us remind , that in a pure `` knot framework '' , one can obtain the @xmath119 invariants of the knots upon the substitution a = q^n hence , the @xmath120 limit corresponds to the situation when all @xmath119 invariants are `` degenerate '' and can not be distinguished . at the critical line in the @xmath121 plane in the ensemble of torus knots , the average winding @xmath122 diverges and the average value @xmath123 diverges as well . on the critical line they are related as n hence , in the ensemble of torus knots , the stable limit ( @xmath56 ) is reached at the critical line , where we also forget about the `` nonabelian character '' of knot invariants . note that the counting problem can be formulated in terms of the seifert surface , which for the @xmath0 torus knots reads as x^n = y^n+1 where @xmath124 . for instance , the alexander polynomials can be formulated in terms of the monodromy in the cohomologies related to the seifert surface . also , the alexander polynomials can be interpreted in terms of the stratification of the algebra of functions on the seifert surface . hence , the critical behavior can be formulated in terms of the seifert surface as well . whether one can relate the type of the seifert surface with the singularity type near the transition point of the generating function , is still an open question . let us comment on the interpretation of the cascades of the phase transitions at `` torus knot side '' . assuming that there are infinite number of phases in the generic situation , it is naturally to conjecture the domination of the different values or windings in different phases . turn finally to the third part of our construction . in the ensemble of instantons , the account of three fugacities @xmath100 corresponds to the account of the coupling constant , mass and parameter of the @xmath1deformation in the unrefined limit . the critical cascades in this case can be derived again from the analytic structure of the @xmath2narayana generating function , @xmath109 . the limit @xmath44 corresponds to switching off the @xmath1deformation . hence , the @xmath102phase space corresponds to the space of the mass , @xmath94 , of the hypermultiplet in the antifundamental representation @xcite , and the instanton charge counting parameter , @xmath125 . the critical line in the physical @xmath102variables is defined by the equation ( 1+q - qe^m_a)^2=4q [ eq : q ] there are two critical lines , @xmath126 and @xmath127 , defined by ( [ eq : q ] ) at the particular relation between the gauge coupling and mass of the antifundamental representation . configurations with different number of instantons dominate in the different regions of the @xmath102space . at the critical line we have the following scaling behavior when the parameter of the @xmath1deformation is switched back again , the cascade of the singular surfaces emerges , dividing the three - dimensional parameter space in the regions . since the parameter @xmath2 corresponds in the 5d sqed to the fugacity for the angular momentum in some plane in @xmath128 , the critical surfaces separate the regions with the different responses of our observable to the rotation . when @xmath75 , the antifundamental becomes massless and we have the critical cascades in @xmath129plane ; their scaling is dictated by the analyitic structure of the @xmath2-airy functions . in the simplest case when only the @xmath47parameter matters , the critical behavior of the @xmath2catalans corresponds to the following values of parameters in the 5d gauge theory _ 1 = 0,_20,s = -8 ^ 2 g^-2 = 42 the general issues concerning the 3rd order phase transitions will discussed in the separate publication @xcite . let us point out that the scaling function @xmath130 , which appeared many times throughout the text ( see , for instance , ( [ eq : asymp1 ] ) , or ( [ ricatti2 ] ) ) , plays also a very important role , connecting be to the integrable systems . it is known ( see @xcite and the references therein ) that @xmath131 , defined in ( [ ricatti2 ] ) , is itself a generating function : w(z)|_z ~_k=0^ _ k z^-(3k-1)/2 , [ eq : gen2 ] where the coefficients @xmath132 have well - defined physical sense : representing @xmath132 in the form @xmath133 where @xmath134 is the gamma - function . one can show @xcite that @xmath135 is the @xmath16th moment of the area under the brownian excursion on the unit interval . defining @xmath136 one sees that @xmath137 satisfy the recursion k_n = k_n-1+_k=1^n-1k_k k_n - k ; k_0=- . [ eq : rec3 ] on the other hand , the function @xmath138 appears it the theory of integrable systems as the solution of the rational painlev ii equation ( @xmath139 ) at @xmath140 : u(x)=2u^3(x)+ 4x u(x)+ 4(+ ) . [ eq : painleve ] the connection of area - weighted generating function @xmath138 with the rational solutions of painlev ii is not restricted exclusively by ( [ eq : painleve ] ) , and can be pushed for any @xmath141 , where @xmath142 . take into account that the rational solutions of ( [ eq : painleve ] ) can be written ( see @xcite ) as @xmath143 , where @xmath144 is the @xmath145-function of the toda system ( see @xcite ) , written as a hankel determinant _ n= ( cccc a_0 & a_1 & & a_n-1 + a_1 & a_2 & & a_n + & & & + a_n-1 & a_n & & a_2n-2 ) [ eq : det1 ] and the entries @xmath146 satisfy the recursion @xcite a_n = 2(n-2 ) a_n-3 + _ k=0^n-2 a_k a_n - k-2 , [ eq : rec2 ] with @xmath147 . the associated generating function , g(x , s)=_j=0^a_j(x ) ( -2s)^-j , [ eq : gen1 ] obeys the riccati equation @xcite ( compare to ( [ ricatti ] ) ) -2 g + g^2 + _ s g - ( 4s^2+s^-1)g + 4 x s^2=0 , [ ricatti3 ] whose solution is g(x , s ) = 2s^2 + ( s^2-x ) . [ eq : airy2 ] the equation ( [ eq : rec2 ] ) at large @xmath16 resembles ( though being different in details ) the recursion ( [ eq : rec3 ] ) for the function @xmath137 . the connection between ( [ eq : rec3 ] ) and ( [ eq : rec2 ] ) can be set by comparing ( [ eq : gen2 ] ) and ( [ eq : gen1 ] ) . finally , we get a_j=(-2)^j_k=0^ _ j=0^ ( c + m ) ( -x)^m _ 3k+2m+1,j , where @xmath148 is the kronecker @xmath149function . thus , we explicitly see how the linear combinations of moments of area - weighted brownian excursions , @xmath132 , are connected to the coefficients @xmath150 in the expansion of the toda @xmath145function . let us also point out the striking similarity of the recursion equation ( [ eq : rec3 ] ) for different momenta of area - weighted dyck paths with the summation over genus , @xmath151 , the partition function of the @xmath152 chern simons matter theory , also known as the abjm theory @xcite . we plan to discuss this question in details in the forthcoming paper @xcite . in this paper , the consideration of the directed random walks with the `` area+length+corners '' statistics allows us to find a new critical behavior in the ensemble of the @xmath0 torus knots and in the ensemble of instantons in 5d sqed . this critical behavior takes place in the space of fugacities and the corresponding parameters in the gauge theory . the result is exact due to the explicit answer of the generating function for paths with these statistics . in the space of fugacities there are patterns of critical lines , and cascades of the phase transitions . the physics of the phase transitions at this critical line for the ensembles of torus knots and instantons is very rich . in both cases at the critical line the contribution to the quantity under consideration with the infinitely large topological number starts to dominate . these quantities are windings in the knot case , or topological charges in the 5d sqed . in the gauge theory we have the critical lines in the ( gauge coupling , mass ) parameter plane . the phase transition in this plane is physically meaningful and certainly deserved detailed investigation . even more profound behavior is seen when all three fugacities are nontrivial . the critical cascades are quite unexpected and physically they take place in the unrefined limit of the @xmath1deformed gauge theory . at the knot side , the critical behavior occurs in the generating function of the @xmath0 knots characterized by the homfly polynomials . the following general comment is in order . one has to distinguish two `` critical phenomena '' : phase transition and the wall crossing phenomena . the latter is typical for the supersymmetric gauge theories when the spectrum of bps states jumps at some curves and surfaces in the moduli space . upon the susy breaking , the curves of marginal stability become the curves of phase transitions . hence literally the wall - crossing should not be considered as a phase transition , since only the spectrum of the stable bps states gets rearranged . however , there is a pattern of real phase transitions , even in susy theory , with the argyres douglas superconformal point in sqcd as the notable example . in this case the particular condensates serve as the order parameter . do we deal with the phase transition or with the wall - crossing phenomena in our case ? formally , the 3rd derivative of the nekrasov partition function in 5d sqed diverges , hence we could speak about the 3rd order phase transition in the instanton ensemble . moreover , the uv properties of the condensate play the key role , and critical lines involve the hypermultiplet masses and the gauge coupling . hence , we could speak on some similarity with the argyres douglas scenario . on the other hand , at the torus knot side , we could remind the interpretation of the knot superpolynomial as the index of the bps states in some theory . hence the jump in the index is better interpreted in terms of wall - crossing . as far as we know , there is no evident notion of the wall - crossing in terms of random walks , however , we cold think about it in terms of the stokes lines in quantum mechanics . summarizing , we think that the critical behavior we have described , admits the interpretation both as the wall - crossing , and as the phase transition , depending on the viewpoint . the critical behavior we have considered in 5d sqed is a nontrivial example of the 3rd order phase transition . the familiar example of the 3rd order phase transition in the gauge theories is the phase transition of douglas kazakov type @xcite in 2d pure yang - mills theory on the sphere characterized by the singularity ( [ eq : asymp1 ] ) . somewhat similar phase transition of douglas kazakov type has been observed in @xcite . in all these situations the phase transitions are driven by instantons . on the other hand , our study , as well as @xcite , yield the clear relation between the torus knots and instantons . this suggests more general viewpoint that there is a knotting process behind the 3rd order phase transitions in the general case and the knot invaraints play the role similar to the central charges of the virasoro algebra in the 2nd order phase transitions . we shall provide the general discussion of the 3rd order phase transition in the gauge theories , knot ensembles and statistical mechanics elsewhere @xcite . there are many challenging issues , related to our work , which deserve further study . for example , it would be interesting to extend the analysis to the whole ensemble of @xmath66 torus knots and the whole set of fugacities . also , it would be important to recognize the counterparts of the `` bounces '' and `` corners '' in the particle path integral in the continuum ( i.e. off lattice ) and obtain the interpretation of the external magnetic field in the be approach as a kind of the berry curvature from a 4d viewpoint . it would be also very interesting to extend our consideration and to relate the spectrum of the full ( i.e. non - chiral ) hofstadter model at be side with knot invariants and 4d instantons . we plan to address all these questions in @xcite . it seems also very important to recognize the critical behavior observed in this paper in the ensembles of branes @xcite , hopfions @xcite and `` ensemble '' of 3d theories classified by the torus knots @xcite . another interesting question concerns the physical interpretation of the critical behavior in terms of the extended objects involved in the torus knot description along the discussion in @xcite . this is the explicit realization of the @xmath153-dual magnetic approach to the evaluation of the knot invariants suggested in @xcite . we are grateful to e. gorsky , a. milekhin and n. nekrasov for the useful discussions . the work of a.g . and k.b . was supported in part by grants rfbr-12 - 02 - 00284 and pics-12 - 02 - 91052 . the work of k.b . was also supported by the dynasty fellowship program and princeton centennial fellowship . thanks scgp at stony brook university where the part of this work has been done during the simons workshop on mathematics and physics 2014 for the hospitality and support . let us associate each entrance of a dyck path at a height @xmath20 with the application of a `` generator '' @xmath154 . define the semigroup @xmath155 with the infinite set of generators @xmath156 , obeying the commutation relations _ j _ k = _ k _ j |k - j|2 [ a:08 ] any dyck path can be uniquely encoded by a `` word '' written in terms of generators @xmath154 ( @xmath157 ) . for example , the dyck path in the fig.[fig:03 ] corresponds to a word = _ 0_3_2_1_2_1_3_3 _ 2 _ 1 _ 1_0 _ 3_2 _ 1_1_1_0 [ a:09 ] ( see also the fig.[fig:04 ] ) . define a `` normal order '' representation of words in the semigroup @xmath155 , which consists in pushing the generators with smaller indices _ as left as possible _ if such a reordering does not violate the commutation relations @xcite . let us note that all dyck paths written in terms of generators @xmath158 ( @xmath157 ) are automatically normally ordered . thus , we can compute the partition function of all such dyck paths via the transfer matrix approach , where the transfer matrix , @xmath159 , defines which particular generator , @xmath160 , can stay next to the previous one , @xmath154 : * if @xmath22 then @xmath161 ; * if @xmath162 then @xmath163 ; * if @xmath164 then @xmath165 graphical representation of the rules ( i)-(ii ) is shown in the fig.[fig:08 ] . if the previous index is @xmath22 ( i ) , @xmath166 ( ii ) and @xmath167 ( iii).,width=226 ] in order to count heights , identify a weight @xmath168 with the generator @xmath154 . then the weighted transfer matrix , @xmath169 , consistent with the rules ( i)-(iii ) has the following form r(q)= ( ccccc 1 & q & q^2 & q^3 & + 1 & q & q^2 & q^3 & + 0 & q & q^2 & q^3 & + 0 & 0 & q^2 & q^3 & + & & & & ) [ a:10 ] define @xmath170 . the number of dyck paths , starting at the height @xmath22 and returning to the height @xmath22 is given by the element @xmath171 of the vector @xmath172 , where w_n(0;q ) = * v * r^n(q ) * v*^ ; = ( 1,0,0 , ... ) [ a:11 ] one can straightforwardly check that w_n(0;q ) = c_n(q)=z_2n(1;q ) [ a:12 ] where @xmath30 is the @xmath173 carlitz riordan @xmath2-catalan number and @xmath27 is given by ( [ eq:03 ] ) . thus , the magnetic random walks can be equivalently described either by the transfer matrix @xmath174 ( eq.([eq:02 ] ) ) , or by the transfer matrix @xmath169 ( eq.([a:10 ] ) ) . the last description can be straightforwardly generalized to account for corners . the approach developed in the section [ sect:2:2 ] can be easily extended on counting of simultaneously area - weighted and @xmath24-corner - weighted @xmath7-step dyck paths shown in the fig.[fig:04 ] . specifically , we are interested in calculation of the partition function , @xmath176 , defined in ( [ eq:1 ] ) , where the summation runs over the ensemble of @xmath7-step dyck paths enclosing the area @xmath14 , controlled by the fugacity @xmath2 , and having @xmath93 @xmath24-corners , controlled by the fugacity @xmath25 ( bounces are not counted , so we set @xmath73 in ( [ eq:1 ] ) ) . note that the @xmath24-corner appears in position @xmath21 if and only if the following condition is fulfilled : @xmath177 . in the opposite situation , i.e. for @xmath178 , the @xmath24-corner is not created . in terms of the transfer matrix approach , which counts the normally ordered words in the semigroup @xmath155 , the @xmath24-corner creation condition means that for all steps which do not decrease the height we should apply the weight @xmath25 ( the fugacity of the corner ) . thus , we get the following @xmath175-deformed transfer matrix @xmath179 , where we have multiplied all entries of matrix @xmath169 , except the lower sub - diagonal , by a factor @xmath25 : r(q , r)= ( ccccc a & aq & aq^2 & aq^3 & + 1 & aq & aq^2 & aq^3 & + 0 & q & aq^2 & aq^3 & + 0 & 0 & q^2 & aq^3 & + & & & & ) [ a:14 ] ( compare to ( [ a:10 ] ) ) . evaluation of the area- and corner - weighted partition function , @xmath180 is similar to ( [ a:11 ] ) , namely : _ n(q , a)z_2n(q , a)=*v * r^n(q , a ) * v*^;=(1,0,0 , ... ) [ a:15 ] in particular , for @xmath181 we get the narayana polynomial , which is the generating function for the narayana numbers @xmath182 , _ n(q=1,a ) = _ k=1^k n_n , ka^k = _ 2f_1(1-n ,- n,2,a)a ; where @xmath183 is the hypergeometric function and _ n , k= ( c n + k ) ( c n + k-1 ) [ a:16 ] for example , for @xmath184 both eq.([a:15 ] ) ( in which we take @xmath42 ) and eq.([a:16 ] ) give z_2n(q=1,a ) = a + 21 a^2 + 105 a^3 + 175 a^4 + 105 a^5 + 21 a^6 + a^7 meaning that in the ensemble of @xmath185step dyck paths has 21 different configurations with 2 @xmath24-corners , 105 configurations with 3 @xmath24-corners etc . for any @xmath175 the recursion relation for @xmath186 reads _ n(q , a ) = a n_n-1(q , a)+_k=1^n-1q^k n_k(q , a)n_n - k-1(q , a ) [ a:17 ] ( compare to ( [ eq:04 ] ) ) . the generating function @xmath187 obeys the functional relation f(q , a , s)=1+(a-1)s f(q , a , s)+s f(q , a , s)f(sq , a , s ) [ a:18 ] ( compare to ( [ eq:05 ] ) ) . these polynomials coincide with the standard @xmath2-narayana polynomials @xcite . for @xmath42 we reproduce the generating function for narayana numbers @xcite : f(a , s)= [ a:18a ] consider the recursion relation _ j+1(q , a ) = r(q , a ) * w*_j(q , a ) [ a:19 ] for the vector @xmath188 , where @xmath189 and the transfer - matrix @xmath179 is defined in ( [ a:14 ] ) . one can easily rewrite ( [ a:19 ] ) for components of the vector @xmath190 : @xmath191 for the generating function , @xmath192 we obtain from ( [ a:20 ] ) s^-1 |f(x ) - s^-1 = q^x-2 |f(x-1 ) + a_y=1^l q^y-1 |f(y ) [ a:21 ] which can be rewritten in the local form relation looks pretty much similar to the recursion relation for orthogonal polynomials associated with rogers ramanujan continued fraction see @xcite : l u_m+1(z;b , c ) = z(1 + b q^m ) u_m(z;b , c ) + - c q^m-1 u_m-1(z;b , c ) ; ( m > 0 ) + u_0(z;b , c)=1 ; u_1(z;b , c)=z(1+b ) [ a:24 ] the function @xmath193 has the following explicit expression u_m(z;b , c ) = _ k=0^[m/2 ] q^k(k-1 ) [ a:25 ] where @xmath194 is the pochhammer symbol . by setting z=1 ; b= ; c= ; m = x [ a:26 ] we get the following expression for the generating function @xmath195 f(x;q , a , s)=u_x(1 ; , ) [ a:27 ] the function @xmath195 can be represented as a continued fraction of rogers ramanujan @xcite : @xmath196 where @xmath197 is the extension of the @xmath2-airy function @xmath32 defined in ( [ eq:07 ] ) . the function @xmath51 reads a_q(s;s(1-a ) ) = _ k=0^ [ a:29 ] at @xmath49 eq.([a:29 ] ) coincides with @xmath32 in eq.([eq:07 ] ) . in this appendix we shall mention that the asymptotics ( [ eq : asymp1 ] ) describes the scaling of top line in a bunch of directed vicious walks . proceeding as in @xcite , take the ensemble of @xmath198 vicious walkers , define the averaged position of the top line and consider its fluctuations near the averaged position . in such a description all vicious walkers lying below the top line play a role of a `` mean field '' , which pushes the top line to some `` atypical '' equilibrium position , around which it fluctuates . it is naturally to suppose that the fluctuations of the top line in a mean - field approximation have the same scaling as the fluctuations of the `` inflated '' brownian excursion with fixed area under the path . one actually can show that , following the line of reasoning of the work @xcite . the solution of the inviscid burgers equation @xmath199 is @xmath200 and gives the wigner semicircle law centered at the point @xmath201 . one can smear the function @xmath202 near the boundary value , @xmath203 , adding by hands the gaussian fluctuations , i.e. passing to the burgers equation with a weak diffusivity ( @xmath204 ) , @xmath205 seeking for weakly fluctuating solutions of viscous burgers equation near the top line , ( @xmath206 ) , in the form @xcite @xmath207 and substituting the ansatz for @xmath208 into the viscous burgers equation , one gets the equation for @xmath209 , which for @xmath210 , @xmath211 and appropriate boundary conditions is transformed in the limit @xmath212 into the dimensionless ricatti equation @xcite -yt^-3/2 + w^2 + _ y w=0 , [ ricatti ] having the solution ( for @xmath206 ) w(z ) = 2(2 ^ -1/3z ) z = yn^-1/2 . [ ricatti2 ] in ( [ ricatti2 ] ) one can recognize the singular part of the grand partition function of the `` area+length''weighted brownian excursion . to make this connection precise , define the partition function @xmath213 of @xmath16-step directed 2d random walk in the upper half - plane of the square lattice ( i.e. the brownian excursion ) with the fixed area , @xmath14 . thus , one can straightforwardly identify @xmath214 with @xmath215 under the following redefinitions : \ { l. carlitz and j. riordan , two element lattice permutation numbers and their @xmath2-generalization , duke j. math . * 31 * , 371 - 388 ( 1964 ) ; j. frlinger and j. hofbauer , @xmath2-catalan numbers , j. comb . a * 40 * , 248 - 264 ( 1985 ) s. mashkevich and s. ouvry , area distribution of two - dimensional random walks on a square lattice j. stat . phys . * 137 * , 71 - 78 ( 2009 ) ; + s. matveenko and s. ouvry , the area distribution of two - dimensional random walks and non - hermitian hofstadter quantum mechanics , j. phys . a : math . theor . * 47 * 185001 ( 2014 ) e. witten , fivebranes and knots , arxiv:1101.3216 [ hep - th ] + d. gaiotto and e. witten , knot invariants from four - dimensional gauge theory , adv . * 16 * , 935 ( 2012 ) arxiv:1106.4789 [ hep - th ] n. nekrasov and a. okounkov , seiberg - witten theory and random partitions , hep - th/0306238 + l. baulieu , a. losev , and n. nekrasov , chern - simons and twisted supersymmetry in various dimensions , nucl . b * 522 * , 82 ( 1998 ) [ hep - th/9707174 ] p.l . ferrari , m. praehofer , and h. spohn , stochastic growth in one dimension and gaussian multi - matrix models , in proceedings of the _ 14th international congress on mathematical physics _ ( icmp 2003 ) , world scientific ( ed . j .- zambrini ) ( 2006 ) , 404 - 411 , arxiv : math - ph/0310053 j .- p . blaizot , m.a . nowak , large @xmath220 confinement , universal shocks and random matrices , lectures given at the 49 cracow school of theoretical physics , may 31 - june 10 ( 2009 ) zakopane , poland , arxiv:0911.3683